Mortality and Cause of Death in the Pacific309197/s41545625_phd_fin… · mortality was high across...

Mortality and Cause of Death in the Pacific Karen Lesley Carter

BAppSci (Env. Health), MPHTM

41545625

A thesis submitted for the degree of Doctor of Philosophy at

The University of Queensland in 2013

School of Population Health

� � �

i

Abstract

Accurate data on mortality levels and causes of death is critical to assist

governments to improve the health of their populations; by identifying priority areas

for intervention, leveraging resources for such interventions, and to monitor the

impact and effectiveness of health programs. To date, there has been little data

available to agencies on mortality and causes of death in the Pacific Islands, and

there is significant disparity in published estimates, with many considered

implausible. Further, there are indications that improvements in life expectancy (LE)

across the region have stagnated in contrast to previously published estimates, most

likely as a consequence of premature adult mortality from non-communicable

diseases (NCDs).

In this research, routine vital registration collections from Pacific Island countries are

examined to determine whether these data can improve our understanding of

mortality and causes of death in the region. The work presents a review of available

data, examines the potential flaws and biases in the data collections, and examines

the methods best suited to working with the identified collections to generate valid,

reliable estimates of mortality level and cause of death, disaggregated by age group

and sex. Due to the scale and diversity of the Pacific Islands, six countries

representing a broad range of population size, level of development, and all three

sub-regions were selected for review. These were: Nauru, Fiji, Kiribati, Palau, Tonga

and Vanuatu.

Reporting systems were evaluated to identify potential impacts on the quality and

completeness of available data using an assessment framework that considers the

� � �

ii

societal, administrative and system (administrative, technical and ownership)

influences on the reporting system. Substantially more data was found to be

available than has been previously analysed or published. Medical certificates of

death are completed for some deaths in all of the study countries, and are required

for all deaths in Nauru, Fiji, Tonga and Palau.

Analytical strategies for deriving estimates of mortality level by age and sex were

selected from available approaches based on the system assessments, data

available and level of migration. Methods employed included direct calculation of

mortality level and evaluation of trends over time (Nauru and Palau), evaluation of

published data to establish plausible trends (Fiji and Tonga), direct demographic

methods to assess completeness - the Brass Growth Balance analysis (Fiji), and

capture-recapture analyses (Tonga and Kiribati). Only Vanuatu had insufficient data

from routine collection systems to reliably generate mortality estimates. Findings

from the use of model life tables in comparison to mortality estimates from empirical

data in the other study countries were therefore reviewed to evaluate the plausibility

of published estimates of mortality level from Vanuatu.

Estimates of cause–specific mortality were calculated directly from the medical

certificate data for Palau where the system review indicated a high level of content

validity. A medical record review was conducted in both Tonga and Nauru, and a

death certificate review conducted in Fiji where access to the medical records was

not possible. Other routine collections for cause of death were explored in Vanuatu

and Kiribati where medical certificates were not available for all deaths.

� � �

iii

Findings from this study demonstrate that LE across the selected countries is lower

than previously reported, but that this indicates a stagnation in LE rather than a

reversal in LE improvements as earlier estimates did not adequately account for

premature adult mortality. Estimates of LE for males ranged from 49 and 58 years in

Kiribati (2005-2009) for males and females respectively, to 65 and 70 years for

males and females respectively in Tonga (2005-2008) and Vanuatu (2005). Adult

mortality was high across all countries, with NCDs the leading cause of death in

adults aged 15-59 years in all countries. In Tonga, it has been possible to

demonstrate significant increases in age-standardised mortality, from the early

2000’s to the present, in ischaemic heart disease, diabetes and lung cancer.

Estimates of IMR and U5 mortality remain fairly high in Kiribati at 52 and 72 deaths

per 1,000 live births respectively, and in Nauru at 28 and 46 deaths per 1,000 live

births respectively. Causes of death in children 0-4 years in both countries reflect a

high proportion of deaths from preventable causes: including perinatal conditions,

malnutrition and infectious diseases. For all other study countries, IMR was

estimated to be below 20 deaths per 1,000 live births. These figures indicate that

while there are still gains to be made in improving child health in the region, infant

and child mortality are not the major influence contributing to the low life

expectancies seen across countries.

Despite a need for improvements in routine reporting systems to address poor

coverage, completeness and reporting practices, locally generated empirical data

from routine reporting systems was found to be able to generate plausible estimates

of mortality and add substantially to our understanding of cause of death in the

region.

� � �

iv

Declaration by author

This thesis is composed of my original work, and contains no material previously

published or written by another person except where due reference has been made

in the text. I have clearly stated the contribution by others to jointly-authored works

that I have included in my thesis.

I have clearly stated the contribution of others to my thesis as a whole, including

statistical assistance, survey design, data analysis, significant technical procedures,

professional editorial advice, and any other original research work used or reported

in my thesis. The content of my thesis is the result of work I have carried out since

the commencement of my research higher degree candidature and does not include

a substantial part of work that has been submitted to qualify for the award of any

other degree or diploma in any university or other tertiary institution. I have clearly

stated which parts of my thesis, if any, have been submitted to qualify for another

award.

I acknowledge that an electronic copy of my thesis must be lodged with the

University Library and, subject to the General Award Rules of The University of

Queensland, immediately made available for research and study in accordance with

the Copyright Act 1968.

I acknowledge that copyright of all material contained in my thesis resides with the

copyright holder(s) of that material. Where appropriate I have obtained copyright

permission from the copyright holder to reproduce material in this thesis.

� � �

v

Publications during candidature

Published

Karen L Carter, Chalapati Rao, Alan D Lopez and Richard Taylor. Mortality and cause-of-death reporting and analysis systems in seven pacific island countries. BMC Public Health 2012, 12:436 doi:10.1186/1471-2458-12-436

Carter K, Cornelius M, Taylor R, Ali SS, Rao C, Lopez AD, Lewai V, Goundar R,

Mowry C. Mortality trends in Fiji. Aust N Z J Public Health. 2011 Oct;35(5):412-20.

doi: 10.1111/j.1753-6405.2011.00740.x.

Carter K, Soakai TS, Taylor R, Gadabu I, Rao C, Thoma K, Lopez AD. Mortality trends and the epidemiological transition in Nauru. Asia Pac J Public Health.

2011 Jan;23(1):10-23.

Karen L Carter, Gail Williams, Veronica Tallo, Diozele Sanvictores, Hazel Madera,

Ian Riley Population Capture-recapture analysis of all-cause mortality data in Bohol, Philippines. Population Health Metrics 2011, 9:9 (17 April 2011).

doi:10.1186/1478-7954-9-9

Hufanga S, Carter KL, Rao C, Lopez AD, Taylor R. Mortality trends in Tonga: an assessment based on a synthesis of local data. Population Health Metrics, 2012

Aug 14;10(1):14.

Carter K, Hufanga S, Rao C, Akauola S, Lopez AD, Rampatige R, Taylor R. Causesof death in Tonga: quality of certification and implications for statistics. Population Health Metrics, 2012 Mar 5;10(1):4.

Kuehn R, Fong J, Taylor R, Gyaneshwar R, Carter K. Cervical cancer incidence and mortality in Fiji 2003-2009. Aust N Z J Obstet Gynaecol. 2012 Aug;52(4):380-

6. doi: 10.1111/j.1479-828X.2012.01461.x.

� � �

vi

Carter, K, Rao, C, Lopez, A and Taylor, R (2011). Routine Mortality and Cause of Death Reporting and Analysis Systems in Seven Pacific Island Countries. In:

IEA World Congress of Epidemiology, Edinburgh, United Kingdom, (A421-A421). 7-

11 August 2011. Journal of Epidemiology and Community Health.

doi:10.1136/jech.2011.142976o.48


Mowry C. An assessment of mortality estimates for Fiji, 1949-2008: findings and life tables. Health Information Systems Knowledge Hub 2010; 12

Submitted papers

Taylor R, Carter KL, Linhart C, Azim S, Rao C, Lopez AD, Wainiqolo I, Jatra N,

Mowry C. Mortality trends by ethnicity in Fiji. ANZ Journal of Public Health. In

review 2012

Conference presentations and posters

Scientific Poster- “Capture-recapture analysis of all-cause mortality data in Bohol,

Philippines” – Global Epidemiological Congress (Edinburgh, 2011)

“Mortality Trends and the Epidemiological Transition in Fiji and Nauru” – Australian

Association for the Advancement of Pacific Studies. (Melbourne 2010)

“Mortality Surveillance Systems in the Pacific Islands” – Western Pacific Regional

Epidemiological Association Meeting (Tokyo 2010)

“Mortality Surveillance Systems in the Pacific Islands” UQ SPH RHD Conference

(2009)

“The Epidemiological Transition in Fiji” – Western Pacific Regional Epidemiological

Association Meeting (Tokyo 2010)

� � �

vii

“The Epidemiological Transition in Fiji” –Queensland Branch Australian Public

Health Association Meeting (Brisbane, 2009),

“Mortality and Cause of Death in the Pacific Islands” – UQ SPH School Presentation

(2009)

Scientific Poster- “Capture-Recapture Analysis of Deaths in Bohol, Philippines” –

Presented at the Queensland Branch Australian Public Health Association Meeting

(Brisbane, 2009),

“Mortality and Cause of Death Surveillance in the Pacific Islands” – Pasifika Medical

Association Conference, Cook Islands, July 2009

Mortality data system reviews- debrief presentations to Government participants

and key in-country stakeholders (i.e. WHO, UNICEF, Save the Children etc) – Fiji,

Tonga, Solomon Islands, Vanuatu, Kiribati, Nauru, Palau. 2007-2009

“The Stagnation of Mortality Decline in Fiji” – Research Higher Degree Conference

2008 (Highly Commended)

(also accepted at the 2008 Australian Public Health Association Conference – but

presented by Prof Richard Taylor due to prior travel commitments)

“Capture-recapture methods and mortality estimation in Bohol, Philippines” –

presentation to Gates GC13 project group in Bohol September 2008.

� � �

viii

Publications included in this thesis

Karen L Carter, Chalapati Rao, Alan D Lopez and Richard Taylor. Mortality and cause-of-death reporting and analysis systems in seven pacific island countries. BMC Public Health 2012, 12:436 doi:10.1186/1471-2458-12-436 -

incorporated into Chapter 4

Contributor Statement of contribution

Karen Carter

(Candidate)

Developed and trialled the system assessment framework and data

collection tools, conducted the field visits and interviews, collated

the findings and extracted key themes, and drafted the analysis and

paper. (65%)

Other Authors CR: assisted with the identification and clarification of key themes,

reviewed the data analysis and interpretation of results; revised the

manuscript critically for structure, clarity and important intellectual

content (25%). AL: revised the manuscript critically for important

intellectual content (5%). RT: reviewed the data analysis and

interpretation of results; and revised the manuscript critically for

important intellectual content (5%). RT and AL both conceived and

coordinated the broader study under which this project was

undertaken. All authors read and approved the final manuscript.

� � �

ix



doi: 10.1111/j.1753-6405.2011.00740.x. – incorporated into Chapter 5


Karen Carter

(Candidate)

KC: Tabulation of MoH data, analysis of MoH data (including

assessment of completeness, Brass Growth Balance assessment,

interpolation of population figures, construction of life tables and

confidence intervals for all periods), Primary collation and analysis

of published cause of death data including conversion to standard

format, assessment of plausibility, tabulation and graphing,

development of criteria for plausibility of mortality level (with RT)

and application, critical interpretation of results and substantial

contribution to writing and shaping of discussion, formatting. (60%)

(60%)

Other Authors MC reviewed the draft paper and results critically for local context

and interpretation (5%): SA, VL and RG: reviewed the draft paper

and results critically for local context and interpretation, and

facilitated the collation and access to local data (2% each); CR

reviewed the draft paper and results critically for methodology (4%):

CM identified and tabulated published data and drafted the

preliminary analysis of trends before censoring for plausibility. RT

(20%) was responsible for the paper concept and original design

(with KC), ADL (5%) and RT conceived and coordinated the

broader study under which this project was undertaken, reviewed

and contributed to the interpretation of results, and revised the

manuscript critically for important intellectual content. RT also

generated the regression models to estimate IMR from childhood

mortality.

� � �

x

Carter K, Soakai TS, Taylor R, Gadabu I, Rao C, Thoma K, Lopez AD. Mortality trends and the epidemiological transition in Nauru. Asia Pac J Public Health.

2011 Jan;23(1):10-23. – incorporated into Chapter 5


Karen Carter

(Candidate)

KLC: Primary collection and collation of 2002-2007 data, analysis

of data (including interpolation of population figures, construction of

life tables and confidence intervals for all periods, child mortality

measures, age standardised rates and confidence intervals, and

confidence intervals for external causes, construction of tables and

graphs,) interpretation of results, contributed to writing and shaping

of discussion, formatting. (55%)

Other Authors RT: Paper concept and original design, collection and collation of

data from 1986-1992, input on analysis methods, interpretation of

results, shaping and writing of all sections. (30%) KT: Collection

and collation of data from 1986-1992, interpretation of results from

early periods (5%). CR: Review of results interpretation, discussion

and epidemiological transition context (2%). SS: Review of results

interpretation, discussion and country context (2%). AL: Review of

results interpretation, discussion and epidemiological transition

context (4%). IG: Collection of population data, review of results

interpretation, discussion and country context (2%).

� � �

xi


Ian Riley Population Capture-recapture analysis of all-cause mortality data in Bohol, Philippines. Population Health Metrics 2011, 9:9 (17 April 2011).

doi:10.1186/1478-7954-9-9 – incorporated into Chapter 5


Karen Carter

(Candidate)

KC designed the study; conceived and conducted the system

review; designed and tested matching criteria; prepared the data

for the three-source analysis; performed the statistical analysis for

the two-source analysis and final model selection including

interpretation of final results; and drafted the manuscript.

(65%)

Other Authors GW made substantial contributions to analysis and interpretation of

data, performed the statistical analysis for the three-source analysis

and revised the manuscript critically (20%). VT coordinated data

collection, collation and matching, and made substantial

contributions to the interpretation of data (4%). DS coordinated

initial data cleaning, management, and tabulation of records, and

made substantial contributions to the interpretation of data (3%).

HM made substantial contributions to the acquisition and

interpretation of data (3%). IR conceived and coordinated the study

and revised the manuscript critically for important intellectual

content (4%).

� � �

xii

Hufanga S, Carter KL, Rao C, Lopez AD, Taylor R. Mortality trends in Tonga: an assessment based on a synthesis of local data. Population Health Metrics, 2012

Aug 14;10(1):14. – incorporated into Chapter 5


Karen Carter

(Candidate)

KC oversaw the project, conducted the system review, oversaw the

design and testing of matching criteria, checked the matching

process, assisted with data matching for the civil registration data

against other sources, performed the statistical analysis for both

the two-source analysis for 2001 to 2004 and the three-source

analysis, including the civil registry data and final model selection

including interpretation of final results and selection of final

estimates, supervised the other analysis, and drafted and led the

revision of the manuscript. (45%)

Other Authors SH conducted the system review, supervised the data collection

and collation processes, designed the matching criteria, conducted

the data matching, managed the ethics process and coordination

with data collection agencies, performed the statistical analysis for

the two source analysis of the HIS and PM data for 2005 to 2009,

completed the search for published estimates, assessed published

data sources for reliability, revised the manuscript critically, and

made substantial contributions to the interpretation of data (35%).

CR (5%) revised the manuscript critically for important intellectual

content. ADL (5%) and RT (10%) conceived and coordinated the


and contributed significantly to the interpretation of results, and

revised the manuscript critically for important intellectual content.

RT also generated the regression models to estimate IMR from

childhood mortality.

� � �

xiii

Carter K, Hufanga S, Rao C, Akauola S, Lopez AD, Rampatige R, Taylor R. Causesof death in Tonga: quality of certification and implications for statistics. Population Health Metrics, 2012 Mar 5;10(1):4. – incorporated into Chapter 6


Karen Carter

(Candidate)

KC designed the study, developed the study forms and procedures,

conducted initial training of data extractors, reviewed trial extraction

forms, arranged coding and collation of data, analysed the

collected data, and drafted the manuscript (55%)

Other Authors SH coordinated data extraction in Tonga and provided significant

input to the manuscript (15%); CR assisted with reviewing trial

extraction forms and revised the manuscript critically for important

intellectual content (5%); SA reviewed the extracted data from both

countries to allocate cause of death from the medical record and

provided significant input into the discussion (5%); RR provided the

second review of deaths for which cause was originally assigned as

unknown and critically revised the manuscript (2.5%). ADL (2.5%)

and RT (15%) conceived and coordinated the broader study under

which this project was undertaken, reviewed and contributed

significantly to the interpretation of results, and revised the

manuscript critically for important intellectual content. RT also

made substantial contributions to the study design.

� � �

xiv


Mowry C. An assessment of mortality estimates for Fiji, 1949-2008: findings and life tables. Health Information Systems Knowledge Hub 2010; 12– incorporated

into the Appendices


Karen Carter

(Candidate)

KC: Tabulation of MoH data, analysis of MoH data (including

assessment of completeness, Brass Growth Balance assessment,

interpolation of population figures, construction of life tables and

confidence intervals for all periods), Primary collation and analysis

of published cause of death data including conversion to standard

format, assessment of plausibility, tabulation and graphing,

development of criteria for plausibility of mortality level (with RT)

and application, critical interpretation of results and substantial

contribution to writing and shaping of discussion, formatting. (70%)

Other Authors MC reviewed the draft paper and results critically for local context

and interpretation (2%): SA, VL and RG: reviewed the draft paper

and results critically for local context and interpretation, and

facilitated the collation and access to local data (2% each); CR

reviewed the draft paper and results critically for methodology (4%):

CM identified and tabulated published data and drafted the

preliminary analysis of trends before censoring for plausibility. RT

(15%) was responsible for the paper concept and original design

(with KC), ADL (3%) and RT conceived and coordinated the


and contributed to the interpretation of results, and revised the

manuscript critically for important intellectual content.

� � �

xv

Contributions by others to the thesis

Significant contributions made by others to the research presented in this thesis are

acknowledged above in the contribution to published papers. Prof Alan Lopez, Prof

Richard Taylor and Dr Chalapati Rao contributed significantly to this work as a whole

by reviewing and commenting on the study design, thesis structure and content,

including providing critical comments on the draft analysis and chapters. This

research was supported under a broader AusAID Australian Development Research

Award project to review vital registration systems in the Pacific Islands.

In addition, the following specific contributions by others were made to work

presented in this thesis other than the published case studies (which are note in the

previous section).

Chapter 5 Kiribati Capture-recapture study: Criteria for data matching between sources were

developed and tested with Tibwataake Baiteke from the Ministry of Health, and

Kabuaua Temaava and Tiensi Teea from the Civil Registration Offices by trialling

sub sets of data for review and discussion. These staff also reviewed the complete

matched data set to identify errors and potential missed errors. Both departments

provided input into the study design through extensive discussions with the

candidate on reporting processes.

Chapter 6 Nauru: Data extraction for the medical record review was completed by a team of

nurses and medical record officers following the protocol provided. Data extraction

was overseen by the director of nursing, periodically reviewed by myself. Location

and extraction of medical records for review was overseen by the manager of

medical records following the list provided. The medical records manager also

maintained the master list containing the identifiable data once study numbers were

assigned to each case in order to de-identify the data. Dr Sione Akuola reviewed the

extracted data from the medical record following the study protocol to complete a

causal sequence of death for the study. Cases for which an underlying cause of

death could not be identified were reviewed by Dr Rasika Rampatige.

� � �

xvi

Fiji: Coding of each line of the certificate and selection of underlying cause of death

according to ICD rules undertaken by me as part of this thesis were checked by the

WHO collaborating centre on coding in Sri Lanka.

Kiribati: Moaiti Nubono from the Kiribati Ministry of Health worked with me to

translate causes of death from the civil registry data and review the collection

procedures to provide local cultural context.

Statement of parts of the thesis submitted to qualify for the award of another degree

None

� � �

xvii

Acknowledgements

Firstly, I would like to thank my family for all their invaluable support during this

process; especially my sister Wendy who kept me supplied with chocolate, provided

distraction when needed, proof read pages and pages of early drafts and provided

unending support and encouragement; and my parents for their ongoing support and

patience.

Many thanks are due to my supervisors Prof Alan Lopez, Dr Chalapati Rao, and Prof

Richard Taylor for their guidance, support and for always challenging me to do

better. In particular Alan for setting me lofty standards to reach for, Chalapati for

keeping my feet on the ground, and Richard for having the patience and

perseverance to try to turn me into a researcher! I greatly appreciate your

persistence. Special mention is also due to Prof Gail Williams and Prof Ian Riley for

their encouragement and support. Many thanks are also due to Dr Deirdre

McLaughlin and Mary Roset the Postgraduate coordinator and administrator who

have shown endless patience and support, and whose dedication and endless good

humour when faced with students who are sleep deprived, stressed and dosed up on

coffee make the whole process work.

I would also like to thank the many colleagues and friends who helped me get here

in the first place especially Heather O’Donnell, Maireng Sengebau and Martyn Kirk;

Hazel Clothier for the nudge that first sent me out into the Pacific and introduced me

to this spectacular part of the world; the friends and colleagues who kept me sane

(or mostly sane) during the process – especially Misha, Heather, Linda, Kath, Jo,

Claire, Audrey, Sally, James, Damian and Danielle; my current boss Dr Haberkorn

for his ongoing support and encouragement; and of course the amazing health and

statistics professionals of the Pacific Islands without whom this research would not

have been possible and who have made the last few years much richer for this

experience.

I could not have done it without you all.

� � �

xviii

Keywords

mortality, cause of death, epidemiology, public health, pacific islands, capture-

recapture, certification of death, civil registration, medical record review.

Australian and New Zealand Standard Research Classifications (ANZSRC)

ANZSRC code: 111706, Public Health: Epidemiology, 65%

ANZSRC code: 160304, Demography: Mortality, 30%

ANZSRC code: 111711, Public Health: Health Information Systems (including

Surveillance), 5%

Fields of Research (FoR) Classification

FoR code: 1117, Public Health, 70%

FoR code: 1603, Demography, 30%

� � �

xix

List of Key Acronyms

5q0 – Under 5 mortality, the probability of dying before age 5

45q15 – Adult mortality, the probability of dying between ages 15-59 inclusive

CEBCS – Children ever born/ Children surviving (an indirect method for obtaining

childhood survival)

CHIPs – Country Health Information Profiles (published annually by WHO)

CI- Confidence interval

CoD – Cause of death

COPD – Chronic obstructive pulmonary disease

CRVS – Civil Registration and Vital Statistics

DHS – Demographic and Health Survey

DSSs – Demographic surveillance sites

ICD – International Classification of Diseases

IMR – Infant Mortality Rate

ITC – Information technology and communication

GBD – Global Burden of Disease Study

HIS- Health Information Systems

HMN – Health Metrics Network

LE – Life expectancy (at birth)

MDG – Millennium Development Goals

MICS – Multiple Indicator Cluster Survey

MoH – Ministry of Health

MMR – Maternal mortality ratio (maternal deaths per 1,000 live births)

STEPS – WHO Stepwise NCD risk factor survey

PATIS – Patient Information System (Fiji’s HIS system)

PICTs – Pacific Island Countries and Territories

PNG – Papua New Guinea

NCDs – Non communicable diseases

WHO – World Health Organisation

UN – United Nations

UNICEF – United Nations Children’s Fund

U5M – Under 5 mortality (deaths in children aged 0-4 years divided by live births)

UQ HIS Hub – University of Queensland Health Information Systems Hub

VA – Verbal autopsy

� � �

xx

Contents

1� Introduction ........................................................................................................ 1�1.1� The importance of mortality and cause of death data ................................... 1�

1.2� The Pacific Islands ........................................................................................ 6�

1.3� Countries ....................................................................................................... 9�

1.3.1� Fiji ......................................................................................................... 10�

1.3.2� Kiribati ................................................................................................... 11�

1.3.3� Nauru .................................................................................................... 12�

1.3.4� Palau .................................................................................................... 13�

1.3.5� Tonga ................................................................................................... 14�

1.3.6� Vanuatu ................................................................................................ 15�

1.4� Measures of mortality and causes of death ................................................. 16�

1.5� An overview of mortality and cause of death data for the Pacific ................ 21�

1.5.1� Historical context – patterns of mortality and causes of death .............. 21�

1.5.2� Previous Data Quality ........................................................................... 24�

1.5.3� Recent data .......................................................................................... 26�

1.6� Research Objectives and Questions ........................................................... 34�

2� An overview of methods for collecting, analysing and assessing the quality of mortality level and cause of death data ........................................................... 38�

2.1� Sources of mortality and cause of death data ............................................. 39�

2.1.1� Census collections ................................................................................ 40�

2.1.2� Periodic Surveys ................................................................................... 45�

2.1.3� Routine vital registration collections ..................................................... 47�

2.1.4� Demographic surveillance sites ............................................................ 52�

2.1.5� Verbal Autopsy ..................................................................................... 52�

2.1.6� Summary of data collection approaches ............................................... 54�

2.2� Evaluating routine data collection and reporting systems ........................... 57�

2.3� Evaluating completeness for mortality level ................................................ 65�

2.3.1� Direct Demographic Methods ............................................................... 66�

2.3.2� Model Life Tables ................................................................................. 69�

2.3.3� Conclusions .......................................................................................... 73�

2.4� Evaluating biases in cause of death data .................................................... 74�

2.4.1� Review of Certification .......................................................................... 74�

� � �

xxi

2.4.2� Assessment and correction of coding practices .................................... 78�

2.4.3� Multiple cause of death analysis ........................................................... 80�

2.4.4� Models for cause of death data ............................................................ 81�

2.5� Final Comments .......................................................................................... 83�

3� Routine reporting systems for deaths and cause of death in selected Pacific Islands ........................................................................................................ 87�

3.1� Assessment Framework for routine reporting systems ............................... 87�

3.2� Common elements of CRVS systems in the region and their effects on data

92�

3.2.1� Case Study 1: Reporting systems across seven PICTs (163) .............. 92�

3.2.2� Data availability and accessibility ....................................................... 103�

3.3� Summary of country specific CRVS systems ............................................ 104�

3.3.1� Fiji ....................................................................................................... 104�

3.3.2� Kiribati ................................................................................................. 105�

3.3.3� Nauru .................................................................................................. 106�

3.3.4� Palau .................................................................................................. 106�

3.3.5� Tonga ................................................................................................. 107�

3.3.6� Vanuatu .............................................................................................. 108�

3.4� Conclusion ................................................................................................ 108�

4� Further analytical considerations – small area estimates, off-island vital events and migration ........................................................................................... 110�

4.1� Introduction ............................................................................................... 110�

4.2� Population size and small-area estimates ................................................. 111�

4.3� Off-Island Events ....................................................................................... 114�

4.3.1� Off-island births .................................................................................. 122�

4.4� Migration ................................................................................................... 123�

4.5� Conclusions ............................................................................................... 128�

5� Assessing mortality level .............................................................................. 132�5.1� System evaluation findings and analytical strategies to assess mortality level

132�

5.2� Data aggregation and trend analysis ......................................................... 137�

� � �

xxii

5.2.1� Case Study 2: Mortality Trends in Nauru. ........................................... 138�

5.2.2� Palau analysis and assessment ......................................................... 155�

5.3� Validity assessment of published data ...................................................... 156�

5.4� Brass growth balance analysis .................................................................. 157�

5.4.1� Case study 3: Mortality Trends in Fiji. ................................................. 161�

5.5� Capture-recapture analysis ....................................................................... 174�

5.5.1� Case study 4: Capture-recapture analysis in Bohol, Philippines – an

application in an “ideal” setting ........................................................................ 176�

5.5.2� Case study 5: Capture-recapture analysis in Tonga – application in a

Pacific Island with multiple data sources ......................................................... 187�

5.5.3� Using Capture-recapture analysis to set plausible mortality limits in

Kiribati 202�

5.6� Mortality Estimation in the absence of routine death reporting .................. 221�

5.7� Summary of method selection and application .......................................... 225�

6� Assessing cause of death distributions ...................................................... 229�6.1� Direct tabulation of cause of death from collected data ............................. 234�

6.1.1� Analysis of cause of death in Palau .................................................... 234�

6.2� Medical Record Review ............................................................................ 246�

6.2.1� Case study 6: Medical Record Review in Tonga ................................ 249�

6.2.2� Medical Record Review in Nauru ....................................................... 266�

6.3� Medical Certificate of Death Review ......................................................... 282�

6.3.1� Review of medical certification of death in Fiji .................................... 282�

6.4� Examining cause of death distributions from alternative data sources:

understanding causal patterns in the absence of nationally representative data

from medical certification .................................................................................... 293�

6.4.1� Comparison of cause of death distributions from hospital and community

data in Vanuatu ............................................................................................... 293�

6.4.2� Cause of death analysis from hospital, community nursing and civil

registry data in Kiribati ..................................................................................... 315�

6.5� Summary of method selection and application .......................................... 328�

7� Mortality patterns in the Pacific Islands ...................................................... 330�7.1� Summary measures of mortality ................................................................ 332�

� � �

xxiii

7.2� Mortality and cause of death distributions by age group ........................... 336�

7.2.1� Infant and child mortality ..................................................................... 336�

7.2.2� Adult mortality 15-59 years ................................................................. 339�

7.2.3� Mortality & Older adults ...................................................................... 352�

7.3� Country summaries ................................................................................... 354�

7.3.1� Fiji ....................................................................................................... 354�

7.3.2� Kiribati ................................................................................................. 356�

7.3.3� Nauru .................................................................................................. 359�

7.3.4� Palau .................................................................................................. 360�

7.3.5� Tonga ................................................................................................. 361�

7.3.6� Vanuatu .............................................................................................. 362�

7.4� Mortality in Transition ................................................................................ 364�

8� Conclusion and recommendations .............................................................. 367�8.1� Synopsis.................................................................................................... 367�

8.2� Recommendations .................................................................................... 373�

8.2.1� Recommendations for system improvement of CRVS systems .......... 374�

8.2.2� Recommendations for further research .............................................. 376�

9� REFERENCES ................................................................................................ 378�

Appendix 1: Brass Growth Balance Assessments for Kiribati ..................................... i�

Appendix 2: Additional country life tables ................................................................... iii�

Appendix 3: Additional analysis on mortality level in Fiji including life tables ........... xvi�

Appendix 4: Additional cause of death distributions by age and sex ........................ lvii�

Appendix 5: Medical record review handbook ...................................................... lxxxv�

� � �

xxiv

List of Tables

Table 1: Uses of cause of death data (11) ................................................................. 3�

Table 2: Land area, population and density by sub region: Pacific Islands (44) ........ 6�

Table 3: Pacific Island Countries and Territories, and their key characteristics .......... 8�

Table 4: Key measures of mortality and causes of death ......................................... 18�

Table 5: Published estimates at 2010 of Infant mortality rates (per 1,000 live births)

(Both sexes combined); Pacific Island Countries and Territories ............................. 27�

Table 6: Published estimates at 2010 of LE at birth (males and females); Pacific

Island Countries and Territories ............................................................................... 29�

Table 7: Extract from Fiji MDG Report: Statistics on Goal 4 - Fiji Millennium

Development Goals 2nd Report, 1990-2009 (113) ................................................... 31�

Table 8: Censuses in the Pacific Islands, 1950-2010 ............................................... 41�

Table 9: Type of data collection and methods of estimation used for calculation of life

tables in the most recent census round for selected Pacific Island Countries and

Territories ................................................................................................................. 44�

Table 10: Key surveys in the Pacific Islands; 1995-2012 ......................................... 46�

Table 11: Summary of Data Collection Approaches for Mortality Data .................... 55�

Table 12: Assumptions and data requirements for direct demographic analysis ...... 68�

Table 13: Routine death reporting system assessment framework outline: Key

aspects for review .................................................................................................... 90�

Table 14: Available routine mortality data collections by country and source ......... 103�

Table 15: Minimum number of years data would need to be aggregated to reach the

minimum person years of data required for acceptable mortality analysis as cited by

Begg et. al. by country (293) .................................................................................. 112�

Table 16: Registered deaths in Australia of new residents from the Pacific Islands by

age at death and duration of residence: 2001-2008 ............................................... 117�

� � �

xxv

Table 17: Deaths in Australia and New Zealand of persons usually resident overseas

by PICT of birth: 2001-2008 ................................................................................... 119�

Table 18: Migration rates by Country for 2010 (110) .............................................. 125�

Table 19: Reported deaths by source, Kiribati: 2000-2009 .................................... 207�

Table 20: Summary measures of mortality for Kiribati by data source and period,

2000-2009 .............................................................................................................. 210�

Table 21: Estimated average deaths per year by data source and method of

estimation: Kiribati, 2000-2009 by age group and sex ........................................... 216�

Table 22: Summary of data sources and methods used in assessment of mortality

level for the Pacific Islands ..................................................................................... 227�

Table 23: Deaths by cause, and proportional mortality for children 0-4 years, by

period for Palau; 1999-2009 (excluding 2007) ....................................................... 238�

Table 24: Cause-specific mortality by ICD chapter and period for adult males 15-64

years: Palau, 1999-2009 (excluding 2007) ............................................................. 244�

Table 25: Cause specific mortality by ICD chapter and period for adult females 15-64

years: Palau, 1999-2009 (excluding 2007) ............................................................. 245�

Table 26: Comparison of underlying cause of death from the medical certificate of

death (including the non-standard first line in the assessment of causal sequence)

against findings of the medical record review: Nauru, 2005-2009 .......................... 272�

Table 27: Comparison of age distribution for deaths with a medical certificate for

which no medical record was found; and deaths with a certificate which were

reviewed against the medical record. Nauru, 2005-2009 ....................................... 274�

Table 28: Deaths by cause (by ICD chapter), proportional mortality and cause-

specific mortality rates for children 0-4 years: Nauru, 2005-2009 .......................... 276�

Table 29: Deaths by cause (by ICD chapter) and proportional mortality for children 5-

14 years: Nauru, 2005-2009 ................................................................................... 277�

� � �

xxvi

Table 30: Deaths by cause (by ICD chapter) and proportional mortality for adults 15-

54 years: Nauru 2005-2009 .................................................................................... 278�

Table 31: Number of selected medical certificates found by age group: Fiji, 2000-

2008 ....................................................................................................................... 285�

Table 32: Reviewed medical certificates by single versus multiple lines completed on

the certificate: Fiji, 2000-2008 ................................................................................ 285�

Table 33: Column of database where underlying cause of death from the original

certificate (specific disease and by chapter) were found, by age group: Fiji 2000-

2008 ....................................................................................................................... 287�

Table 34: Distribution of deaths by broad cause where underlying cause of death

was not recorded in the Ministry of Health database. Fiji 2000-2008 ..................... 288�

Table 35: Deaths, proportional mortality and death rate by selected cause (total

mentions on the certificate) for children 0-4 years: Fiji 2000-2008 ........................ 289�

Table 36: Deaths and proportional mortality by mentioned cause (total mentions on

the certificate) for selected causes for children 5-14 years: Fiji 2000-2008 ............ 290�

Table 37: Deaths and proportional mortality by underlying cause (by ICD chapter) for

adults 15-59 years: Fiji 2005-2008 ........................................................................ 291�

Table 38: Deaths and proportional mortality by mentioned cause (total mentions on

the certificate) for selected causes for adults 15-64 years: Fiji 2000-2008 ............. 292�

Table 39: Proportion of deaths (%) attributed to Ill-defined or unknown causes by

age group, gender and source: Vanuatu 2001-2007 .............................................. 299�

Table 40: Deaths, proportional mortality and cause-specific mortality rates (by

ICDv10 General Mortality List 3) for children 0-4 years: Vanuatu 2000-2007 ....... 301�

Table 41: Deaths, proportional mortality and cause-specific mortality rates (by

ICDv10 General Mortality List One) for children 5-14 years: Vanuatu 2000-2007 .. 304�

� � �

xxvii

Table 42: Deaths and proportional mortality (excluding ill-defined causes) by cause

(ICDv10 General Mortality List 1) for adults 15-59 years: Vanuatu 2000-2007 ...... 307�

Table 43: Proportion of ill-defined or unknown causes and adjusted proportional

mortality (%) by age group and period: Kiribati males, 2000-2008 ........................ 321�

Table 44: Proportion of ill-defined or unknown causes and adjusted proportional

mortality (%) by age group and period: Kiribati females, 2000-2008 ...................... 323�

Table 45: Proportion (%) of hospital deaths of children aged 1-4 years where

malnutrition was reported as either the primary diagnosis or contributing factor,

Kiribati 2000-2008 .................................................................................................. 325�

Table 46: Minimum suicide rates (per 100,000 people) based on uncorrected data,

Kiribati 2000-2008 .................................................................................................. 326�

Table 47: Estimated Age-specific mortality rate from Diabetes (as reported through

civil registration and health system reporting) per 100,000 people ........................ 327�

Table 48: Summary of data sources and methods for used in assessment of cause

of death patterns for selected Pacific Islands ......................................................... 329�

Table 49: Summary of key measures of mortality for selected Pacific Islands ....... 333�

Table 50: Comparison of ‘Best estimates’ of life expectancy at birth (years) from this

research against Taylor et. al. 2005 review (2) ...................................................... 334�

Table 51: Key measures of childhood mortality (0-4 years), both sexes, for selected

Pacific Islands ........................................................................................................ 338�

Table 52: Adult mortality and proportional mortality (%) for key causes (by ICD

chapter) for selected Pacific Islands; Males ........................................................... 345�

Table 53: Cause-specific mortality rates for key causes (by ICD General Mortality

List 1) for selected Pacific Islands; Males 15-59 years ........................................... 346�

Table 54: Age-standardised incidence rates of specific cancer types 1980s-1990s

(cases per 100,000 population) by country extracted from Palafox et. al (368) ...... 347�

� � �

xxviii

Table 55: Adult mortality and proportional mortality (%) for key causes (by ICD

chapter) for selected Pacific Islands; Females ....................................................... 350�

Table 56: Cause-specific mortality rates for key causes (by ICD General Mortality

List 1) for selected Pacific Islands; Females 15-59 years ....................................... 351�

Table 57: Life expectancy at 65 years for selected countries ................................. 353�

� � �

xxix

List of Figures

Figure 1: Map of selected Pacific Island Countries included in this research ............. 9�

Figure 2: International standard medical certificate of death format (68).................. 19�

Figure 3: Recent published estimates of male LE at birth, Pacific Island Countries

and Territories .......................................................................................................... 30�

Figure 4: Recent published estimates of female LE at birth, Pacific Island Countries

and Territories .......................................................................................................... 30�

Figure 5: Thesis structure and content ..................................................................... 37�

Figure 6: Level of certainty of underlying cause of death by data source ................. 56�

Figure 7: The two source capture-recapture model .................................................. 69�

Figure 8: Sources of errors in medical certification of death from Maudsley and

Williams, 1994 (249) ................................................................................................. 75�

Figure 9: Routine death reporting system assessment framework ........................... 89�

Figure 10: Major medical referral pathways in the Pacific Islands .......................... 116�

Figure 11: Recommended analysis approach for assessment of mortality level in the

Pacific Islands ........................................................................................................ 133�

Figure 12: Age-specific mortality for Nauru 2002-2007, by sex .............................. 138�

Figure 13: Age-specific mortality for Palau 2004-2009 (excluding 2007), by sex ... 155�

Figure 14: Brass Growth Balance analysis of reconciled data from Tonga; Males

2005-2009 .............................................................................................................. 159�

Figure 15: Brass Growth Balance analysis of reconciled data from Tonga; Females

2005-2009 .............................................................................................................. 160�

Figure 16: Brass Growth Balance analysis for mortality data completeness: Fiji,

Males 2002-2004 ................................................................................................... 163�

Figure 17: Brass Growth Balance analysis for mortality data completeness: Fiji,

Females 2002-2004 ............................................................................................... 164�

� � �

xxx

Figure 18: Reporting completeness of deaths in Tonga by age group, sex and data

source (2001-2009) ................................................................................................ 189�

Figure 19: Differences in total number of reported deaths by age group between Civil

registry data and MoH data for Kiribati, 2000-2009 ................................................ 211�

Figure 20: Estimated completeness of reported deaths from 2-source capture-

recapture analysis by source, sex and age group for Kiribati, 2000-2009 .............. 212�

Figure 21: Estimated average deaths per year (2000-2009) for Kiribati from census

data and routinely collected data ............................................................................ 214�

Figure 22: Empirical age-specific mortality compared with predictions from one and

two parameter models for Fiji, Males: 2002-2004 .................................................. 222�

Figure 23: Empirical age-specific mortality compared with predictions from one and

two parameter models for Fiji, Females: 2002-2004 .............................................. 223�

Figure 24: Collection and suggested review process for cause of death data ........ 233�

Figure 25: Proportional mortality by period for children aged 0-4 years both sexes:

Palau, 1999-2009 (excl 2007) ................................................................................ 237�

Figure 26: Proportional mortality by cause and period for Palauan males: Adults 15-

64 years for 1999-2009 (excluding 2007) ............................................................... 240�

Figure 27: Proportional mortality by cause and period for Palauan females: Adults

15-64 years for 1999-2009 (excluding 2007) .......................................................... 241�

Figure 28: Proportion of deaths due to external causes, by type of external cause for

males aged 15-64 years: Palau, 2004-2009 (excluding 2007) ............................... 243�

Figure 29: Age-distribution of deaths in Tonga by global burden of disease* category;

2005-2009 .............................................................................................................. 250�

Figure 30: Medical record review flowchart for Nauruan study ............................... 267�

Figure 31: Comparison of cause of death distributions by broad category for

reviewed* and non-reviewed^ deaths: Nauru, 2005-2009 ...................................... 273�

� � �

xxxi

Figure 32: Age distribution of reported deaths by data source and gender; Vanuatu,

2001-2007 .............................................................................................................. 298�

Figure 33: Proportional mortality in Kiribati 2005-2008 for children 0-4 years,

excluding ill-defined causes, as reported through the Ministry of Health (hospital and

community health facilities) .................................................................................... 318�

Figure 34: Proportional mortality in Kiribati 2005-2008 for children 5-14 years,



Figure 35: Proportional mortality in Kiribati 2005-2008 for males 15-59 years,



Figure 36: Proportional mortality in Kiribati 2005-2008 for females 15-59 years,



� � �

1

1 Introduction

The international community (1), researchers (2-4) and national governments

themselves (5, 6) have frequently lamented the lack of data, and the poor quality and

difficulties of accessing existing data on mortality and cause of death (CoD) data for

Pacific Island countries and territories (PICTs). This work will examine the empirical

data available on mortality and CoD from selected PICTs, the quality of these data,

and how they may be used to improve our knowledge of mortality and causes of

death in the region.

1.1 The importance of mortality and cause of death data

Mortality and CoD data are essential to quantify the health of a population and for

government planning. Statistics on mortality level have uses in generating population

estimates, assessing population dynamics and societal structure, workforce and

social security planning, economic development, and ensuring the effectiveness and

coverage of health systems (7). Registration of deaths through a civil registration

process also has benefits in terms of “legal status, and the protection of social, and

human rights” (8). For example, a legally recognised death certificate allows

remaining family members to access bank accounts, and transfer land or other

property ownership. Death registration flags some deaths as suspicious, requiring

investigation by police and civil authorities. A formal registration of death is also a

means of recognising the life which has passed as important, providing a measure of

societal accountability for how and when people die (7).

� � �

2

Governments are tasked with supporting their populations to achieve an acceptable

health status both through broad environmental (including physical, social, and

economic) impacts, and the provision of quality preventative health services and

care. Complete and accurate statistics on mortality level and causes of death are

core elements of the information required to fulfil these functions (7, 9, 10). As noted

by the Health Metrics Network (HMN) “lost data, poor documentation, lack of access

to available knowledge, and reliance on memory all impede the delivery of high-

quality health care services” (8, 11) (12). Sufficient details to accurately disaggregate

data by key population characteristics (e.g. age group, ethnicity, sex, and location)

are critical to identify priority health gaps and target interventions effectively. While

predicted or modelled data have an important role in addressing data gaps where

empirical data are not available or reliable (13), they are not a substitute for empirical

data that has been assessed for validity and corrected for potential biases (14).

Predicted statistics are unsuitable for “monitoring progress towards agreed targets

and assessments of what works and does not” (15), (13, 16). By their nature,

predicted statistics cannot identify in smaller populations localised events or issues

affecting mortality patterns that are not represented in the assumptions or “usual”

patterns. It is these issues and events which are of greatest interest to local health

managers.

A useful summary of the applications of CoD data was developed by Byass in 2007

(11) and is reproduced below.

� � �

3

Table 1: Uses of cause of death data (11)

Source(11): Byass P. Who needs cause of death data? PLoS medicine. 2007;4(11):1715

The routine collection and publication of CoD data began with the English Bills of

Mortality to monitor plague in the 1600’s and subsequent analysis by John Graunt

(17). Early efforts to standardise the causes of death (the precursors to the

International Classification of Diseases) are attributed to Sauvages, Linnaeus, and

Cullen in the 1700’s (17), with the quality of CoD statistics recognised as an issue of

scientific investigation dating from as early as 1940 (18). International interest in

mortality statistics and CoD certification procedures was renewed in the 1990’s with

the rise of the HIV epidemic in Africa and was solidified in 2000 with two events: a

legal case in the United Kingdom and the adoption of the Millennium Development

Goals (MDGs) by the United Nations (UN). The Shipman case involved a general

practitioner convicted of murdering 15 patients, but who was thought to have killed

up to 200 patients during his 24 year career (19, 20). That so many deaths could go

unnoticed sparked a major review of the medical certification and review process in

the UK and greater scrutiny of these processes more broadly (20-26). Among the

targets adopted under the MDGs were commitments to reduce childhood mortality

� � �

4

by two thirds from 1990 levels, reduce maternal mortality, and reduce the incidence

(and impact) of malaria and HIV (27, 28). Billions of dollars of funding were

subsequently invested in addressing these challenges through structures such as

Global Fund (GFATM) established in 2002 (29) and The United States President’s

Emergency Plan for AIDS Relief (PEPFAR) established in 2004 (30).

The MDGs and subsequent investment in health infrastructure and programs has led

to significant pressure for governments to demonstrate progress towards mortality

related targets, and growing support to improve civil registration and vital statistics

(CRVS) systems (and more generally, national statistical systems and health

information systems (HIS)) to enable this. The Partnership in Statistics for

Development in the 21st Century (PARIS 21) established by the UN in 1999, adopted

the Marrakech Action Plan for Statistics in 2004 which, although focussed on

population and economic statistics, included the routine collection of vital events

(births and deaths) as a core element of a national statistical system (31). In 2005

the HMN was formed under the auspices of the World Health Organisation (WHO),

and in 2007 co-sponsored a series of articles in the Lancet on vital statistics and civil

registration. The “Who Counts” series (7, 32, 33), (34) highlighted the lack of reliable

and timely statistics on births and deaths (including causes of death) from low and

middle income countries (7) and emphasised the importance of these statistics in

ensuring government and international community accountability in supporting health

improvements where they are most needed (7, 35).

Mortality and CoD data is of particular importance in a setting such as the Pacific

Islands where most PICTs would easily be classified as low or middle income

settings (36). Only the Territories, Palau, and Tuvalu are classified as upper-middle

� � �

5

or high income settings by the World Bank (37). Governments must deal with

logistical issues of remote communities, high transport costs and limited resources

(38). Health budgets and priorities are heavily influenced by major funding agencies

and interest groups. Local empirical data is critical to ensure key issues (such as the

high premature mortality from NCDs (39)) are not overlooked if local health concerns

do not closely reflect those of the broader international community (40).

There are indications life expectancy (LE) has, in contrast to previously published

estimates, shown no improvement in several counties across the region including

Fiji, Tonga and Nauru (14), most likely as a consequence of premature adult

mortality from NCDs. Despite the region declaring an NCD crisis in 2011, the relative

contribution of specific NCDs to mortality levels, or how rapidly they are changing, is

not well understood. The Pacific Islands are also one of only two regions worldwide

(the other being Eastern Europe) where there appears to be excessively high levels

of adult compared to child mortality (41). These factors must be better understood for

policy action to improve health in the PICTs. To enable this, more reliable estimates

of underlying mortality patterns and trends are needed.

WHO reports “Most countries in the region are taking steps to align their information

systems to public health functions and needs. More active and rational use of

information is being observed in health programme management and health

planning. The majority of countries now also appreciate the need to monitor their

health information systems [including mortality and CoD data] continuously to

improve the quality and use of data” (42).

� � �

6

1.2 The Pacific Islands

The PICTs consist of 21 countries and territories across three sub-regions:

Melanesia, Micronesia and Polynesia. Ranging from just over 1000 people (Tokelau)

(43) to 6.7 million (Papua New Guinea) (43), no two islands share a land border, and

most are separated by thousands of kilometres of ocean. While most have relatively

small land areas (as shown in Table 2), they all manage considerable marine areas

and many are spread over vast reaches of ocean (27, 36).

Table 2: Land area, population and density by sub region: Pacific Islands (44)

Region Land area (km²)

Estimated number of people Mid-year 2011

Population density (people/km²)

Melanesia 542,377 8,797,410 1,909,113 (excluding PNG)

16 (24 excluding PNG)

Micronesia 3,156 546,491 173

Polynesia 7,986 668,470 84

TOTAL 553,519 10,012,371 3,124,074 (excl. PNG)

18 (34 excl. PNG)

Over 800 languages are spoken across the region, with English and French the

primary official languages. There is significant cultural and political diversity (Table

3); from Tonga which has recently shifted power to a democratically elected

parliament, but remains the only Kingdom with an independent monarchy to have

avoided colonization; to Fiji which is currently under the rule of a military selected

government; to the territories of the United States (Guam and the Northern Mariana

Islands), France (New Caledonia and French Polynesia) and New Zealand

(Tokelau). There is also a vast Pacific diaspora in Australia, New Zealand and the

United States, with many communities maintaining close links with their home

� � �

7

islands and significant co-habitation across countries. Health services in the PICTs

are almost exclusively provided by the government sector.

PICTs are at different stages of the demographic transition, with populations such as

Kiribati and Vanuatu still experiencing relatively high mortality and fertility, while

others such as Palau manifest lower mortality and declining fertility (2). PICTs are

also passing through the epidemiological transition, with a progressive increase in

proportionate mortality from NCDs and consequent widening in the sex differential of

death rates (2).

� � �

8

Table 3: Pacific Island Countries and Territories, and their key characteristics

Country or Territory

Status Sub-region Estimated population(2011) (44)

Land area (km2)(44)

No. of islands (45)

GDP(Constant)per capita USD - 2012 (46)

Fiji Islands Independent Melanesia 851,745 18,273 >320 5,233 New Caledonia

Territory (France)

Melanesia 252,331 18,576 12

Papua New Guinea

Independent Melanesia 688,297 462,840 Mainland + 600 islands

Solomon Islands

Independent Melanesia 553,254 30,407 992 (347 inhabited)

Vanuatu Independent Melanesia 251,784 12,281 >80 89,825 Fed States of Micronesia

Independent (Compact of association with USA)

Micronesia 102,360 701 607 2,115

Guam Territory (USA)

Micronesia 192,090 541 1

Kiribati Independent Micronesia 102,697 811 34 Marshall Islands

Independent (Compact of association with USA)

Micronesia 54,999 181 34

Nauru Independent Micronesia 10,185 21 1 N Mariana Islands

Semi-independent territory (Self-governing Commonwealth in union with the USA)

Micronesia 63,517 457 17 16,494

Palau Independent (Compact of association with USA)

Micronesia 20,643 444 >200 (8 inhabited)

American Samoa

Territory (USA)

Polynesia 66,692 199 7 7,874

Cook Islands

Independent (Compact of association with New Zealand)

Polynesia 15,576 237 15 11,109

French Polynesia

Territory (France)

Polynesia 271,831 3,521 118

Niue Independent (Compact of association with New Zealand)

Polynesia 1,446 259 1 16,494

Samoa Independent Polynesia 183,617 2,935 8 5,766 Tokelau Territory

(Administered by New Zealand)

Polynesia 1,162 12 3 atolls

Tonga Independent Polynesia 103,682 650 170 (36 inhabited)

3,677

Tuvalu Independent Polynesia 11,206 26 9 (8 inhabited)

1,770

Wallis and Futuna

Territory (France)

Polynesia 13,193 142 3

� � �

9

Note: GDP not available for other countries.

1.3 Countries

The scale and diversity of the region (Table 3) means it is not possible to focus on all

of the PICTs in this research. Consequently the thesis will focus on six countries: Fiji,

Kiribati, Nauru, Palau, Tonga and Vanuatu. These countries represent a range of:

population and land sizes, government structures, centralised and decentralised

health services, and economic situations. They also emanate from all three sub-

regions (Melanesia, Micronesia and Polynesia).

Figure 1: Map of selected Pacific Island Countries included in this research

Source: Pacific Islands Forum Secretariat (47)

� � �

10

1.3.1 Fiji

Fiji is comprised of 322 islands, of which around two thirds are inhabited. The

population of Fiji is 837, 271 (2007) (48). It is the largest population in the PICTs

outside of PNG. Fiji is a predominantly mixed society of Indigenous Fijians

(Melanesians) or i-Tauki (475,739 people in 2007) and Indo-Fijians now called

Fijians of Indian descent (313,798 people in 2007) (48). Growth is slow due to a

falling birth rate and high levels of emigration. (48)

The population is concentrated on the two largest islands of Viti Levu and Vanua

Levu. Both Suva (the capital) and Nadi (the second largest city and site of the

international airport) are located on Vitu Levu. The main islands are mountainous.

Major industries include agriculture (particularly sugar cane), tourism, fishing and to

a lesser extent manufacturing. Traditional community structures (chiefs and

extended family relationships) remain very important in many areas.

Fiji has a centralized health system. There are hospitals in each of the four divisions

(central, east, west, and north). There are two government hospitals in Suva

including the Colonial War Memorial Hospital, the largest hospital in Fiji. Health

programs are managed predominantly from Suva and through Division offices.

Health Centres at the Provincial level (the next level down) provide basic inpatient

(and some basic surgical) care. Fiji also has a strong community health nurse

program, with community health nurses and midwives assigned to communities

throughout the country. Retention of staff (particularly skilled staff) is a major issue.

Pay-cuts introduced for all government employees have led to many nurses and

some doctors leaving Fiji for higher wages elsewhere.

� � �

11

1.3.2 Kiribati

Kiribati is a group of 33 coral atolls spread across 4 million square kilometres of the

Pacific Ocean, spanning both the equator and the International Date Line (49). There

are three major island groups: the Gilbert Islands, the Phoenix Islands and the Line

Islands. While there are regular flights from the capital Tarawa to other islands of the

Gilbert Island group, there are no routine flights to the Phoenix or Line Islands.

International flights stop on Kirimati Island (in the Line Islands Group) between New

Zealand and Hawaii, however, all other access is by Sea. Kiribati gained

independence in 1971, having previously been a British Colony. The islands were

captured by the Japanese in World War II and were the site of several major battles

before being returned to British rule.

The population is approximately 100,000, with over one third of the population under

15 years of age (50). South Tarawa is home to 44% of the population and has a

population density of 2558 people per km square (50). This would be the 6th highest

population density in the world if South Tarawa Island was treated as a country (51).

Religion is important, with the population split approximately evenly between

Protestant and Catholic affiliations. Agriculture and industry are limited by the small

land area and topography. As atolls, soil quality is poor and predominantly sand,

limiting cropping opportunities. Average height above sea-level is less than 0.6m.

Today, Kiribati is heavily reliant on foreign assistance, and classified amongst the

least developed countries in the world (52, 53).

Health Services are provided through Tungaru Hospital, the national referral hospital

on South Tarawa Island, and two smaller hospitals at Betio (South Tarawa) and

Kirimati Island. Cuban doctors currently assist local doctors to staff the Tunguru

� � �

12

Hospital and provide consultation services at a small number of health clinics in

South Tarawa. There are 96 health clinics throughout the islands staffed by a

medical assistant (equivalent to a nurse practitioner), registered nurses or nurse

aides. Traditional birth attendants operate throughout the islands but are not

regulated through the Ministry of Health (MoH). There are no private health facilities,

mortuary or morgue facility. There are 24 local doctors (in 2012), of whom

approximately 90% have trained through the Fiji School of Medicine. The remainder

were mostly trained in PNG.

1.3.3 Nauru

One of the smallest countries in the Pacific, Nauru has a land area of just 24 square

kilometres with a population of just 9,547 people, along with an expatriate workforce

of around 4002 (2006) (54).This is changing rapidly due to the re-opening of the

immigration detention facility in conjunction with Australia, and the associated

support personnel. The island is a raised coral atoll, with phosphate mining the only

major industry. Mining commenced early in the 1900s but ceased in 2003 with easy

to access deposits exhausted. Mining operations recommenced under a national

corporation in 2006 at a significantly reduced scale (55).

Originally annexed by Germany and subsequently becoming a protectorate under

the then League of Nations following World War One (WWI); the island was

occupied by Japan then Australia during World War Two (WWII), and subsequently

placed under the joint management of Australia, New Zealand and Britain. Nauru

gained independence in 1968. Although one of the most prosperous countries in the

region in the early 1990’s (based on GNP per capita), Nauru is now heavily reliant on

foreign aid (56) to support government functions and health care.

� � �

13

There is one hospital on the island, following the 2005 merger of the public hospital

with the private hospital previously run by the mining corporation. The previous

public hospital site has been renovated and now functions as the public health and

dialysis clinic. All health services operated from these two facilities.

1.3.4 Palau

Palau is a chain of islands in the Marianas group in the northern Pacific. It stretches

from Helen Reef in the south near the equator, to Kyangel in the north. Of the

hundreds of islands and islets, only a few are permanently inhabited. The bulk of the

population of approximately 20,000 are based in Koror and Babeldaob (57). The

three main islands, Koror, Babeldaob and Pelelieu, lie within a single barrier reef

system, with smaller centres at Anguar and Kyangel (just south and north of this reef

respectively) and in the South-West Islands (some 600 kilometres from the main

group). Previously occupied by Spain then Germany, Palau was the Japanese

Pacific headquarters during WWII before it became a trust territory of the United

States. The country obtained independence in 1994 under a “Compact of Free

Association” with the USA. Funding under the original agreement expired in 2010

(58), with a revised agreement currently before the US House of Representatives for

approval following formal endorsement of the revised funding arrangements by the

Palau Parliament and US House of Representatives sub-committee in 2012 (58).

Palau has 16 states, each with its own government, along with a strong network of

traditional chiefs and women’s associations. The national government offices moved

to Melekeok state on the large island of Babeldaob in 2005, while the business

capital remains in Koror (state and island). The society is matrilineal with property

� � �

14

passed from mother to daughter, and traditional (male) chiefs selected by the senior

women. The major industry is tourism, with the country being renowned for its diving

sites. Palau also licenses a number of commercial fishing operations within its

waters.

Health services are provided through the national hospital and dispensaries in each

state. Each dispensary is staffed by at least one registered nurse, and is visited

monthly by a team (including doctors and other health staff) on a monthly basis to

conduct clinics.

1.3.5 Tonga

Tonga has four main island groups spread over a large expanse of ocean. These are

the Tongatapu group (where the capital Nuku a’lofa is situated), the Ha’apai group,

the Vava’u group and the two islands of the Niuas. Of the 171 islands, fewer than 40

are inhabited. Tonga has a population of 103, 036 (59). Most of the population (73%)

lives on the island of Tongatapu, with the second largest centre Vava’u located

approximately 200 kilometres to the north (57). Roughly one third of the population

lives in urban areas, with tourism and farming the major industries. The population is

Polynesian, with a small Chinese minority (many of whom have Tongan citizenship).

Women traditionally have a great deal of influence within the family, however only

men can own or inherit land.

The population is very mobile, both within Tonga and to and from other countries in

the region. There is a high rate of out-migration to Australia and New Zealand.

Subsequently the overall population growth rate is extremely low given the high

fertility rate (57).

� � �

15

Tonga is a Kingdom with a monarch as the head of the government. Government is

managed by the Prime Minister. It is a structured, hierarchical society with strong

traditional expectations regarding both communication and roles. The first

democratic elections were held in Tonga in 2010 resulting in a parliament comprised

of 26 elected representatives and nine nobles nominated by the monarch (60) .

The main hospital is a 200 bed facility in Nuku a’lofa, which underwent extensive

renovation in 2011-2012. Health centres provide basic medical facilities and limited

inpatient services at the next level. Of these there are seven on Tongatapu, two in

Hapai group, four in the Vavau group and two in the Nua’s. Below these there are

approximately 30 maternal and child health clinics. A strong community

midwife/nursing program is accessible through most of the country.

1.3.6 Vanuatu

Vanuatu is a volcanic chain of approximately 82 islands, making up a total land area

of 12190 square kilometres. Vanuatu has a population of 234,023 people (2009)

(61), with only 21% living in urban areas. Agriculture (including beef cattle) and

tourism are the major industries. Offshore financial services are also a substantial

source of income. Vanuatu remains amongst the least developed countries in the

world under the UN development index (62). Managed jointly by both France and

England for 76 years prior to independence in 1980 (62), both languages continue in

official use. Bislama, a form of pidgin English, is the common language spoken. Most

people have either English or French as a third language, along with one of the

many indigenous languages that remain in use.

� � �

16

Government is largely decentralised to the six provinces. The capital Port Vila is on

the island of Efate, in Shefa province. The provinces are reflected in the health

administration districts. There is one referral hospital, with 4 provincial hospitals and

25 private health care facilities. Local health services are provided through Aid posts

staffed by nursing aides, and area health centres staffed by enrolled nurses. Health

services are party to a Sector Wide Activity Plan (SWAP) negotiated with partners

such as AusAID, Global Fund and the World Bank. Measures of mortality, including

childhood mortality rate (under 5 years) (U5M), infant mortality rate (IMR) and

maternal mortality ratio (MMR) are considered critical performance evaluation

measures under key funding agreements such as the SWAP plan.

1.4 Measures of mortality and causes of death

The simplest measure of mortality level is a direct count of the crude number of

deaths in a given population over a given time. While counts of reported deaths (by

age group and cause) may be useful in small stable populations where little change

is anticipated, mortality levels are usually measured as rates or proportions to allow

comparison over time and between different populations. Crude death rates similarly

do not readily allow comparison between populations due to the effect of age

structure. Important measures of mortality level used for health planning include sex

and age-specific rates, IMR and under 5 mortality rate (probability of dying between

0-4 years inclusive, 5q0). Summary measures of mortality such as LE, adult mortality

(the probability of dying between ages 15-59 inclusive, 45q15) and age-standardised

rates are also important for comparison between countries, regions and sub-

populations to identify inequalities, evaluate trends over time, and to present

mortality trends in a meaningful way for policy makers (Table 4). There are much

� � �

17

greater social and financial incentives for reporting deaths in productive adult age

groups, and therefore under-estimation of the total number of deaths tends to have a

greater effect on early childhood deaths and older adults (63, 64). As LE is

calculated from mortality rates disaggregated by age group rather than as a total

population, unless IMR is very high LE estimates tend to be much less sensitive to

errors in estimation of completeness than crude death rates (65) and are therefore a

more reliable measure of mortality.

� � �

18

Table 4: Key measures of mortality and causes of death

Measure Definition Infant mortality rate (IMR) Deaths in children aged less than 1 year divided by the

total number of live births. Usually reported per 1,000 live births.

Under 5 mortality rate (U5M)

Deaths in children aged less than 5 years divided by the total number of live births. Usually reported per 1,000 live births. This is a probability of dying rather than a population based rate.

Age-specific mortality rates

Number of deaths in a specific age group for a defined period divided by the total (mid-period) population in that age group. Usually reported per 1,000 population.

Age-standardised mortality rate

The number of deaths that would be expected to occur in a given population based on local empirical age-specific mortality rates if the age structure reflected a defined standard (usually the WHO world standard population), divided by the standard population. Usually reported per 1,000 population.

Adult mortality (45q15) The probability of dying between ages 15 and 59 years of age inclusive. Expressed as deaths per 1,000 population or as a percentage.

Life expectancy at birth (LE0)

The average number of years a person born at a defined point in time would be expected to live if exposed to the age specific mortality rates of that time throughout their lifetime.

Proportional mortality by cause

The proportion of deaths (as a percent) attributed to a specific underlying CoD (as defined by the International Classification of Diseases version 10, ICDv10)

Cause-specific mortality rate

Number of deaths in a specific age group for a defined period attributed to a specific underlying CoD (as defined by ICDv10) divided by the total (mid-period) population in that age group. Usually reported per 100,000 population.

(66, 67)

Cause of death may be expressed as age/sex/cause specific death rates and as

proportional mortality by cause if data are under-enumerated. CoD is conventionally

reported as a single underlying cause; derived from the medical certificate of death

� � �

19

(hereafter called the medical certificate) based on a prescribed set of rules outlined

in the International Classification of Diseases (ICD) (68) The international standard

medical certificate has two parts. Part I is the causal sequence leading backwards

from the immediate CoD in line one. Part II is for conditions that were present at the

time of death, but did not directly lead to the death.

Figure 2: International standard medical certificate of death format (68)

Changes in the ICD rules over time have resulted in some shifts in patterns of

underlying cause, making comparison between versions problematic (69). The

current revision in use is Version 10 (ICDv10) (68). There is an Australian

modification (ICDv10AM) (70) used by several PICTs for morbidity coding which is

where major differences in these two versions predominantly occur. Mortality coding

is essentially comparable. Changes from ICD version 9 (ICDv9) to ICDv10 has

� � �

20

resulted in a significant shift in the recording of diabetes, leading to a higher

attribution of deaths to diabetes as the underlying cause in some countries (69, 71).

In ICDv9, diabetes was considered the underlying CoD only if the preceding causes

were directly attributable to complications of diabetes or diabetic renal disease (71).

ICDv10 recognizes the causal effect of diabetes in certain types of cardiovascular

disease. It allows diabetes as the underlying cause for deaths attributed to

cardiovascular causes in the preceding lines on the medical certificate, provided

diabetes is listed in Part I. Diabetes as a risk factor for coronary atherosclerosis

producing ischaemic heart disease (listed in Part II) may not qualify. As such,

diabetes rates as CoD are expected to be significantly higher under ICDv10 than

previous versions (69, 71).

The ICD attributes each death to one underlying cause (68). This ensures the sum of

all cause-specific deaths equals but does not exceed the total number of deaths.

Cause-specific death rates may be calculated for multiple cause analysis. In this

case, the numerator is the total deaths with any mention of the disease of interest

(total mentions) anywhere on the medical certificate (Part I or Part II). This is useful

for measurement of the total burden of mortality attributable to each cause, but is

less readily understood by policy makers, and less comparable.

� � �

21

1.5 An overview of mortality and cause of death data for the

Pacific

1.5.1 Historical context – patterns of mortality and causes of death

Mortality associated with devastating epidemics of infectious diseases such as

measles, whooping cough, and influenza (including the 1918-19 pandemic) was

recorded throughout the region as a result of European and Asian contact and

continued through to the 1940’s (72, 73). Infectious disease, nutrition and maternal

causes were the major causes of death noted at this time (72, 73). Infant mortality

was high, estimated at 150 deaths per 1,000 live births for Fijians and 176 deaths

per 1,000 live births in Papua New Guinea in the 1920’s (74),(72).

Major land and sea battles occurring in Guam, the Solomon Islands, the Marshall

Islands, Kiribati (then known as the Gilbert Islands), Nauru, and Palau during WWII

(72, 75-79). Mass depopulation occurred as islands were affected by the fighting,

and troop movements spread dengue mosquitoes around the region (72, 73). At the

end of the war, most of the PICTs were under colonial control by British, French,

Australian, New Zealand and United States governments. During this period, public

health measures focussed on malaria prevention and improved sanitation (72).

Immediately post WWII, the major causes of death in Vanuatu, the Solomon Islands

and PNG were reported as malaria, pneumonia, tuberculosis, dysentery and

malnutrition. In the non-malarious areas of Melanesia these were pneumonia,

dysentery, enteric fevers, filariasis, yaws, tuberculosis, leprosy, syphilis, and

childhood infections (72). Death rates from infectious diseases gradually declined

from WWII to the end of 1950’s (3),(80), although they remained the leading CoD

(72).

� � �

22

Between WWII and the turn of the century, many of the islands regained their

independence (43). This was a period of rapid transition in many of the Pacific

Islands. Populations grew from continued high fertility and declining premature

mortality, and age structures shifted as people lived longer. Changes in sanitation,

health care and nutrition saw patterns of mortality shift. From the 1950’s, public

health programs continued to expand at both national and regional levels. The

Expanded Programme for Immunization, managed through WHO, began in the

Pacific in the 1970's and achieved 80% coverage of children receiving the basic set

of vaccines by 1988 (81). By the early 1960’s, infectious diseases no longer

predominated as the leading CoD in many PICTs (39, 80).

The first cases of HIV were reported in the mid 1980’s in all three sub-regions of the

Pacific (82). By 2000, HIV had become a generalised epidemic (self-sustaining

through heterosexual transmission (83) ) in PNG and moving towards a generalised

epidemic in Tuvalu and Kiribati. Other countries remain within the threshold for a

concentrated epidemic (with prevalence <1% in the general population but more than

5% in at least one high risk groups (83)).

It is evident that there was no single pathway through this demographic and

epidemiological transition. Early transition populations, characterised by continued

high infant and maternal mortality and infectious diseases remaining a leading CoD,

include PNG, Kiribati, Vanuatu, and the Solomon Islands (2, 3, 84-89), In countries

such as Fiji, Tonga and Samoa infant and childhood mortality rates have fallen and

chronic NCDs have become more important as a CoD (39). Increases in proportional

mortality from NCDs in these countries, when coupled with the lack of ongoing

� � �

23

improvements in LE beginning to be reported (89), suggest real age-specific

increases in NCD related deaths rather simply reflecting a decreasing impact from

infectious diseases. Obesity rates in the Pacific islands are now amongst the highest

in the world, with 75% of adults in seven PICTs either overweight or obese (40).

Six countries were considered to have high (>15%) proportional mortality from

cancer by 1985 (Guam, French Polynesia, Palau, Tonga, American Samoa and

Cook Islands) (90). As early as the 1970’s, diseases of the circulatory system were a

leading CoD in many PICTs (91). Reports note that despite the obvious health risks

associated with atmospheric nuclear testing carried out in French Polynesia from

1966 to 1974, the biggest health impact was from increased affluence and the

associated lifestyle and environmental changes (39).

By the 1980’s, the first suggestions of a stagnation in health improvement were also

noted, reflected in very minimal or no improvement in LE from previous periods. This

stagnation noted at the 1986 Solomon Islands census with LE estimated at 59.9

years for males and 61.4 years for females (92). Stagnation in LE has also been

more recently demonstrated in Fiji (93). This demonstrates the ‘double burden of

disease’, where increases in the incidence, prevalence and mortality of NCDs occur

alongside existing health problems of infectious diseases such as malaria,

respiratory infections and gastroenteritis (40).

NCDs are the leading cause of hospital based deaths in many of the islands,

including FSM (1990, (94)), Fiji (1996 (91)), and Tonga (2005 and 2007 (95, 96)) .

However, information on CoD in the community remains limited for most PICTs. In

� � �

24

2011 Health Ministers declared an NCD crisis in the region, further highlighting the

need for improved information for evaluation and monitoring (5).

‘Best Estimates’ of LE for 2000 published by Taylor et. al. for the PICTs were as low

as 51-58 years in PNG and as high as 73-77 for Guam and 70-76 for New Caledonia

(2). These estimates were strikingly lower than those reported by WHO for the same

period, and suggest mortality in the PICTs was in fact higher than had previously

been considered (2). A range of available data sources were evaluated for quality,

identifying wherever possible the original data and analytical method and considering

these in light of their known strengths and weaknesses. A key finding of this work

was that many of the available sources provided limited information on how data had

been obtained and estimates derived. As such, for many of the countries the authors

were able to provide only a broad range of values in which LE was likely to fall. This

indicates a critical need to re-examine sources of primary data in the PICTs to

establish more current estimates that will withstand similar scrutiny of their methods

and to provide greater certainty around the current mortality levels in the Pacific.

1.5.2 Previous Data Quality

An assessment of available data by Taylor et. al. in 1989 (3) highlighted significant

variation in published estimates of LE and a lack of COD data (3). This assessment

was repeated by Taylor et. al. in 2005 for estimates around the year 2000 (2) as

noted in the previous section, resulting in a similar conclusion. Wide disparities in

published estimates of mortality across the region were noted, suggesting many

published estimates over-estimated LE and indicating a continuing stagnation or

decline in health status in the PICTs (2). Plausible estimates of LE around 2000 were

derived predominantly from census analyses. Vital registration was considered

� � �

25

reliable for the territories and small countries of Nauru, Palau and the Cook Islands.

The assessment of data sources is largely consistent with other reviews on reporting

completeness available for this period. The Solomon Islands MoH is reported to

have captured less than 10% of deaths through routine reporting in 1985-1986 (92).

The Republic of the Marshall Islands reported “significant under-registration” of

deaths in 1995 (97). In 1997, death registration data from the MoH in the Federated

States of Micronesia was estimated to be less than 65% complete (98).

The absence of reliable data was highlighted repeatedly throughout the 1990’s and

2000’s (4), (2), (99, 100). The first Global Burden of Disease (GBD) study, published

in 1996 (101-104), found there was inadequate mortality data available for most

PICTs, and therefore data from other countries in the broader WHO Western Pacific

region were used to develop mortality level and CoD estimates (104),(105). This

process was repeated in the 2000 update released in 2005 (105). Regional

aggregations, largely dominated by the health situation in PNG as used in the GBD

studies, are of little use in health planning at a country or local level in other PICTs

due to the relative population sizes.

Even in PICTs where mortality data was considered reliable, there was a notable

lack of data on causes of death. At the end of 2003, only Niue was reported to have

recent reliable CoD data (4). “Unclassified” deaths or “others” were reported as the

highest CoD (20.4%) in the Marshall Islands in 1999 (106), while “unknown causes,

signs and symptoms” were the second and third highest CoD in Tonga in 2005 (95)

and 2007 (107) respectively. Where CoD was recorded, tabulations of deaths were

often based on immediate rather than underlying CoD (106). Recent published CoD

data for the PICTs is fairly limited, consisting predominantly of leading CoD lists by

� � �

26

all ages and both sexes combined, as published in the WHO Country Health

Information Profiles (89), and some national MoH annual reports (95),(107)

1.5.3 Recent data

Mortality estimates for the PICTs are currently published regularly by a range of

international agencies, with additional reporting through national statistics offices and

Ministries of Health. Tables 5 and 6 provide IMR and life expectancies at birth for the

PICTs as published in 2010 by WHO Country Health Information Profiles (CHIPs)

(108), the Secretariat of the Pacific Community’s (SPC) PRISM website (46), and the

UN statistical yearbook (109). Male and female LE at birth as published by these

sources is shown in Figures 3 and 4.

The CHIPs report is the primary health status publication in the Western Pacific

Region and has been published annually since the early 1980’s (108). Data are

derived from a combination of country MoH sources sent directly to WHO by

member states on request, and WHO program data. The PRISM website is

managed by SPC on behalf of the governments of the PICTs. The program

maintains close links with the statistical agencies for each country. Estimates are

derived from country reports and census analyses (mostly conducted by SPC in

partnership with national statistical agencies). The UN estimates shown here are

derived from the 2010 statistical yearbook and report the most recent empirical data

held by the UN statistical agency for each country at that time. IMR is not published

in the UN yearbook.

� � �

27

Table 5: Published estimates at 2010 of Infant mortality rates (per 1,000 live births) (Both sexes combined); Pacific Island Countries and Territories

Country WHO (CHIPs) (108) SPC (PRISM) (43)�Reference Year

IMR Reference Year

IMR

AMERICAN SAMOA 2007 11.8 2006-08 11.3 COOK ISLANDS 2007 10.5 2005-09 11.6 FIJI 2008 16 2000 37.5 FRENCH POLYNESIA 2005 5.3 2006-08 17.0 GUAM 2004 12.3 2007-08 5.8 KIRIBATI 2005 52 2005-07 11.7 MARSHALL ISLANDS 2007 23 2003 52.0 MICRONESIA, FS of 2006 33 2003-07 21.0 NAURU 2002 12.7 2006-07 45.8 NEW CALEDONIA 2007 6.6 2007 6.1 NIUE 2003 29.4 2001-06 7.8 NORTHERN MARIANA IS

2005 7.11 2006-08 4.9

PALAU 2004 16.22 2004-06 20.1 PAPUA NEW GUINEA 2006 49 2001-06 56.7 SAMOA 2005 13.7 2006 20.4 SOLOMON ISLANDS 2008 31.4 2002-07 24.3 TOKELAU 2000 33 2000-03 31.3 TONGA 2005 11.8 2004-05 19.0 TUVALU 2003 21.6 2006-08 17.3 VANUATU 2006 30 2009 21.0 WALLIS AND FUTUNA 2003 5.9 2005-08 5.2

As can be seen in the above table, even allowing differences in reference years,

there are significant disparities between estimates from CHIPs and PRISM. Notable

examples include French Polynesia, Kiribati and Tonga. The data suggest IMR has

fallen substantially in Fiji, Niue, the Marshall Islands and Vanuatu and is increasing

in French Polynesia, Nauru, Palau, Samoa, the Solomon Islands and the Federated

States of Micronesia. These apparent changes are reasonably unlikely and the

differences suggested by these data require investigation.

� � �

28

The disparity in published estimates of LE is even more apparent (Table 6, Figures 3

and 4). LE in Kiribati ranges by more than 13 years for males and 16 years for

females. In part, this disparity is generated by the extreme difference in reference

time periods; however, as all of these publications purport to be the most recent data

available, there is significant potential for these to be misleading. Much of the data is

also implausible, with estimates from Marshall Islands showing an 8 year

improvement in LE for males between 2003 and 2004 (108, 110), and Kiribati

showing a significant decline in LE over a five year period from 2000-2005 (108,

110).

Notably there are missing and dated estimates from the UN. This suggests the most

recent mortality data available for Kiribati at 2010 was for the 1950’s and 1960’s.

None of the data post dates 1995. SPC included data as recent as 2009 while WHO

had data as recent as 2008 in their 2011 CHIPS report, although most estimates

were for more than five years prior to the publication. Despite being significantly

more recent than the UN reports, this highlights a delay of at least two to three years

in the most recent data available to international agencies.

� � �

29

Table 6: Published estimates at 2010 of LE at birth (males and females); Pacific Island Countries and Territories

Country

LE at birth (years, Males) LE at birth (years, Females) Reference Years

WHO (CHIPS)

SPC (PRISM)

UN stats

WHO (CHIPS)

SPC (PRISM)

UN stats

WHO (CHIPS)

SPC (PRISM)

UN stats

AMERICAN SAMOA 72.27 69.3 65 79.62 75.9 69.1 2005 2000 1969-

1971 COOK ISLANDS 65 69.5 63.17 73 76.2 67.09 2005 2001-6 1974-

1978

FIJI 68 67.4 69.5 72 68.0 73.7 2008 2000 1990-1995

FRENCH POLYNESIA 73 63.8 68.3 76.9 67.7 73.8 2006 2001 1990-

1995

GUAM 75.34 72.0 72.2 81.64 76.6 76 2005 2005-07

1990-1995

KIRIBATI 58.9 71.1 56.9 63.1 76.1 59 2005 2000 1958-1962

MARSHALL ISLANDS 67 58.9 59.06 70.6 63.1 62.96 2004 2003 1989

MICRONESIA, FS of 67 63.7 64.4 70 67.4 66.8 2006 1999 1991-

1992 NAURU 58 55.2 61 57.1 2004 2006 NEW CALEDONIA 71.9 71.8 67.7 78.6 80.3 73.9 2006 2007 1994

NIUE 67.0 76 76.0 2004 2001-6 NORTHERN MARIANA IS 73.3 73.5 78.61 77.1 2005 1999-

01 PALAU 67.8 66.3 75.68 72.1 2004 2005 PAPUA NEW GUINEA 52.5 53.7 55.2 53.6 54.8 56.7 2000 2000 1990-

1995

SAMOA 71.8 71.5 65.9 73.8 74.2 69.2 2001 2006 1990-1995

SOLOMON ISLANDS 62.2 60.6 68.4 64.3 61.6 72.7 2005 1999 1990-

1995 TOKELAU 68.4 67.8 71.3 70.4 2000 1990

TONGA 70 67.3 72 73.0 2005 2004-05

TUVALU 61.7 61.7 65.1 65.1 2002 1997-02

VANUATU 67 69.6 63.5 70 72.7 67.3 2006 2009 1990-1995

WALLIS AND FUTUNA IS 73.1 72.7 75.5 75.9 2003 2005-

08

� � �

30

Figure 3: Recent published estimates of male LE at birth, Pacific Island Countries and Territories

Figure 4: Recent published estimates of female LE at birth, Pacific Island Countries and Territories

� � �

31

In addition to the WHO CHIPs and UN statistical reports discussed previously, both

agencies, along with the United National Children’s Fund (UNICEF), and the United

Nations Development Programme (UNDP) maintain databases of annual estimates

of mortality for all countries including LE (all agencies), U5M and IMR. In most of

these databases, the original data source or analytical method are not explicitly

stated. The UNDP database uses predicted estimates based on an assumed

improvement each year from earlier data (108).

Despite studies that suggest estimates derived from local census analyses are the

most reliable data on mortality for much of the region (excluding the small islands

and territories) (2, 93, 111, 112), country generated reports frequently rely on

estimates from international agencies. Table 7 is derived from the most recent MDG

progress report published by the Fijian government. The IMR estimates shown were

obtained from the World Bank, despite local data being available for the period

shown. There was no information available regarding the original data or methods

used to generate these estimates.

Table 7: Extract from Fiji MDG Report: Statistics on Goal 4 - Fiji Millennium Development Goals 2nd Report, 1990-2009 (113)

� � �

32

Additionally, Table 7 demonstrates the use of annual estimates of mortality

measures for small populations of the Pacific. This potentially leads to significant

over-interpretation of data, resulting in annual estimates that are highly susceptible

to stochastic variation. Annual estimates are used in a range of annual reports from

Ministries of Health, the CHIPs reports (as shown in Tables 5 and 6) and the

statistical databases maintained by the UN and UNICEF (109). Only the estimates

published by SPC (Table 5 and 6) show any consistent use of aggregation over time.

Similarly, measures of maternal mortality in the PICTs are highly unstable due to

stochastic variation, particularly when generated for single years. The regional MDG

progress report published by SPC in 2004 stated the MMR was ‘not a suitable

measure’ for maternal mortality in the Pacific Islands due to the small population

sizes (114). Despite this, MMR continues to be used almost exclusively by both

country health departments and international agencies.

Information on the top ten leading causes of death is produced by several PICTs and

collated in the WHO CHIPs (108). As noted previously, estimates are submitted

directly by countries or WHO programmes. Data from the CHIPs publications show a

number of issues. Firstly, cardiovascular diseases, cancer and even diabetes feature

prominently in recent lists, demonstrating the importance of these chronic and

lifestyle disease. Secondly, as noticed with PICTs such as Kiribati, Palau and

Vanuatu, unknown and ill-defined causes appear in the list of top ten causes of

death (108). Finally, the lack of age-specific data on CoD makes it difficult to assess

the causes that contribute to the early adult mortality that must be present given the

stagnation in LE reported by Taylor et.al (2) and confirmed in studies presented in

this thesis (93).

� � �

33

Comparison of the CoD distribution between PICTs has the potential to provide

valuable information on major regional problems, and to highlight emerging issues in

the region or where individual countries and territories are being left behind in terms

of health status. It is not possible to make comparisons in CoD distributions between

PICTs from the CHIPS data as there is no standard format, resulting in various

combinations of specific diseases and categories of diseases to be used. That ill-

defined and unknown deaths feature in these lists of leading causes of death also

indicates a high proportion of deaths in the region do not have a specific CoD

assigned, and the proportional mortality reported may not reflect the true distribution

of causes.

While the PICTs have been undergoing an epidemiological transition from a high

proportional mortality from infectious disease to increasing proportional mortality

from NCDs, this has not been even across the region, and uncertainty remains over

the effect of this transition on mortality levels. Previous studies have demonstrated

stagnation in LE across the 1980’s and 1990’s (2, 3, 115, 116). Other than the work

presented here, there has been no comprehensive assessment of mortality level or

CoD since these studies. It is apparent that both the lack of data and wide disparities

noted twenty years ago remain. The disparity in current mortality estimates is

potentially highly misleading for decision makers, and the estimates themselves are

of little use in evaluating health status without further assessment of their validity,

and the ability to disaggregate CoD data by age and sex.

The growing importance of NCDs increases the need for reliable ongoing mortality

and CoD data, to guide health planning, prioritise funding and evaluate interventions.

This data can be produced through well functioning routine CRVS systems. At

� � �

34

present, little is known regarding the operation and potential biases of many of the

CRVS systems in the PICTs. Taylor et. al.’s assessment of data from the 1980’s and

1990’s (3, 115) suggest only the small islands (such as Nauru and the Cook Islands)

and territories were able to generate valid mortality estimates from routine data.

More recent analysis by Mathers et al. in 2003 (4) suggests not even these smaller

countries are producing reliable data.

Mortality data (both level and CoD) are fundamental information required to assess

the health of a population, plan appropriate services, evaluate programs and make

decisions around how limited resources are best directed. Additionally, an

understanding of how mortality has changed over time allows an assessment of

where each country or territory is in terms of epidemiological transition. If sufficient

information can be obtained to determine if there is a “typical” pattern of

epidemiological transition within the PICTs, countries still in the midst of this

transition will be better equipped to predict the speed and distribution of these

changes for health planning purposes. As these “early transition” Islands are also

predominantly those most reliant of international assistance to meet health funding

needs, the ability to plan for the future is vital to ensure required services can be

met.

1.6 Research Objectives and Questions

Investigating mortality and CoD draws on research methods from both epidemiology

and demography. This thesis aims to define and investigate the current gaps and

uncertainties in mortality and CoD profiles for the Pacific Islands by addressing the

following primary research questions for selected islands:

� � �

35

o Based on empirical data (from routinely collected data where possible), what are

the current mortality levels in the Pacific Islands by age and gender?

o What are the major causes of death contributing to these mortality levels?

o How can existing data collections and CRVS systems be better used to generate

valid, reliable data on mortality and CoD, significantly disaggregated by age group

and cause to inform health planning and evaluate health programs?

In order to answer these primary research objectives, this thesis also addresses

three secondary research questions.

1) what are the current empirical mortality data collections available for the Pacific

Islands and

2) what biases are inherent in these collections?

3) what methods are suitable to assess the quality and completeness of the data

available from the Pacific Islands?

As noted earlier, due to the scale and diversity of the region, it is not possible to

review all the PICTs in this research, and these questions are investigated in relation

to six selected countries that reflect this diversity.

Chapter two provides a literature review that is essentially the “toolbox” of methods

available for this work. The review covers the primary methods of data collection for

mortality and CoD data, methods for evaluating the design and performance of

routine CRVS systems and their impact on data quality, and the methods for

assessing the quality of data and for generating estimates of mortality and CoD

distribution from flawed data. Strengths and weaknesses of these methods are

discussed, and their potential applicability to the PICTs reviewed.

� � �

36

Chapter three examines the routine CRVS systems for deaths in the selected

countries to address the first two secondary questions above, 1) what are the current

empirical mortality data collections available for the PICTs and 2) what biases are

inherent in these collections? The chapter also provides recommendations for how

routine data collections could be improved to provide more reliable estimates in the

future. Potential biases are further discussed in Chapter four which outlines the

potential impacts of small population sizes and migration on mortality and CoD

estimates for the PICTs, in order to identify where additional care or corrections may

be required.

Chapter five provides an overview of methods used to analyse mortality levels in

selected countries, based on the tools and data collections available and discusses

their strengths and weaknesses in this setting. Best-estimates for age-specific

mortality are subsequently generated. Causes of death are similarly dealt with in

Chapter six through examination of validity and the need for data correction.

The mortality and CoD profiles for selected countries are summarised in Chapter

seven, with a discussion of the data in a regional context examining how patterns of

mortality have changed and implications of this for public health priorities.

This thesis concludes in Chapter eight, returning to the original research questions,

and providing recommendations for additional research.

� � �

37

An overview of how these chapters fit together in this structure is given in the

following diagram (Figure 5).

Figure 5: Thesis structure and content

�

� � �

38

2 An overview of methods for collecting, analysing and assessing the quality of mortality level and cause of death data

Literature relevant to this thesis falls into two broad categories:

1/ data collection approaches for mortality and CoD data and the types of data

collected through these systems, including an overview of their strengths and

weaknesses; and

2/ the methods available for assessing the quality of the data from CRVS systems

and deriving estimates of mortality level and CoD distributions.

� � �

39

Methods can be further categorized by:

a/ methods for evaluating and assessing routine CRVS systems for mortality and

CoD data, and the impact of system design and performance on data availability and

quality;

b/ methods for evaluating (and correcting as necessary) mortality level data for

quality and validity; and

c/ methods for evaluating (and correcting as necessary) CoD data for quality and

validity.

A literature review was conducted in Pubmed and Medline, and using the reference

lists of key articles. Data were also sought directly from country holdings, and key

historical collections held by the SPC library and Prof. Taylor. Documents were

critiqued for content and evaluated by theme, with key findings presented below.

2.1 Sources of mortality and cause of death data

Vital statistics may be collected through censuses, surveys, or routine vital

registration collections such as civil registration or registries operated through the

MoH (34). The UN identified the importance of taking into account the distribution of

mortality when designing a system to measure it, and that the three collection

methods are complementary, rather than alternative options (117). Additional

administrative data collections including records of deaths may also exist, and be

used for analysis of mortality level and CoD distribution. These include Health

Information Systems (HIS) and hospital discharge data, police records for attended

deaths, and social security records.

� � �

40

2.1.1 Census collections

The primary source of mortality data in the PICTs has historically been through

periodic censuses (Table 8). The first census in Fiji was conducted in 1879 (118).

Regular censuses were held across the region as early as the 1920’s (119).

Undercount was a noted concern with these early censuses, with mortality measures

derived from direct retrospective questions on deaths in the household over a

nominated reference period (75, 119-123).

��

�

41

Tabl

e 8:

Cen

suse

s in

the

Paci

fic Is

land

s, 1

950-

2010

C

EB

CS

+ R

egis

tratio

n da

ta -

mod

elle

d

No

mor

talit

y da

ta re

leas

ed

� � �

42

In more recent census analyses, a more common approach is to derive measures of

childhood mortality, or both childhood and adult mortality, from indirect demographic

methods that distribute deaths over time based on the age of the respondent (124-

126) (Table 9). These measures have then been used as inputs for model life tables

that generate estimates of age-specific measures of mortality and LE. Models that

use a single input parameter based on childhood mortality have been shown to have

a tendency to under-estimate mortality in the PICTs where low infant mortality does

not necessarily imply low adult mortality, as demonstrated for Fiji in case study

three (included in Chapter five) (93). Models that use both childhood and adult

mortality generally perform better (1, 93). Childhood mortality can be derived using

the children ever born/children surviving technique (CEBCS), where women of child

bearing age are asked how many children they have ever had and how many are still

alive. Deaths are then distributed in time according to national fertility patterns and

the age of the respondent (127). Adult mortality can be estimated using sibling

survival (how many brothers and sisters the respondent has had, how many are still

alive, and age of respondent), widowhood (has the respondent ever been married, is

that spouse alive, and age of respondent), and/or orphanhood techniques (is the

respondent’s mother or father alive, and age of respondent) with deaths distributed

according to the age distribution of the respondents (127). The UN manual X (127)

provides a good overview of calculations associated with these methods.

Indirect methods, based on the age distribution of respondents, are generally

preferred by statistical agencies over direct questions related to the number of

deaths in the previous twelve months which have “not proved to be very satisfactory”

(117) due to serious concerns around recall bias and subsequent under-enumeration

of deaths (11). In particular, the UN notes experience with the widowhood technique

� � �

43

suggests it should not be relied upon for accurate estimation (117). Although the

other indirect methods are considered more reliable (128-131), they are also affected

by issues of recall, can be affected by bias due to cultural issues around discussing

deaths, and rely on assumptions around the relationship between the distribution of

deaths over time and respondents age (66). Estimates of adult mortality may also be

derived from changes in population (by age and sex) based on survival and

migration recorded between multiple census rounds. The technique of inter-censal

survival is particularly reliant on accurate migration data, which is frequently

unavailable at the level of disaggregation required in the PICTs (132).

Enumeration of deaths through a census is noted to be more difficult in developing

countries as there tends to be a wider family structure (less nuclear families) (117),

increasing the level of uncertainty in data captured through questions about family

(such as how many siblings do you have) or households (such as have there been

any deaths in the household in the previous 12 months). Some Pacific cultures also

have social conventions that mean death cannot be freely spoken about, which may

introduce response bias. Despite these concerns, censuses have remained the main

source of mortality data for the region over the past 60 years due to the absence of

reliable CRVS systems and concerns about data quality (38, 50, 133). Censuses

have been held in most PICTs every five or ten years since 1946 (Table 8) (48, 50,

57, 59, 61, 123, 134-147), facilitated by assistance from SPC in Melanesia and

Polynesia, and the US Census Bureau in Micronesia. Limitations of demographic

analyses of census data have been highlighted on many occasions. For example,

IMR estimates for the 1967 and 1979 Vanuatu censuses were reported as

implausible (122, 148, 149) while the 1989 Vanuatu census reported difficulty in

establishing an appropriate measure of adult mortality (144). Censuses in the

� � �

44

Solomon Islands in 1986 and 1999 also reported difficulties establishing adult

mortality, with models used to derive these measures from the data providing

implausible results (92, 142). Until 2000, most of the censuses conducted in the

PICTs continued to use life table models with a single input parameter of IMR or

child mortality to estimate adult mortality and LE (141-144, 149-152). Although more

recent censuses have demonstrated a shift to including two input parameter life

tables where possible, mortality estimates for several countries including Tonga (59,

136) and Vanuatu (135) continue to be generated from single input parameter

models. Methods used in the most recent census round are outlined in Table 9.

Censuses do not collect data on causes of death and therefore cannot be used to

derive population based measures of cause-specific mortality.

Table 9: Type of data collection and methods of estimation used for calculation of life tables in the most recent census round for selected Pacific

Island Countries and Territories

Country Census Year

Method of estimation

Fiji 2007 No mortality data released (48) Kiribati 2005 Model life tables using single input parameter

(CEBCS) (50) Nauru 2011 Vital registration data smoothed using model life

tables (134) Palau 2005 Model life tables (input unclear – smoothed vital

registration data or CEBCS input parameter – both collected) (57)

Tonga 2006 Model life tables using single input parameter (CEBCS) (136)

Vanuatu 2009 Model life tables using single input parameter (CEBCS)

� � �

45

2.1.2 Periodic Surveys

There are several standard survey formats commonly used to collect data as inputs

for mortality estimates. Unlike the census that collects information on the whole

population, surveys collect data for a proportion of the population considered to be

representative of the broader population of interest. As with censuses, surveys may

collect data to undertake direct estimation of mortality from reported deaths, or

indirect estimation based on the age distribution methods described previously.

Repeated household surveys may also be used to estimate mortality based on

survival and migration.

Surveys fill an important role in the provision of mortality estimates in the absence of

routine data collection. The reliability of estimates from survey data is driven by how

well the sample selection reflects the broader population of interest (66). Surveys are

also subject to recall bias as they collect information on events that may have

happened several years prior (153, 154), and response bias (where people that

complete the survey are different in some way to those who opt out or cannot be

contacted). As with censuses, further response bias may be introduced by cultural

issues around discussing deaths. The use of surveys to collect data for monitoring

health status and evaluating health programs can also prove costly (38).

Historically, surveys in the PICTs have generally been more concerned with infant

and child mortality, or deaths from specific causes such as maternal deaths than all-

age all-cause mortality (155, 156). Two important survey types for mortality

estimation in the PICTs are Demographic and Health Surveys (DHS) (38, 43, 133)

and the UNICEF Multi-Indicator Cluster surveys (MICS) (157).

� � �

46

Table 10: Key surveys in the Pacific Islands; 1995-2012

DHS surveys have been implemented across the region since the early 2000’s

(Table 10). These are retrospective cohort studies that use a birth history (all births

with dates and outcomes) from respondents to collect inputs for estimation of infant

and child mortality. Estimates often become less plausible for earlier periods (5-10

and 10-15 years prior to the survey) (158, 159) due to recall bias (153, 155). To date,

the only MICS in the PICTs was conducted in Vanuatu in 2007 (157), although more

are planned for the region over the next several years. These collect data for

estimation of infant mortality using the CEBCS techniques as employed in the

censuses. Although the Vanuatu MICS achieved a response rate of 89%, the

response rate from 15-19 year old women was particularly poor, and one of the most

noted reasons for non-participation of households as a whole was the death of a

family member (157). These issues raise concerns that childhood mortality figures,

potentially more so than other indicators derived from this survey, may be biased

and therefore not nationally representative. The smaller sampling frame employed in

� � �

47

the MICS (157) compared with the universal data collection employed in the census

obviously also generates a greater range of uncertainty around the subsequent

estimates of infant and child mortality generated from the CEBCS method.

Both surveys can be used to derive life tables from the estimates of infant and child

mortality using model life tables as for censuses; however this is routinely done only

for DHS’s.

2.1.3 Routine vital registration collections

An efficient routine CRVS system, with medical certification of CoD, provides

ongoing and relatively low cost data collection and therefore timely mortality data for

decision making (34, 35, 117, 160). The medical certificate is considered the

reference standard for CoD information, in the absence of an autopsy, as a qualified

practitioner is required to assess the case and make an informed decision

concerning the sequence of events that led to the death (68, 117). This is particularly

attractive in the PICTs where small population sizes make complete routine

registration feasible, despite the challenges of dispersed populations and limited

infrastructure.

Routine data collections for deaths are usually managed through civil registration or

health system reporting systems, but may also rely on lay reporting by a local chief

or leader (161). Civil registration is managed by the Ministry of Justice (or local

equivalent) with local functions often delegated to local government (such as Town

or Island Councils). The process comprises of two major components, recording the

death in the official records and therefore establishing the civil status of the

� � �

48

deceased person, and the issuance of a legal document, the death certificate,

certifying the death for legal purposes such as insurance and land inheritance. A

death certificate may also be required prior to burial (117).

Official registration of death requires the family to provide proof of the death (often in

the form of a medical certificate completed by a qualified doctor) (161), attend the

local civil registration office to complete the required paperwork, and pay a

nominated fee. Although the civil registration process may record CoD, this is not the

primary purpose of the system (other than to rule out unnatural causes that may

require criminal investigation (7)), and may vary in usefulness according to the

source of the information. Systems that require a medical certificate are likely to be

more reliable than systems that record CoD as reported by family members. As the

legal record, civil registration of vital events is frequently legislated as the official

source for mortality data (34) despite not always being the most reliable data source

in country.

While civil registration is frequently upheld as the “gold standard” (7, 33-35, 127,

160, 162, 163) of mortality data due to the relatively low cost of data collection, the

ongoing nature of collection (and therefore timeliness) and the ability to combine this

function with medical certification to obtain CoD (117); only seven of the 23 countries

of the WHO Western Pacific Region were assessed by Mathers et. al. to have

complete data collection systems in 2003 (4). The 2004 assessment of mortality data

in the region by Taylor et. al. (2) also found mortality from civil registration was

implausible and demonstrated significant under-recording. As Byass succinctly

states “the chances of a death being registered and documented as to cause

depends strongly on the socioeconomic status of the community and the nation in

� � �

49

which it occurs” (11). In many developing countries, a substantial proportion of

deaths do not appear in the final tabulations, either because they are simply not

registered (163), or because of losses in the administrative chain from a local

registrar to the central processing unit (164). Countries with low civil registration are

also subject to substantial sudden change in completeness, often due to subtle

process changes (117).

Vital events such as births and deaths are also often recorded through routine data

collections within the health system. “Complete and reliable mortality systems are

the cornerstone of national health information systems” (34). In comparison to the

legal focus of the civil registration system, health data collections are primarily to

inform operational decisions, and CoD is central to this purpose. Health systems for

reporting deaths may include a vital registration system (a record of all deaths in

both the health facilities and the community) based on medical certificates, hospital

separation data (hospital records that indicate whether the patient was discharged

home from hospital, transferred to another facility or died), and/ or community

nursing data.

The medical certificate requires doctors to record “to the best of their ability” the

causal sequence that lead to the death (starting from the immediate cause and

working back through the chain of events) (165-167). There is also space to record

any other diseases that contributed to the death but were not strictly causal. In

practice, certification has been found to be affected by many factors including the

specialty of the physician, their experience and beliefs, as well as their education

and training on the subject (168-172).

� � �

50

The medical certificate as recommended by WHO is divided into two parts (as shown

in Figure 2). Part I has three lines and requires the physician to list the CoD starting

with the immediate cause followed by the preceding cause and finally the underlying

CoD. Part II of the certificate provides space for other conditions that did not directly

cause the death to be recorded. Coding practice varies internationally as to whether

the immediate cause, underlying cause or all causes are coded for analysis (161).

Many have argued that even with doctors’ certification, accurate CoD reporting

requires routine autopsy to verify CoD patterns. Various studies show the rate of

error when medical certificates are compared to autopsy data (173, 174). These

include hospital based studies in the US that found 48% of cases reviewed missed

“myocardial infarction” as a CoD, while 25% of the deaths reviewed were falsely

attributed to this cause (175). Very few PICTs have routine autopsy procedures, with

most either performing no autopsies at all, or flying a pathologist in only for high

profile medico-legal cases. There have been suggestions that while these errors are

significant at the individual cause level, they are less significant when grouped to

broader categories. Studies of autopsied cases in the 1930’s in the US showed that

correlation between the medical certificate and the autopsy finding rose from 79% to

90% when assigned to broad category of disease (ICD chapter rather than specific

causes) and that careful recording of deaths shows mistakes go a long way to

cancelling each other out at the population level (10). This is supported by a more

recent study of asthma in the UK (176).

Numerous studies, including Iran (177), India (178) and China (63, 179, 180), found

major discrepancies may occur between the medical certificate and information from

the medical record, highlighting the importance of training and periodic verification of

� � �

51

certification practices. Despite these concerns, in the absence of an autopsy,

certification by a medical practitioner is recognised as the most reliable source of

CoD data.

In the hospital setting, unit record data on deaths and CoD may be collected through

hospital separation data. In these collections CoD is based on the principal diagnosis

at the time of death (and hence discharge from hospital) as recorded by the

physician on the front sheet of the medical record (117). While these data are based

on cause as nominated by a trained physician, the reason for treatment is not always

the underlying cause that led to death (117) and therefore there will be some small

degree of error in using these data in place of a medical certificate.

Countries may also collect data on deaths through primary or community health care

nursing programs, with deaths recorded on a separate form to be sent to the local

area nursing manager and/or recorded on monthly reports (161, 164). Details

collected about each death vary between country and system, but usually include

name, place of death (or report), age at death, and may include CoD as reported by

the local health worker nurse or family members (181). These systems may focus on

specific categories of deaths such as maternal or child deaths. There has been

limited analysis (at least in the published literature) of CoD based on this type of data

collection (182), and the usefulness of these data may be limited, or restricted to

very broad categories of cause. Similarly lay reporting by community leaders, such

as utilised in Tonga (164), can provide accurate collection of mortality level but is

limited in scope for collecting anything more than broad CoD. CoD is collected

through these systems essentially to rule out whether a death is suspicious and

requires investigation (14).

� � �

52

2.1.4 Demographic surveillance sites

In the absence of universal civil registration, countries such as India and China have

demonstrated success with sample demographic surveillance sites (DSSs) (63, 117,

180, 183-186). These use smaller populations to provide a representative overview

of both mortality level and CoD distribution from medical certification. Jha notes

“sample registration systems are probably the most useful method to obtain CoD

information” in the absence of medical attention at death (187), and the Chinese

system is an excellent randomised cluster selection of urban and rural areas

accounting for around 1% of the total population and providing robust information for

decision makers (179, 180). This technique has been used in PNG (188) but it is

unlikely to be a suitable solution in other PICTs where the small population numbers

mean rates are often quite unstable and already need to be averaged over several

years to be interpretable (189)). A further disadvantage of DSSs is that focussing

enumeration in these areas over a number of years could result in a shift in

community knowledge and attitude to health interventions and reporting that may not

be reflected in the broader population. These sites may therefore become less

representative of the broader population over time (190).

2.1.5 Verbal Autopsy

The shortcoming of using CoD information from medical certificates is that in many

of the PICTs, a large proportion of deaths occur outside the hospital system without

a medical attendant (72). This means either CoD is not provided, or is based on the

determination of someone without a health background (such as the police or local

registrar). The verbal autopsy (VA) arose out of work in “population laboratories” in

India and Bangladesh in the 50’s to 70’s to address this issue. It is now recognised

internationally as a viable alternative for collection of accurate CoD statistics, at a

� � �

53

population level, where universal certification is not practiced (181, 191-193). They

are usually associated with either demographic surveillance sites or surveys as they

are both labour and resource intensive (14).

The term “verbal autopsy” was first proposed by Keilman and associates working in

the population laboratory of Narangwal India as a “procedure to exploit the

information provided by the relatives of a deceased person” (194) in order to

establish a medically acceptable CoD, and was further developed by the MATLAB

site in Bangladesh (195). The technique has since been substantially refined to

reconcile differing approaches (190). A standardised questionnaire on symptoms

displayed by the deceased prior to death is completed by a trained interviewer who

visits family members, with CoD established by analysis of the questionnaire (either

by a physician, or more recently by automated analysis) (194).

Studies such as carried out in Tanzania by Setel et. al. in 2006 (196) found VA

effective for determining CoD from a list of locally important diseases (such as HIV,

malaria, TB and cardiovascular diseases among others) for both adults and children,

as verified against medical records. Limitations of VA are the focus on specific

common causes of death by broad age group (rather than all causes), and reliance

on well trained interviewers (often not medically trained) (195). VA has been utilised

in PNG using a questionnaire that pre-dates the current international standard (197).

There is no literature suggesting the technique has been applied elsewhere in the

region, however it may have application for the larger populations in the PICTs such

as the Solomon Islands and Vanuatu, where universal certification is currently not

feasible.

� � �

54

In 2012, WHO also released a short version of the standard VA tool designed for use

in conjunction with routine reporting. This however is yet to be tested in the field

(198) . Standardised approaches with trained interviewers are a significant

improvement on earlier systems of lay reporting of CoD through community leaders

which had limited quality control around recording or standardisation of collection.

2.1.6 Summary of data collection approaches

As outlined in the preceding sections, and summarized in Table 11, while all of the

data collection approaches have strengths and weaknesses, routine vital

registration, done well, should be the preferred approach for mortality data in the

PICTs due to the continuous nature of collection (7, 117), ability to generate data on

both mortality level and CoD distributions (7, 117), and the small populations sizes

which result in a great deal of uncertainty when dealing with rare events such as

deaths, especially when disaggregated by age and sex as required to generate

meaning data for policy. Although censuses remain the primary data collection in the

region, they are not able to capture CoD, and in most instances rely on extrapolation

of adult mortality and LE from childhood mortality based on model life tables, rather

than measuring actual events. DHS and censuses will continue to be an important

source of data in the absence of reliable routine CRVS systems, and for generating

data to assess their performance.

Direct methods of estimating mortality from routine reporting have the advantage of

timeliness, can better recognise differentials, and are better suited for providing

estimates for infant and child mortality (127); although indirect methods are less

expensive (199).

��

�

55

Tabl

e 11

: Sum

mar

y of

Dat

a C

olle

ctio

n A

ppro

ache

s fo

r Mor

talit

y D

ata

Dat

a So

urce

Pe

riodi

city

Sam

ple

fram

e Pe

riod

of

inte

rest

Dat

a co

llect

ion

(mor

talit

y le

vel

data

)C

oD D

ata

Col

lect

ed?

Cen

sus

Per

iodi

c –

5-10

yea

rs

Who

le P

opul

atio

n R

etro

spec

tive

Dire

ct -

(dea

ths

in th

e ho

useh

old)

In

dire

ct –

par

tial b

irth

hist

ory

(CE

B/C

S) &

orp

hanh

ood

data

No

Sur

vey

DH

S

Per

iodi

c –

~ 5

year

s S

elec

ted

sam

ple

– re

pres

enta

tive

of w

hole

po

pula

tion

Ret

rosp

ectiv

eD

irect

– c

ompl

ete

birth

his

tory

N

o

MIC

S

Per

iodi

c –

~ 5-

10

year

s

2 st

age

clus

tere

d sa

mpl

e –

repr

esen

tativ

e of

who

le

popu

latio

n

Ret

rosp

ectiv

eIn

dire

ct -

par

tial b

irth

hist

ory

(CE

B/C

S)

No

Oth

er h

ouse

hold

ba

sed

surv

eys

Usu

ally

on

ce-o

ff V

arie

s R

etro

spec

tive

Var

ies

Pos

sibl

e -

usin

g ve

rbal

aut

opsy

R

outin

e vi

tal

regi

stra

tion

Civ

il R

egis

tratio

n C

ontin

uous

Who

le p

opul

atio

n (d

epen

ding

on

cove

rage

) C

urre

nt

Dire

ct re

porti

ng o

f eve

nt

Yes

Hea

lth v

ital

regi

stra

tion

Con

tinuo

usW

hole

pop

ulat

ion

(dep

endi

ng o

n co

vera

ge)

Cur

rent

D

irect

repo

rting

of e

vent

Y

es

Hos

pita

l di

scha

rge

reco

rds

Con

tinuo

usH

ospi

tal c

ases

onl

y C

urre

nt

Dire

ct re

porti

ng o

f eve

nt

Yes

Oth

er

rout

ine

data

base

s

Var

ious

C

ontin

uous

Var

ies

– us

ually

targ

ets

sub-

popu

latio

n of

spe

cific

in

tere

st.

Cur

rent

D

irect

repo

rting

of e

vent

U

sual

ly li

mite

d

Dem

ogra

phic

sur

veill

ance

S

ites

Con

tinuo

usS

elec

ted

area

s –

usua

lly n

ot

repr

esen

tativ

e of

who

le

popu

latio

n ov

er ti

me.

Cur

rent

D

irect

repo

rting

of e

vent

Y

es

� � �

56

As noted in this chapter, CoD can be obtained through medical certification, hospital

discharge records, verbal autopsy, nursing reports or lay reports from family and

other community representatives. As doctors are trained in diagnosis of disease,

sources that collect data based on direct observation by the doctor, provided the

doctor is aware of the patients history, are obviously likely to be more accurate than

reporting by nurses or family members who do not have the same level of training.

While both medical certificates and diagnosis for hospital discharge records are

completed by doctors, the medical certificate is specifically structured to capture

underlying CoD and there is a clear set of coding rules in order to identify this (ICD

10). In comparison the hospital discharge record captures the “primary diagnosis”

which may not necessarily reflect the cause that initiated the sequence of conditions

that ultimately resulted in the death occurring.

Figure 6: Level of certainty of underlying cause of death by data source

Highest certainty

Although medical certification is therefore accepted as the preferred source of CoD

data, the quality of this information varies significantly, with numerous studies

� � �

57

identifying discrepancies between the medical certificate and information from

medical records. This is discussed further in section 2.4.

2.2 Evaluating routine data collection and reporting systems

Although routine vital registration through either health or civil registry offices has

been identified as the preferred data source for vital statistics and CoD data in the

Pacific, both the design of the data collection processes and system performance

can have a profound effect on the quality of the data collected, and subsequently on

the validity of the estimates generated. It is therefore necessary to assess the

strengths and weaknesses of routine data collection and reporting systems to

understand how these may bias the data available.

The most comprehensive guide to the elements and processes necessary for

reliable and valid mortality level and CoD statistics was published by the UN in 1987

(and updated in 2001) (117). Simultaneously, the UN completed a survey of the

status of vital registration in member states. Although this forms the foundation for all

future work in system assessment, little action was taken at the time, and there was

little connection between this work and the subsequent information technology and

communication (IT) approaches.

Early approaches to assess HIS and CRVS systems developed from the burgeoning

field of IT applications for managing clinical health information. It was recognised that

an initial assessment of processes was needed before existing systems could be

replaced or adapted to computer-based applications. Craig et. al. (200) describe

assessment as a process that “looks at information in terms of needs and availability

� � �

58

and its gathering, recording, transmission, processing, reception and use. These

questions are primarily examined with respect to the questions who, what, where,

when, why, how and how much” (200). He also describes the need to obtain all

documentation relating to the system. This approach recognised interviews as critical

to understanding the “informal processes” at work within the system, the way things

actually work as opposed to the way they are designed to work.

Interviews and document review remain at the heart of all future approaches to

record how data moves through the system (179, 186, 201). While further tools were

developed to “quantify” some of the aspects of the system operation, the qualitative

information obtained through this approach is critical to developing a rich

understanding of how and why the processes work the way they do. However, there

remained a need to organise key findings of the system assessment into a set of

coherent themes and ideas that could be used to focus further work. As later stated

by HMN “What has been missing in these efforts has been an overall vision of a

comprehensive health information system and the interlocking of the various parts”

(7).

A number of different frameworks were developed and trialled during the 1990’s and

2000’s to address this gap, most focussing on HISs more broadly rather than vital

registration or CoD specifically. The HMN was established in 2005 under WHO to

assist countries to improve the use of data in health planning by supporting

improvements in HISs (202). They describe the HIS in terms of demand (who needs

the data and what for), supply (the tools and methods available to obtain this data)

and level (the level of the system at which data are generated and used) (8). This

categorisation would later be developed into a comprehensive assessment tool (the

� � �

59

HMN assessment) designed to assist countries to conduct their own HIS review. The

HMN assessment (203) covers key areas of:

• Content (Cost efficiency and effectiveness);

• Capacity to collect data, manage and analyse the results;

• Application of agreed standards;

• Dissemination and analysis; and

• Linkages to other data sources.

Whilst mortality data are collected though multiple different systems and

organisations, the HMN assessment is primarily focussed on health system

reporting. It does however make a specific point of emphasising the linkage to other

sources. Macfarlane (204) argues a key measure of success of a HIS is the

integration of information systems from other sectors. While fairly non specific about

what this should look like, she suggests possible means to achieve integration

including conformity of sources, definitions and reporting periods; rationalisations

and sharing of software, boundaries and maps; agreement on minimum health

related requirements for household surveys; and coordination with non-governmental

information systems.

Applied specifically to mortality data in Thailand, a workshop of key participants

reported the HMN tool provided a useful assessment framework (203). Interestingly,

this assessment found mortality data in Thailand to be of moderate or good quality,

whereas an assessment of the same data conducted shortly before found it to be of

poor quality (9). The HMN assessment relied on measures of completeness from a

national household survey (conducted 10 yearly) which asks for the registration

document for all deaths noted from the household. This is likely to be highly biased

� � �

60

due to recall and the fact unregistered deaths may be less likely to be reported in the

survey, especially if participants know registration documents will be requested. The

95% completeness noted in the national survey seems unlikely, especially as the

demographic surveillance site noted in the same article suggests 12.5% under-

reporting in a site that has been closely monitored for over 10 years (203).

The key strength of the HMN assessment is that it generates consensus for areas to

focus on improvements, however the questions include predominantly subjective

criteria that would be difficult to apply uniformly given it is a self assessment. In order

to minimise possible bias the approach relies heavily on the formation of a

committee to conduct the assessment. The assessment tool has been applied in

both Fiji and Tonga (205, 206). Although both countries identified a need to improve

CoD data collection and reporting, given their broad focus of the assessment, there

is little discussion regarding the specific barriers or review of options for

improvement.

Concurrently with an increased focus on HIS, the 1990’s and 2000’s saw an increase

focus on assessing the IT used for health information. In particular, why so many HIS

had failed or were performing poorly. Most evaluations focussed on why a particular

software was chosen, or presented only a very general picture of costs and benefits,

driven primarily on the economic aspects of the system (207). These were

conducted by panel discussions, structured questionnaires or “expert” evaluation. A

literature review in 2007 found none addressed broader implications of IT on the

CRVS system as a whole (207). One assessment however advocated the use of a

framework to assess the “design-reality” gap (208). This framework considered

elements of information, technology, processes, objectives and values, staffing and

� � �

61

skills, and other resources (i.e. time and cost). Although designed as a risk-

assessment tool to consider whether systems were likely to fail, rather than

specifically looking at the impact of the system on the data collected, this approach

highlights the effect of differences between system design and the “informal

processes" that develop within the system.

Following the first questionnaire from the UN in 1987 (161), it was not until the mid

2000’s that researchers and institutions began to develop more targeted approaches

to assessing routine reporting systems for deaths and CoD in their own right as

rather than simply as part of the broader HIS. Early approaches by Ruzicka and

Lopez (209), guided in large by the original UN civil registration guidelines (117),

were developed by Rao et. al. into a comprehensive framework to organise and

present key system aspects for collecting and reporting mortality and CoD data.

Applied to Africa (64) this framework divides the system into the following

components: administrative (legal framework, organisational issues, system design);

technical issues (data standards, capacity building, quality control aspects); and

societal issues (political will, public awareness).

Rao et. al.’s framework (64) assumes the person conducting the assessment has an

in-depth knowledge of mortality data collection and analysis questions to tease these

broad guidelines into a comprehensive assessment. The major advantage of this

approach is the flexibility of the framework to accommodate the different reporting

structures that may be encountered. The themes are also substantially consistent

with the major components considered essential in an assessment of public health

surveillance systems as outlined by the US Centres for Disease Control (210):

system considerations (purpose, stakeholders, and operation), outbreak detection

� � �

62

(timeliness, validity) and system experience (usefulness, flexibility, acceptability,

portability, stability – resilience to changes, costs). While there are obvious

differences in data collections for mortality data and outbreak investigation (primarily

in terms of timeliness and strength of evidence required), both data sets are primarily

to inform decision making in health. This common purpose is reflected in the

similarity of the key issues considered important in system assessment.

Over the same period, the earlier work of Ruzicka and Lopez (209) was developed

further in parallel by Prof. Lopez and others at the University of Queensland; initially

to develop a set of questions that could be used by researchers to guide in-country

assessments (as carried out in Oman (211)), and subsequently into a series of tools

published in conjunction with WHO to guide countries through a self-assessment

process. These include a rapid assessment of 25 questions (212), and a broader

comprehensive assessment (213). The WHO assessment framework is based on

five key areas: legal basis and resources for civil registration; registration practices,

coverage and completeness; certification and CoD; ICD mortality coding practices;

data access, use and quality checks (212, 214). The prescriptive questions make the

assessment more appropriate for self-assessment, but lose a degree of flexibility that

Rao et. al.’s more thematic framework is able to accommodate when assessing

approaches and systems that do not fit the expected norms upon which the

questions in the WHO tool were based. Both have an important place in assessing

routine CRVS systems, and despite their differences share many similarities given

their common origins. While both approaches are focussed on identifying key areas

for improvement from a system perspective in order to improve future data

collections, only Rao et. al.’s framework attempts to relate these system

� � �

63

characteristics to the effect on the data collected and subsequently how data can be

used.

One of the differences in approach is who should undertake the assessment. While

self-assessment may result in stronger ownership of the results, and therefore

possibly a greater likelihood to invest in recommended improvements, it requires a

high level of understanding of both the current CRVS system operations and best

practice CRVS system theory to move beyond very broad findings useful at an

advocacy level to specific actionable steps. Given the small specialist workforce

available in many PICTs, and the long and ongoing calls for improvements, it is felt

an external assessment process would be more appropriate in this setting, although

this could subsequent feed into a country led process.

Although implied under the various approaches, the need to document the flow of

information through a mapping exercise or flow chart describing what information is

collected and how it moves through the system, is not explicitly described. This is a

critical step in obtaining this rich qualitative data, as it allows the assessor to identify

breaks in process where data may be lost, delayed or changed.

Another concept gaining importance over recent years is the link between HIS and

health governance. This is broadly referenced to some degree under “administrative”

aspects in Rao et. al.’s framework, but is not covered beyond legislation in the WHO

tools. None of the approaches outlined previously focus specifically on the questions

of Who owns the data, Who is responsible for the data, For what purposes will the

data be used, If the data will (or may) be used, and what controls will be put in place

to govern that use (215). Siddiqi (216) applied the following questions in an

� � �

64

assessment of health governance in the Eastern Mediterranean under the heading

“Intelligence and Information”:

• How efficient and up to date are communication processes (forms etc)?

• What information is available about the health system and health in the

country and how accessible is it?

• What is the reliability of information available for development of policies?

• What evidence is there for the use of information in the decision making

process?

• How is the relevant information about health generated?

• How is implementation of health policies monitored?

As noted by McGrail (215), the experience of developed countries demonstrates

failure to address these questions early in CRVS system development can result in

situations where data are not available or can only be accessed by a few specific

people, even when the system as a whole is working well, thus negating much of the

value of the collected data. The integration of some components of the frameworks

that have been developed for assessing health governance, such as the WHO

“Domains for Stewardship” (217) , into the assessment of CRVS systems may begin

to address this gap. Existing approaches also did not adequately highlight the

important relationship-driven personal elements of the systems which are critical to

the success or failure of CRVS systems in this region.

An amalgam of existing approaches, rather than one specific tool would therefore be

considered the best approach to adequately assess system issues that affect data

collection, capture, analysis and reporting in the PICTs in order to make plausible

� � �

65

judgements about the validity and reliability of published data, and to understand the

bias inherent in the raw data.

The approaches presented here all address system issues rather than providing a

structure for assessing data independently to the CRVS system through which it was

gathered. This has been covered thoroughly by a further data quality assessment

framework developed by Rao et. al. (218) which considers nine criteria under four

key data attributes. These are: generalizability (coverage and completeness); validity

(use of non-specific codes, content validity, incorrect or improbable age and sex

patterns); reliability (general mortality level and consistency of cause-specific

proportions over time); and policy relevance (timeliness and geographical

disaggregation) (179, 218). This has been applied successfully in both China (179)

and Brazil (218). Alone, this approach may have limited application in the PICTs, as

most of the data are expected to be of “poor” quality and may be considered of

limited use without a further understanding of how the data were obtained. However,

it has enormous potential to add to the value of a full system assessment by adding

a quantitative understanding of exactly how system issues are affecting estimates to

add to the qualitative understanding of the CRVS system review. The components of

this data assessment also provide the link from the system understanding to making

appropriate “corrections” to the data as demonstrated by Rao et. al. in Brazil (218).

2.3 Evaluating completeness for mortality level

Means of measuring mortality level fall into two broad categories; direct and indirect

methods. Direct methods record every death, either as it occurs or retrospectively,

such as through routine CRVS systems, demographic surveillance sites or surveys

� � �

66

and census that record each death in detail. These can then be assessed for

completeness and adjusted accordingly. Indirect methods collect information from

relatives regarding which family members have died and use age distributions from

the remainder of the population to distribute these deaths over time and age group,

subsequently generating age-specific mortality rates as described in Chapter one. As

noted in an earlier section of this chapter, direct reporting of deaths through vital or

civil registration processes is the preferred source of data for measuring mortality

and causes of death when data have been obtained from a well functioning CRVS

system (where biases in the data collection can be assessed) and the data have

been assessed and corrected if necessary for completeness. Preston notes the fact

empirical data are often “incomplete” tends to be used to justify ignoring findings

based on these data (219, 220). He goes on to note that for many data sets, if the

data can be assessed in terms of its completeness it is nearly as useful as a

complete data set. He sets the threshold for a data set being useful at around 60%

completeness (221), a limit also recognised by the Australian Bureau of Statistics.

2.3.1 Direct Demographic Methods

Methods to assess the completeness (or consistency between the death and

population data) of the reported deaths (through vital registration, census or survey)

include those developed by Brass, Preston, Bennet and Horiuchi (131, 199, 222-

225). The Brass “Growth Balance Method” assumes population stability, so that

growth rate is constant with age. It is based on the true growth of a population aged

x and over is equal to partial birth rate (people who reach age x in a given year

divided by the person years lived above age x in that year) minus the partial death

rate (deaths in a year in persons aged x at last birthday and above divided by person

years lived above age x in that year). Adapting this equation so the recorded number

� � �

67

of deaths equals the completeness (C) of the data set multiplied by the true number

of deaths provides a linear equation where the Partial Birth rate = growth ( r) + 1/C *

(partial death rate). The slope of the line formed by this equation is 1/C and can be

used to both assess the data and correct for under-reporting. This method is very

sensitive to error in the data which changes the slope (and therefore the estimation

of completeness of the data source) and to changes in the assumption of stability

(225, 226). The selection of age groups to “Crop” in order to fit the line may also

have a significant impact on the estimate of completeness (124). Murray et.al.

examined optimal age ranges against developed country data in 2009 (124). While

they found the optimum cut as 40-70, the results from the age trims were found to be

highly susceptible to systematic age misreporting, and while not specifically noted in

the paper, this would also imply susceptibility to differentials in reporting

completeness by age, a scenario likely in the PICTs. As there were several age

trims found to perform well, the results indicate a need to consider optimal age trims

to fit the line in the context of local data. (227)

Modifications include the Preston-Coale method, which uses growth rates instead of

the age distribution of deaths used by Brass, and the Bennet and Horiuchi method

which discards the assumption of stability (225). These methods require inter-censal

data which is often not available (227).

An overview of the assumptions and data required for each method is shown in

Table 12 below.

� � �

68

Table 12: Assumptions and data requirements for direct demographic analysis

Method Data Required Assumption Time period Brass Growth Balance

Age-distribution of population

Stable population

Can be matched to available data

Preston-Coale Age-distribution of population and population growth rate

Stable population

Can be matched to available data

Generalised Growth Balance

Age-distribution of population from two censuses

Closed population

Inter-censal

Bennett-Hounichi Age-distribution of population from two censuses

Closed population

Inter-censal

A second approach for assessing completeness is direct comparison with another

source or “capture-recapture” techniques (127). These methods, adapted originally

from population biology, compare the overlap between the different sources to

estimate the number of deaths missed by all sources (228-230) (see Figure 7). Two

source, or dual record methods such as described by Sekar and Deming (231), also

known as the “Peterson method” (232), assume independence between the two

sources (228). That is, reporting the death to one source should not affect the

probability that the death will be reported to the other source, a scenario unlikely to

hold true in most settings. Log-linear methods that use three or more sources

address this issue by producing multiple models with different combinations of

dependence between the data sources. The most appropriate model is selected

based on Bayesian selection criteria such as the Bayesian Information Criterion or

Aikaike Information Criterion, and on a qualitative understanding of the relationship

between the data sources (229, 230). Although the three source model is more

robust (233), it is uncommon (especially in the PICTs) to have access to multiple

data sets.

� � �

69

Figure 7: The two source capture-recapture model

(229, 230).

Other assumptions of the capture-recapture method include a closed population and

equal “catchability” of deaths. That is, every death has an equal probability of being

reported. While this is unlikely across an entire population given the incentives for

reporting a death for an adult (who has property etc) are likely to be greater than for

a child, the advantage of the capture-recapture method is that analysis can be done

by sub-group (such as by age group) where these assumptions are more likely to

hold.

2.3.2 Model Life Tables

Model life tables may be used to impute age-specific mortality and LE where

empirical data is unavailable or incomplete using information from childhood

mortality (from CEBCS) alone, or with adult mortality (from methods such as sibling

survival). This involves fitting the known (or estimated) input parameter(s) to a model

life table such as the Coale and Deming life tables (234) derived primarily from

Europe and the Far East, or the UN life tables which use a slightly wider group of

countries (including some developing country life tables from Africa and eastern

Europe) (13, 66, 117). Although the UN life tables, and the newer tables designed by

� � �

70

Murray et.al. (235) are an improvement on earlier tables as they include a wider

variety of sources including both developed and developing country data, in the main

these tables reflect country experiences where a low U5M equals a low adult

mortality rate and subsequent high LE. This association between U5M and adult

mortality levels was observed repeatedly across Europe, parts of Asia and Africa

(where data were available and subsequently used as inputs to the various model

life table systems (234) and reflects the earlier varieties of the epidemiological

transition where decreases in mortality from infectious diseases, primarily due to

improved nutrition and sanitation, resulted in marked improvements in longevity and

LE (234). Reductions in mortality from infectious diseases over the years (as

described in Chapter one) have largely (or in some cases completely (112)) been

offset by increasing mortality from NCDs, resulting in high adult mortality from NCDs

with lower IMR as infectious diseases have diminished in importance (2).

Subsequently, single input parameter models such as used by census estimates (61,

135) are likely to underestimate adult mortality and overestimate LE, although the

choice of model family (reflecting the broad geographical areas from which the

source data for the model was drawn) and subsequent ratio U5M to adult mortality

will affect to what extent. Modelled estimates based on single parameter input

models may be more suitable for the few PICTs which still report relatively high

IMRs, such as PNG, Kiribati, and the Solomon Islands, and have a high prevalence

of infectious diseases (108). However even Kiribati and the Solomon Islands are

beginning to note a dual burden of disease as NCDs such as diabetes and cancer

become more important (as noted through the reported leading causes of

hospitalisation (108) which would be expected to adversely affect adult mortality.

� � �

71

Model systems such as the modified Brass logit system developed by Murray et al

(235) and developed into the WHO Modmatch software, may have greater

application for the region as they allow two input parameters (usually 5q0 and 45q15,

both of which can be derived from indirect demographic methods such as in a

census) forcing the selection of models that more accurately reflect the adult

mortality pattern.

Model life tables are generally calculated using input parameters from indirect

measures of mortality obtained through a census or survey as described in part one

of this chapter (127). In this case, final age-specific estimates of mortality are

affected by the uncertainty of both the model selection, and the distribution of events

over time based on age patterns inherent in the indirect estimates (127). Models may

also be used to “smooth out” age-specific mortality estimates that are unstable, or

show somewhat unexpected patterns due to stochastic variation from small

population sizes such as used in Niue and Nauru censuses (137). Given the small

population sizes in the PICTs, this application of model life tables may have

application in generating a valid, reliable age and sex specific mortality profile for

countries even where data collections are relatively complete.

Another method used to provide mortality estimates in the absence of recent data is

to use the last available empirical estimates and extrapolate to a current mortality

level. This is inherently flawed and as noted by Murray (13) when the change in LE is

examined there is “little left except the UN assumed rate of improvement”.

Unfortunately this approach is used to provide estimates many years distant to the

timeframe for which valid data are available, and does not account for the known

effects on LE if economic and social situations do not improve, if there is a HIV

� � �

72

epidemic that takes hold, or if NCDs in adults cause premature deaths. Murray notes

in late 1980’s annual estimates by the WHO and UN for many PICTs were still based

on data from the early 60’s. (13)

A step further than the model life table systems are the more advanced models

developed by researchers such as Lopez, Rajatnaram and Murray to estimate

mortality trends over time for infant and child mortality(236, 237) , adult mortality

(41) and specific cause mortality ( such as maternal mortality) (238), in large

developed for the 2005 GBD study (239). These multivariate models use a wider

range of input data including economic measures such as GDP, patterns of mortality

in neighbouring countries with known data, as well as available country mortality

data fitted using Gaussian regression to estimate mortality trends over time. Such

models have an important role in filling information gaps and providing an indicative

assessment of the long term trends (upwards of 20 years) , however the wide

uncertainty shown in such estimates and the complexity of the approach limit the

usefulness of these methods at country level. They are therefore best suited to

academic assessments at a broad regional level, or to extrapolate trends in countries

where multiple reliable data sources are available at various points over a long

period of time. The validity of these models has been demonstrated successfully

multiple times at a broad regional level (240). However, as the models rely on

neighbouring country data to successfully provide context, their validity in the Pacific

may be limited given the diversity in the PICTs and documented lack of data upon

which to draw. Resultant estimates need to be further evaluated against empirical

estimates for the region where possible.

� � �

73

2.3.3 Conclusions

In spite of the shortcomings outlined, model life tables based on single input

parameters generated though census collections (Table 9), remain the staple

method of generating age-specific mortality measures for many PICTs including

Tonga, Vanuatu, Kiribati and the Solomon Islands, highlighting a need to further

explore the potential for using direct demographic methods and capture-recapture

analysis to assess and correct mortality data from CRVS systems in the region

wherever there is sufficient data to do so. While the Brass-Growth Balance method

(241) has limitations in relation to its ability to account for changes in population

structure, the simplicity of approach and the subsequent greater availability of the

necessary input data mean it could prove a useful first “screening” tool in populations

for which migration is low (migration is discussed further in Chapter four) and the

system assessment supports an assumption of a reasonably high level of

completeness of reported deaths. Neither the Brass Growth Balance nor capture-

recapture methods are suitable for application in populations with high migration, and

in the absence of more suitable methods, modelled estimates will remain important

where CRVS systems generate incomplete data or the completeness of reporting

cannot be adequately anticipated through a review of the system design and

performance. Where model life tables do remain necessary, the use of two input

sources (both child and adult mortality) should be routine to avoid misleading

estimates.

The importance of assessing the plausibility of mortality estimates regardless of the

method of data collection and analysis, such as through consideration of the design

and performance of the CRVS system as noted above, has been highlighted by a

range of papers. These include the UN (117) and Sibbai (242) who noted a range of

� � �

74

considerations such as access and timeliness which should be examined. In 2005,

Rao et. al. (179) developed this concept further generating a rating framework for

quality of CoD data for application in China. This was based on generalizability

(coverage and completeness), reliability (consistency of cause data with mortality

pattern by age and consistency of cause specific rate over time), validity (content

validity, use of ill-defined categories, incorrect/implausible age or sex dependency),

and policy relevance (timeliness and geographical disaggregation). This framework,

which extends to data on cause as well as mortality level, is useful as the measures

can be applied to the data themselves without a lot of additional information about

the system. The examination of consistency with long term trends will be particularly

important in the PICTs where small population sizes are likely to result in substantial

stochastic variation in mortality estimates.

2.4 Evaluating biases in cause of death data

The quality of CoD data as collated from medical certificates is dependent upon not

just the practice of certification, but how this information is then coded according to

the ICD rules and tabulated.

2.4.1 Review of Certification

As noted by Morton, Omar et al.(243) the medical certificate was not originally

intended for epidemiological research, however it remains the primary source of data

on CoD worldwide.

Determining the underlying CoD is an inherently difficult task, particularly if the

patient did not die while under medical care. A range of factors determine how

� � �

75

closely the recorded sequences of deaths match the “truth” (Figure 8). These include

differences in practices by country (244) education and language (242). Local names

may exist for specific diseases or symptom complexes (242). While education can

improve accuracy of certification (245, 246), experience has been found not to be a

major factor (247). Studies based on case vignettes in Taiwan suggest individuals

produce very different results from the same information, especially when choosing

between an acute complication of a chronic disease and the chronic disease itself,

suggesting variations in certification were due to differences in interpretation of the

available information not knowledge of the process itself (248)

Figure 8: Sources of errors in medical certification of death from Maudsley and Williams, 1994 (249)

Some causes of death are more likely to be correctly identified than others. In

England and Wales, studies comparing medical certificates with autopsy results has

shown high levels of accuracy for ischemic heart disease and unnatural causes but

poor accuracy for a range of other diseases (250). In countries with good diagnostic

facilities, medical certificate studies have found high accuracy with recording

cancers. In the UK, studies have found breast cancer to be underreported by only

4% (251), and colorectal cancer by only 5% (252). In comparison, a study of brain

donors in the UK found less than half the patients who were clinically diagnosed with

� � �

76

Parkinson’s disease prior to their death had this recorded on the certificate (either in

the causal sequence or in the contributory causes) (253).

Even common causes of death are often misidentified. A community hospital in the

US found 48 % of deaths due to myocardial infarction did not have this recorded in

Part I of the medical certificate and 25% of deaths that did have this recorded should

actually have been attributed to a different cause (173). Another study in the US

found medical certificates missed 35% of cancers known to the cancer registry in

four US states (254). In an interesting study that looked at how certain doctors felt

about their ability to identify the CoD, Karwinski found increased certainty of

diagnosis (due to better diagnostic technology) did not result in increased accuracy

on the medical certificate (255).

Age also affects the likelihood of an accurate CoD being recorded on the certificate.

As chronic disease advances there is also more likelihood of co-morbidity making

the underlying cause more difficult to determine (256). Acute conditions with high

case fatality are likely to be reported more accurately in death statistics than chronic

conditions which may progress slowly or produce non-specific symptoms. This has

been demonstrated in studies of asthma in the UK (257) and diabetes in France

(258).

Causes of death are even more difficult in neonatal deaths. Studies in an Australian

hospital neonatal intensive care unit show that even using just seven broad

categories, the main CoD was incorrectly recorded for 34% of deaths (259).

� � �

77

Interestingly, causes of death may be deliberately incorrect on the medical

certificate. Legislation in the UK (no longer in force) requiring alcohol related deaths

be reported to the coroner appeared to result in under-reporting, with doctors likely to

amend the CoD to something less controversial (260, 261). Also in the UK, a study

of hospital doctors and GP’s found that of those who said they would always use the

best possible CoD statements more than half the doctors would modify the “true”

CoD to avoid distressing relatives (247). As noted by (262) while the information on

medical certificates remains non confidential, inaccuracy in certification will occur.

In busy hospitals in the developed world, certification often falls to the most junior

member of the team (263). Although this may not be the case in the PICTs, chronic

understaffing and extremely high workloads mean doctors are unlikely to complete

certificates with the care required for accurate statistics unless they understand the

importance of the data created. As Robertson said about the physician in his

address to the American Public Health Association in 1929, “beyond all he must be

admitted as a partner into this vast undertaking” (264). This remains true today.

There are two options for reviewing the quality of death certification, depending on

the data to which the researcher has access. The first is a review of the medical

certificates themselves, in terms of plausibility of sequences, the number of causes

listed, consistency with age and sex of the deceased etc. This is useful in reviewing

the quality of the processes employed in completing the certificate and the

plausibility of the results, but cannot give an indication of the consistency of the

findings with the medical history of the deceased and is therefore limited in terms of

being able to apply correction factors to the collected data. The second is a review of

the certificate against the medical record for the deceased (265-271), in which a

� � �

78

medical practitioner or team of practitioners reviews the case history and completes

a “mock’ medical certificate to be compared against the original. This approach relies

on access to identifiable data, and assumes the reviewing physician or team is both

independent and is more accurate than the original physician. While this approach

has merit, in the PICTs the small medical community on island and often poor state

of the medical records mean findings from a review are not necessarily more

accurate than the original certificates. Implementation of such an investigation would

therefore require some modification to this approach.

2.4.2 Assessment and correction of coding practices

Coding is inherently affected by the quality of the medical certificates. Incomplete

data and inappropriate sequences are major problems (272). Standard practice is for

written causes of deaths (from certificates, verbal autopsy etc) to be coded using the

WHO ICD. The current version of this classification system is ICDv10. Deaths are

then tabulated by code and other disaggregations (such as age and sex). WHO

publishes guidelines on appropriate tabulations based on the level of coding

undertaken (273).

Although coding should be standardised, several studies have noted local practices

may develop over and affect coding and the international comparability of findings

(274-276). Even developed countries with well developed CRVS systems need

periodic review to correct systematic misclassification that can occur (187). Both

Kelson (275) and Mackenbach (276) found coding practices across Europe had

developed to “account for” common problems in the certification of cancer and went

some way to evening these discrepancies out.

� � �

79

Changes in coding practice can also have a significant affect on cause specific

mortality. In the US, ischemic heart disease increased by a factor of 1.15 due to

changes in coding when ICDv7 was replaced with ICDv8 (171). These changes are

not consistent across countries due to the differences in certification and coding

practice. The change from ICDv9 to ICDv10 created a fall in COPD deaths in France

from 1998/99-2000/02 and in increase of around 10% in the US (69).

CoD has historically been measured and reported according to the underlying cause,

either by cause-specific mortality rates, or by proportional mortality. In order to

calculate these rates and/or proportions, the effect of errors in certification and the

often substantial number of deaths assigned to “ill-defined and unknown” causes

must be accounted for. In European data, inverse correlations have been found

between the proportion of “unknown” category deaths and suicides (277) There has

also been suggestion that cardiovascular disease in particular may be affected by

the proportion of deaths considered to be “unknown”. (276).

Some papers class errors in certification and coding into major and minor, although

this has been disputed as subjective (278). Given the state of data in PICTs much of

the CoD data cannot be coded to specific subsections and must therefore be

analysed at a broader category (i.e. diabetes rather than subtypes of diabetes).

Therefore there is value in differentiating between errors that result in classification

to a different category as opposed to those which affect coding only at sub-category

level.

While the mistakes are important, at a population level there is some evidence to

suggest inaccuracies balance out. A study of asthma mortality in Northern Ireland

� � �

80

found false positives to be balanced by the number of false negatives when reviewed

against medical records and relative interviews about patient history (176).

2.4.3 Multiple cause of death analysis

While analysis of CoD has traditionally been carried out in relation to the underlying

cause, there has been a trend towards analysis of multiple causes of death in

recognition that several diseases may contribute to a patient outcome. This is

perhaps more relevant for public health practice, as it more accurately reflects the

actual burden of a disease on the population. Whilst there has been work put forward

to examine the use of multiple COD models to determine causality (including

frequency with which specific causes are recorded on the same certificate and odds

ratios) (279) this is fraught with danger, as the diagnosis is the physicians “best

guess” and substantial selection bias would be expected (i.e. causes already not

considered to be causal would not be reported as such). The most straight forward

analysis of multiple CoD data is the “total mentions”. That is, all deaths that have the

disease of interest on the certificate (in part I or part II) are counted (279). Of deaths

autopsied in Georgia (US), only 42 % of certificate had the correct underlying cause

– but when analysed using multiple causes of death, 72% had the correct underlying

cause somewhere on certificate (280).

Although multi cause data may improve the usefulness of the statistics, it does not

solve the issue of international comparison as the ratio between total mentions and

deaths with the cause of interest as the underlying cause may differ, reflecting the

different reporting practices of different areas (281). The frequency between which a

cause is reported (at all) and reported as an underlying CoD is affected by disease

type with acute conditions having a higher ratio (282). This suggests estimates of

� � �

81

death due to acute conditions may provide similar estimates using either underlying

CoD data or multiple CoD data, but that multiple CoD analysis may be particularly

useful for chronic conditions.

2.4.4 Models for cause of death data

In many places medical certificates or other CoD data are often only available for the

proportion of the deaths that occur in or near a hospital or major health facility.

Murray and Lopez (283) have recently developed a method for using facility based

proportional mortality to estimate national proportional mortality by age and cause.

This still depends on the availability of the requisite data, and is highly context

dependant (that is a significant proportion of deaths need to occur in hospital) (11).

The method uses cause specific mortality fractions for the proportion of deaths that

occur in hospital, using a subset of the population for which there is good VR data, or

a similar population. The method was found to be fairly robust when tested in Mexico

and was not overly sensitive to the size of the population used to estimate

proportions. (15). The method could prove useful in PICTs where many areas have

poor data outside the hospital system, but may face difficulties due to different

access to health services by different populations and the lack of comparable

populations with good vital registration data.

Salomon and Murray (284) note current estimates indicate around 30% of all deaths

worldwide are covered by CRVS systems. In the complete absence of empirical

data, indirect methods have been developed for assigning CoD from the age

distribution of deaths. The best known of these methods were developed for the

GBD studies by Salomon and Murray (284). The “CODMOD” program is a

spreadsheet that uses probability distributions based on regression models to

� � �

82

determine the proportion of deaths by cause from all-cause mortality and income per

capita.

Morris and colleagues (285) describe a similar regression method specifically for

estimating the distribution of deaths by cause in children under five years of age

when there is insufficient empirical data. This does not distinguish between neonatal

causes and essentially looks at diarrhoea, malaria, measles and pneumonia (as well

as undetermined deaths). More recently, Rao et al (286) have developed a model

which addresses these gaps but excludes malaria, an important CoD in children in

parts of Melanesia, from the generated CoD distributions.

The model by Morris was developed on data from SE Asia and Africa where studies

were available and child mortality was greater than 26 per 1,000. As data were not

available from other regions, they were not incorporated into the design, and the

model could not be validated against these other populations (285). Arguably the

same would be true of the Solomon and Lopez model, and when examined closely

there has been little empirical data for the PICTs used as input data, with included

data from the early 1990’s in one or two countries only; thus calling into question the

relevance of these models on the small scale applicable to the PICTs

Models may provide causal distributions by broad cause group (such as infectious

and maternal illnesses, NCDs and external causes) yet they cannot provide the

detail that health planners and governments need to address the very real health

issues in the PICTs. Taking just one group of NCDs as an example, the risk factors

and subsequently the health interventions required at a population level to address

rheumatic heart disease, ischaemic heart disease and cerebrovascular diseases are

� � �

83

very different. Without the ability to disaggregate at this level, health planners are

mostly guessing as to where interventions should best be targeted.

2.5 Final Comments

As presented in this chapter, there are a range of methods available for collecting

and analysing mortality and CoD data. Censuses and surveys are well accepted in

the region as a means of collecting mortality data, and provide periodic data for

analysis that may simply be unavailable through routine CRVS systems. The use of

census data to generate mortality estimates across all age groups is however simply

no longer suitable (if it ever was) in the PICTs in the face of the changing

epidemiology of premature adult mortality and mortality from NCDs. Model life tables

based on single input parameters of child or infant mortality do not adequately

capture premature adult mortality, as this “uncoupling” of child and adult mortality

levels was not the experience of most countries from which the data used in the

models was drawn. While this could potentially be overcome using a second adult

mortality input in the model life tables, generated from indirect methods such as the

orphanhood or widowhood methods; these have had limited use in the region and

are known to be prone to both recall and selection bias as outlined earlier. Similarly,

methods reliant on inter-censal population changes to assess mortality by age group

are limited by the poor quality of migration data for much of the region, and may be

affected by poor recording of age of respondent. As such, there is a need to focus on

alternative means of generating reliable mortality estimates by age and sex through

the use of routinely collected data, wherever possible.

� � �

84

As discussed in Chapter one, routine civil and vital registration collections in the

PICTs are known to be incomplete and may have issues around coverage,

completeness, timeliness and quality. In order to use the collected data to generate

reliable estimates of mortality, a qualitative and quantitative assessment of these

aspects is required. The system assessment approaches described together

address the range of system aspects that may have a bearing on data quality, while

both direct demographic and capture-recapture approaches methods have merit in

the PICTs. Other than the Brass Growth Balance method, the direct demographic

methods rely on data periods that align with an inter-censal period or require

additional data on migration, while the capture-recapture methods require multiple

data sets and assume all deaths have the same probability of being reported. The

choice of method to assess the data from each PICT will depend on the best fit of

available data with the data required and assumptions made for each approach.

What is apparent in the methods described in this chapter is the assumptions (such

as whether the population is stable or whether there is dependence between

sources), which are an inherent part of each of these approaches, have a significant

impact on the final estimates of mortality by age. Selections regarding the approach

used must therefore be made in the context of a qualitative system assessment to

ensure appropriate selections are made and final results are plausible.

Censuses and surveys are also clearly limited in their ability to collect CoD data.

While causal distributions by age and sex can be estimated from models based on

mortality level, this is not an adequate substitute for nuanced empirical data that is

able to monitor real rather than anticipated changes over time. Further, as described

in this section, the final estimates of CoD distributions generated in this manner may

be a long way from any original data and incorporate a substantial range of

� � �

85

assumptions. They are used in scenarios where routine collection of CoD is simply

not available or complete and in this way, CoD estimates may be modelled from age

and sex specific mortality rates that were themselves derived from models using only

U5M as an input. CoD distributions derived from models therefore have a potential

role in evaluating priorities at a very broad scale where no empirical data is available,

but limited value in monitoring trends over time or for health planning at a detailed

level. As such, there is no real substitute for medical certification of death data

routinely collected at a nationally representative level.

The quality of CoD data relies on the diagnostic ability of the person reporting the

data in selecting the underlying CoD and recording the sequence of events leading

to the death. As noted in this chapter, this is an inherently difficult task even for

trained medical practitioners, and is cause is frequently misreported. Systematic

biases in certification, coding or tabulation may need to be assessed through a

medical record or certificate review according to the level of access to original source

data. The medical record review is more robust in its ability to assess and correct

reported CoD as it is possible, through this process, to evaluate each of the steps

from diagnosis and certification, to coding, tabulation and reporting. There is

however an assumption in this approach, as employed in Iran (185), that the medical

record itself is properly documented and complete, a scenario which is known to be

untrue in a number of PICTs including Tonga and Nauru (14). A death certificate

review does not allow the diagnosis to be reviewed beyond the plausibility of the

recorded sequence, and therefore allows only some of the revision possible through

the medical record review. While the use of a systematic VA may provide useful CoD

data, the size of Pacific Island populations means this would be unlikely to be a

viable option for data collection at a nationally representative level, although there

� � �

86

may be scope for this approach to be used in the future to improve data collection

through community nursing reports from remote islands where medical certification is

not possible.

The limitations of models for CoD data as described above mean even for countries

where there is no representative national collection of CoD data from medical

certificates or verbal autopsy, empirical data collections on cause that do not

conform to these quality standards, such as hospital records, lay reporting through

civil registration or community nursing reports should be explored. While these

cannot be used to make national estimates, they may provide additional depth of

information for policy makers about key diseases and conditions of public health

interest, and what age groups these are impacting, that are not otherwise available.

This thesis will therefore focus on improving understanding and use of routinely

collected data to generate estimates of mortality and CoD profiles by age and sex.

Methods of evaluating data quality and correcting for systematic biases will be

employed from the methods described above as determined by both the data

available and access to the original source material. Available data for selected

countries will be examined more closely in the following chapter. As noted by the

UN, whichever method of data collection and analysis is selected, the final analysis

must be considered in the face of inherent plausibility (117).

� � �

87

3 Routine reporting systems for deaths and cause of death in selected Pacific Islands

Examination of mortality level and CoD profiles should begin with a clear

understanding of the data sources available and how the data collection, collation,

analysis and reporting processes may affect these data. This is essential for

selecting appropriate methods to work with the data and critical assessment of the

plausibility of mortality level and CoD estimates derived from it. As discussed in the

previous chapters, this assessment will focus on routine government CRVS systems.

3.1 Assessment Framework for routine reporting systems

CRVS systems are frequently complex, spanning multiple departments or

organisations, and influenced by a diverse array of factors from community attitudes

through to available infrastructure. As discussed in the previous chapter, while there

� � �

88

are a range of existing approaches to assessing CRVS systems, it was felt an

amalgam of these existing approaches was required to adequately capture all the

major features of the system that may affect data availability and quality.

The framework presented in this chapter was developed to provide a logical means

of identifying both the planned and actual operation of the systems and subsequent

effects on data quality and availability; and presenting these in a manner to allow key

strengths and weaknesses to be identified, and broad comparisons to be made

across systems and countries whilst maintaining flexibility in the approach. The

framework was developed specifically for application in the PICTs. It draws on

lessons from previous approaches, but is primarily based on previous work by Rao

et. al.(63, 64) presented in Chapter two; incorporating questions from tools such as

the HMN HIS assessment (202), the IT focused assessments of Heeks, Kaufman,

and Yusof (287-291) and previous work by Ruzicka and Lopez (209) under Rao’s

thematic headings (64). Initially developed as a data collection tool, the framework

was refined during and applied to country assessments for Fiji, Kiribati, Nauru,

Palau, Tonga, the Solomon Islands and Vanuatu between 2008 and 2010.

Amendments were also incorporated to reflect the layers of administration and

governance that dictate the way CRVS systems operate and interact, and the

importance of social structures and interpersonal relationships that are the

cornerstone of successful systems in this region. Key revisions, drawn primarily from

the body of work on health governance (216), included the separation of

administrative issues at the system or organisational and national levels, and the

addition of concepts of responsibility and authority (under administration aspects).

This is particularly important in the PICTs where seniority may be heavily influenced

� � �

89

by community standing and age, rather than job title alone. A separate theme to

address staff ownership of the CRVS systems or the acceptability of the systems to

those directly involved in their operation was also added.

The framework presented here structures system findings into five key aspects: the

socio-political environment, national administrative framework, system

administration, system technical characteristics and system ownership (Figure 9).

The first two aspects refer to the overall context in which routine death reporting is

carried out within the country, while the final three are system specific. A summary of

key aspects for review is shown in Table 13. CRVS systems may be contained

entirely within a single organisation, or operated by a combination of government

and non-government agencies. Each distinct reporting system should be assessed

separately, and in the context of its relationship to other existing systems.

Figure 9: Routine death reporting system assessment framework

��

�

90

Tabl

e 13

: Rou

tine

deat

h re

port

ing

syst

em a

sses

smen

t fra

mew

ork

outli

ne: K

ey a

spec

ts fo

r rev

iew

Cat

egor

y To

pic

Key

Des

ign

aspe

cts

for r

evie

w

Key

Per

form

ance

asp

ects

for r

evie

w

Soc

ieta

l iss

ues

Fune

ral

arra

ngem

ents

Ti

me-

fram

e to

bur

ial

Bur

ial l

ocat

ion

(priv

ate

vs. C

emet

ery)

Pub

lic s

uppo

rt R

equi

rem

ents

for s

ocia

l sec

urity

pay

men

t R

ole

of th

e ch

urch

lead

ers

Com

mun

ity a

war

e of

the

need

to re

port

deat

hs

Com

mun

ity a

ccep

tanc

e of

gov

ernm

ent r

ole

Pol

itica

l su

ppor

t P

rese

nce

of in

tere

st g

roup

s/ in

tern

atio

nal

stak

ehol

ders

M

orta

lity

data

use

d to

info

rm p

oliti

cal d

ecis

ions

Adm

inis

trativ

e en

viro

nmen

t G

eogr

aphy

G

eogr

aphi

cal d

istri

butio

n of

ser

vice

s re

ason

able

acc

ess

to c

ivil

regi

stra

tion

cont

act

Infra

stru

ctur

e av

aila

ble

Nat

iona

l int

erne

t net

wor

k R

elia

bilit

y of

pow

er s

uppl

y A

cces

s (tr

ansp

ort a

nd m

ail r

oute

s)

Com

mun

icat

ion

chan

nels

ava

ilabl

e

Rea

sona

ble

syst

ems

are

suita

ble

for t

he in

frast

ruct

ure

avai

labl

e

Lega

l fra

mew

ork

Per

sons

requ

ired

to re

port

Tim

efra

me

for r

epor

t E

vide

nce

requ

ired

for r

egis

tratio

n S

till-b

irths

incl

uded

in c

olle

ctio

n

Lega

l ins

trum

ent e

xist

s (y

ear)

R

epor

ting

of d

eath

s re

quire

d R

epor

ting

befo

re b

uria

l req

uire

d O

vers

eas

deat

hs a

nd s

tillb

irths

cle

arly

iden

tifie

d O

rgan

izat

iona

l ro

les

Gov

ernm

ent a

nd s

take

hold

er o

rgan

isat

ions

in

volv

ed in

rout

ine

repo

rting

or u

se o

f dat

a V

ital r

egis

tratio

n fu

nctio

ns c

lear

ly a

ssig

ned

Inte

rnat

iona

l con

tact

poi

nts

clea

rly d

efin

ed

Coh

esio

n /

inte

grat

ion

Coo

rdin

atio

n be

twee

n ag

enci

es

Nat

iona

l sta

tistic

s co

mm

ittee

mee

ts re

gula

rly

Qua

lity

cont

rol c

heck

s be

twee

n ag

enci

es

Form

al a

gree

men

ts fo

r dat

a sh

arin

g / a

cces

s S

taff

know

thei

r cou

nter

parts

in re

late

d or

gani

satio

ns

Sys

tem

issu

es

– adm

inis

tratio

n

Des

ign

and

inte

nt?

Pur

pose

of s

yste

m (i

.e. m

easu

ring

CoD

, civ

il st

atus

etc

) S

yste

m is

sui

tabl

e fo

r the

pur

pose

and

ava

ilabl

e in

frast

ruct

ure

Org

anis

atio

nal s

truct

ure

Pro

vinc

ial p

roce

sses

are

alig

ned

with

org

. stru

ctur

e M

anag

emen

t su

ppor

t

Dat

a us

ed to

info

rm m

anag

emen

t dec

isio

ns

Dat

a m

anag

emen

t sta

ff ha

ve a

cces

s to

man

ager

s R

espo

nsib

ility

Res

pons

ibili

ty fo

r all

proc

esse

s cl

early

des

igna

ted

Res

pons

ibili

ties

docu

men

ted

and

agre

ed w

ith s

taff

Aut

horit

y A

ppro

vals

requ

ired

to a

cces

s an

d re

leas

e da

ta

Dat

a m

anag

emen

t sta

ff ab

le to

che

ck a

nd c

orre

ct d

ata

Sta

ff ab

le to

rai

se p

robl

ems

outs

ide

imm

edia

te u

nit

Res

ourc

es

A

dequ

ate

staf

f and

fund

ing

reso

urce

s fo

r sys

tem

as

desi

gned

.

��

�

91

Sys

tem

issu

es

– te

chni

cal

Pro

cess

es

(incl

udin

g qu

ality

con

trol)

Det

erm

inat

ion

of c

ause

of d

eath

R

ole

of th

e po

lice

in c

ertif

ying

dea

ths

Fiel

d co

llect

ion

and

reco

rdin

g pr

oces

s C

olla

tion

and

entry

pro

cess

es

Exi

sten

ce o

f reg

iste

rs /o

ther

reco

rds

Dis

ease

spe

cific

regi

ster

s R

outin

e vs

. on-

dem

and

anal

ysis

Pro

cess

doc

umen

tatio

n ex

ists

and

mat

ches

pra

ctic

e P

roce

sses

avo

id d

uplic

atio

n be

twee

n pr

ogra

ms

Res

pons

ibili

ty fo

r dat

a ch

ecki

ng a

ssig

ned

and

cond

ucte

d Q

ualit

y co

ntro

l / s

yste

m re

view

car

ried

out r

egul

arly

R

epor

ting

timef

ram

es a

nd fo

rmat

C

onsi

sten

cy o

f pra

ctic

e D

ata

secu

rity

and

stor

age

prac

tices

To

ols

Use

and

sui

tabi

lity

of a

utom

ated

cod

ing

tool

s C

ompl

exity

of d

atab

ases

and

spr

eads

heet

fo

rmat

s

Med

ical

cer

tific

ate

com

plie

s w

ith s

tand

ards

C

onte

nt/ f

orm

of c

ertif

icat

e ap

prop

riate

H

ospi

tal a

dmis

sion

and

dis

char

ge fo

rms

Sui

tabi

lity

of IT

tool

s fo

r ava

ilabl

e re

sour

ces

Rep

ort f

orm

ats

suita

ble

for p

urpo

se

App

ropr

iate

lang

uage

use

d in

tool

s an

d do

cum

ents

D

ata

stan

dard

s C

odin

g sy

stem

use

d C

oD c

lass

ifica

tion

syst

em

Off-

shor

e de

aths

and

med

ical

refe

rral

s

Leve

ls o

f agg

rega

tion

Cod

ing

is c

onsi

sten

t bet

wee

n pr

ogra

ms

Geo

grap

hica

l bou

ndar

ies

are

cons

iste

nt

Tech

nica

l su

ppor

t In

frast

ruct

ure

Com

mun

icat

ion

betw

een

syst

em le

vels

A

bilit

y to

acc

ess

and

upgr

ade

softw

are

and

prog

ram

s K

now

ledg

e an

d sk

ills

Exi

stin

g ca

paci

ty b

uild

ing

prog

ram

s e.

g. c

linic

al

revi

ew m

eetin

gs/ i

n-ho

use

train

ing

/ for

mal

pr

ogra

ms

etc.

Le

vel o

f edu

catio

n / t

rain

ing

Kno

wle

dge

of c

ertif

icat

ion

and

codi

ng s

tand

ards

S

taff

have

app

ropr

iate

IT s

kills

S

taff

have

app

ropr

iate

dat

a m

anag

emen

t ski

lls

Res

pons

ible

sta

ff ha

ve a

naly

tical

ski

lls

Trai

ning

in p

lace

to a

ddre

ss id

entif

ied

need

s S

yste

m is

sues

- o

wne

rshi

p S

taff

owne

rshi

p an

d in

tere

st

A

ppro

pria

te fe

edba

ck to

sta

ff at

all

leve

ls

Sta

ff vi

ew th

eir r

ole

as im

porta

nt

Sta

ff vi

ew d

ata

as a

n im

porta

nt to

ol

� � �

92

Key system aspects can be classified by performance and design elements.

Performance elements are those for which there are clearly recognised international

standards or best practices with which to compare, such as the use of the

internationally recognised format of the medical certificate (68) or coding for CoD

according to the ICDv10 (68). Design elements of the system encapsulate the

suitability of the system for the environment, infrastructure and administrative

framework under which it operates. The design elements are neither inherently

“good” or “bad” but have a significant influence on the ability of a CRVS system to

generate high quality accessible mortality level and CoD data.

3.2 Common elements of CRVS systems in the region and their

effects on data

An overview of the key data sources available and the major biases associated with

these data collections across the study countries is presented in case study one. A

brief summary of country specific findings follows the case study.

3.2.1 Case Study 1: Reporting systems across seven PICTs (164)

Karen L Carter, Chalapati Rao, Alan D Lopez, Richard Taylor. Routine death

reporting systems and implications for data in seven Pacific Island countries BMC

Public Health 2012, 12:436 doi:10.1186/1471-2458-12-436

This case study presents an assessment of the routine CRVS systems in Fiji,

Kiribati, Nauru, Palau, the Solomon Islands, Tonga and Vanuatu, based on the

� � �

93

assessment framework described earlier in this chapter. Assessments were

conducted to achieve three primary aims:

� To identify the key routine data collections available in the study countries

and where these data are held.

� To identify aspects in the design and performance of the CRVS systems

that could affect data accessibility and quality (including coverage and

completeness potential biases such as age or gender reporting

differentials); and

� To identify key priorities for system improvement based on the strengths

and weaknesses common across the countries to improve future data

collections.

Staff interviews, observation, document review and mapping of all processes (both

as they are meant to operate and how these actually functioned were key aspects o

the approach used to conduct the assessments.

While substantial differences in the CRVS systems (and subsequently the available

data) were evident across the study countries, common themes affecting how well

the systems functioned emerged. Significant biases that may affect mortality and

CoD data were present in all systems and included issues such as coverage of outer

islands, differentials in incentives for reporting deaths at different ages, and off-island

referrals for high risk cases.

Carter et al. BMC Public Health 2012, 12:436http://www.biomedcentral.com/1471-2458/12/436

CORRESPONDENCE Open Access

Mortality and cause-of-death reporting andanalysis systems in seven pacific island countriesKaren L Carter1,2*, Chalapati Rao1*, Alan D Lopez1 and Richard Taylor1,3

Abstract

Background: Mortality statistics are essential for population health assessment. Despite limitations in dataavailability, Pacific Island Countries are considered to be in epidemiological transition, with non-communicablediseases increasingly contributing to premature adult mortality. To address rapidly changing health profiles,countries would require mortality statistics from routine death registration given their relatively smallpopulation sizes.

Methods: This paper uses a standard analytical framework to examine death registration systems in Fiji, Kiribati,Nauru, Palau, Solomon Islands, Tonga and Vanuatu.

Results: In all countries, legislation on death registration exists but does not necessarily reflect current practices.Health departments carry the bulk of responsibility for civil registration functions. Medical cause-of-death certificatesare completed for at least hospital deaths in all countries. Overall, significantly more information is available thanperceived or used. Use is primarily limited by poor understanding, lack of coordination, limited analytical skills, andinsufficient technical resources.

Conclusion: Across the region, both registration and statistics systems need strengthening to improve theavailability, completeness, and quality of data. Close interaction between health staff and local communitiesprovides a good foundation for further improvements in death reporting. System strengthening activities mustinclude a focus on clear assignment of responsibility, provision of appropriate authority to perform assigned tasks,and fostering ownership of processes and data to ensure sustained improvements. These human elements need tobe embedded in a culture of data sharing and use. Lessons from this multi-country exercise would be applicable inother regions afflicted with similar issues of availability and quality of vital statistics.

Keywords: Mortality, Death registration, Vital statistics, Cause of death, Pacific islands.

BackgroundAccurate mortality statistics are essential for populationhealth assessment, and to design and monitor healthintervention programs. Despite extensive collection ofhealth data in Pacific island countries, these data arerarely analysed or utilised [1], as they are considered tobe incomplete or unreliable [2-4].Despite the data shortcomings, there is a general ap-

preciation that non-communicable diseases are increas-ingly contributing to premature adult mortality inPacific Island Countries [2,5]. Cancer and cardiovascular

* Correspondence: [email protected]; [email protected] of Population Health, University of Queensland, Herston, Australia2Statistics for Development Programme, Secretariat of the PacificCommunity, Noumea, New CaledoniaFull list of author information is available at the end of the article

© 2012 Carter et al.; licensee BioMed Central LCommons Attribution License (http://creativecreproduction in any medium, provided the or

disease have been recognised as important causes ofdeath in these countries since the 1980s [6-10], andobesity rates are amongst the highest in the world [10].Timely and reliable data on mortality and causes ofdeath is needed to effectively monitor and combat thisimpending epidemic. The ideal mortality data source isan efficient death registration system with medical certi-fication of cause-of-death. In Pacific Island Countries,small population sizes make complete routine registra-tion an attractive proposition, but there is a concomitantchallenge of servicing widely dispersed, low densitypopulations.Since the late 1990’s there has been an increasing

international emphasis on improving health informationsystems and mortality reporting [11-13]. More recently,

td. This is an Open Access article distributed under the terms of the Creativeommons.org/licenses/by/2.0), which permits unrestricted use, distribution, andiginal work is properly cited.

mailto:[email protected]


http://creativecommons.org/licenses/by/2.0

Carter et al. BMC Public Health 2012, 12:436 Page 2 of 9http://www.biomedcentral.com/1471-2458/12/436

the Health Metrics Network has highlighted the paucityof vital statistics in developing countries [14]. The Mil-lennium Development Goals (MDG’s) have also pres-sured governments to demonstrate progress towardsthese targets [15]. Despite this increased attention, littleinformation is available on the registration and statisticalsystems in Pacific Island Countries, and the influence ofthese processes on data quality.This paper examines the structure and operations of

routine death reporting systems in seven Pacific IslandCountries; Fiji, Kiribati, Nauru, Palau, Solomon Islands,Tonga and Vanuatu. Strengths and limitations commonacross national systems are identified, and system char-acteristics are related to data availability and quality.This review also makes specific recommendations tostrengthen these systems and improve data availabilityand quality at the national and regional level.

MethodsCountries were selected in order to represent Pacific Is-land States with a range of: population and land sizes,government structures, distribution of health services,and economic development [1] [see Additional file 1].These countries represent all three regions of the PacificIslands: Melanesia (Fiji, Solomon Islands, Vanuatu, andKiribati), Micronesia (Palau and Nauru) and Polynesia(Tonga). A standard analytical framework [16] was usedto assess practices for data collection, management, ana-lysis, reporting and utilisation in each country.Assessments were conducted through collaboration

between the authors and key national stakeholders, in-cluding representatives from the Ministry of Health andBureau of Statistics. Other stakeholders representedincluded: government departments such as the civilregistry office, finance, and central planning agencies;local non-government organisations; and country officesof WHO, UNICEF and UNFPA.The assessment comprised a field visit guided by a

detailed data collection tool [16], and involved compre-hensive document review, observation of procedures andprocesses, discussions with staff across organisations andlevels, and examination of available mortality data. Infor-mal interview respondents were asked to discuss theirrole in data collection, management or use, and percep-tions of the strengths and weaknesses of the system.Where practices differed from policy or managementperspectives, both were documented. For each country,a flow chart was assembled to depict the reportingprocess from the initial recording of each death to itsrepresentation in a national statistical dataset.All information was subsequently reviewed using a

standard assessment framework that classifies perform-ance and design aspects [16] of the system according tothe socio-political environment, national administrative

framework, system administration, technical characteris-tics and ownership [16]. The first two refer to the overallcontext in which routine death reporting is carried outwithin the country, while the final three are system spe-cific. This framework was adapted from earlierapproaches [17-19] with the additional assessment ofthree human elements that are central to system per-formance in the Pacific Islands; responsibility and au-thority under administrative aspects of the system, andorganisational ownership [16].Country and system specific findings were distilled

into key themes that reflect strengths and weaknessescommon to multiple systems across the countriesreviewed, that have the potential to significantly impactthe coverage, completeness and quality of data, or its ac-cess and utility. Finally, potential impacts of these sys-tem characteristics were reviewed in relation to availabledata.

Results and discussionCommon elements of routine reporting systemsRoutine reporting of deaths in the Pacific Island coun-tries is predominantly managed by civil registration sys-tems or Health departments (Figure 1). The importanceof health reporting systems in supporting civil registra-tion was evident across countries. In particular, despiteexisting legislation, civil registration of deaths was notactive in the Solomon Islands or Vanuatu.Within health departments, deaths are reported by med-

ical practitioners and community nursing staff (Figure 1).Medical certification of death is practiced in hospitals inall countries, however data from death certificates is rou-tinely collated only in Fiji, Nauru, Palau, Tonga, and inPort Vila hospital in Vanuatu. Collated data is codedaccording to the International Classification of Disease(ICD) version 10 in Fiji, Tonga and Port Vila hospital,while Nauru has subsequently started coding since thistime. Until recently only hospital discharge data wascoded in Palau (using ICD version 9), although recentchanges subsequent to this review have seen the introduc-tion of coding for medical certificates of death using ICD9. In Kiribati, the Solomon Islands, and other hospitals inVanuatu, data for hospital deaths is compiled from dis-charge data and coded according to ICD version 10.In Fiji, Palau, and Nauru, medical certification is also

conducted for deaths occurring in the community. In allcountries except Nauru, community deaths are reportedthrough monthly nursing reports. These are only rou-tinely collated in Fiji, Kiribati and Vanuatu. Reportingthrough the hospitals accounts for nearly all reporteddeaths in the small islands of Palau and Nauru, and ahigh proportion of reported deaths in Tonga; while amuch larger proportion are reported through monthlynursing reports in the other countries. Only in Kiribati

Figure 1 Routine Death Reporting Pathways in the Pacific Islands. Dotted lines indicate link does not apply to all study countries.


is there a large proportion of deaths reported throughcivil registration that are not also reported via the healthsystem.Other routine death recording systems in these coun-

tries are operated by churches, cemeteries and the po-lice. However, these additional sources have limitedapplication in improving the completeness of deathrecords as they pertain only to specific sub-groupswithin the population, the records are of doubtful qual-ity, or the data cannot be easily extracted.

Key strengths and weaknessesStrengths and weakness of the routine reporting systemsare reviewed below, with key themes common acrosscountries reviewed outlined in Table 1.

Societal issuesInitial reporting and formal registration of deaths reliesprimarily on family members. Social factors such as bur-ial arrangements, public support, and political interest indata for planning purposes therefore have a significantbearing on the likelihood of the death being reported,and the quality of the data captured.Burial in cemeteries (parts of Vanuatu) require specific

approvals, which promotes registration. Such approvals

are not necessarily required for burial on private lands(as in Kiribati). Funeral customs that require extendedfamily to gather, as observed in Palau and Tonga alsoencourage reporting. In these cases, the death certificatefacilitates airline reservations, and the body is held forseveral days at a mortuary. Social incentives for report-ing are limited, although access to pension plans or landtransfers may encourage reporting, particularly for adultmale deaths. A notable exception is Nauru where a gov-ernment funeral assistance payment strongly encouragesuniversal reporting. Local health personnel are closelyintegrated into the community across these countries,and are an important foundation for developing publicsupport. Across the region, there is increasing evidenceof data use in deciding national health priorities [20-22],with the Pacific Health Ministers meeting in 2009recommending that countries “improve the quality andreliability of data” [23].

National administrative environmentThe collection and compilation of national mortality andcause-of-death statistics involves multiple governmentagencies (Figure 1). There is significant potential for du-plication of effort and inconsistencies in the data, in theabsence of appropriate coordination at the central level.

Table 1 Key Strengths and Weaknesses of Mortality Reporting Systems in the Pacific Islands

Category Strengths Weaknesses

Societal Issues Social incentives for registration Private land burials without officialapproval

AdministrativeEnvironment

Existing legal framework Inadequate/inconsistent implementation of laws

Health systems involvement inCivil registration and vitalstatistics operations

Passive registration i.e. onus to reporton citizen

Registration process for “off-island”deaths unclear

National statistics committees Complex statistical reportingrequirements

System Issues –Administration

Community nurses formally taskedto notify vital events

Improper emphasis on communitynurses to report cause of death

Routine compilation of mortalitydata by different healthdepartments

Need for better coordination togenerate one reconciled mortalitydataset from the health system

Private health intuitions rarelyintegrated adequately into reportingsystems

No clear delineation of responsibilityacross institutions, leading to taskduplication

Personnel lack authority toquery/clarify data

System Issues –Technical

Standard international medicaldeath certificate (except Nauru)

Lay reporting of cause for deathsoutside facilities in some countries

Medical certificates of death notroutinely tabulated in all countries

Trained ICD coders High turnover of trained staff

Key personnel adequately skilledfor data management at nationallevel

Statistical analysis limited to tenleading causes of death

Insufficient data quality assessmentand control

Initiatives to set data standards forhealth information

Dysfunctional/outdated softwareprograms not amenable to modificationor upgrade

System Issues –Ownership

Strong ownership ofsystems/interest at national levelsthat has contributed substantiallyto ongoing survival of the systems

Generally poor feedback to local levelstaff

Many systems are highly dependent onone or two key individuals with astrong interest in providing health data

Source data: [16].


Equally, inadequate delineation of organisational respon-sibilities could result in significant data gaps.Legislation requiring registration of deaths exists for

all islands [24-31], although not fully implemented, suchas in rural areas of Vanuatu [24]). Current practices arealso not necessarily reflected in the legislation, as in Fijiwhere deaths cannot be registered prior to burial or cre-mation. While all countries have a process for holdingan inquest or inquiry for a death suspected to have

resulted from a criminal act, fire or other causes of spe-cific interest, only Fiji and Kiribati have legislation thatdefines responsibility for ensuring findings are subse-quently forwarding to the civil registry and used tocomplete the registration process [25,26]. In Vanuatuand the Solomon Islands, health systems have evolved toperform the role of civil registration. Compilation ofcause-of-death data was generally recognised as a role ofthe health department despite not being defined in


legislation, except in Fiji where coding was duplicated atthe Bureau of Statistics. All countries currently havefunctional inter-agency committees to meet reportingrequirements for the Millennium Development goals[15]. These have had a clear impact in improving coord-ination and collaboration between departments.

System issues – administrationCompilation of mortality statistics also involves a widearray of organisational units within and across agencies(Figure 1). In the larger countries, provincial governmentstructures add to this complexity (Additional file 1). Theoperations and subsequent data quality are thereforeinfluenced by administrative aspects within the system.In most countries, the system design was appropriate

for available resources, with a clear mandate for datacompilation. However, in Vanuatu, an over-reliance ontechnology was noted, that did not have the requisite in-frastructure. In most departments, responsibilities forreporting and analysis of data were not clearly assigned;or, were assigned to staff without adequate training,resulting in poor use of existing data sets. A strong de-mand for health information in Palau, Tonga and Kiri-bati translates to better access for data managers toquery or clarify issues with doctors and senior staffregarding the data. In contrast, data managers in Vanu-atu, the Solomon Islands, and Fiji had limited authority,either real or perceived, to raise concerns. In decentra-lised civil registration systems, local registrars are notdirectly under the authority of the national Civil Regis-trar, leading to confusion in the processes for resolvingissues.

System issues – technicalDuplication of data compilation was observed in allcountries. This was particularly apparent between pro-vincial and national systems (for both health and civilregistration); and in the parallel reporting systems forspecific health programs (Figure 1). For example, com-munity nurses in the Solomon Islands use up to nineforms to report a death. Databases that cannot be modi-fied, or which use obsolete software for which technicalsupport is no longer available, as seen in Vanuatu andKiribati, contribute to the proliferation of multiple data-bases to meet changing needs. The international stand-ard medical certificate of death [32] is used in allcountries except Nauru. Much of the collected data wasnot available for analysis, such as in Kiribati where thehealth department does not retain a complete copy ofthe certificate. Quality control processes are limited, ex-cept in Palau where completed medical certificates areroutinely reviewed by senior doctors resulting in a lowproportion of deaths being assigned to ill-defined causes.Recording of infant deaths was inconsistent as result of

unclear procedures, a lack of data standards, and com-plicated forms; with early deaths potentially missed(Kiribati) or stillbirths included in the data (Nauru).In Fiji and Tonga, data entry and coding practices

leading to alterations in the causal sequence on the med-ical certificate result in difficulties in applying ICD rulesto select the underlying cause [32]. A high staff turnoverfor doctors (Nauru, Solomon Islands and Kiribati),nurses (Fiji) and key analytical roles (Tonga and Fiji) hasalso adversely affected knowledge of reporting processesand depleted the skill set available in some areas.

System issues – ownershipA sense of ownership in individuals as well as institu-tions is essential for smooth operation of vital registra-tion systems. In general, a broader sense of sharedownership provides the system with greater flexibility torespond to changed circumstances such as staff turnoveror natural disasters.Feedback to staff is an important mechanism to foster

ownership among personnel at different levels. This hasbeen provided through annual public health conferences,(e.g. Palau, Tonga and the Solomon Islands); and fieldvisits by data managers (e.g. Tonga and the SolomonIslands). In Tonga, such visits include local civil registrarstaff and community leaders. At local levels, there isconsiderable variation in the quality of registers and sub-mission of statistical returns by community nurses inKiribati, Solomon Islands and Vanuatu, which reflectsthe degree of ownership perceived by them. At the sametime, a strong sense of ownership at the national level ofthe health information system in Vanuatu helped sustainthe overall data compilation process. In health facilities,physicians tended to store medical records for deathsseparately, to preserve them from loss or destruction.This practice limits access to such data for reporting andanalysis.

Data availabilityAs shown in Table 2, significantly more data was identi-fied as available for reporting and analysis than previ-ously recorded by international sources. Additional datasources (Figure 1) that were not tabulated (and thereforenot accessible for analysis) were also found.System characteristics indicate that under-registration

of deaths is likely from civil registration data sets exclud-ing Nauru [42] and Palau. Health department deathreporting systems were estimated to be more than 90%complete in Fiji [41] and Palau (Table 2), and sufficientlycomplete in Tonga to produce reliable estimates of mor-tality when corrected for under-reporting [43]. ICDcoded [32] cause-of-death data based on medical deathcertificates were available for both hospital and commu-nity deaths in Fiji, and Tonga, with un-coded data based

Table 2 Tabulated mortality data available for selected Pacific Islands, 2000–2009: WHO databases, as compared withlocally available data

Mortality data by age and sex Cause of death data

Country EstimatedDeaths(2011)#

WHO*(Year/Completeness)

Locallyavailable data+

(estimated completeness$)

WHO}

(Year/Quality)Locallyavailabledata+

(frommedicalcertificates) +

Fiji 7185 No data(1999/90-100%)

MoH reports (>95%) 2000/Low MOH reports

Kiribati 827 2001/>75% MoH reports (40-60%)Civil registration (40-60%)

2002/Low Not tabulated

Nauru 88 No data Civil Registration (>95%) 1996/not rated MoH reports

Palau 158 No data MoH reports (>95%) No data MoH reports

Solomon Islands 4039 No data MoH reports (not estimated) No data Not tabulated

Tonga 683 No data MoH^ reports (>80%) 1998/Low MoH reports

Vanuatu 1311 No data MoH reports (not estimated) No data MoH reports(hospitaldeaths only)

# Estimated deaths extracted from Secretariat of the Pacific Islands Population data [33].* Data extracted from World Health Organisation Statistical Information System [34], [35-40]: data last updated March 2011.+ Most complete source listed only. MoH=Ministry of Health. Most complete available source or reconciled data: Fiji (MoH), Kiribati (reconciled MoH and civilregistry data), Solomon Islands (MoH), Vanuatu (MoH), Nauru (MoH/Civil registry data), Palau (MoH), Tonga (MoH).$ Estimates of completeness source: Fiji [41], Kiribati (capture-recapture analysis, unpublished), Solomon Islands (system assessment), Vanuatu (system assessment),Nauru [42], Palau (Brass analysis), Tonga [43].} WHO assessment, 2003 [3].^ MoH data from 2009 onward is routinely reconciled against civil registry data.


on medical certificates available for Nauru and Palau.Very few deaths in Palau are assigned to ill-definedcauses, due to quality control measures described previ-ously. Data from both civil registration and healthreporting systems in Kiribati appear incomplete, al-though a reconciled data set derived from both systemswould be more plausible. Only Vanuatu and the Solo-mon Islands did not have routine reporting systems cap-able of generating estimates of the mortality level. Allthree had systems from which cause-of-death could beextracted for hospital deaths; with some cause-of-deathinformation available for deaths in the community, ei-ther based on family or community nurses reports.

ConclusionsDespite the data gaps identified in this analysis, signifi-cantly more information on mortality of populations inthe Pacific is available than currently used (Table 2).Findings suggest that while significant investment isrequired to improve capture of cause-of-death, it shouldbe possible for all the countries reviewed to be able togenerate reliable data on mortality level within 4–5 years.Causes of death reported by nurses are currently of lim-ited value due to use of broad categories and high pro-portions of "ill-defined" causes. The current system isappropriate to record only demographic aspects ofdeath, unless procedures such as verbal autopsy [44] areintroduced. This should be considered for systems such

as Kiribati, Solomon Islands and Vanuatu where univer-sal medical certification is not currently feasible. All ofthe health department reporting systems reviewed wereaffected by problems with data extraction, duplicate datasets, coding and data entry issues and difficulties acces-sing records in the database that adversely affect dataquality.Legislation requiring registration of deaths exists for

all islands, but does not necessarily reflect current prac-tices. Health departments are carrying the bulk of re-sponsibility for supporting routine data collection andcivil registration functions, and their importance cannotbe overemphasised. Efforts to improve data collection,use, and acceptability to decision makers will need to ei-ther focus on these systems, or ensure their integrationinto official reporting processes. A clear strength in allcountries reviewed was the close interaction betweenhealth staff and local communities, including the oppor-tunities this creates for building strong reporting rela-tionships at the local level.Significant duplication of data collection and entry

exists across all systems, reflecting issues with data own-ership, obsolete and unresponsive technology and frac-tured management of the reporting systems. Theassessment identified three critical human elements thatinfluence the effectiveness of civil registration and vitalstatistics operations. These are responsibility, authorityto effectively undertake assigned responsibilities, and


ownership or the value that staff place on the reportingsystem and their role within it.The approach used in this study shares many similar-

ities with the comprehensive assessment tool for vitalstatistics systems developed by researchers at the Uni-versity of Queensland and WHO [19] which is primarilya policy tool that aims to lead countries through a crit-ical self-appraisal of their systems as a basis for develop-ing a planning document for system improvement. Theframework presented here was developed from a similarhistory, however is intended to provide a structured crit-ical evaluation of system characteristics to inform the in-terpretation of available and published data from theregion, yet also allows findings to be used in a policy set-ting to guide decisions around future system improve-ments required. Although conducted in partnership withthe Ministry of Health and Bureau of Statistics, andthrough close collaboration with other local partners;the assessment reported here was led by researchers ex-ternal to the systems being evaluated. This could poten-tially result in less ownership of the findings by those ina position to effect system improvements. However thisapproach allows an assessment by objective observers,with a broad range of experience across different report-ing systems, and using a standardised framework; thussubsequently allowing for a regional comparison and forcommon themes to be distilled from individual countrydata. Further, this approach allows staff at all levels to

Table 3 Key Regional Priorities for Action

Assessment Category Recommendations

Societal Issues Recognise the importance of communthe local community) and provide suphealth planning. [Long term]

Administrative Environment Conduct operational research to validainternal management support and con

Formalise MOU or similar between heaand cause of death information. [Short

System Issues – Administration Identify and minimise duplication betwthrough data sharing agreements. [Sho

Assign staff (and adequate resources) t

Investigate ways of better engaging prwith civil registration, and support this

System Issues – Technical Support countries to identify staff respothem to meet minimum skill and know

Encourage tabulation and use of medihospital discharge data. [Short Term]

Design databases to capture medical ccause to avoid changes to recorded se

Encourage use of standard statistical msystem limitations to aid in appropriate

Assess and build staff capacity, particul

System Issues –Ownership Build supportive relationships betweenmanagers to develop sufficient authori

Source data: [16].

discuss their views privately, and provides an opportun-ity for in-depth review where resources may not other-wise be available.While many of the systems reviewed were highly com-

plex, the framework used in this study provides a usefulmeans of identifying themes for further consideration.This paper is designed to provide an insight into themortality and cause-of-death reporting in the region.Several of the issues are identified here for the first time.Although previous studies have been limited, the issuesraised are consistent with those noted in this assessment.These include little incentive to register vital events [45]particularly in the case of children [46], a lack of any “at-tempt to estimate and correct for under-enumeration”[2], and referral biases in cause-of-death analysis due tothe reliance on hospital data in the absence of robustcommunity based reporting [47].Across the region, both registration and statistics sys-

tems need strengthening to improve the access, com-pleteness, and quality of data. Close interaction betweenhealth staff and local communities provides a good foun-dation for further improvements in death reporting. Sug-gested priorities to strengthen systems are noted inTable 3. Addressing issues of authority and support, andreducing the duplication of functions across systemswould assist in alleviating resource pressure. Longerterm priorities relate to broader shifts in practice andwould require significant investment.

ity nurses in the reporting process (including links and influence withport for increased training and feedback on how their data is used in

te reporting completeness and quality in functioning systems to buildfidence to use results. [Short Term]

lth departments and civil registration offices on data sharingTerm]

een parallel systems (particularly within health departments),rt Term]

o analysis role. [Short Term]

ovincial governments (island councils, municipal governments etc.)with feedback at a provincial level. [Long Term]

nsible for each process and provide in-country training to supportledge level. [Short Term]

cal certificates as a source of information rather than relying solely on

ertificates as written. Create an additional field for selection of underlyingquence. [Short Term]

easures (confidence intervals, and rolling averages) and reporting ofdata interpretation. [Short Term]

arly in data analysis and certification practices. [Short Term]

data managers, senior management and doctors that will allow dataty to question data as needed. [Long Term]


System strengthening activities must focus on ele-ments such as clear assignment of responsibility,provision of appropriate authority to perform assignedtasks, and fostering ownership of processes and data toensure sustained improvements. These human elementsneed to be embedded in a culture of data sharing anduse. Although there are common priorities for systemimprovement across the region, the diversity of infra-structure, social context and existing system capacity willrequire action to address priorities to be locally appro-priate. Governments and international agencies shouldalso support the use of local data wherever possible. Les-sons from this multi-country exercise would be applic-able in other regions where vital statistics systems aresimilarly characterised by duplication, inadequatelytrained staff, and poor use of good quality data on birthsand deaths in informing policy decisions.

Additional file

Additional file 1: Appendix 1. Country Characteristics

AcknowledgementsThis work was conducted in conjunction with many government and non-government agencies, and was made possible by the efforts of manyindividuals within those departments. In particular the authors would like toacknowledge the participation and assistance of the following organisations(and specific individuals in particular): Fiji Ministry of Health (Laite Kavu,Shareen Ali); Fiji Islands Bureau of Statistics (Vasemaca Lewai); Fiji CivilRegistration Office, Ministry of Justice; Fiji School of Medicine; Kiribati Ministryof Health; Kiribati Bureau of Statistics; Kiribati Civil Registration Office; NauruMinistry of Health (Gano Mwareow, David Dowiyogo); Nauru Bureau ofStatistics (Ipia Gadabu); Palau Ministry of Health (Losii Samsel, MairengSengebau); Palau Bureau of Statistics (Rhinehart Silas); Solomon IslandsMinistry of Health (Baakai Iakoba and Christine Bishop); Tonga Ministry ofHealth (Sione Hufanga); Tonga Bureau of Statistics (Sione Lolohea); TongaCivil Registry Office; Vanuatu Ministry of Health (Joao Costa); World HealthOrganization (Country Liaison Officers and Western Pacific Regional Office);and the Secretariat of the Pacific Community. Funding for this research wasprovided under the AusAID Australian Research Development Awards(HS0024). The funding body had no role in study design; in the collection,analysis, and interpretation of data; in the writing of the manuscript; or inthe decision to submit the manuscript for publication.

Author details1School of Population Health, University of Queensland, Herston, Australia.2Statistics for Development Programme, Secretariat of the PacificCommunity, Noumea, New Caledonia. 3School of Public Health andCommunity Medicine, University of New South Wales, New South Wales,Australia.

Conflicts of interestThe authors declare that they have no conflicting interests.

Authors’ contributionsKLC: developed and trialed the system assessment framework and datacollection tools, conducted the field visits and interviews, collated thefindings and extracted key themes, and drafted the analysis. CR: assisted withthe identification and clarification of key themes, reviewed the data analysisand interpretation of results; revised the manuscript critically for structure,clarity and important intellectual content. AL: revised the manuscript criticallyfor important intellectual content. RT: reviewed the data analysis andinterpretation of results; and revised the manuscript critically for importantintellectual content. RT and AL both conceived and coordinated the broader

study under which this project was undertaken. All authors read andapproved the final manuscript.

Received: 6 September 2011 Accepted: 13 June 2012Published: 13 June 2012

References1. Kuartei S: Health Care Plans and Dust Collection in the Pacific. Pacific

Health Surveillance and Response 2005, 12(2):155–158.2. Taylor R, Bampton D, Lopez AD: Contemporary patterns of Pacific Island

mortality. Int J Epidemiol 2005, 34(1):207–214.3. Mathers CD, Fat DM, Inoue M, Rao C, Lopez AD: Counting the dead and

what they died from: an assessment of the global status of cause ofdeath data. Bull World Health Organ 2005, 83(3):171–177.

4. Haberkorn G: A sea of islands – a myriad of indicators: on the interfacebetween demography and planning in the Pacific Islands. Asia PacificViewpoint 1997, 38(3):219–236.

5. Khaleghian P: Non-communicable diseases in Pacific Island Countries: Diseaseburden, economic costs and policy options. Noumea: Secretariat of the PacificCommunity and World Bank; 2003.

6. Taylor R, Henderson B, Levy S, Kolonel L, Lewis N: Cancer in Pacific IslandsCountries. Noumea: South Pacific Commission; 1985.

7. Ministry of Health: Health Status Report - Fiji. Suva: Ministry of Health,Government of Fiji; 1996.

8. Ministry of Health: Report of the Minister of Health for the year 2005.Nuku'alofa: Government of Tonga; 2005.

9. Ministry of Health: Report of the Minister of Health for the year 2007.Nuku'alofa: Government of Tonga; 2007.

10. Coyne T: Lifestyle diseases in Pacific communities. Noumea: Secretariat of thePacific Community; 2000.

11. Hill K, Lopez AD, Shibuya K, Jha P: Interim measures for meeting needs forhealth sector data: births, deaths, and causes of death. Lancet 2007,370:1726–1735.

12. AbouZahr C, Cleland J, Coullare F, Macfarlane SB, Notzon FC, Setel P, SzreterS: and on behalf of the Monitoring of Vital Events (MoVE) writing group.The way forward. The Lancet 2007, 370:1791–1799.

13. Pacific Islands Health Officers Association: Declaring a Regional State ofHealth Emergency Due to the Epidemic of Non-Communicable Diseases.: BoardResolution #48-01; http://new.pihoa.org/files/resolutions/NCD_Emergency.pdf Accessed [2 April 2011].

14. Setel Philip W, Macfarlane Sarah B, Simon Szreter, Lene Mikkelsen, PrabhatJha, Susan Stout, Carla AbouZahr, on behalf of the Monitoring of VitalEvents (MoVE) writing group: Who Counts? 1: A scandal of invisibility:making everyone count by counting everyone. Lancet 2007, doi:doi:10.1016/S0140-6736(07)61307-5.

15. Secretariat of the Pacific Community: Pacific Islands Regional MillenniumDevelopment Goals Report 2004. Noumea: Secretariat of the PacificCommunity in cooperation with the United Nations; 2004. http://www.arab-hdr.org/publications/other/undp/mdgr/regional/asia-pacific/mdg-pacific-04e.pdf Accessed [2 April 2011].

16. Carter KL, Rao C, Taylor R, Lopez AD: Routine mortality and cause of deathreporting and analysis systems in seven Pacific Island countries. University ofQueensland: Health Information Hub; 2010:8. http://www.uq.edu.au/hishub/index.html?page=154695&pid=116774 Accessed [2 April 2011].

17. Rao C, Osterberger B, Anh TD, MacDonald M, Chuc NT, Hill PS: Compilingmortality statistics from civil registration systems in Viet Nam: the longroad ahead. Bull World Health Organ 2010, 88(1):58–65.

18. Rao C, Bradshaw D, Mathers CD: Improving death registration andstatistics in developing countries: Lessons from sub-Saharan Africa.SAJDem 2004, 9(2):79–97.

19. Mikkelsen L, Lopez AD: Improving the quality and use of birth, death andcause-of-death information: guidance for a standards-based review of countrypractices. Geneva: Health Information Hub, University of Queensland andWorld Health Organization; 2010.

20. Ministry of Health: Corporate Plan 2007–2009. Nuku'alofa: Government ofTonga; 2007.

21. Ministry of Health: Annual Corporate Plan for the Financial Year ending on31st December 2010. Suva: Government of Fiji; 2010.

22. Ministry of Health and Medical Services: Strategic Plan 2008–2011. Tarawa:Republic of Kiribati; 2009.

http://www.biomedcentral.com/content/supplementary/1471-2458-12-436_S1.doc

http://new.pihoa.org/files/resolutions/NCD_Emergency.pdf

http://new.pihoa.org/files/resolutions/NCD_Emergency.pdf

doi:10.1016/S0140-6736(07)61307-5

http://www.uq.edu.au/hishub/index.html?page=154695&pid=116774

http://www.uq.edu.au/hishub/index.html?page=154695&pid=116774


23. Pacific Islands Ministers of Health: Eight Meeting of the Ministers of Health ofthe Pacific Island Countries.: World Health Organization and Secretariat of thePacific Community. PNG; 2008. DPM/03-E http://www.wpro.who.int/internet/resources.ashx/PIC/2009/8thPIC_meeting+report.pdf Accessed[2 April 2011].

24. Civil Status (Registration) Act (Chapter 61). Republic of Vanuatu. 1971.25. Births, Deaths and Marriages Registration Act. Government of Fiji. 1975.26. Kiribati Births, Deaths and Marriages (Amendment) Act. Republic

of Kiribati. 1997.27. Nauru Births, Deaths and Marriages Ordinance. Republic of Nauru. 1957.28. Palau National Code. Republic of Palau. 1985.29. Solomon Islands Births and Deaths (Registration) Act. Government of the

Solomon Islands; 1988.30. Births, Deaths and Marriages Registration (Amendment) Act. Kingdom of

Tonga; 1991.31. Registrar General's Births and Deaths Regulations. Kingdom of Tonga; 1979.32. World Health Organization: International Statistical Classification of Diseases

and Related Health Problems. Geneva: Tenth Revision; 2007. http://www.who.int/classifications/icd/en/Accessed [26/11/2010].

33. Secretariat of the Pacific Community: Pacific Island Populations - Estimatesand projections of demographic indicators for selected years. http://www.spc.int/sdp/Accessed [23 April 2012].

34. World Health Organization: World Health Statistics. 2010. http://www.who.int/whosis/whostat/2010/en/index.html Accessed [2 April 2011].

35. SPC: Pocket Statistical Summary. Noumea: Secretariat of the PacificCommunity; 2009. http://www.spc.int/sdp/index.php?option=com_docman&task=cat_view&gid=28&Itemid=42 Accessed [2 April2011].

36. ABC Island descriptions. http://www.abc.net.au/ra/pacific/places/country/palau.htm Accessed [2 April 2011].

37. Stanley D: South Pacific Organizer., http://www.southpacific.org/guide/tonga.html Accessed [2 April 2011].

38. Ramsar Sites Information Service. http://ramsar.wetlands.org/Portals/15/Republic_of_Kiribati.pdf Accessed [2 April 2011].

39. UNDP: Human Development Report. 2009. http://hdrstats.undp.org/en/indicators/87.html Accessed [2 April 2011].

40. World Health Organization, Western Pacific Regional Office: Western PacificCountry Health Profiles (CHIPS) 2010 Revision. Manila: World HealthOrganization; 2008.

41. Carter K, Cornelius M, Taylor R, Ali SS, Rao C, Lopez AD, Lewai V, Goundar R,Mowry C: Mortality Trends in Fiji. Aust N Z J Public Health 2010,35(5):412–20. In press.

42. Carter KL, Soakai TS, Taylor R, Gadabu I, Rao C, Thoma K, Lopez AD:Mortality Trends and the Epidemiological Transition in Nauru. Asia Pac JPublic Health 2011, 23:10.

43. Hufanga S: Evaluation of published estimates of mortality measures forTonga, and Capture-recapture analysis of mortality data in Tonga.University of Queensland: School of Population Health; 2010. WorkplaceProject Portfolio [Masters dissertation].

44. Setel PW, Sankoh O, Rao C, Velkoff VA, Mathers C, Gonghuan Y, Hemed Y,Jha P, Lopez AD: Sample registration of vital events with verbal autopsy:a renewed commitment to measuring and monitoring vital statistics. BullWorld Health Organ 2005, 83(8):611–617.

45. Taylor R, Lewis ND, Levy S: Societies in transition: mortality patterns inPacific Island populations. Int J Epidemiol 1989, 18:634–646.

46. Finau SA: Pacific Health: An Analysis for Training New Leaders. Asia Pac JPublic Health 1992, 6(2):46–53.

47. Hufanga S, Bennett V: Mortality Data in the Kingdom of Tonga: A Reviewof Changing Trends over the Ten Years since 1996. Health InformationManagement Journal 2007, 36(2):43–48.

doi:10.1186/1471-2458-12-436Cite this article as: Carter et al.: Mortality and cause-of-death reportingand analysis systems in seven pacific island countries. BMC Public Health2012 12:436.

Submit your next manuscript to BioMed Centraland take full advantage of:

• Convenient online submission

• Thorough peer review

• No space constraints or color figure charges

• Immediate publication on acceptance

• Inclusion in PubMed, CAS, Scopus and Google Scholar

• Research which is freely available for redistribution

Submit your manuscript at www.biomedcentral.com/submit

http://www.wpro.who.int/internet/resources.ashx/PIC/2009/8thPIC_meeting+report.pdf

http://www.wpro.who.int/internet/resources.ashx/PIC/2009/8thPIC_meeting+report.pdf

http://www.who.int/classifications/icd/en/


http://www.spc.int/sdp/


http://www.who.int/whosis/whostat/2010/en/index.html

http://www.who.int/whosis/whostat/2010/en/index.html

http://www.spc.int/sdp/index.php?option=com_docman&task=cat_view&gid=28&Itemid=42


http://www.abc.net.au/ra/pacific/places/country/palau.htm

http://www.abc.net.au/ra/pacific/places/country/palau.htm

http://www.southpacific.org/guide/tonga.html

http://www.southpacific.org/guide/tonga.html

http://ramsar.wetlands.org/Portals/15/Republic_of_Kiribati.pdf

http://ramsar.wetlands.org/Portals/15/Republic_of_Kiribati.pdf

http://hdrstats.undp.org/en/indicators/87.html

http://hdrstats.undp.org/en/indicators/87.html

� � �

103

3.2.2 Data availability and accessibility

Significantly more data on mortality level and CoD is available from routine CRVS

systems than has previously been used or documented, although accessing these

data remains problematic. An overview of the available data sources is given in the

Table 14.

Table 14: Available routine mortality data collections by country and source

Data Source Fiji Kiribati Nauru Palau Tonga SolomonIslands Vanuatu

Civil Registration Available Available Available Available Available Not available

Limited

Health Reporting

Medical Certificate Data Hospital Deaths Community Deaths

Available Available

Available No

Available Available

Available Available

Available Available

Available Not available

Available Not available

Hospital Separation Data

Available Available Available Available Available Available Available

Community Nursing Data

Available Available Not available

Available Available Available Available

Other reporting systems

Police (accidents)

NA Treasury Mortuary Prime Minister’s Office

NA NA

Access to data was hampered by complex and slow approval and ethics review

mechanisms, a lack of clarity around who could approve access to data (and hence

a propensity to refer all decisions to a ministerial level), and in some cases an

unwillingness of individual data managers to allow scrutiny or use of the data,

despite higher approvals. In part, this was explained by concerns around privacy of

the data, which were able to be addressed through study protocols and relationship

building with the officers involved. Other cultural and historical influences were more

� � �

104

difficult to identify or address. This includes the concept that knowledge should only

be passed to worthy individuals as reward or currency for a political or social role,

rather than being generally available; and an exhaustion with a legacy (real or

perceived) of past research that has limited country involvement or ownership of

outcomes.

3.3 Summary of country specific CRVS systems

A brief overview of routine CRVS systems in each country is given below.

3.3.1 Fiji

All deaths in Fiji are required to have a medical certificate. If the death occurs

outside the hospital, the body is either brought in by the family or police, or a doctor

is required to attend.

The medical certificates of death are collated nationally and entered into the HIS,

which also houses the hospital separation data entered at the hospital. Community

deaths are also reported through monthly nursing reports which are checked against

the medical certificates received.

To register a death through the civil registry, the family is required to present a

medical certificate, coroner’s or police report; and provide details of the burial site to

the local government offices. Deaths are entered into the registry’s database which

is managed by the government IT organisation (ITC). Deaths are registered by

� � �

105

completing a paper form which is sent as hard copy to the provincial office where the

death certificate is issued, and the death is either entered onto the national registry

database or details are forwarded to the national civil registry office. Periodically, the

Bureau of Statistics requests details of the registered deaths through ITC, which are

coded against the ICD and re-entered into a separate database.

3.3.2 Kiribati

Although incomplete, civil registration is operational in Kiribati. Families are required

to present evidence (either a medical certificate, radio or newspaper notice of the

death or statutory declaration from a church minister) to the registrar to register the

death. Provincial offices complete a hard copy monthly report for the national civil

registry office, where all deaths are entered into a database.

Within the health system, medical certificates of death are issued for deaths in the

national referral hospital and major provincial hospitals staffed by doctors. They are

not collated centrally, and much of the data collected under this process is lost.

Hospital separation data (and therefore deaths in hospital) are recorded in the

electronic HIS. Deaths in the community are recorded on monthly reports completed

by nurses or nurse aids at the community health centres and aid posts. Both the

hospital separation data and monthly reports are entered into a database at the

national health information section. The Bureau of Statistics currently relies on

census analyses and does not use any routinely collected data from either the MoH

or Civil Registrar for mortality estimation.

� � �

106

3.3.3 Nauru

All deaths in Nauru are required to have a medical certificate and are recorded

through the MoH. If the death occurs outside the hospital, the body is either brought

in by the family or police, or a doctor is required to attend. The medical certificate is

entered by nursing staff into a spreadsheet, as well as into the civil registry

database. Both the spreadsheet and database were established recently, with data

prior to these maintained only in hard copy. A hard copy of the certificate is

forwarded to the Civil Registry Office, with the original kept at medical records. Data

prior to 2005, before the two hospitals merged, is not available at the MoH. A

notification of death is also completed by nursing staff and forwarded to the Minister

of Health.

The government of Nauru provides a payment to the family to assist with expenses

when a death occurs. As such, it is also possible to obtain an almost complete list of

deaths (although name details are often unclear and the records do not include age

or date) from the treasury records. There is no routine analysis or publication of

mortality data other than with the national census, managed by the Bureau of

Statistics, which utilises data derived from civil registration records.

3.3.4 Palau

Like Nauru, Palau has a small population, and all deaths are required to have a

medical certificate. If the death occurs outside the hospital, the body is either brought

in by the family or police, or a doctor is required to attend. The medical certificate is

entered by the Public Health unit into a spreadsheet and separate database. Hard

copies of the certificate remain with hospital medical records and a copy is sent to

� � �

107

the Ministry of Justice for civil registration purposes (although the family is still

required to attend the office to complete the registration process). Hospital

separation data (and therefore deaths) are recorded in the electronic HIS.

Community deaths are also reported through monthly nursing reports, although

these are not further collated. The Bureau of Statistics obtains data for reports

directly from the MoH upon request. Data are routinely analysed by the public health

epidemiologist for health planning and reporting.

3.3.5 Tonga

Tonga has multiple systems for routine reporting of deaths. Civil registration may be

completed through the national civil registry office or in provincial centres. Deaths

are recorded manually in the registry book, with those registered in Nuku’alofa (the

capital) also entered into an electronic database. Registry books from provincial

centres are forwarded to the national office when full, where they are entered into the

database. Deaths are also reported through local unpaid leaders called “Town

officers”, who are required to record all deaths in their area on a monthly report form,

covering a range of local issues, that is forwarded to the Prime Minister’s office.

All deaths are required to have a medical certificate. These are forwarded and

entered nationally by the Health Information Section into a spreadsheet and separate

database. Deaths in the community are recorded on “Notice of death” forms and

monthly reports completed by nurses or nurse aides at the community health centres

and aid posts, which are forwarded to the closest hospital where a medical certificate

should be prepared; before being forwarded to the Public Health nursing coordinator

at the national hospital.

� � �

108

3.3.6 Vanuatu

Deaths in Vanuatu are only required to be registered when they occur within the

urban areas of Port Vila and Luganville. In practice, the civil registration system is

only recording births, with fewer than 20 deaths recorded in 2008. Routine reporting

of deaths is solely through the MoH. Medical certificates of death are issued for

deaths in the national referral hospital and the provincial hospital in Luganville where

there are doctors (there is a single doctor permanently located at only one of the

remaining provincial hospitals). Only medical certificates for deaths occurring at the

Port Vila hospital are collated electronically (as an excel spreadsheet). Hospital

separation data (and therefore deaths) are also recorded in the electronic HIS at the

major provincial hospitals. This information is sent electronically (or collected on disc

periodically) to the national HIS. Deaths in the community are recorded on “Notice of

death” forms and monthly reports completed by nurses or nurse aides at the

community health centres and aid posts. CoD is given as free text on a single line.

Notice of deaths forms are sent to the provincial offices as hard copy for entry into

the database, with data forwarded to the national health information office either as

electronic files or hard copy. Most data are entered (or re-entered) manually at the

national level.

3.4 Conclusion

As described in this chapter, each of the study countries is currently collecting some

data on deaths and CoD through routine CRVS systems. The extent of these data

collections and the quality of the data collected do however vary significantly

� � �

109

between countries due to the different systems in place and the environments in

which they operate. This necessitates the use of a range of evaluation and correction

measures as described in Chapter two, rather than using a single approach across

the region.

While each country has a unique set of challenges and potential biases that should

be considered when evaluating the plausibility of reported measures of mortality, or

when estimating mortality and CoD patterns from raw data; there were a number of

common issues identified. These include greater social incentives for registering

deaths in wage earners and property holders (usually adult males) that could lead to

under-representation of women and children in the reported data, coverage issues

away from the major centres of population, significant issues around late reporting

that may affect the timeliness of the data for analysis, and a propensity to use

hospital discharge records for CoD tabulations (based on primary cause or

immediate CoD) rather than underlying cause, which could skew CoD patterns

towards broad categories such as cardiovascular diseases (from deaths reported

with an immediate cause such as “heart failure” or “heart attack”). Other potential

biases on mortality estimates include the different procedures for dealing with vital

events that occur off-island and the impact this can have in small populations, as in

these study countries. The impact of off-island events, migration and small

populations are discussed further in the next chapter.

�

� � �

110

4 Further analytical considerations – small area estimates, off-island vital events and migration

4.1 IntroductionIn addition to the system biases outlined in Chapter three, any discussion of mortality

patterns in the Pacific Islands must specifically consider the impact of small

population sizes, off-island events and migration. In the context of generating

measures of mortality, PICTs have inherently small populations. The number of

deaths occurring in these countries each year is very low, and subsequently

population measures of mortality may be significantly affected by both stochastic

variation and small changes both in the numerator (the number of deaths) and

denominator (population, or for some measures, the number of live births). Changes

to the numerator (through deaths happening off-shore for example) would have the

greatest effect given the comparative size of the small number of events and the

population.

� � �

111

4.2 Population size and small-area estimatesPopulation size in PICTs ranges from just over 1000 people in Tuvalu and Niue to

more than 1.5 million in PNG. Excluding PNG, Fiji has the largest population in the

Pacific with just over 800,000 people (44). The small population sizes mean

stochastic variation can have a significant impact on mortality rates, with one or two

deaths substantially changing mortality rates. In Tonga for example, a difference of

2 maternal deaths more than doubles the MMR (292). It is therefore critical data are

aggregated over multiple years to provide greater stability, and estimates are

reported with appropriate confidence intervals. The small population sizes mean

even at a national level, mortality estimates for PICTs can be regarded as small area

estimates. The field of small area estimation has several lessons that may prove

useful for understanding mortality in the PICTs.

Examining minimum sample size within demographic surveillance sites Begg et. al.

(293) cite a minimum of 852,900 person years of observation in a high mortality

(adult and child) population as necessary to obtain acceptable cause-specific data

(data with sufficiently narrow confidence intervals for most age groups to enable use

to inform planning decisions). Outside PNG, only Fiji has a sufficiently large

population to generate sufficient person years of exposure from routine registration

of deaths to allow data to be examined on an annual basis; although much

uncertainty remains and age groups must be kept very wide. For all of the other

study countries, multiple years must be aggregated before analysis by age group is

possible, as shown in Table 15. The years shown assume universal coverage and

� � �

112

completeness of the CRVS system, and are therefore the minimum aggregation of

years required.

Table 15: Minimum number of years data would need to be aggregated to reach the minimum person years of data required for acceptable mortality

analysis as cited by Begg et. al. by country (293)

Country Population (2011)(44) Minimum aggregation of years

Fiji 851,745 1 Kiribati 102, 697 9 Nauru 10, 185 84 Palau 20,643 42 Solomon Islands 553, 254 2 Tonga 103,682 9 Vanuatu 251,784 4

Table 15 shows few PICTs could consider annual estimates of mortality and CoD to

be stable, and highlights the need to aggregate across several years in order to

reach sufficient person years of data to generate reliable measures. There is

however a “trade-off” between generating reliable data, and producing timely data to

meet policy and planning needs. Aggregating over more than 80 years for example,

as would be required in Nauru (Table 15) is simply not useful or possible in practical

application. Additional approaches to interpreting data, such as trend tests across

multiple years will need to be considered.

Small area analysis has been used extensively in Australia for mortality and cancer

analysis (189). At a sub-national level, difficulties generating mortality estimates for

small populations are likely to be compounded by issues such as numerator and

denominator sources which have incongruent geo-coding where data comes from

different sources (i.e. civil registration and census areas). Taylor (189) further notes

� � �

113

that aggregation between adjoining areas often combine districts with very different

characteristics. This is particularly true in the PICTs, where customs, transport,

access to health services and lifestyle may vary significantly between adjoining

islands and regions. Aggregation over several years to provide sufficient data for

analysis may therefore be preferable. Despite these problems, Taylor argues small

area estimates are sufficiently useful that synthetic estimates are justifiable to

produce them (189).

More recent work in the UK on small area analysis has demonstrated that summary

measures of mortality, such as LE and adult mortality, can be generated reliably from

smaller population sizes using standard life table methods (without smoothing by

model life tables). Testing small random subsets of larger populations demonstrated

life tables can be generated with reasonable validity (results close to those

generated by the larger data set) and reliability (sufficiently narrow confidence

intervals to be useful) for populations of at least 5,000 people (over a single year),

even when there were zero deaths recorded in some age groups. These results

were found to be conditional on some deaths being recorded in the open ended

(oldest) age group (294, 295). This approach could be replicated in the PICTs by

aggregating over time to ensure 5,000 person years of data. The UK team also

evaluated several approaches to calculating confidence intervals for small area

estimates (294), and found the Chiang method (296) most appropriate, as it coped

best with missing values that may occur with small population sizes (294). This

approach has been adopted in Australia (297) and New Zealand (298, 299) for small

area analysis, and will subsequently be adopted in this work for the PICTs.

Stochastic variation will be minimised by aggregation over multiple years, with long

� � �

114

term trend analysis applied where possible. Age groups of 5 or 10 years will be used

to ensure sufficient data in each in age group, with the open end age group set

according to expected mortality levels (for most countries at 65 years of age) to

ensure sufficient deaths recorded in this age group for reliable analysis.

4.3 Off-Island Events

As noted previously, while changes to both the numerator and denominator may

affect mortality rates, changes in the numerator produce the greatest effect due to

the relative sizes of the number of events and the population. Of the issues

presented in this chapter, deaths that occur off-island and missed from numerator of

mortality measure calculations would have the greatest impact on final estimates of

mortality.

There are three categories of travellers of interest. In decreasing order of risk of

dying, these are:

� People referred off-island for medical treatment through official

channels;

� People who “self-refer” and travel at their own expense (and without

any official record of the reason for travel) for off-island medical

treatment; and

� People who travel for non-medical reasons (holiday, employment,

visiting relatives etc).

� � �

115

The last two categories are extremely difficult to measure as there is often no official

record of reason for travel. Anecdotal evidence from contacts at health departments

across the region suggest most long term residents who die overseas are returned

home for burial due to cultural affiliations and land issues; however there is often no

accessible record of these deaths (described in Chapter three). Where recorded,

these deaths could be included in the calculations, as with any other event in the

resident population. Where data are not available, deaths in the first category could

be used as “proxy events” to signal the level of impact off-island deaths may likely

have on total mortality and causes of death.

Deaths are most likely to occur amongst travellers referred overseas through official

channels for medical treatment due to the severity of their illness or condition. In

theory, data on these events should be captured through government reporting

systems (since the government has funded the travel and in some cases, the

treatment). In practice, many countries do not capture information on these deaths,

or are unable to collate these data with their primary death reporting processes (see

Chapter three).

All PICTs have a medical referral program, although the extent and destination

country vary. Primary destination countries include Australia, New Zealand, the USA

(primarily the Pacific Islander Health Care program at Tripler Army Hospital in

Hawaii) and the Philippines. Fiji is also an important referral destination for

surrounding PICTs. More recently, rising costs in traditional destination countries and

the burgeoning medical tourism market have seen India and Thailand added to the

list of potential destinations (used by Tokelau and Tuvalu respectively). Although

� � �

116

exact numbers of referrals are difficult to obtain, these account for up to 20% of the

health budget in some PICTs (36). Major patient referral pathways, updated from

earlier work by Pryor et. al. (300) are shown in Figure 10.

Figure 10: Major medical referral pathways in the Pacific Islands

(adapted from Pryor, 2000 (300))

Registered deaths of Pacific Islanders in Australia for less than two years are shown

in Table 16.

�

�

�

�

Samoa

Wallis�&�Futuna

American�Samoa

French�Polynesia

Cook�Islands

Niue

Tonga

Fiji Tuvalu

Kiribati

Nauru PNG

Solomon�Islands

Vanuatu

New�Caledonia

Australia

New�Zealand

Guam

F.S.M

Palau

Tokelau

Hawaii

C.N.M.I.

RMI

Philippines

France

Polynesia

Micronesia

Melanesia

� � �

117

Table 16: Registered deaths in Australia of new residents from the Pacific Islands by age at death and duration of residence: 2001-2008

Duration of residence <1 year

Duration of residence 1 to <2 years

Country of birth 0-14 years

15-54 years

55+ years Total

0-14 years

15-54 years

55+ years Total

New Caledonia 6 24 24 54 4 11 6 21 Fiji np np 35 54 0 7 21 28

Total Melanesia, Micronesia and Polynesia 11 64 83 158 5 32 49 86 np not available for publication but included in totals where applicable, unless otherwise indicated

Data provided by ABS (unpublished)

The bulk of referred patients to Australia from the PICTs, and subsequently deaths

as shown in the table, come from New Caledonia due to agreements between the

Australian and New Caledonian governments regarding treatment costs. Fiji is the

only other sending country for which there were more than 5 cases in any age group

(the ABS criteria for disaggregation). The table shows most deaths were of people

aged greater than 55 years, where there would be minimal impact on life tables and

LE. A substantial proportion (39%) also occurred in younger adults, which would

have some impact on estimated deaths rates, with (7%) of these deaths in children.

These data are noted to be affected by issues of under-reporting (due to poor

capture of duration of residence) and reflect ethnicity rather than country of previous

residence. Although the total number of deaths of interest is likely to be higher, these

data provide an indication of the age groups that may be most affected. Similar data

was not available by age group for New Zealand. Deaths in Australia and New

Zealand for 2001-2008 of visitors (non-residents) of Pacific Islander background are

shown in Table 17. As with Table 16, these figures are likely to be affected by

� � �

118

difficulties capturing usual place of residence and therefore may be somewhat

under-representative. They do however provide an indication of the scale of the

issue. The average number of deaths in Australia and New Zealand reported for

non-residents (visitors) from the PICTs was 42 in Australia and 38 in New Zealand.

On average, a further 30 deaths could be expected per year in Australia amongst

those who were recorded as having resided in Australia for less than two years

(short term residents) (Table 16). Excluding New Caledonia, this equates to an

average of 46 deaths reported deaths amongst visitors and new residents in

Australia who were born in the PICTs (across all PICTs).

While there are problems with recording residence that mean this data could be

under-reported, as country of origin is only available as country of birth, this data is

likely to include a substantial number of visitors and immigrants from New Zealand

who were born in the PICTs. The average number of deaths per year of those who

have arrived directly from the PICTs would be smaller again and the number of

deaths this equates to for each PICT would be negligible.

� � �

119

Table 17: Deaths in Australia and New Zealand of persons usually resident overseas by PICT of birth: 2001-2008

Country or Territory of Birth

Australia New Zealand

New Caledonia 137 1 Papua New Guinea 35 Fiji 84 56 Solomon Island 4 Vanuatu 7 Kiribati 6 5 Nauru 13 Cook Islands 5 28 French Polynesia 9 39 Samoa 15 68 Tonga 12 36 Wallis and Futuna 8 1 Niue 7 Tokelau 2 Polynesia 339 243

Data supplied by ABS (301), and New Zealand Statistics (302)

Most deaths of non residents in New Caledonia between 2000 and 2007 were from

the French territories (55%) and mainland France (21%). All other nationalities (not

just PICTs) contributed to only 24%, or between 2-5 deaths per year (303). The

overall impact of these deaths on national reporting for PICTs would therefore be

negligible when spread across the sending countries.

While off-island deaths are likely to affect all mortality estimates to some extent, the

greatest impact will be to skew age-specific measures such as IMR and causal

patterns of death. High risk pregnancies are more likely to be referred off-shore, and

these births represent the greatest risk of early neonatal and maternal deaths.

� � �

120

Although probably representing an extremely small number of total events, if the

outcomes of overseas referrals are not captured, the rarity of these events could

have a significant impact on underestimation of key measures. Similarly, referral

scheme criteria would result in a greater likelihood of specific causes of death in

overseas referrals.

For example, of the American associated territories, only Guam and to a limited

extent CMNI, have chemotherapy treatment available on-island. None have

radiological treatment available (304). Therefore, cancer patients are more likely to

be referred overseas for treatment than a patient with diabetes for example; which

may result in specific causes being under-represented in the national statistics. A

2000 review of the Guam cancer registry demonstrates the biases that may arise

from overseas referrals, both in terms of total number of cases (and subsequent

deaths) and the types of cancer recorded. As a receiving destination for cancer

treatment, the Guam cancer registry records cases of cancer diagnosed on Guam

for both residents and non-residents. For 1995-99, 10% of cases were non-

residents, primarily from other Micronesian PICTs. There were 66 cases of cancer

recorded in Micronesians who were non-residents in Guam between 1970-2000, an

average of 2.2 cases per year. Of these non-resident cases, 55% were Chamorro

(primarily resident in CNMI), 30% were from FSM, 14% from Palau and 1% from the

Marshall Islands. Thirty three percent of cases were for cancers of the digestive

tract, with a further 6% recorded as cancers of the mouth and pharynx. Respiratory

cancers accounted for 27%, cancer of the female genital organs was 8% and breast

cancer was 6% of cases diagnosed (305, 306). These figures demonstrate the type

of cases that may not be adequately captured in the source country data. Although

� � �

121

overall they represent a very small number of events, they represent both common

and rarer causes of death, such as mouth cancer.

While it is difficult to obtain reliable figures by type of illness or treatment required,

the two major receiving centres in Hawaii (Tripler and Shiners) report the majority of

patients seen under the Pacific Referral Program are treated by the surgical

departments; primarily for cardiac care, orthopaedic procedures, cancer treatments,

emergency care and paediatrics (305-308).

While an important issue, the proportion of cases referred overseas has decreased

over recent years for many countries due to the costs involved, financial pressure on

governments (309-311). The rise of telemedicine, including rapid advances in

imaging (312, 313), has also allowed a greater number of conditions to be remotely

treated with expert advice.

The findings presented here indicate off-island events should have minimal impact

on estimates of mortality level, and errors introduced from these occurrences are

likely to be of smaller magnitude than the uncertainty in these figures due to the

small population sizes. Estimates of mortality level will therefore not be adjusted, but

will be reported with uncertainty analysis wherever possible. In comparison, while

off-island events may have little impact on cause-specific rates for some broad

categories, the patterns of referral may have significant impact for specific causes.

As there is no reliable way to correct for these biases without adequate data on

medical referrals and outcomes, estimates will be made primarily for broad cause

� � �

122

categories. Reporting for specific causes should be considered for plausibility in light

of a system review identifying whether systematic biases in the findings are likely.

4.3.1 Off-island births

Off-island births are important, as they may be excluded from denominators for

mortality measures if the birth is not reported in the family’s country of residence. If

these births are not recorded (or reported too late for inclusion in current data sets)

both infant and childhood mortality may be inflated from true levels if births are used

as a denominator. Travelling overseas to give birth is common in several PICTs,

including Palau (with many deliveries occurring in birthing centres in Guam or the

United States) and the Cook Islands (with medical services sought in New Zealand

(314)). The majority of off-island births are arranged privately rather than through

official medical referral programs and as such, there is no simple way of tracking the

number that may be included in this category.

Overseas births affect not only the denominators of IMR and U5M, but also the

numerators, as they tend to include a higher proportion of high-risk births (hence the

referrals) and may therefore be more likely to be associated with infant deaths. As

discussed for overseas deaths generally, changes to numerators could potentially

have a significant impact on mortality estimates. It is therefore important to capture

these events wherever possible. Changes to denominators would have less impact,

and would only be likely to have a substantial effect where populations are extremely

small and overseas referrals common (such as in Niue or Tokelau), where primary

data capture of these events is critical to generating reliable mortality estimates. In

most of the PICTs, the impact is likely to be much less substantial. Although the

� � �

123

Cook Islands, for example, is considered to have a high proportion of overseas births

due to both the government referral program and self referral (as Cook Island

residents qualify for government health care in New Zealand), overseas births are

estimated to be less than 5% of total births to resident women (314). The Cook

Islands reported an IMR of 7.8 deaths per 1,000 live births in 2009 (315), based on

255 reported births. If 5% of all births were off-island and not reported, the total

estimated births would be 268.42 and IMR would be 7.4 deaths per 1,000 live births.

Given IMR ranged from 3.8 to 21.4 over the period 2005 to 2009, the error

introduced by off-island births is negligible when compared to the stochastic error

due to the small population size and is therefore not a significant problem. Deaths

associated with overseas births would remain a greater issue.

4.4 MigrationAs noted at the introduction to this section, changes to the numerator (through

deaths happening off-shore for example) would have the greatest effect on mortality

rates given the comparative size of the small number of events and the population.

Nevertheless, changes to the denominator (population) through migration would also

potentially have an impact on mortality rates if:

a) the proportion of the population that is either removed (through out-

migration) or added (through in-migration ) to the denominator is different to the

proportional change in the number of deaths. That is, if the migrant population was

significantly more or less likely to die than the background resident population; or

� � �

124

b) population estimates used in the denominator did not adequately

account for the extent of migration. This may occur if migration patterns have

changed substantially since the previous inter-censal period.

Migration figures show three distinct country groupings. New Caledonia and Guam

both have net in-migration (Table 18). The high rate of migration into Guam in 2010

(13 per 1,000 population) is a substantial increase from previous years and reflects

the significant up-scaling of US defence force personnel in the territory, while in

migration to New Caledonia is largely driven by the mining boom. In both cases,

migrants entering the country are more likely to be healthy individuals moving for

work, and could therefore be expected to have a lower mortality rate (at entry) than

the broader resident population. In-migration would therefore be expected to

increase the population denominator to a greater extent than it would contribute

additional deaths to the numerator. Thus decreasing calculated mortality rates and

potentially masking health concerns particularly within sub-groups of the population

who may have different health and social characteristics than represented in the

influx of migrants. Correlating trends in summary measures of mortality against

changes in migration would be important in interpreting mortality estimates for these

countries.

� � �

125

Table 18: Migration rates by Country for 2010 (110)

Migration Category and Country

Crude net migration rate (‰) per 1000 pop

Number of net migrants

Annual population growth (%)

Net In-migration New Caledonia 4.65 1173.33 1.55 Guam 13.00 2400.00 2.67 Net Out-migration Niue -28.14 -42.11 -2.31 Marshall Islands -18.43 -1000.00 0.69 Samoa -16.68 -3050.00 0.30 Tonga -16.58 -1711.36 0.33 Tokelau -16.22 -18.91 -0.18 Northern Mariana Islands -15.85 -1000.00 -0.06 Federated States of Micronesia

-14.69 -1632.65 0.42

Wallis and Futuna -13.16 -175.00 -0.64 Tuvalu -8.78 -97.67 0.51 Fiji Islands -7.67 -6488.89 0.46 American Samoa -7.10 -465.12 1.20 Cook Islands -6.30 -97.61 0.32 Negligible migration – Kiribati -1.00 -100.00 1.85 Papua New Guinea 0.00 0.00 2.13 Solomon Islands 0.00 0.00 2.69 French Polynesia 0.00 0.00 1.16 Vanuatu 0.00 0.00 2.54 Palau 0.00 0.00 0.59 Nauru 0.00 0.00 2.08

The largest group of PICTs have net out-migration. This includes Niue, Marshall

Islands, Samoa, Tonga, Tokelau, Northern Mariana Islands, Federated States of

Micronesia, Wallis and Futuna, Tuvalu, Fiji Islands, American Samoa and the Cook

Islands (Table 18). All have out-migration rates higher than six per 1,000 with eight

PICTs having rates of more than 13 per 1,000 population.

� � �

126

The effect of out-migration is to decrease population denominators, the result of

which is greater uncertainty around estimated rates. If the migrating population is

predominantly healthier than average (i.e. those with a high level of education and

income and most likely to quality for visas or employment etc), the number of deaths

is less likely to fall as people move away, thus artificially increasing calculated

mortality rates and potentially over-stating health concerns. Of course, if people

leaving the country through out-migration have a similar or higher risk of death than

the population average (people with illness going to receive treatment elsewhere for

example as discussed in the previous section), then the numerator (number of

deaths) would be reduced in proportion to the smaller denominator and an inflation

of mortality rates would not be expected. In countries where there are no reciprocal

residency arrangements in place such as in Tonga, Samoa, and Fiji etc the former

scenario is far more likely as people need to meet visa and employment conditions in

their new countries, whereas outmigration from countries such as the Cook Islands

and Niue which have automatic residency rights in New Zealand, or American

Samoa with residency rights in the US are more likely to see a broader spectrum of

migrants and therefore a smaller “inflationary” effect on mortality rates.

As under-reporting of deaths is likely in these countries (Chapter three), the impact

of migration on mortality estimates is likely to be far less significant than the under-

reporting of deaths. One is likely to completely cancel the other, resulting in empirical

estimates without correction that form the lower boundary of a “plausible envelope of

mortality”. For example, approximately 4,000 to 6,000 people have migrated from Fiji

each year between 2001-2009 (110). This equates to approximately 2.5 to 7.5% of

the population. Ideally, mortality estimates could be adjusted for the impact of

� � �

127

migration, however, this would require much more detailed data on migration

patterns by age group than is available for the study countries (given migration is not

consistent by age and tends to be highest in young adults and their children) (66). In

the absence of more detailed migration data, it is not possible to exactly calculate the

potential impact on mortality rates. An indication could be made by adjusting

denominators of calculated mortality rates to increase the population based on

migration rates and making the (very poor) assumption that age differences are

minimal. In the case of Fiji, with a moderate level of migration (as above), increasing

the denominator by 7% across all groups would give a LE estimate of 65.8 for males

and 70.0 for females for 2002-2004, compared with the estimated LE of 64.3 (64.0-

64.6) and 69.1 (68.8-69.4) calculated from empirical data and presented in case

study three (Chapter five). As seen in this example, the difference in estimated

population resulted in estimates nearly two years higher for males and one year

higher for females. This is slightly outside the calculated confidence intervals,

although is without any adjustment for undercount in the Fijian data; as would be

expected in collections for most other countries. For most of the other study

countries, greater adjustments were made for undercount in the CRVS system as

presented in Chapter five (including Tonga which has the highest out-migration of

the selected study countries). As such, the unadjusted figure would provide a

plausible lower limit to mortality levels, even allowing for the impacts of migration.

The results show while migration should be considered in the interpretation of results

for countries where migration is moderate or high, there is insufficient data to

accurately correct estimates for the effect of migration and insufficient impact to

warrant doing so in the absence of detailed data. The greater effect of migration is

therefore in limiting the analytical options available for assessing and correcting

� � �

128

mortality data, as methods may assume closed or stable populations such as the

Brass Growth Balance method and capture-recapture method.

Net migration was negligible in seven countries: Kiribati, Papua New Guinea,

Solomon Islands, French Polynesia, Vanuatu, Palau, and Nauru. Migration was -1

per 1,000 population for Kiribati, one of the larger populations, while remaining

countries were reported as zero. At a national level, overall migration should

therefore have minimal impact on mortality estimates and allows many of the

methods to correct under-reporting to be applied. Caution must still be applied at the

sub-national level of estimation for the larger countries of this group, including

Kiribati, PNG, Solomon Islands and Vanuatu, where there is substantial internal

movement from rural areas and outer islands to the urban centres (50, 61, 133, 135).

4.5 Conclusions Direct calculations of mortality (from empirical data) can be affected through impacts

on both the numerator and denominators of mortality measures, both in terms of

overall numbers (summary measures) and through the introduction of biases in the

data, particularly in relation to age group and causes of death. Deaths that occur off-

island and subsequently not counted in the national statistics may reduce mortality

rates. In the few PICTs that receive people travelling for medical treatment such as

Fiji, New Caledonia and Guam, deaths from this non-resident population may

artificially increase rates, biasing the results, particularly for specific causes of death

which reflect the treatments available. Movement of large numbers of people through

migration changes the denominators for mortality rates and if large enough, may

impact summary measures of mortality. Net migration away from the country may

� � �

129

reduce denominators and increase rates. Migration may also have more subtle

effects on mortality estimates by introducing biases due to the selection criteria

generally applied to immigration applications.

In addition to the impacts on direct calculation of measures of mortality, population

movement defines the scope of methods of estimation available to derive mortality

estimates where empirical data is either not available or not complete.

The work presented in this chapter demonstrates the importance of data aggregation

over time, although this alone is not sufficient to generate reliable estimates for

disaggregated data in many of the smaller countries. Summary measures of

mortality, such as LE are therefore critical in understanding mortality profiles in the

smaller countries, and data for all countries should be aggregated over several years

to improve reliability. Even in the larger populations of Fiji and PNG, where

population sizes are sufficiently large to generate annual estimates of mortality

based on Begg’s calculations (293) this would assume complete reporting which is

unlikely, and a minimum aggregation of 3-5 years is recommended. For smaller

countries, collapsing age groups from 5 years and older to ten year groupings (0-4,

5-14, 15-24...etc) or even to broad groups of 0-4, 5-14, 15-59 and �60 may be

required to generate more robust estimates.

While it is not practical to capture all off-island events, it would be possible in most of

the study countries to count vital events (births and deaths) in the high risk medically

referred population in national statistics. The significant impact of these events on

mortality indicators, particularly for cause of death, would highlight a need to do so.

� � �

130

In countries that already count these events to some extent (Nauru and Palau),

these will be included (where possible) in the analysis in the following chapters.

Nauru and Palau are in fact the smallest of the study countries, where the off-island

events are likely to have the greatest impact on estimates. In the remaining study

countries where these events are not currently captured, their impact on crude and

most age-specific mortality level estimates is likely to be declining due to the

increase of telemedicine and cost of patient referrals leading to reductions in these

programs, and small number of events. While capture of these events should be

encouraged for future data collections, in overall numbers, they are likely to have

less impact than stochastic variation and under-reporting (as outlined in Chapter

three). Where these events are more likely to have a significant impact is in the

estimation of infant and child mortality, and cause-specific mortality measures

(including MMR) which are affected by differential referral patterns. Findings should

be considered in light of the referral program, and interpreted with care when

discussing implications and trends.

Although potentially of less impact than changes to the numerators of mortality

estimates from off-island deaths, migration remains a potentially important influence

on mortality measures in several countries. With limited data on migration, both from

the outbound and destination countries, there is limited adjustment that can be made

to direct calculations of mortality to account for this, and improvements in data

capture are required. Methods to evaluate and correct mortality level and CoD data

can be carefully selected to ensure the most suitable method for the context is used.

Selection of the population of interest may also be possible to limit the effect of

� � �

131

migration. For example, deaths in Nauru have long been separated by Nauruan vs.

non-Nauruan ethnicity, due to the much greater mobility in the expatriate community.

Assessment of plausibility is critical in working with the data limitations identified.

One potential source of information to help make this assessment is mortality

estimates by ethnicity within receiving countries such as for New Zealand (316-318),

and these should be reviewed for comparison wherever possible.

� � �

132

5 Assessing mortality level

5.1 System evaluation findings and analytical strategies to assess

mortality level

Study countries can be grouped into three broad categories based on the system

assessments in Chapter three: those expected to have complete or near complete

reporting of deaths; those with partial reporting that may be able to be corrected

sufficiently to generate estimates from the empirical data; and those where data

collection is mostly absent and insufficient for use in establishing measures of

mortality. The analytical strategies selected for each country are dependent upon

which category the country falls into, whether more than one data source is

available, and whether the population is affected by high levels of migration or off-

island events. Approaches for deriving mortality measures for each of these

� � �

133

categories have been tested in this chapter and developed into the following flow

diagram (Figure 11).

Figure 11: Recommended analysis approach for assessment of mortality level in the Pacific Islands

The system assessments described in Chapter three identified the small countries,

Nauru and Palau, were likely to have complete or almost complete recording of

deaths due to their small population size, relatively limited geographical distribution,

social incentives for reporting and reporting processes. Estimates of mortality can

thus be derived directly from the routinely collected data; provided care is taken

where more than one collection exists to select the most complete (as in Palau) and

Single data source only

� � �

134

to avoid over interpretation of data generated over short periods of time due to

stochastic variation. Mortality estimates presented in this chapter for Palau were

derived from the medical certificate data collated by the MoH Epidemiologist, as this

is the most timely, complete data source. Analysis of data from Nauru is presented in

case study two and demonstrates the critical importance of evaluating mortality

data in small populations in the context of long term trends to avoid over

interpretation. While no data “correction” is required before analysis for PICTs in this

category, resultant estimates should be critically considered for consistency with

expected values based on long term trends, comparable overseas values and

internal consistency between results. Frameworks for assessing data quality, such

as outlined by Rao et. al. for use in China (179) or the more recent work by UQ HIS

Hub, may be useful for this purpose. Other PICTs reasonably expected to be

included in this first category, with reasonable mortality level data sources, are the

Cook Islands, Niue, CNMI and Guam (319).

Although routinely collected data for Fiji, Tonga and Kiribati, are considered less

likely to be fully enumerated than for the small islands, there are sufficient data

available to warrant further examination of completeness to determine if corrected

estimates of mortality could be derived. A range of methods for assessing and

correcting incomplete data are outlined in Chapter two. While there are strengths

and weaknesses to each of these, two are considered particularly useful in the

PICTs, as demonstrated in this chapter.

The analysis for Fiji in case study three presents the results of the Brass Growth

Balance analysis of completeness. As discussed in Chapter two, the Brass Growth

� � �

135

Balance method requires less data than other direct demographic methods that

require analysis to match an inter-censal period (131), and provides a useful

screening tool for level of completeness. As with the small island countries with

good data, derived mortality estimates should be considered in light of longer term

trends derived from plausible estimates. The Brass Growth Balance method was

also applied to assess data completeness from unit record sources in Kiribati

(Appendix one). Results from this assessment showed the data sources were too

incomplete to be reliably adjusted in this manner.

The second method of assessing the completeness of reported deaths presented in

this chapter is capture-recapture analysis. Strengths and weaknesses of this are

examined through two case studies: firstly application of the method in an ideal

island setting in Bohol in the Philippines (case study four), where there are three

good quality data sources; then in Tonga, which shares many of the strengths of the

Bohol study (case study five). Although two source capture-recapture analysis

should not be used to generate definite estimates of all-cause mortality, it may be an

effective means of setting boundaries around the plausibility of estimates produced

by indirect estimation where there are insufficient data to generate estimates directly

from empirical sources. This approach is explored further for Kiribati, where

complete three source analysis has not been possible. Instead a two source capture-

recapture analysis is used to examine existing mortality estimates.

A system assessment would be required to determine other PICTs could be included

in this category of having some data that could be used to improve current estimates

� � �

136

of mortality level. This could reasonably be expected to include Samoa, which

operates both health reporting and civil registration systems (319).

Finally, the system analysis presented in Chapter three identified that reporting in the

remaining study country, Vanuatu, was far less likely to be complete, and deaths

would subsequently be significantly under-enumerated. The civil registration of

deaths is virtually non-functional in Vanuatu and the health system captures only

around 60% of the deaths expected according to census estimates. Routinely

collected data are therefore considered too incomplete to be used as the basis for

estimates of mortality level. Coverage was also considered insufficient to be

representative at a national level. In larger countries, and where sufficient data exists

to do so, it may be possible to focus on smaller geographical areas to develop

estimates at a sub-national level using the techniques described for the previous

category. Otherwise, other sources of mortality data, such as census data and

survey data, must be used. Estimates from indirect methods and recall methods

employed in census and survey analyses must be assessed for validity in light of

what is known for other PICTs to determine the plausible range of estimates. For

example, it is unlikely Vanuatu, with more limited health facilities, would in 2010 have

had lower infant or child mortality than a non-malarious country such as Fiji with

more universal access to health care (108), while it is entirely plausible there would

be lower cause-specific mortality from NCDs, consistent with Vanuatu being at an

earlier stage of the epidemiological transition (2). Both PNG and the Solomon

Islands would also be expected to also fall within this category with incomplete

routine reporting of deaths (319).

� � �

137

5.2 Data aggregation and trend analysis

As discussed in Chapter four, population sizes in the PICTs are small and thus

stochastic variation can have a significant impact on mortality rates. It is therefore

critical that data are aggregated over multiple years to provide greater stability and

reported with appropriate statistical confidence intervals.

For many countries, aggregation alone is insufficient to reach the population of

852,900 person years required to generate reliable estimates by age, sex and cause

as cited by Begg et. al. (293). In Nauru for example, this level of certainty would

require aggregation of 88 years of data (Table 15), a situation that is neither

plausible in regard to data availability nor useful for planning and policy. Therefore,

while aggregation remains important, another approach is needed to assess the

plausibility of measures of mortality level from the smaller PICTs, even when

complete data is available. A review of long term trends and how these relate to the

most recent estimates is one way of providing this context, as demonstrated in case

study two for Nauru.

� � �

138

5.2.1 Case Study 2: Mortality Trends in Nauru.

Carter KL, Soakai TS, Taylor R, Gadabu I, Rao C, Thoma K, Lopez AD. Mortality

trends and the epidemiological transition in Nauru. Asia Pac J Public Health. 2011

Jan;23(1):10-23

This study provides an example of the use of data aggregation and assessment of

long term trends in understanding mortality levels in a small population.

Age-specific mortality rates, as shown for 2002-2007 in Figure 12 demonstrate a

plausible pattern of mortality level (320) albeit with significantly higher mortality in

males across all adult groups up to 65 years of age.

Figure 12: Age-specific mortality for Nauru 2002-2007, by sex

As demonstrated in this case study, neither childhood mortality nor adult mortality

(for males and females) has improved over a period of approximately 50 years.

0.1

1

10

100

1000

0�4 5�14 15�24 25�34 35�44 45�54 55�64 65�74

Mortality�rate(deaths�per�1,000�pop)�

Age�group

Males

Females

� � �

139

Despite the lack of improvement in child mortality, U5M is sufficiently low (46.0

deaths per 1,000 live births) that this is not the primary factor influencing the low LE

estimated in this study. LE at birth for 2002-2007 is estimated as 52.8 years for

males, and 57.5 years for females.

LE for males remains one of the lowest in the region, with decreases in mortality

from external causes offset by increases in NCDs (112). Adult mortality (15-59

years) (at 510 deaths per 1,000 population for males and 401 deaths per 1,000

population for females) is three to four times higher than in Australia or New

Zealand. These levels of LE at birth are comparable with South Africa, which has a

high mortality from HIV/AIDs (321). Life tables from Nauru which are not included in

the paper are presented in Appendix two. Although not specifically reported in the

paper, the plausibility of the data across all broad age groups suggest reported

deaths could also be used to measure IMR with a reasonable level of reliability

across this period. IMR was subsequently estimated from the U5M using a

regression analysis based on international data as 28 deaths per 1,000 live births

(95% CI: 12 – 55 deaths per 1,000 live births).

http://aph.sagepub.com/Asia-Pacific Journal of Public Health

http://aph.sagepub.com/content/23/1/10The online version of this article can be found at:

DOI: 10.1177/1010539510390673

2011 23: 10Asia Pac J Public HealthAlan D. Lopez

Karen Carter, Taniela Sunia Soakai, Richard Taylor, Ipia Gadabu, Chalapati Rao, Kiki Thoma andMortality Trends and the Epidemiological Transition in Nauru

Published by:

http://www.sagepublications.com

On behalf of:

Asia-Pacific Academic Consortium for Public Health

can be found at:Asia-Pacific Journal of Public HealthAdditional services and information for

http://aph.sagepub.com/cgi/alertsEmail Alerts:

http://aph.sagepub.com/subscriptionsSubscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

http://aph.sagepub.com/content/23/1/10.refs.htmlCitations:

What is This?

- Dec 17, 2010Version of Record >>

at UQ Library on February 4, 2013aph.sagepub.comDownloaded from

http://aph.sagepub.com/

http://aph.sagepub.com/content/23/1/10

http://www.sagepublications.com

http://www.apacph.org/site/index.php

http://aph.sagepub.com/cgi/alerts

http://aph.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.com/journalsPermissions.nav

http://aph.sagepub.com/content/23/1/10.refs.html

http://aph.sagepub.com/content/23/1/10.full.pdf

http://online.sagepub.com/site/sphelp/vorhelp.xhtml


Asia-Pacific Journal of Public Health23(1) 10 –23

© 2011 APJPHReprints and permission: http://www. sagepub.com/journalsPermissions.nav

DOI: 10.1177/1010539510390673http://aph.sagepub.com

Mortality Trends and the Epidemiological Transition in Nauru

Karen Carter, BAppSci (Env Health), MPHTM1, Taniela Sunia Soakai, Dip Health Admin, MBA2, Richard Taylor, PhD, FAFPHM1,3, Ipia Gadabu4, Chalapati Rao, MBBS, PhD1, Kiki Thoma, DSM2, and Alan D. Lopez, PhD, HonFAFPHM1

Abstract

This article aims to examine the epidemiological transition in Nauru through analysis of available mortality data. Mortality data from death certificates and published material were used to construct life tables and calculate age-standardized mortality rates (from 1960) with 95% confidence intervals. Proportional mortality was calculated from 1947. Female life expectancy (LE) varied from 57 to 61 years with no significant trend. Age-standardized mortality for males (15-64 years) doubled from 1960-1970 to 1976-1981 and then decreased to 1986-1992, with LE fluctuating since then from 49 to 54 years. Proportional mortality from cardiovascular disease and diabetes increased substantially, reaching more than 30%. Nauru demonstrates a very long period of stagnation in life expectancy in both males and females as a consequence of the epidemiological transition, with major chronic disease mortality in adults showing no sustained downward trends over 40 years. Potential overinterpretation of trends from previous data due to lack of confidence intervals was highlighted.

Keywords

mortality, life expectancy, Nauru, cause of death, Pacific Islands

Introduction

Nauru is a small (24 km2) raised atoll in the mid-Pacific. Originally colonized by Germany in 1888, it was administered by Australia, Britain, and New Zealand from 1914 until independence in 1968.1 Phosphate mining began in the early 1900s, but by the early 1990s most deposits had been extracted. Mining ceased entirely between 2003 and 2006,1 before recommencing at a

1University of Queensland, Brisbane, Queensland, Australia2Nauru Ministry of Health, Yaren, Nauru3University of New South Wales, Sydney, New South Wales, Australia4Nauru Bureau of Statistics, Yaren, Nauru

Corresponding Author:Richard Taylor, PhD, University of New South Wales Main Campus, Samuels Building, Level 2, Room 223, Botany St, Gate 11, Randwick, Sydney, New South Wales 2052, AustraliaEmail: [email protected]



Carter et al. 11

significantly reduced scale. The Nauruan (Micronesian) population in 2006 was approximately 9547, along with an expatriate workforce of around 400.2 This expatriate workforce is signifi-cantly reduced from the 1000 or so phosphate workers (primarily from other Pacific Islands) present in 1992.3 Although entirely economically self-sufficient in the early 1990s, Nauru is now almost entirely dependent on foreign aid.1

Nauru gained notoriety following independence due to the considerable revenues from phos-phate mining. Profits were used for running government and providing services, invested in vari-ous enterprises, and paid to land owners.4 Annual per capita income implied from the value of phosphate exports during the 1970s and 1980s was relatively high at approximately US$10 000 to US$20 000.4,5 The ready availability of cash enabled considerable consumption expenditure on imported items such as food, alcohol and sweetened beverages, cigarettes, vehicles, and ste-reo and video equipment. References to adverse effects on health and high accident rates as a result of this affluence were noted beginning in the mid-1970s.3-5 Community surveys revealed an extraordinary high prevalence of diabetes in adults—30% in a 1976 survey6,7 and 24% in 19828—as well as considerable hypertension,9 hyperuricaemia,10 coronary heart disease,11 obe-sity,12 and microalbuminuria.13 Extensive morbidity and mortality14-16 associated with these con-ditions was evident, including diabetic eye disease,17 lower limb amputations,18 and higher mortality in diabetics.19 Obesity was common but not seemingly independently related to mortality.12 Genetic and environment factors are thought responsible for the prevalence of diabetes.17 A decline in the incidence of diabetes during the 1980s may indicate a partial exhaustion of the genetically susceptible.20 Tobacco smoking was frequent21 and related to higher mortality.12 Continuing high levels of diabetes, hypertension, serum cholesterol, obesity, and tobacco and alcohol consumption were demonstrated in a 2004 noncommunicable disease risk factor survey.22

Nauru provides an example of a deviation from the classical epidemiological transition described by Omran,23 in that mortality increased with changes of causes of death to noncom-municable disease and injuries.14-16 This study examines the unique epidemiological transition in Nauru by a secular analysis of mortality data to 2007—from 1947 for cause of death and from 1960 for level of mortality.

MethodsMortality data have only been sporadically available from Nauru, and the present compilation contains gaps where studies or assessments were not available. Data from 1986-1992 and 2002-2007 are reported here for the first time. Analysis is limited to Nauruan nationals because of significant changes in expatriate populations over time. Given the extensive timeframe covered, it is inevitable that there is variation in data sources and collection methods, as outlined below.

Population data for 2002-2007 were obtained by linear interpolation from the Nauruan popu-lations from the 2002 census3 and the 2006 Household Income and Expenditure Survey (HIES).2 The HIES data included people of unknown ages, and populations were redistributed to the age categories according to proportions reported. Population data for 1992-2002 were obtained from data published with the 2002 census.3 Population data for 1986-1992 were obtained by linear interpolation from the 1983 census24 and the 1992 census,25 whereas population data for previous periods were obtained from published sources.14-16 Population data for 1959 and 1972-1974 used in the examination of external causes of death (injuries, poisonings, suicide, homicide, etc) were extrapolated from population data for 1960-197026 based on the average growth rate of the adja-cent 5 years (1960-1964 and 1968-1971, respectively).

Details of the number of deaths for 2002-2007 were extracted (by KC) from death certifi-cates held at the national hospital and the National Registration Office, which receives the certificates from the hospital. Nauruan deaths were identified by recorded ethnicity on the death



12 Asia-Pacific Journal of Public Health 23(1)

certificates. Details of sex, age at death, year of death, and causes of death were collected. Details of Nauruan deaths and mortality rates by age group for the periods 1997-2002 and 1992-1997 were obtained from data extracted from death certificates and registers of the Secretariat of the Pacific Community (SPC).3,27,28 Prior to 2000 there were 2 hospitals in Nauru, and data of the deaths for the years 1986-1992 were extracted from registers at both hospitals in 1993 by 2 of the authors (RT and KT); Details of Nauruan deaths were identified by name since all Nauruan families were known to one of the authors (KT). Variables were abstracted as for later data. Mortality data for periods before 1986 were obtained from published material,14-16,26 with 1976-1981 deaths extracted by RT and KT. For the period 1982-1985, mortality rates for ages less than 15 years were not available,16 and these were interpolated using 1976-198114-26 and 1986-1992 data. For all periods prior to 1986, the oldest open age group was more than 65 years.

Age-specific mortality rates, for age groups of 0-4 years then decades (because of small num-bers) to 74 years for each sex, were obtained by dividing deaths by population for each of the peri-ods analyzed. The mortality rate for age group 65 to 74 years before 1986 was estimated by linear extrapolation of the logarithm of the age-specific death rate from the age range of 35 to 64 years.

Data from 1992-1997 and 1997-2002 were published as 5-year age groups.3,27,28 These were converted to decade age groups for consistency with the other periods. Childhood mortality (0-4 years) was calculated for both sexes combined to reduce stochastic variation due to the small population size using age-specific mortality rates.

Life tables were constructed according to standard methods29 from age-specific mortality rates by sex for each period and life expectancy (LE) at birth reported. Confidence intervals (CIs) for both LE and childhood mortality (probability of dying aged 0-4 years) were calculated according to the Chiang method.30 Age-standardized mortality rates for adults (15-64 years) were calculated using the direct method based on the 2000-2007 population (both sexes). CIs were calculated using the Poisson distribution for all rates due to the small number of deaths, except for age-standardized mortality rates where there were sufficient deaths to use the normal approx-imation of the binomial.31 LE and age-standardized mortality rates were graphed to illustrate trends (Figures 1 and 2). The linear trend in death rates for males for the 3 time periods from 1976 to 1992, adjusted for age, was assessed using Poisson regression31: loge(deaths/population) was modeled on age group and period.

Causes of death for 2002-2007 and 1986-1992 were obtained from death certificates and/or death registers, with data for 1986-1992 supplemented by information from the patients’ files. A single underlying cause of death was determined according to the International Classification of Diseases (ICD), 9th revision,32 to maintain consistency between time periods and reported according to ICD9 chapter headings.

Cause of death data for other periods were obtained from published material,14-16,33 with 1976-1981 data extracted by RT and KT. Data from early periods are given by major cause ICD chapter. Deaths from external causes for the period 1982-1985 were obtained by inflating the deaths ≥15 years using the ratio of all external deaths to external deaths <15 years deaths for both the previous and succeeding periods (inflation factor 1.19).

Proportional mortality by ICD chapter was graphed over time to identify trends. Disease chapters were subsequently grouped further to demonstrate major trends from those that are predominately infectious in nature (infectious diseases, respiratory diseases, and diseases of the digestive tract), major chronic diseases (diseases of the circulatory system, neoplasms, and endo-crine, metabolic, and nutritional diseases), external causes, and all others (Figure 3).

Mortality rates due to external causes (all available years) were calculated for the total popu-lation and for males only where data were available, with Poisson CIs.31 Data for external causes is more specific by nature allowing this analysis, whereas disaggregation of other causes is made difficult by the nonstandard format of the medical certificate of death used in Nauru.



Carter et al. 13

0

2

4

6

8

10

12

14

16

1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Time Period+

Dea

ths

per

1,0

00 p

op

ula

tio

n

Figure 1. Child mortality (deaths per 1000 population), 1960-2007, 0-4 years, both sexesNote: See Table 1 for age-specific mortality and probability of dying by period.+Results were plotted at the midpoint of each time period as follows: 1960-1970 (1965), 1976-1981 (1978.5), 1982-1985 (1983.5), 1986-1992 (1989), 1992-1997 (1994.5), 1997-2002 (1999.5), and 2002-2007 (2004.5).

0.0

5.0

10.0

15.0

20.0

25.0

1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Females

Males

Time Period+

Mo

rtal

ity

rate

per

100

,000

Figure 2. Nauruan age-standardized mortality rate (per 100 000), 1960-2007, 15-64 years, by sex+Results were plotted at the midpoint of each time period as follows: 1960-1970 (1965), 1976-1981 (1978.5), 1982-1985 (1983.5), 1986-1992 (1989), 1992-1997 (1994.5), 1997-2002 (1999.5), and 2002-2007 (2004.5).




Results

Child mortality (deaths <5/1000 population) in Nauru fluctuated over the time periods exam-ined from 6.0 to 10.8; all 95% CIs overlap. In 2002-2007, the child mortality rate was 9.5 deaths per 1000 population, equivalent to a probability of dying (0-4 years) of 46 per 1000 live births (Table 1, Figure 1).

40

45

50

55

60

65

70

1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Males

Females

Lif

e ex

pec

tan

cy (

year

s)

Time Period+

Figure 3. Nauruan life expectancy at birth, 1960-2007, by sex+Results were plotted at the midpoint of each time period as follows: 1960-1970 (1965), 1976-1981 (1978.5), 1982-1985 (1983.5), 1986-1992 (1989), 1992-1997 (1994.5), 1997-2002 (1999.5), and 2002-2007 (2004.5)

Table 1. Child Mortality in Nauru, 1960-2007, 0-4 Years, Both Sexesa

Period Deaths for PeriodDeaths per 1000

PopulationProbability of Dying 0-4

Years per 1000 Live Births (5q0b)

1960-1970 46 7.5 (5.5-10.0) 36.7 (26.3-47.1)1976-1981 45 10.2 (7.5-13.7) 49.4 (35.4-63.5)1982-1985 35 8.6 (6.0-12.0) 41.9 (28.3-55.5)1986-1992c 49 6.0 (4.5-8.0) 29.6 (21.4-37.8)1992-1997 47 6.6 (4.9-8.8) 32.4 (23.3-41.6)1997-2002 70 10.8 (8.4-13.6) 52.1 (40.3-64.0)2002-2007d 62 9.5 (7.2-12.0) 46.0 (34.8-57.1)

aProbability of dying calculated from age-specific mortality rates. Source: 1960-197041; 1976-198123,24; 1982-198525; 1992-19973,42,43; 1997-2002.42,43

b5q0 is the demographic notation for the probability of dying before the age of 5 years.cData extracted by RT and KT in 1993 (previously unpublished).dData extracted and analyzed by authors.



Carter et al. 15

For Nauruan males, LE at birth between 1960-1970 and 1976-1981 declined from 58 to 47 years, and the age-standardized mortality rate (15-64 years) more than doubled, with 95% CIs indicating significant differences (Table 2, Figures 2 and 3). From 1976-1981 to 1986-1992, mortality progressively decreased (Table 2), as evidenced by increasing LE to 50 years in 1982-1985 and 54 years in 1986-1992 and a decline in age-standardized mortality rate (15-64 years; c2

(1) = 15.4, P < .001; Figure 3). Comparison of 1976-1981 to 1986-1992 indicated that the larg-est decline in mortality occurred in the 25- to 34-year age group, with the mortality rate reduc-ing by two thirds; this is statistically significant according to 95% CIs (Table 2). Mortality again increased from 1986-1992 to 1997-2002, with male LE at birth in 1997-2002 falling signifi-cantly (according to 95% CIs) to 49 years,3 before improving again between 1997-2002 and 2002-07 to reach 53 years, a level similar to 1986-1992 but still less than the LE recorded in 1960-1970 (Table 2). The increase in LE between 1997-2002 and 2002-2007 was not statisti-cally significant. Female LE in Nauru has been higher than for males since 1976-1981, but no improvement in female LE occurred between 1960-1970 (57 years) and 2002-2007 (57 years; Table 3, Figures 2 and 3).

Major trends are discernable in causes of death over the past 60 years (1947-2007; Table 4, Figure 4). Infection was a major cause of death in the 1940s and 1950s but decreased signifi-cantly thereafter. Proportional mortality from respiratory diseases and diseases of the digestive tract, often substantially associated with infectious causes, show a similar pattern, although the decline is less dramatic.

Cardiovascular disease and diabetes increased from a small proportion of deaths to comprise 20% to 30% of deaths from the 1970s. Although deaths from endocrine and metabolic diseases (including diabetes) rose from 5% in 1976-1981 to almost 10% in 2002-2007, the proportion of deaths with diabetes recorded anywhere on the death certificate rose from 26% to 36% over the same period. Cancer was not reported in the early periods, but constituted around 10% to 12% of deaths by the 1980s.

External causes increased from less than 10% prior to 1970 to peak at 24% in the mid-1980s, then declined to the current level of 10%. The mortality rate from external causes (mostly in males) was significantly lower in 1986-1992, compared with previous periods (Table 5).

DiscussionAlthough Nauruans are a small population, mortality analysis can be undertaken by aggregation of several years’ data, and 95% CIs assist with interpretation of trends. The mortality data reported in this and previous studies over 4 decades3,14-21,24-28,33 are likely to be complete because of the small number of deaths, all well known to medical staff, and centralized hospital-based death certification. A review of the death reporting systems confirmed very little potential for underreporting of deaths to have occurred. Furthermore, cause of death data should be reason-ably accurate since they emanate from death certificates, supplemented by other relevant infor-mation. Systematic biases in either death counts or population counts are likely only among the highly unstable expatriate community in Nauru, and this was dealt with by restricting the analy-sis to Nauruans only. Records were sufficiently detailed to confidently exclude non-Nauruan deaths. There is reasonable consistency in the estimates over time when individual years were examined within the aggregated periods analyzed. Estimates of mortality from registration data 1992-2002 are congruent with indirect demographic methods of estimation from the 2002 Census.3 The orphanhood method revealed a higher survival of mothers (78%) than fathers (64%), and the widowhood method showed only 19% of men ≥60 years were widowed, whereas 61% of women in the same age group were widowed.3 The recently published Demographic and Health Survey for Nauru gives a 0 to 4 years mortality of 38 (95% CI = 18-57) per 1000 live births for



Tab

le 2

. Mal

e N

auru

an M

orta

lity

(Age

s 0-

74)

and

Life

Exp

ecta

ncy

at B

irth

, 196

0-20

07a

Age

G

roup

(Y

ears

)

1960

-197

019

76-1

981

1982

-198

519

86-1

992b

1992

-199

719

97-2

002

2002

-200

7c

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

0-4

[26]

e8.

0 (5

.3-1

1.8)

2310

.1 (

6.4-

15.1

)[1

8]8.

7 (5

.1-1

3.6)

22

5.2

(3.3

-7.9

) 2

46.

5 (4

.2-9

.6)

45

13.4

(9.

7-17

.9)

39

12.1

(8.

5-16

.3)

5-14

[3]

0.5

(0.1

-1.6

)10

2.8

(1.3

-5.1

)[2

]1.

3 (0

.2-4

.7)

9

1.4

(0.6

-2.6

) 1

21.

9 (1

.0-3

.3)

5

0.8

(0.2

-1.8

)

60.

9 (0

.3-1

.9)

15-2

4[9

]3.

5 (1

.6-6

.8)

226.

9 (4

.3-1

0.4)

2310

.5 (

6.7-

15.8

) 2

24.

9 (3

.1-7

.5)

14

3.2

(1.7

-5.4

) 3

06.

1 (4

.1-8

.7)

13

2.3

(1.2

-3.8

)25

-34

[9]

4.5

(2.1

-8.6

)27

16.7

(11

.0-2

4.4)

1610

.1 (

5.8-

16.4

) 1

85.

5 (3

.2-8

.6)

20

6.4

(3.9

-9.8

) 2

67.

9 (5

.2-1

1.6)

24

6.3

(4.0

-9.3

)35

-44

[15]

9.0

(5.0

-14.

8)23

19.6

(12

.5-2

9.5)

711

.6 (

4.7-

23.9

) 2

512

.9 (

8.3-

19.0

) 4

018

.8 (

13.4

-25.

5) 4

620

.0 (

14.6

-26.

6) 4

216

.1 (

11.4

-21.

5)45

-54

[9]

14.5

(6.

4-26

.7)

2931

.5 (

21.1

-45.

2)16

28.4

(16

.2-4

6.1)

24

29.3

(18

.8-4

3.7)

28

36.3

(24

.1-5

2.5)

57

52.5

(39

.8-6

8.0)

55

33.1

(24

.6-4

2.7)

55-6

4[2

2]49

.0 (

31.3

-75.

5)25

73.3

(47

.4-1

08.2

)20

66.4

(40

.6-1

02.5

) 4

274

.5 (

53.7

-100

.7)

27

56.6

(37

.3-8

2.3)

29

66.7

(44

.3-9

5.1)

36

68.8

(47

.7-9

4.2)

65-7

499

.9{2

6}13

3.1

(86.

9-19

5.0)

{17}

106.

1 (6

1.8-

169.

9) 1

797

.1 (

56.6

-155

.5)

29

154.

7 (1

03.6

-222

.1)

15

72.5

(41

.2-1

21.3

) 1

779

.6 (

45.8

-126

.0)

Age

-sta

ndar

dize

d ra

tef

15-6

464

7.7

(4.2

-11.

2)12

617

.1 (

11.9

-22.

2)82

16.2

(11

.1-2

1.2)

131

11.9

(7.

5-16

.2)

129

12.6

(8.

1-17

.1)

188

16.6

(11

.5-2

1.8)

170

11.8

(7.

5-16

.1)

Life

exp

ecta

ncy

(yea

rs)

At

birt

h58

.1 (

55.7

-60.

5)47

.4 (

45.1

-49.

8)50

.3 (

47.4

-53.

2)54

.0 (

52.0

-56.

1)52

.3 (

50.3

-54.

3)49

.0 (

47.4

-51.

6)52

.8 (

51.1

-54.

5)

Abb

revi

atio

n: C

I, co

nfid

ence

inte

rval

.a S

ourc

e: 1

960-

1970

41; 1

976-

1981

23,2

4 ; 19

82-1

98525

; 199

2-19

973,

42,4

3 ; 19

97-2

002.

42,4

3

b Dat

a ex

trac

ted

by R

T a

nd K

T in

199

3 (p

revi

ousl

y un

publ

ishe

d).

c Dat

a ex

trac

ted

and

anal

yzed

by

auth

ors

2008

.d A

vera

ge d

eath

rat

e pe

r 10

00 p

opul

atio

n pe

r an

num

.e [

] D

eath

s im

plie

d by

pop

ulat

ion

and

deat

h ra

te. {

} D

eath

s ag

ed ≥

65 y

ears

. 200

2-20

07 d

eath

s in

clud

e de

aths

of u

nkno

wn

age

redi

stri

bute

d by

age

gro

up.

f Age

-sta

ndar

dize

d ra

te b

ased

on

2002

-200

7 po

pula

tion

and

by d

irec

t m

etho

d.

16



Tab

le 3

. Fem

ale

Nau

ruan

Mor

talit

y (A

ges

0-74

) an

d Li

fe E

xpec

tanc

y at

Bir

th, 1

960-

2007

a

Age

G

roup

(Y

ears

)

1960

-197

019

76-1

981

1982

-198

519

86-1

992b

1992

-199

719

97-2

002

2002

-200

7c

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)d

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)

Dea

ths

for

Peri

odR

ate

per

1000

(9

5% C

I)

0-4

[20]

e7.

0 (4

.2-1

0.7)

2210

.4 (

6.5-

15.8

)[1

7]8.

7 (5

.1-1

3.9)

276.

9 (4

.6-1

0.1)

236.

8 (4

.3-1

0.2)

258.

0 (5

.2-1

1.8)

237.

0 (4

.4-1

0.4)

5-14

[5]

1.0

(0.3

-2.3

)7

2.1

(0.8

-4.2

)[2

]1.

3 (0

.2-4

.7)

30.

5 (0

.1-1

.4)

40.

7 (0

.2-1

.7)

61.

0 (0

.4-2

.1)

20.

3 (0

.0-1

.2)

15-2

4[1

2]5.

5 (2

.7-9

.2)

82.

6 (1

.1-5

.0)

62.

7 (1

.0-5

.9)

61.

4 (0

.5-3

.1)

61.

5 (0

.5-3

.2)

81.

7 (0

.7-3

.3)

81.

4 (0

.6-2

.8)

25-3

4[1

0]6.

0 (2

.8-1

0.6)

63.

9 (1

.4-8

.4)

85.

1 (2

.2-1

0.0)

61.

8 (0

.7-3

.9)

92.

9 (1

.3-5

.4)

144.

5 (2

.5-7

.6)

123.

3 (1

.7-5

.8)

35-4

4[5

]4.

0 (1

.2-8

.8)

33.

3 (0

.7-9

.6)

59.

6 (3

.1-2

2.4)

199.

7 (5

.8-1

5.1)

2812

.2 (

8.1-

17.7

)28

10.8

(7.

2-15

.6)

2810

.2 (

6.8-

14.7

)45

-54

[18]

29.0

(16

.9-4

5.1)

911

.5 (

5.2-

21.8

)11

20.3

(10

.1-3

6.3)

1720

.5 (

11.9

-32.

8)16

18.9

(10

.8-3

0.6)

3123

.6 (

16.0

-33.

3)55

27.0

(20

.3-3

5.0)

55-6

4[9

]35

.9 (

17.0

-70.

6)10

25.5

(12

.2-4

6.9)

1127

.4 (

13.7

-49.

0)23

37.0

(23

.5-5

5.6)

1939

.6 (

23.9

-61.

9)18

41.6

(25

.0-6

6.8)

2644

.4 (

28.9

-64.

8)65

-74

93.1

{23}

76.5

(48

.5-1

14.8

){1

5}49

.9 (

27.9

-82.

3)16

48.3

(27

.6-7

8.5)

1652

.5 (

30.0

-85.

2)33

126.

0 (8

8.0-

179.

6)33

136.

2 (9

3.5-

190.

7)A

ge-s

tand

ardi

zed

rate

f

15-6

454

10.0

(6.

1-14

.0)

365.

3 (2

.1-8

.2)

419.

1 (5

.3-1

2.9)

717.

3 (3

.9-1

0.6)

787.

9 (4

.4-1

1.4)

998.

9 (5

.2-1

2.6)

129

9.0

(5.3

-12.

7)Li

fe e

xpec

tanc

y (y

ears

)A

t bi

rth

57.0

(54

.2-5

9.8)

59.5

(56

.9-6

2.1)

57.2

(54

.1-6

0.4)

61.5

(58

.9-6

4.0)

60.4

(57

.8-6

2.9)

56.9

(55

.4-6

0.0)

57.5

(55

.9-5

9.0)

Abb

revi

atio

n: C

I, co

nfid

ence

inte

rval

.a S

ourc

e: 1

960-

1970

41; 1

976-

1981

23,2

4 ; 19

82-1

98525

; 199

2-19

973,

42,4

3 ; 19

97-2

002.

42,4

3

b Dat

a ex

trac

ted

by R

T a

nd K

T in

199

3 (p

revi

ousl

y un

publ

ishe

d).

c Dat

a ex

trac

ted

and

anal

yzed

by

auth

ors

2008

.d A

vera

ge d

eath

rat

e pe

r 10

00 p

opul

atio

n pe

r an

num

.e [

] D

eath

s im

plie

d by

pop

ulat

ion

and

deat

h ra

te. {

} D

eath

s ag

ed ≥

65 y

ears

. 200

2-20

07 d

eath

s in

clud

e de

aths

of u

nkno

wn

age

redi

stri

bute

d by

age

gro

up.

f Age

-sta

ndar

dize

d ra

te b

ased

on

2002

-200

7 po

pula

tion

and

by d

irec

t m

etho

d.

17




Table 4. Proportional Mortality by Major Cause Group in Nauruans, 1947-2007, Both Sexes Combineda

Period 1947-1958 1959-1968 1972-1974 1976-1981 1982-1985 1986-1992b 2002-2007c

Deathsd 207 144 114 273 157 317 459Proportional mortality (%) Infection 41 4 15 9 13 7 Cancer 2 8 12 10 13 Endocrine\

metabolic 4 5 14 10 9

Nervous system diseasee

7 13 3 2 2

Cardiovascular disease

11 15 28 21 29 26 22

Respiratory disease 7 10 11 6 5 7 8 Digestive diseasef 7 15 5 6 5 5 Genitourinary

disease 5 7 3 1 3 8

Pregnancy and child birthg

5 3 1

Perinatal conditions 2 5 7 h 6 6 External causesi 4 8 15 20 24 10 10 Other causesj 11 16 44 7 8 7

aMajor cause according to ICD9 chapters. Source: 1947-195823; 1959-196823; 1972-197448; 1976-198123,24; 1982-198525; 1992-19973,42,43; 1997-2002.42,43

bData extracted by RT and KT in 1993 (previously unpublished).cData extracted and analyzed by authors 2008.dIncludes diabetes; allocation of deaths to diabetes is sensitive to the coding criteria employed.eIncludes ear and eye disease.fIncludes liver disease.gThese percentages should be doubled to obtain the approximate proportional mortality for females.h1982-1985 data are for age ≥15 years only.iMost deaths from external causes are in males.jIncludes unknown and ill-defined.

2003-2007 and 41 (95% CI = 22-60) for 1998-2007, based on retrospective maternal history and child survival from a sample of women aged 15 to 49 years in 2007.34 These figures are some-what less than recorded mortality for similar periods for the entire population of 52 (1997-2002) and 46 (2002-2007), but 95% CIs overlap.

There is substantial stochastic variation apparent in all the mortality measures reported here. This demonstrates the importance of avoiding overinterpretation of data from small populations, such as Nauru, by considering mortality measures in light of secular trends and their statistical CIs. Despite aggregation of data over several years, the stochastic variation in these measures has previously led to suggestions of significant recent increases in childhood and infant mortality,3 which are not supported by the analysis presented here.

Changes in proportional mortality for diseases of the circulatory system and endocrine/metabolic diseases (including diabetes) are likely to be affected by certification practices. Although this may have resulted in fewer deaths coded to diabetes, the proportion of deaths with diabetes noted anywhere on the certificate has increased dramatically since 1976-1981, confirming the continu-ing importance of this condition in Nauru. Recent declines in proportional mortality from cardio-vascular disease may largely be attributed to improvements in certification resulting in more specific diagnoses in other chapters. Cause of death was reported by ICD chapter to provide



Carter et al. 19

Time Period+

0

10

20

30

40

50

60

Pro

po

rtio

nal

Mo

rtal

ity

Infectious, Respiratory andDigestive Tract Diseases

Endocrine\metabolic, Cardiovascular diseases and Cancer*

External causes~

All other causes

Figure 4. Trends in major causes of death in Nauru, 1947-2007, all ages, both sexes combinedNote: See Table 4 for numbers of deaths by period. Deaths from infection interpolated for 1972-1974. Unknown and ill-defined included in “All other causes.”*Endocrine deaths not included 1947-1958, and 1959-1968 cancers not listed separately (included in “Other”).~External causes include injury and accidents (including road accidents, falls, burns, etc), homicide, and suicide.+Results were plotted at the midpoint of each time period as follows: 1947-1958 (1952.5), 1959-1968 (1963.5), 1972-1974 (1973), 1976-1981 (1978.5), 1982-1985 (1983.5), 1986-1992 (1989), and 2002-2007 (2004.5).

Table 5. Mortality in Nauruans From External Causes, 1947-2007, All Agesa

Period

Both Sexes Males

Deaths for Period

Rate per 100 000b (95% CIs)

Deaths for Period

Rate per 100 000b (95% CIs)

1947-1958 8 43 (19-85)1959-1968 12 45 (23-79)1972-1974 17 138 (80-221)1976-1981 54 208 (156-272) 42 315 (227-425)1982-1985 [45]c 249 (181-333) [34]c 376 (261-526)1986-1992d 31 71 (49-101) 25 114 (74-168)2002-2007e 45 92 (66-122) 33 134 (92-189)

Abbreviation: CI, confidence interval.aAge adjusted using the indirect method to the 2002-2007 rates as a standard. Confidence intervals calculated according to the Poisson distribution. Source: 1947-195823; 1959-196823; 1972-197448; 1976-198123,24; 1982-198525; 1992-19973,42,43; 1997-2002.42,43

bAverage death rate per 1000 population per annum.c[ ] Deaths less than age 15 years estimated (see Methods).dData extracted by RT and KT in 1993 (previously unpublished).eData extracted and analyzed by authors 2008.

consistency with previously published data14-16 and allow the assessment of trends over time. As Nauru uses a death certificate that is not consistent with international guidelines, aggregating the data by ICD chapter also avoids comparability difficulties for specific causes created by the certificate format.




The major finding of this study is that there has been no sustained reduction in mortality in Nauru over the past half century (1960-2007) in children or adults. For child mortality, neither the decrease from 10.2 deaths per 1000 population in 1976-1981 to 6.0 in 1986-1992 nor the subsequent increase to 9.5 in 2002-2007 was statistically significant. These variations are likely to represent stochastic fluctuations due to the small population.

After an increase from 1960-1970, Nauruan male mortality declined significantly between 1976-1981 and 1986-1992, partly due to a decrease in deaths from external causes, but gains in LE were not sustained. Improvement in the latest male LE (2002-2007) is within expected sto-chastic variation given the population size and is not statistically significant using 95% CIs compared with the previous period. Female Nauruan LE has shown no improvement over the study period with LE in both 1960-1970 and 2002-2007 at 57 years. Fluctuations in LE over this period (between 57 and 61 years) were not statistically significant.

Prior to the 1950s, infectious diseases (particularly tuberculosis) were the major cause of death, and injury accounted for only a small proportion of reported deaths. Hepatitis B, including infection with delta agent,35-37 was prevalent during the 1980s; however, universal immunization has since been implemented.37 As infectious diseases fell, chronic diseases began to increase in importance, and by 1976-1981, diseases of the circulatory system was the most significant cause of all deaths, followed by external causes, although the latter were the major cause of years of life lost.14-16 Diabetes makes a considerable contribution to mortality in Nauru, both directly and as a major risk factor for cardiovascular disease.12,14-16

Despite declines in proportionate infectious disease mortality, male LE (at birth) declined by 10 years from 57.6 years (1960-1970) to 47.2 years (1976-1981). This was the lowest national LE for males in the Pacific Island region during this period.38,39 The decline in LE in Nauru in concert with the epidemiological transition was reported prior to declines in LE in the former Soviet Union and Eastern European countries, also mainly due to noncommunicable disease and injuries in adults.40-43

The decline in male mortality between 1976-1981 and 1986-1992 was led by the reduction in deaths due to external causes in adults in the period 1986 to 1992. Previous analyses indicated that most external cause deaths at that time were motor vehicle accidents (MVA), 5 times that of Australian males for the same period,14,15 and frequently associated with alcohol intoxication, although other injuries, also often alcohol related, make up this category.14,15 Although it was not possible in this analysis to reliably separate MVAs from other external causes, the decline in external cause mortality by more than 50% in more recent periods implies a substantial reduction in MVA mortality. Evidence from key informants (including one of the authors, KT) suggests that driving motor vehicles or riding motor cycles while intoxicated with alcohol has declined substantially from the mid-1980s. This important change in behavior is attributed, in part, to the effect of the clergy, both Catholic and Protestant, who used the funerals of those who died from MVAs to deliver dire warnings from the pulpit on the dangers of drink driving. The awareness of the clergy was no doubt derived from their own observations but probably reinforced by analyses documenting the massive mortality from this cause.14-16 The message to avoid drunk driving would have fallen on fertile ground, particularly among women, who recognized that Nauru had become a “land of widows.”44 There is no evidence that the fall in external cause mortality was due to any legal or policing actions, restrictions on alcohol supply or sale, or organized health promotion interventions—since there were none at that time. Anecdotal evidence is that alcohol consumption did not substantially change during the period of rapid decline in external cause mortality and the loss in economic prosperity and hence purchasing power for both vehicles and alcohol did not occur until later.

Nauru demonstrates unusual aspects of the health transition. First, the emergence of non-communicable disease and injuries vastly exceeded the decline in infectious disease mortality in males, initially leading to a reduction in LE; this phenomenon has now been documented in



Carter et al. 21

other populations.41-43 In females, the decline in infections and maternal mortality was approximately balanced by the rise in noncommunicable diseases and accidents. While stagnation in growth of LE as a consequence of the substitution of infections by noncommunicable conditions occurred in Australia and similar countries around 1950-1970, especially in males,45,46 the severity of the effect in Nauru was much greater and occurred at a lower LE. Declines in LE have been noted in populations of Eastern Europe and the former USSR since the 1960s and continuing through the 1990s, especially in males, as a consequence of cardiovascular disease, cancer, and other tobacco- and alcohol-related conditions and injuries—but at a higher LE than in Nauru.41-43 During the 1990s, Indigenous Australians (particularly males) demonstrated low life expectan-cies compared with the general population, predominately as a consequence of high mortality from noncommunicable disease and injury in adulthood, at comparable life expectancies to Nauru.47-50 Nevertheless, the predominant pattern in the Pacific Island region, as in many other developing countries such as the newly industrializing countries of Asia51 and historically in Western Europe, North America, and Australasia, is for continual improvements in LE despite the emergence of noncommunicable diseases as the major cause of morbidity and mortality, but with initial periods of stagnation in growth of LE.27,28 Additional work is currently underway to examine trends in the epidemiological transition in other Pacific Islands as recent empirical data on mortality levels and cause of death by age group is largely unpublished or unavailable. Second, the Nauru experience demonstrates that external cause mortality in males (the major category being MVAs while intoxicated) can be reduced through actions by civil society in the absence of specific interventions. Third, Nauru demonstrates a very long period of stagnation in LE in both males and females as a consequence of the epidemiological transition, with major chronic disease mortality in adults showing no sustained downward trends over the past 40 to 50 years. In Australia, the period of stagnation in LE for males was 25 years (1945-1970) and for females 10 years (1960-1970)45 as a consequence of increased risks of death from noncom-municable disease and injuries, before progressing to the fourth stage of the epidemiological transition,52 characterized by delayed mortality from noncommunicable diseases to older ages and increases in LE. Evidence from populations such as Nauru, Indigenous Australians, and former socialist economies in Europe suggest that these populations may endure longer peri-ods of stagnation of life expectancy unless effective public health measures are urgently implemented.

Acknowledgments

The authors would like to thank Andreas Demmke from the Secretariat of the Pacific Community (SPC) Statistics and Demography Programme for comments on the article and the staff of the Ministry of Health, the Registrar’s office, and the Bureau of Statistics in Nauru for their assistance.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interests with respect to the authorship and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research and/or authorship of this article:

Funding for this work was provided under an AusAID Australian Development Research Award (Grant no: 44738)

References

1. Department of Foreign Affairs and Trade. Republic of Nauru country brief. http://www.dfat.gov.au/geo/nauru/nauru_brief.html. Accessed December 15, 2008.




2. Bureau of Statistics. Nauru 2006 HIES Report. Yeren, Nauru: Ministry of Finance, Republic of Nauru; December 2006.

3. Demography/Population Programme and Bureau of Statistics. 2002 Nauru Census Main Report and Demographic Profile of the Republic of Nauru 1992-2002. Noumea, New Caledonia: Secretariat for the Pacific Community; 2006.

4. McQuade W. The smallest richest republic in the world. Fortune. December 1975:132-140. 5. Holmes M. This is the world’s richest nation—all of it. Natl Geogr Mag. 1976;150(3):344-353. 6. Zimmet P, Taft P, Guinea A, Guthrie W, Thoma K. The high prevalence of diabetes on a Central Pacific

Island. Diabetologia. 1977;13:111-115. 7. Zimmet P. Epidemiology of diabetes and its macrovascular manifestations in Pacific populations: the

medical effects of social progress. Diabetes Care. 1979;2:144-153. 8. Zimmet P, King H, Taylor R, et al. The high prevalence of diabetes mellitus, impaired glucose tolerance

and diabetic retinopathy in Nauru—the 1982 survey. Diabetes Res. 1984;1:13-18. 9. Zimmet P, Jackson L, Whitehouse S. Blood pressure studies in two Pacific populations with varying

degrees of modernisation. N Z Med J. 1978;91:249-252.10. Zimmet PZ, Whitehouse S, Jackson L, Thoma K. High prevalence of hyperuricaemia and gout in an

urbanised Micronesian population. Br Med J. 1978;1:1237-1239.11. Li N, Tuomilehto J, Dowse G, Virtala E, Zimmet P. Prevalence of coronary heart disease indicated

by electrocardiogram abnormalities and risk factors in developing countries. J Clin Epidemiol. 1994;47:599-611.

12. Hodge AM, Dowse GK, Collins VR, Zimmet PZ. Mortality in Micronesian Nauruans and Melanesian and Indian Fijians is not associated with obesity. Am J Epidemiol. 1996;143:442-455.

13. Collins V, Dowse G, Finch C, Zimmet P, Linnane A. Prevalence and risk factors for micro- and macro-albuminuria in diabetic subjects and entire population of Nauru. Diabetes. 1989;38:1602-1610.

14. Taylor R, Thoma K. Nauruan Mortality 1976-81 and a Review of Previous Mortality Data. Nouméa, New Caledonia: South Pacific Commission; 1983.

15. Taylor R, Thoma K. Mortality patterns in the modernized Pacific Island nation of Nauru. Am J Public Health. 1985;75:149-155.

16. Schooneveldt M, Songer T, Zimmet P, Thoma K. Changing mortality patterns in Nauruans: an example of epidemiological transition. J Epidemiol Community Health. 1988;42:89-95.

17. King H, Balkau B, Zimmet P, et al. Diabetic retinopathy in Nauruans. Am J Epidemiol. 1983;117:659-667.18. Humphrey AR, Dowse GK, Thoma K, Zimmet PZ. Diabetes and nontraumatic lower extremity ampu-

tations. Incidence, risk factors, and prevention—a 12-year follow-up study in Nauru. Diabetes Care. 1996;19:710-714.

19. Balkau B, King H, Zimmet P, Raper LR. Factors associated with the development of diabetes in the Micronesian population of Nauru. Am J Epidemiol. 1985;122:594-605.

20. Dowse GK, Zimmet PZ, Finch CF, Collins VR. Decline in incidence of epidemic glucose intolerance in Nauruans: implications for the “thrifty genotype.” Am J Epidemiol. 1991;133:1093-1104.

21. Tuomilehto J, Zimmet P, Taylor R, Bennet P, Wolf E, Kanakaanpää J. Smoking rates in Pacific Islands. Bull World Health Organ. 1986;64:447-456.

22. Ministry of Health, World Health Organisation. Nauru NCD Risk Factors STEPS Report. Suva, Fiji: Republic of Nauru and World Health Organisation; 2007.

23. Omran AR. The epidemiological transition: a theory of the epidemiology of population change. Mill-bank Q. 1971;49:509-538.

24. South Pacific Commission. Population 1983. Nouméa, New Caledonia: South Pacific Commission; 1985. (Statistical Bulletin No. 26)

25. Bureau of Statistics. Nauru Population Census 1992. Main Report. Yeren, Nauru: Ministry of Finance, Republic of Nauru; December 1993.

26. Palmer R, Trahair G, Owen W, Whittle B. Report to the Chief Secretary on Vital Statistics of Nauru Population. Melbourne, Australia: Government of Australia; 1972.



Carter et al. 23

27. Demmke A, Haberkorn G, Gadabu, Rakaseta V, Beccalossi G. Nauru: A Population Profile. Population/Demography Programme. Noumea, New Caledonia: Secretariat of the Pacific Community; 1999.

28. Demmke A. Nauru Life Tables 1992-1997 and 1997-2002. Noumea, New Caledonia: Secretariat of the Pacific Community.

29. Pollard A, Yusuf F, Pollard G. Demographic Techniques. Sydney, Australia: Pergamon; 1991.30. Chiang CL, Variance and covariance of life table functions estimated from a sample of deaths. Vital

Health Stat 2. 1967;20:1-8.31. Armitage P, Berry G. Statistical Methods in Medical Research. 3rd ed. Oxford, England: Blackwell

Scientific; 1994.32. World Health Organization. International Classification of Diseases, 9th Revision. Geneva, Switzerland:

World Health Organization (WHO); 1977.33 Guinea A. Morbidity, Mortality, and Epidemiological Statistics. Yeren, Nauru: Department of Health,

Republic of Nauru; 1975.34. Gust ID, Dimitrakakis M, Zimmet P. Studies on hepatitis B surface antigen and antibody in Nauru. I.

Distribution amongst Nauruans. Am J Trop Med Hyg. 1978;27:1030-1036.35. Dimitrakakis M, Crowe S, Gust I. Prevalence of delta infection in the western Pacific region. J Med

Virol. 1986;18:335-339.36. Brown P. The seroepidemiology of hepatitis A and B in the Asia-Pacific region. Asia Pac J Public

Health. 1987;1:62-78.37. Taylor R, Lewis N, Levy S. Societies in transition: mortality patterns in Pacific Island populations. Int

J Epidemiol. 1989;18:634-646.38. Taylor R, Lewis N, Sladden T. Mortality in Pacific Island countries around 1980: geopolitical, socio-

economic, demographic, and health service factors. Aust N Z J Public Health. 1991;15:207-221.39. Uemura K, Pisa Z. Trends in cardiovascular disease mortality in industrialized countries since 1950.

World Health Stat Q. 1988;41:155-178.40. Thom TJ. Stroke mortality trends. An international perspective. Ann Epidemiol. 1993;3:509-518.41. Watson P. Explaining rising mortality among men in Eastern Europe. Soc Sci Med. 1995;41:923-934.42. Chenet L, McKee M, Fulop N, et al. Changing life expectancy in central Europe: is there a single reason?

J Public Health Med. 1996;18:329-336.43. Connell J. Nauru: The First Failed Pacific State? The Round Table. 2006;95:47-63.44. Taylor R, Lewis M. The Australian mortality decline: all-cause mortality 1788-1990. Aust N Z J Public

Health. 1988;22:27-36.45. Taylor R, Lewis M, Powles J. The Australian mortality decline: cause-specific mortality 1907-1990.

Aust N Z J Public Health. 1998;22:37-44.46. Thomson NJ. Recent trends in aboriginal mortality. Med J Aust. 1991;154:235-239.47. Hogg RS. Indigenous mortality: placing Australian aboriginal mortality within a broader context. Soc

Sci Med. 1992;35:335-346.48. Veroni M, Gracey M, Rouse I. Patterns of mortality in Western Australian aboriginals, 1983-1989. Int

J Epidemiol. 1994;23:73-81.49. Unwin CE, Gracey MS, Thomson NJ. The impact of tobacco smoking and alcohol consumption on

aboriginal mortality in Western Australia, 1989-1991. Med J Aust. 1995;162:475-478.50. Olshansky SJ, Ault AB. The fourth stage of the epidemiological transition: the age of delayed degenera-

tive diseases. Milbank Q. 1986;64:355-391.51. Phoon WO. Epidemiological Transition in Asian Countries and Related Health Policy Issues. Asia Pac

J Public Health. 1989;3:139-144.52. Nauru Bureau of Statistics, Secretariat of the Pacific Community (SPC), Macro International Inc.

Republic of Nauru: Demographic and Health Survey 2007. Final Report. Noumea, New Caledonia: Nauru Bureau of Statistics, Macro International Inc, Secretariat of the Pacific Community, Asian Development Bank; 2009.



� � �

155

5.2.2 Palau analysis and assessment

As for Nauru, the system assessment in Chapter three demonstrated reporting in

Palau is likely to be mostly complete for the data held by the MoH Epidemiologist.

Data included in this assessment were aggregated over the period from 2004-2009,

excluding 2007 (when there was no epidemiologist on island) which was found to

contain a number of irreconcilable errors. Deaths were also aggregated using 10

year age groups (for deaths at age 5 and over) to calculate age-specific mortality

and life tables directly. Life tables for Palau are attached in Appendix two with age-

specific mortality shown in Figure 13.

Figure 13: Age-specific mortality for Palau 2004-2009 (excluding 2007), by sex

Age-specific mortality shows a plausible pattern of deaths with relatively low infant

and child mortality, and high adult mortality, with age-specific mortality rates rising

substantially in age groups from 25 years of age and older. As the number of births

0.1

1

10

100

1000

0-4 5-9 15-24 25-34 35-44 45-54 55-64 65-74 75+

Deaths per 1,000

population

Age group

MalesFemales

� � �

156

was not available, IMR for 2004-2009 was estimated from the population mortality

rate (population <1year) and converted to a probability using life table methods.

Confidence intervals were derived using the Poisson distribution applied to infant

deaths. IMR is therefore estimated at 16.7 deaths per 1,000 live births (95% CI:

10.2-25.3). Twenty five deaths were recorded in children less than 5 years of age for

2004-2009 (excluding 2007), with U5M estimated from the life table (using Chiang

calculation of confidence intervals) as 17.9 deaths per 1,000 live births (95% CI:

10.9-24.8). The excessive premature mortality seen in the age-specific mortality

rates is reflected in a very high adult mortality (15-59 years), estimated at 596 deaths

per 1,000 population for males and 431 deaths per 1,000 population for females.

This is consistent with the high prevalence of traffic related deaths and diabetes

prevalence in Palau. The high adult mortality, particularly for males, is reflected in LE

which is estimated at 65.8 years (95% CI: 64.4-67.2 years) for males and 71.2 years

(95% CI: 69.9-72.5 years) for females. These results are similar to the 2005 census

which estimated LE at 66.3 years for males and 72.1 years for females (no

confidence intervals given), calculated by smoothing empirical data with a Coale-

Demeny West model life table.

5.3 Validity assessment of published data

Case study two demonstrates the importance of evaluating data in the context of

long term trends in Nauru. Records from vital registration with the health department

and civil registry were available for much of the previous 50 years, allowing this

assessment to be conducted using primary data and previously published estimates

generated using a very similar methodology to the current analysis. For many other

� � �

157

PICTs, it is unlikely that primary data, especially primary data of reliable quality,

would be available for such lengths of time; thus requiring researchers to turn to

secondary data sources such as censuses and surveys to establish these longer

term trends.

As noted in Chapter one, many of the published estimates of mortality level for the

PICTs are either implausible and/or show substantially different results for similar

periods. These estimates should not be used to establish trends in mortality levels

without first establishing whether individual estimates are likely to be plausible.

Reliable published sources can then be used to establish mortality trends for

comparison to current calculated measures. Case study three, demonstrates such

an assessment for published data in Fiji. A similar assessment was performed for

Tonga (case study four), however there were so few data points considered reliable

after the assessment that it was only possible to draw very general conclusions

regarding trends in respect to infant mortality.

5.4 Brass growth balance analysis

The system assessment in Chapter three indicated that although most deaths were

likely to be reported in Fiji, an assessment of completeness would be required to

evaluate and correct the available data as necessary. In accordance with Figure 11,

direct demographic methods (as outlined in Chapter two) were considered to

complete this assessment. As only one primary source of data, the Fijian MoH

records, was identified as being likely to be substantively complete, case study

three includes a Brass Growth Balance analysis of routinely collected mortality data.

� � �

158

As described in Chapter two, the Brass Growth Balance method uses the internal

consistency of the data to assess whether the age distribution of deaths is plausible.

To conduct this analysis, the partial (by age) death rate is compared at each age

with the partial birth rate, as shown in Figures 12 and 13. Completeness of the

reported deaths is then given by the inverse of the slope of the line fitted to these

points.

The Brass Growth Balance Method does not rely on having data for a full inter-

censal period and is therefore the most accessible method of assessing

completeness where migration is low to moderate. Other direct demographic

methods account better for population movement (14) if it is significant. The Brass

Growth Balance method was used as the additional information (such as migration

data) required for the other direct demographic methods was not readily available,

and the time period of the data collection available did not fit an inter-censal period.

Fiji has moderate out-migration of approximately 7.6% in 2010 (110). This appears to

have a minimal effect on the analysis for the total population. Despite this, the Brass

Growth Balance method has proven unsuitable for calculation of mortality rates by

ethnicity in Fiji due to the significant out-migration in the Indian population (322).

Implausible results were also obtained from a Brass Growth Balance analysis of the

Tongan data (where migration is -16.9% 2010 (110) (Appendix one), and population

has not increased significantly for around 30 years despite high fertility) and were

subsequently disregarded.

� � �

159

Both data from the MoH and reconciled data (Figures 14 and 15) from the MoH and

civil registry data (2005-2009) (case study five) in Tonga were assessed using a

Brass Growth Balance analysis. Analysis of the MoH data alone indicated a reporting

completeness as >99% for males and females, and the assessment of the reconciled

data indicated a reporting completeness as >99% for males but only 63% for

females. These results are clearly incompatible, both in regard to the single data

source, compared with reconciled data, and in relation to the sex differential in

reporting in Tonga. While some difference in reporting completeness by sex is

expected based on the system assessment (in terms of social incentives such as

land inheritance), there is no indication that suggests it should be this large.

Figure 14: Brass Growth Balance analysis of reconciled data from Tonga; Males 2005-2009

� � �

160

Figure 15: Brass Growth Balance analysis of reconciled data from Tonga; Females 2005-2009

As seen in Figure 14, for males, data did not fit comfortably on the straight line of

partial death rate graphed against the partial birth rate, indicating problems with the

age distribution of the source data. The Brass Growth Balance analysis performed

better (in terms of fit), and gave results similar to the capture-recapture analysis

(described later in this chapter) when limited to deaths on Tongatapu (the main

island) only, where there is perhaps less impact (on a per capita basis) from

migration (59, 136).

These examples demonstrate the problem of using a method of assessing reporting

completeness which is based on an internal consistency and age distribution only.

There is no clear “point” at which migration becomes too high for this method to be

used, rather it is a gradient where the results may have less and less reliability with

increasing migration. The Brass Growth Balance method provides the only option for

direct assessment of completeness in countries with a single accessible data source

� � �

161

and limited access to age-specific population and migration estimates, as required

by the other direct demographic methods. This approach provides a useful overview

of whether the data are likely to be complete, but should be used cautiously when

correcting data.

5.4.1 Case study 3: Mortality Trends in Fiji.

Carter KL, Cornelius M, Taylor R, Ali SS, Rao C, Lopez AD, Lewai V, Goundar R,


doi: 10.1111/j.1753-6405.2011.00740.x

Case study three presents a similar assessment of trends over time as for Nauru in

case study two, but adds existing published estimates and demonstrates the

importance of critical appraisal of published data. A set of core criteria assessing the

validity of published data were developed and applied to the estimates of IMR and

LE in Fiji. These criteria are as follows:

o the source publication noted the data were derived from estimation models

that assume a given improvement in LE or IMR by year, or the data source

showed a perfectly linear improvement in LE or IMR by year when graphed

indicating use of models that incorporate the above assumption;

o multiple incompatible estimates were given by the same source for a single

year or incompatible estimates were given by the same source for adjacent

years;

o the source data included estimates that were implausible given accepted

levels of LE and IMR in other countries (for example – sources that listed IMR

� � �

162

in Fiji well below the level in Australia or New Zealand at an equivalent point in

time were excluded); and/or

o calculations were documented or strongly suspected to be based on

uncorrected vital registration data where no assessment of completeness of

registration was provided.

As demonstrated in this case study, many of the published estimates are either

unclear regarding their primary data source and method, employ a series of

assumptions to extrapolated data from previous years, or estimate mortality levels

based on patterns of mortality from accepted model life tables. The empirical data

illustrates a rather different picture of the health status in Fiji than that presented by

many of the published estimates; demonstrating the risk of overlooking serious

health concerns when assumptions of improvement are made. The full list of sources

presented in the paper and reasons for exclusion are provided in Appendix three.

As discussed in the paper, many published sources excluded from the analysis were

from international agencies. This highlights one of the key problems facing decision

makers in small countries, where senior managers who are removed from the actual

collection and analysis of data in their own country may be influenced to accept

these estimates due to the authority with which they are presented; rather than local

data that may in fact be more reliable.

Case study three also presents a Brass Growth Balance analysis of the MoH data,

as the system assessment (Chapter three) demonstrated although CRVS systems

were relatively robust, there was some potential for under-reporting. Alternative

� � �

163

analyses such as the Bennet-Hounichi (199, 224) and capture-recapture methods

were not possible, as death data could not be accessed for the full inter-censal

period and only one data source was available (reporting through the civil registry

was found to be extremely incomplete).

The Brass Growth Balance analysis for Fiji (Figures 16 and 17, plus Appendix one)

shows the data from the MoH were more than 90% complete for most years, and

more than 80% complete for females for 2002 to 2004. With such high levels of

completeness, mortality estimates can be derived directly from the reported deaths

for comparison to long term trends.

Figure 16: Brass Growth Balance analysis for mortality data completeness: Fiji, Males 2002-2004

� � �

164

Figure 17: Brass Growth Balance analysis for mortality data completeness: Fiji, Females 2002-2004

Completeness is estimated by 1/K where K=the slope of the line

As shown in the figures, the slope indicates reporting for males is more than 99%

complete (1/K). As over-reporting is unlikely given the CRVS system (as described in

Chapter three), this can be interpreted as essentially complete reporting, indicating

the data does not need to be adjusted for undercount.

Key findings from the analysis in Fiji as outlined in the case study are that LE for

both males and females has shown no improvement over the last 20 years,

remaining stable at around 64 years for males and 69 years for females. Infant

mortality is stable below 20 deaths per 1,000 live births and therefore cannot account

for this low LE, which is being driven primarily by premature adult mortality.

412 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2011 vol. 35 no. 5© 2011 The Authors. ANZJPH © 2011 Public Health Association of Australia

As c o u n t r i e s m ove t h r o u g h

demographic and epidemiological

transitions, declines in under-

nutrition and infectious diseases are reflected

in improvements in survival which result in

improvements in life expectancy (LE) at birth.

However, real increases in adult mortality

from non-communicable diseases (NCDs)

and injury can slow or even reverse mortality

decline, if of sufficient magnitude.1-3 This

requires urgent action to reduce premature

adult mortality, such as occurred in Australia

and New Zealand.4,5

Fiji is a multi-ethnic society (Melanesian

57%, Indian 38%) of 827,900 people (2007).6

NCDs such as cardiovascular disease and

diabetes have been documented as public

health concerns since the 1970s.7,8 Mortality

Mortality trends in Fiji

Karen CarterSchool of Population Health, University of Queensland

Margaret CorneliusFiji Ministry of Health

Richard TaylorSchool of Population Health, University of Queensland and School of Public Health and Community Medicine, University of New South Wales

Shareen S. AliFiji Ministry of Health

Chalapati Rao, Alan D. LopezSchool of Population Health, University of Queensland

Vasemaca LewaiFiji Islands Bureau of Statistics

Ramneek GoundarFiji School of Medicine

Claire MowrySchool of Population Health, University of Queensland

Submitted: September 2010 Revision requested: January 2011 Accepted: February 2011Correspondence to: Richard Taylor, University of New South Wales Main Campus, Samuels Building, Level 2, Room 223, Botany St, Gate 11, Randwick, NSW 2052; e-mail: [email protected]

Abstract

Objectives: Mortality level and cause

of death trends are evaluated to chart

the epidemiological transition in Fiji.

Implications for current health policy are

discussed.

Methods: Published data for infant

mortality rate (IMR), life expectancy

(LE) and causes of death for 1940-2008

were assessed for quality, and compared

with mortality indices generated from

recent Ministry of Health death recording.

Trends in credible mortality estimates

are compared with trends in proportional

mortality for cause of death.

Results: IMR declined from 60 deaths

(per 1,000) in 1945 to below 20 by 2000.

IMR for 2006-08 is estimated at 18-20

deaths per 1,000 live births. Excessive LE

estimates arise by imputing from the IMR

using inappropriate models. LE increased,

but has been stable at 64 years for males

and 69 years for females since the late

1980s and early 1990s respectively.

Proportional mortality from diseases of

the circulatory system has increased from

around 20% in the 1960s to more than

45%. Extensive variation in published

mortality estimates was indentified,

including clearly incompatible ranges of

IMR and LE.

Conclusions: Mortality decline has

stagnated. Relatively low IMR and

proportional mortality trends suggest this is

largely due to chronic diseases (especially

cardiovascular) in adults.

Implications: Reconciliation of mortality

data in Fiji to reduce uncertainty is urgently

needed. Fiji’s health services and donor

partners should place continued and

increased emphasis on effective control

strategies for cardiovascular disease.

Key words: Cardiovascular disease,

epidemiological transition, Fiji, mortality

Aust NZ J Public Health. 2011; 35:412-20

doi: 10.1111/j.1753-6405.2011.00740.x

measures for Fiji are published by government

sources and a range of international agencies.

This paper evaluates mortality levels to chart

the progress of the epidemiological transition

in Fiji by assessing available information for

1940-2008. Levels of adult mortality are

compared with similar measures for Australia

and New Zealand. Published data on causes of

death are used to explain trends in all-cause

mortality, and implications of these findings

for current health policy are discussed.

MethodsData Collection

Measures of level of mortality were obtained

from two sources: (1) previously unpublished

data on reported deaths aggregated by age

group and sex for 1996-2004 from the Fiji

Trends Article

2011 vol. 35 no. 5 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 413© 2011 The Authors. ANZJPH © 2011 Public Health Association of Australia

Ministry of Health (MoH); and (2) published reports on all-

cause and cause-specific mortality from 1940-2008 by local and

international agencies. Data from published reports7,9-69 were sought

through direct contact and website searches from Fiji government

departments, United Nations agencies, the Secretariat of the Pacific

Community (SPC), the World Bank (WB), Asian Development Bank

(ADB) and non-government organisations. A literature search was

undertaken in PubMED and Medline.

For published data on mortality levels, the method of data

collection and analysis was rarely available. Limited data on causes

of death have been published for Fiji, many of which included only

leading causes of death, usually not tabulated by age group or sex,

or ill-defined and unknown causes could not be excluded from

proportional mortality.

Data Assessment All-cause mortality

The infant mortality rate (IMR) and life expectancy (LE) were

chosen as measures of all-cause mortality due to availability. IMRs

from Fiji MoH death registration and published reports were graphed

over time by major source of data (Figure 1). Data sources were

assessed for reliability and plausibility of estimates on the basis of

method of estimation, original source of data, and data consistency.

Unreliable sources were censored from further analysis if they

met any of the following criteria: (a) data were derived, or were

considered likely to have been derived, from models assuming

a given improvement by year, as evidenced by a perfectly linear

improvement in LE or IMR by year; (b) multiple incompatible

estimates were given by the source for a single year or adjacent

years; (c) source data included implausible estimates (based on

equivalent measures for developed countries); (d) calculations were

based on uncorrected vital registration data known to be significantly

under-reported or with no assessment of reporting completeness.

An exponential trend was fitted to the average IMRs of each year

from data remaining after censoring (Microsoft Excel). Childhood

mortality (probability of dying before five years of age) was largely

unavailable from published reports. Recent measures were estimated

by applying the proportion of childhood mortality accounted for

by infant mortality (78%) from credible data sources for which

both measures were available (analyses from the 198637 and 199638

censuses, and MOH data 1996-2004), to credible estimates of IMR

for 2006-08.54

Age-specific mortality rates and life tables70 were calculated

from the MoH death data and populations from SPC.71 Data

were aggregated by triennia to reduce stochastic variation (1996-

98, 1999-2001 and 2002-04). Mortality rates were analysed for

completeness using the Brass growth-balance method70 for age

groups 35-74 years, a method that has been recently validated as

a way of measuring reporting completeness.72 No adjustment was

made to reported deaths as registration was assessed to be over

85% complete (more than 95% for all except females 2002-04).

LE and adult mortality (probability of dying between 15 and 59

years of age inclusive) were calculated (Table 1). Adult mortality

for 2002-04 was compared to measures derived from published

life tables for Australia and New Zealand for the same period73-76

(Table 1). LE from all sources were graphed over time by major

source of data (Figures 2-3), with unreliable estimates censored from

further analysis. A quadratic function was fitted to the remaining

life expectancy estimates (Microsoft Excel) (Figures 2-3) to obtain

final estimates and overall trend.

Empirical measures of LE derived from MoH data for 1996-

2008 were compared with LE predicted by the WHO Model Life

Table system77. First, LE was predicted using empirical childhood

mortality as a single input; and, second, using probabilities of

childhood mortality and adult mortality (two parameters).

Causes of death Available data on cause-specific mortality were examined using

a range of measures and cause aggregations. Data are presented

as proportional mortality (PM) by International Classification of

Disease (ICD) chapter78 as several sources did not provide total

deaths and data are frequently under-enumerated. Cause of death

was calculated for the total population as there was insufficient

information available to examine by age group, sex, or ethnicity. PM

excluding unknown and ill-defined cases was calculated for sources

where this was possible. PM, and corrected PM, were graphed over

time for each ICD chapter.

ResultsAll-cause mortality

Extensive variation was found in published estimates of IMR

(Figure 1), with several reporting levels below those of developed

countries in the region.74,76 Of 16 IMR sources, 12 were censored

from further analysis with a further two partially censored. The trend

Table 1: Probability of dying between ages 15 and 59 years (%) by age group in Fiji, New Zealand and Australia, males and females, 1996-2004.Age group (years) 15-34 35-59 15-59Sex Males Females Males Females Males Females

Fiji

1996-1998 3.6 2.5 26.8 18.5 29.5 20.5

1999-2001 4.1 2.9 27.5 18.4 30.5 20.8

2002-2004 3.7 2.9 24.9 17.9 27.6 20.3

New Zealand 2002-2004 2.0 0.9 7.9 5.5 9.7 6.4

Australia 2002-2004 1.7 0.7 7.2 4.4 8.9 5.1Source data: Fiji – calculated from deaths reported to MoH (previously unpublished); New Zealand 75,76; Australia 73,74

Trends Mortality trends in Fiji


is consistent with IMR falling significantly over the last several

decades (1940–2006) to a level below 20 deaths per 1000 live births.

Credible sources indicate IMR for 2006-08 was 18 -20 deaths per

1000 live births, suggesting s a childhood mortality rate (<5 years)

of 23-26 deaths per 1000 live births (see methods).

Analysis of the MoH reported deaths shows LE was stable

between 1996 and 2004, as was adult mortality which was 2 to

4 times higher in Fiji than in Australia or New Zealand (Table

1). LE estimates from MoH data were notably lower than most

published estimates for the same period, but similar to estimates

from demographic analysis of censuses.38 Published estimates of LE

at birth for 1995-2000 vary by 11 years in males (61 to 72 years),

and by eight years in females (68 to 76 years) (Figures 2 and 3).

Of the 16 sources identified for LE data, 13 were censored from

the final analysis. Remaining estimates suggest improvements in

LE began to stall in the 1980s for both sexes, and that it has not

Carter et al. Article

Figure 1: Estimated infant mortality rate in Fiji, by data source, 1940-2004.

1a) All estimates

0

10

20

30

40

50

60

70

80

90

1940 1950 1960 1970 1980 1990 2000Year

Infa

nt M

orta

lity

Rate

(per

100

0 bi

rths

)

Fiji Fertility Survey (Total) (33)

Vital Registration (Fijian Male) (12,19,40)

Vital Registration (Indo-Fijian Male) (12,19,40)

Vital Registration (Fijian Female) (12,19,40)

Vital Registration (Indo-Fijian Female) (12,19,40)

Census (Total) (13,37,38,40,47)

Census (Male) (13,37)

Census (Female) (13,37)

Census (Fijian) (13,37)

Census (Indo-Fijian) (13,37)

MOH Reported Deaths (Total) (10,14-16,20,21,25,27,29,34,35,43,44,54)MOH Reported Deaths (Fijian) (14-16)

MOH Reported Deaths (Indo-Fijian) (14-16)

Adjusted MOH Reported Deaths (total) (21,26)

Adjusted MOH Reported Deaths (Fijian) (1,26)

Adjusted MOH Reported Deaths (Indo-Fijian) (21,26)

UNDP (total) (36,53,55)

UNICEF (total) (41,43,52,64,65,67)

Fiji Ministry of Information (total) (42)

SPC (total) (39,62,63)

Fiji Co-ordinating Committee on Children (Total) (31)

Fiji Government (Total) (28)

World Bank (Total) (32,63)

Asian Development Bank (59,60)

WHO (49,51,56)

WHO - WPRO (46,50,57)

UN ESCAP (69)

1b) Estimates censored by method

0

10

20

30

40

50

60

70

80

90

1940 1950 1960 1970 1980 1990 2000

Year

Infa

nt M

orta

lity

Rate

(per

100

0 bi

rths)

Average of censored sources by year

1a) All estimates

0

10

20

30

40

50

60

70

80

90

1940 1950 1960 1970 1980 1990 2000Year

Infa

nt M

orta

lity

Rate

(per

100

0 bi

rths)






Census (Total) (13,37,38,40,47)





MOH Reported Deaths (Total) (10,14-16,20,21,25,27,29,34,35,43,44,54)MOH Reported Deaths (Fijian) (14-16)

MOH Reported Deaths (Indo-Fijian) (14-16)

Adjusted MOH Reported Deaths (total) (21,26)

Adjusted MOH Reported Deaths (Fijian) (1,26)

Adjusted MOH Reported Deaths (Indo-Fijian) (21,26)

UNDP (total) (36,53,55)

UNICEF (total) (41,43,52,64,65,67)


SPC (total) (39,62,63)


Fiji Government (Total) (28)

World Bank (Total) (32,63)

Asian Development Bank (59,60)

WHO (49,51,56)

WHO - WPRO (46,50,57)

UN ESCAP (69)

1b) Estimates censored by method

0

10

20

30

40

50

60

70

80

90

1940 1950 1960 1970 1980 1990 2000

Year

Infa

nt M

orta

lity

Rate

(per

100

0 bi

rths)

Average of censored sources by yearAverage of censored sources by year

Year

Year


improved (Figures 2 and 3). LE for the period 2006-08 is estimated

at 64 years for males and 69 years for females.

LE and adult mortality predictions from model life tables using the

childhood mortality and adult mortality from MoH death recording

varied less than one year from results calculated using age-specific

mortality rates (not shown). However, LE predicted from childhood

mortality alone over-estimated LE by 4-5 years (similar to censored

estimates) and were excluded from further analysis.

Cause specific mortality The proportion of deaths coded to ill-defined and unknown causes

since 1960 generally fluctuated below 8% of deaths, with higher

peaks (up to 18%) in the early 1980s and early 2000s (Figure 4). PM

attributable to ‘Diseases of the Circulatory System’ (ICD Chapter

IX) shows a significant and ongoing linear increase over 1960-2000,

from around 20% of deaths to more than 45% of all deaths (Figure

4), apparent before and after ill-defined and unknown causes are

excluded. Early data relating to 1960 were only available as PM


Figure 2: Male life expectancy at birth, in Fiji, by data source, 1940-2006.

ADB – Asian Development Bank; BoS – Fiji Bureau of Statistics; MoH – Fiji Ministry of Health; SPC – Secretariat of the Pacific Community; UNDP – United Nations Development Program; WHO – World Health Organization; WPRO – WHO Western Pacific Regional Office; VR – Vital registration. NB: Census data co-published by SPC with Fiji Bureau of Statistics listed under BoS only. References shown in brackets after each source within figure.

2a) All estimates

40

45

50

55

60

65

70

75

80

1940 1950 1960 1970 1980 1990 2000 2010

Period (year)

Life

Exp

ecta

ncy

at B

irth

(yea

rs)

Registrar General - VR (12,19,40)

MOH Reported Deaths (Brass adjustment) - Fijian (21,26)

MOH Reported Deaths (Brass adjustment) - Indo-Fijian (21,26)BoS Census (13,37,38,40,47)

BoS Census - Fijian (13)

BoS Census - Indo-Fijian (13)

United Nations Population Division (36,53,55)

Fiji Coordinating Committee on Children - Male / Femalecombined total (31)World Bank - Male/ Female combined total (32,63)

Fiji Department for Women and Culture (30)

SPC (39,62,63)

Fiji Ministry of Information (42)

WHO - World Health Statistics (49,51,56)

MOH reported deaths (unpublished data)

WHO - WPRO (46, 50, 57)

ADB (59,60)

UNFPA (69)

UN ESCAP (58, 68)

2b) Estimates remaining after censoring by method

40

45

50

55

60

65

70

75

80

1940 1950 1960 1970 1980 1990 2000 2010

Year

Life

Exp

ecta

ncy

at B

irth

(yea

rs)

MOH Reported Deaths (Brass adjustment) (21,26)

BoS Census (13,37,38,40,47)


Year

Year


>5 years7; even though this would over-estimate all-ages PM from

cardiovascular disease, the observed proportion (15-18%) was the

lowest ever observed.

PM due to infectious diseases showed no significant trend,

fluctuating between 5% and 12% of all deaths. Respiratory diseases

fell linearly from 14% to 7%, and injuries varied between 4% and

10%, with a gradual decline in evidence since the mid 1980s. PM

from conditions arising in the perinatal period fell from 8% in the

1970s to 2 %. There were no clear trends for any other ICD disease

categories; including neoplasms, endocrine and metabolic diseases

or diseases of the digestive tract and genitourinary tract.

DiscussionCredible data sources demonstrate that IMR has declined

steadily to below 20 deaths per 1000 live births. While IMR has

not fallen significantly since the late 1980s, at these levels it is a

minor influence on the overall LE70. The decline in infant and child

mortality since 1990 is insufficient for Fiji to meet the Millennium

ADB – Asian Development Bank; BoS – Fiji Bureau of Statistics; MoH – Fiji Ministry of Health; SPC – Secretariat of the Pacific Community; UNDP – United Nations Development Program; WHO – World Health Organization; WPRO – WHO Western Pacific Regional Office; VR – Vital registration.

Figure 3: Female life expectancy at birth, in Fiji, by data source, 1940-2006.

3a) All estimates

40

45

50

55

60

65

70

75

80

1940 1950 1960 1970 1980 1990 2000 2010

Year

Life

Exp

ecta

ncy

at B

irth

(yea

rs)

Registrar General - VR (12,19,40)

MOH Reported Deaths (Brass adjustment) - Fijian (21,26)

MOH Reported Deaths (Brass adjustment) - Indo-Fijian (21,26)BoS Census (13,37,38,40,47)

BoS Census - Fijian (13)

BoS Census - Indo-Fijian (13)

United Nations Population Division (36,53,55)

Fiji Coordinating Committee on Children - Male / Femalecombined total (31)World Bank - Male/ Female combined total (32,63)

Fiji Department for Women and Culture (30)

SPC (39,62,63)

Fiji Ministry of Information (42)

WHO - World Health Statistics (49,51,56)


WHO - WPRO (46, 50, 57)

ADB (59,60)

UNFPA (69)

UN ESCAP (58, 68)

3a) Estimates remaining after censoring by method

40

45

50

55

60

65

70

75

80

1940 1950 1960 1970 1980 1990 2000 2010

Year

Life

Exp

ecta

ncy

at B

irth

(yea

rs)

MOH Reported Deaths (Brass adjustment)(21,26)

BoS Census (13,37,38,40,47)



Year

Year


Development Goals by 2015;79 meeting these targets would require

an IMR comparable to that of Australia and New Zealand in 199074,75

where specialist medical services and technology were already

available – and nutritional and environmental conditions were better.

LE rose through the 1970s and 1980s, then stabilised around 64

years for males and 69 years for females.

The most reliable all-cause mortality estimates are those

from demographic analyses of censuses and recent MoH death

registration. LE produced from MoH data were amongst the lower

estimates found in available data. No adjustment for under-reporting

was made to the MOH data as they were found to be substantially

complete, however it this had been performed it would have resulted

in an even lower measure of LE. LE calculated from MoH death

registration (direct methods) correlates closely with published

Census mortality estimates (indirect methods), thus providing

independent verification of these figures, and by implication,

completeness of death reporting by Fiji MoH.

Bias was minimised by selecting the most commonly reported

mortality measures (i.e. IMR rather than childhood mortality) and by

converting all-cause of death information to proportional mortality

where possible; although this provides only an approximate picture

of cause of death patterns. As changes in the proportion of deaths

due to ‘Ill-defined’ and ‘Unknown’ causes may significantly affect

PM trends, PM excluding these categories was calculated where

possible. The increasing trend in circulatory diseases was evident

both before and after this refinement. As PM was tabulated by ICD

chapter, changes in coding over time should have minimal impact

as these were predominantly within chapters.80

In Australia, LE for males did not improve from 1945 to 1970

as increased mortality, mainly from cardiovascular diseases, offset

declines in death from infectious and related diseases.1,2,81 Similar

stagnation in LE has been reported in the 20th Century in Western

Europe, North America and New Zealand in association with the

epidemiological transition.3 Subsequent declines in cardiovascular

mortality and improvements in LE in Australia and New Zealand

from the 1970s have been driven by decreases in serum cholesterol,

blood pressure and cigarette smoking,4,5 as well as improved health

care, even though body weight has increased.82

Adult mortality was lower than recently published predictions for

201083 based on adjusted civil registration data from 1970-1988.

The adult mortality levels in Fiji for 2002-04 were 2-4 times higher

than in Australia or New Zealand. The probabilities of dying (for

2002-04) between ages 35 and 59 years in Fiji were 25% for males

and 18% for females, compared to 7–8% for males and 4–6% for

females in Australia and New Zealand.73-76 This segment of the adult

age group is where cardiovascular disease is a significant cause of

death. PM in Fiji from cardiovascular disease increased from just

below 20% of deaths around 1960 to more than 45% of deaths in

2001. While this shift in PM may also be influenced by aging of the

population, the age structure and LE in Fiji suggest that population

aging alone does not explain this trend. The PM from cardiovascular

diseases is substantially higher than that for infectious diseases

and respiratory illnesses. While a similar stagnation in LE may be

driven by increasing mortality from HIV/AIDS, Fiji is estimated

to have only 0.1% HIV prevalence in the adult population, and a

low mortality from AIDS.50

The increase in PM from cardiovascular diseases and levels of

adult mortality in Fiji are consistent with documented trends since

the 1970s, in both hospital admissions for cardiovascular disease

and risk factor exposure for these conditions7,82. A national survey

Figure 4: Proportional mortality from diseases of the circulatory system in Fiji, 1950-2010.


Source data: 7-9, 11, 14-17, 20, 25,

26, 29, 35, 44, 51.


of risk factor prevalence in 19807 showed adult prevalences of

hypertension (5.1-9.9%) (using ≥160 mm hg systolic and/or ≥95

mm hg diastolic and/or on treatment for hypertension) and diabetes

prevalence of 1.1-11.8% (using two hour post 75 g glucose load

blood glucose of ≥11.1 mm/L and/or on treatment for diabetes),

with the lowest prevalences in rural melanesians. The Fiji STEPS

survey in 200284 indicated adult prevalence of hypertension of 19%

(using ≥140 mmhg systolic and /or ≥90 mmhg diastolic and/or on

treatment for hypertension) and diabetes prevalence of 12% (using

fasting blood glucose of ≥7.0 mm/l and/or on treatment for diabetes)

Cigarette sales in Fiji rose 273% from 1956 to 1984, three times

the population increase (88%)7.

Dietary intake surveys in Fiji and elsewhere in the 1980’s

demonstrated that urban populations were more obese, had

higher prevalence of diabetes and hypertension, higher salt intake

and generally higher serum cholesterol levels, than their rural

counterparts, despite a lower overall calorie intake; indicating

the impact of physical exercise and dietary differences on

cardiovascular risk factors.85 The relationship of Fiji adult mortality

to cardiovascular risk factors was demonstrated in an 11-year follow-

up81 of the 1980 national risk factor survey,7 which found systolic

blood pressure, total serum cholesterol and two hour (post load)

plasma glucose associated with cardiovascular disease mortality,

with differences by sex and ethnicity. These studies suggest that the

prevalence of circulatory disease risk factors in Fiji is consistent

with the high adult mortality and increasing PM from cardiovascular

diseases. It is unclear when decline in cardiovascular disease and

subsequent increase in LE could be expected in Fiji as previously

experienced in Australia and New Zealand, however, on-going

monitoring of risk factors in the population by the Fiji Ministry of

Health84 will provide indications.

This study shows a range of eight years between various published

estimates of LE for Fiji in the early 2000s. Spurious increases in

LE imply that the health situation is improving when this is clearly

unsupported by more reliable data. Extensive variation in published

estimates of IMR by source (Figure 1) was also noted, with several

reporting improbable levels below those of developed countries in

the region such as Australia (IMR of 5.0 in 2007)67 and New Zealand

(IMR of 6.0 in 2007)67 (Figure 1). A previous study found substantial

uncertainty about mortality conditions in Pacific Island populations

with variations of 10 years or more in LE in the 1990s, depending

on the source.86 Principal issues include: under-enumerated vital

registration data; annual stochastic fluctuations in mortality in small

populations; errors in the imputation of adult mortality from infant

and childhood rates; implausible results from indirect demographic

methods; use of inappropriate model life tables; and inadequately

described and implausible projections.86

Difficulties in assessment of published data were the lack

of information concerning primary data sources, methods of

calculation, and assumptions. In order to deal with this, strict

exclusion criteria were specified and unreliable estimates censored

from further analyses. Many of the previously published estimates of

LE are implausible, and it is likely that many were produced through

the use of single parameter models (using infant or childhood

mortality) which under-estimate adult mortality in Pacific Island

populations, as shown in these results. These subsequently over-

estimate LE, thus “masking” the importance of cardiovascular

disease as a public health issue.

The considerable variation in mortality estimates for Fiji is a

significant concern for public policy. This also occurs in other

Pacific Island states.86 Less than seven of the 27 countries of the

WHO Western Pacific Region, including Fiji, have reliable mortality

data.87 These findings suggest that research is urgently required

to determine whether this pattern of premature adult mortality

has become the ‘typical’ picture of epidemiological transition

throughout the Pacific Islands. Reconciliation of mortality data to

reduce uncertainty is urgently needed. The quality assessment in

this study indicates that measures provided by Fijian government

sources based on indirect demographic methods from censuses and

direct MoH death recording provide the most reliable estimates of

mortality and LE at birth in Fiji. As such, these directly observed

data should be used by decision makers as key tools in health

planning rather than estimates from other sources.

The stagnation in life expectancy, at a relatively low IMR,

combined with trends in cardiovascular disease mortality, morbidity

and risk factors, suggests that NCDs are having a profound limiting

effect on further mortality decline in Fiji. Other contributing factors

need to be considered and require more investigation, including

out-migration of skilled workers (who may be healthier than non-

migrants), social dislocation, declines in health expenditure, and

effects of unemployment. While further research is required, there

is an urgent need for health services and donor partners to place

increased emphasis on effective prevention and treatment strategies

for cardiovascular disease.

AcknowledgementsThe authors would like to acknowledge the assistance of the Fiji

Ministry of Health and Bureau of Statistics.

References1. Taylor R, Lewis M, Powles J. The Australian mortality decline: all-cause

mortality 1788 – 1990. Aust N Z J Public Health. 1998;22(1):27-36.2. Taylor R, Lewis M, Powles J. Australian mortality decline: cause-specific

mortality 1907-1990. Aust N Z J Public Health. 1998;22(1):37-44. 3. Thom T, Epstein F, Feldman J, Leaverton P, Wolz M. Total Mortality and

Morbidity from Heart Disease, Cancer and Stroke from 1950 to 1987 in 27 Countries. Washington (DC): National Institutes of Health, National Heart, Lung and Blood Institute; 1992. Report No.: 92-3088.

4. Taylor R, Dobson A, Mirzaei M. Contribution of changes in risk factors to the decline of coronary heart disease mortality in Australia over three decades. Eur J Cardiovasc Prev Rehabil. 2006;13(5):760-8.

5. Tobias M, Taylor R, Yeh L, Huang K, Mann S, Sharpe N. Did it fall or was it pushed? The contribution of trends in established risk factors to the decline in premature coronary heart disease mortality in New Zealand. Aust N Z J Public Health. 2008;32(2):117.

6. Fiji Bureau of Statistics. Census 2007 Results: Population Size, Growth, Structure and Distribution. Suva (FIJI): FIBOS; 2008.

7. Tuomilehto J, Ram P, Eseroma R, Taylor R, Zimmet P. Cardiovascular diseases and diabetes mellitus in Fiji: analysis of mortality, morbidity and risk factors. Bull World Health Organ. 1984;62(1):133-43.

8. Li N, Tuomilehto J, Dowse G, Virtala E, Zimmet P. Prevalence of coronary heart disease indicated by electrocardiogram abnormalities and risk factors in developing countries. J Clin Epidemiol. 1994;47(6):599-611.



9. Savou SF. Magnitude of Cancer Problem in the South Pacific and Possible Aetiologic Factors (Fiji Islands Context) – Draft Document. Suva (FJI): Fiji Ministry of Health; Undated.

10. Biumaiwai M, Bavadra T, Olakowski T. Health Situation and Trend Analysis in Fiji Based on Hospital Morbidity and Mortality by Causes. Suva (FJI): World Health Organization, Regional Office for the Western Pacific; 1984

11. Ministry of Health. The Nation’s Health. Suva (FJI): Health Information Unit, Ministry of Health; 1997.

12. McArthur N. Annual Report of the Registrar-General 1967. Suva (FJI): Fiji Islands Bureau of Statistics; 1967

13. Zwart, F. Report on the Census of the Population 1976, Volume II: Demographic Characteristics. Suva (FJI): Fiji Islands Bureau of Statistics; 1979.

14. Ministry of Health. Annual Tabulation 1982. Suva (FJI): Government of the Republic of Fiji; 1982.

15. Ministry of Health. Annual Reports for the Years 1979 and 1980. Paper No.: 49. Suva (FJI): Government of the Republic of Fiji; 1982.

16. Ministry of Health. Annual Report for the Year 1981. Paper No.: 25. Suva (FJI): Government of the Republic of Fiji; 1983.

17. Biumaiwai M, Bavadra T, Olakowski T. Materials for WHO Intercountry Seminar on Strengthening of Epidemiological Surveillance System and Utilisation of Existing Health Board in Pacific Island Countries. Suva (FJI): World Health Organization, Regional Office for the Western Pacific; 1984.

18. Taylor R, Henderson B, Levy S, Kolonel L, Lewis N. Table 13: Proportion of Mortality Due to Cancer in Pacific Island Countries. In: Cancer in Pacific Island Countries. Noumea (NCL): Secretariat of the Pacific Community; 1985.

19. Fiji Bureau of Statistics. Social Indicators for Fiji. Suva (FJI): Government of the Republic of Fiji; 1986. p. 5.

20. Ministry of Health. Annual Reports for the Year 1984. Suva (FJI): Government of the Republic of Fiji; 1986

21. Taylor R, Naroba V. Analysis of Mortality in the Population of Fiji 1981-83, and a Review of Previous Mortality Data. Noumea (NCL): South Pacific Commission; 1986.

22. Fiji Bureau of Statistics. Vital Statistics in Fiji (Statistics of Births, Deaths and Marriages) – A Report for the Year 1979-80. Suva (FJI): Government of the Republic of Fiji; 1988.

23. Ministry of Health and Social Welfare. Table 4 – Five major causes of hospital mortality 1980 – 1987. In: AIDAB Pacific Multicounty Health Initiatives Program: Health Sector Desk Review (draft working paper). Suva (FIJI): Australian International Development Assistance Bureau; 1989.

24. Taylor R, Davis Lewis N, Levy S. Societies in Transition: Mortality Patterns in Pacific Island Populations. Int J Epidemiol. 1989;18(3):634-46.

25. Ministry of Health. Fiji Ministry of Health Annual Report 1988 and 1986. Paper No.: 7. Suva (FJI): Government of the Republic of Fiji; 1989.

26. Balkaran S, Taylor R, Naroba V. Mortality trends and differentials in Fiji (Chapter 5). In: Chandra R, Bryant J, editors. Population of Fiji. Noumea (NCL): Secretariat of the Pacific Community; 1990.

27. Ministry of Health. Table: 81. In: Ministry of Health Annual Report for the Years 1987 and 1988. Paper No.: 13. Suva (FJI): Government of the Republic of Fiji; 1991.

28. Parliament of Fiji. Opportunities for Growth: Policies and Strategies for Fiji in the Medium Term. Parliamentary Paper No.: 2. Suva (FJI): Government of the Republic of Fiji; 1993.

29. Ministry of Health. Appendix IX. In: Ministry of Health Annual Report 1990. Paper No.: 49. Suva (FJI): Government of the Republic of Fiji; 1993.

30. Department for Women and Culture. Women in Fiji: A Statistical Gender Profile. Suva (FJI) Government of the Republic of Fiji; 1994.

31. Ministry of Women, Social Welfare and Poverty Alleviation. The Rights of the Child. Suva (FJI): Government of the Republic of Fiji; 1995.

32. World Bank. Fiji: Restoring Growth in a Changing Global Environment. Washington (DC): The World Bank Country Operations Division, East Asia and Pacific Region; 1995. Report No.: 13862-FIJ.

33. Fiji Bureau of Statistics. Fiji Fertility Survey. Suva (FJI): Government of the Republic of Fiji; 1996.

34. Ministry of Health. Figure 1993: 33. In: Chaimberlain K, Kamm D. A Situation Analysis of Children and Women in Fiji 1996. Suva (FJI): Government of the Republic of Fiji; 1996.

35. Ministry of Health. Annual Tabulation 1997. Suva (FJI): Government of the Republic of Fiji; 1997.

36. United Nations Development Programme. Table 9a – Expectation of life at specified ages for each sex: 1948-1997. In: United Nations: Demographic Yearbook, Historical Supplement. New York (NY): United Nations Statistical Division; 1997

37. Fiji Bureau of Statistics. 1986 Fiji Census of Population and Housing: Analytical Report, Part 1 Demographic Characteristics. Suva (FJI): Government of the Republic of Fiji; 1998.

38. Fiji Bureau of Statistics. 1996 Fiji Census of Population and Housing: Analytical Report, Part 1 Demographic Characteristics. Suva (FJI): Government of the Republic of Fiji; 1998.

39. Secretariat of the Pacific Community. Pacific Island populations – revised edition. In: International Conference on Population and Development; 1994; Cairo, Egypt. Noumea (NCL): SPC; 1998.

40. Rakeseta V. Fiji Islands Population Profile Based on the 1996 Census: A Guide for Planners and Policy Makers. Noumea (NCL): Secretariat of the Pacific Community; 1999.

41. United Nations Children’s Fund. Fiji Mid-Term Review 1999. Suva (FJI): UNICEF Pacific; 2000.

42. Ministry of Information. Fiji Today. Suva (FJI): Government of the Republic of Fiji; 2000.

43. United Nations Children’s Fund. Country Statistics [Internet]. New York (NY): Statistics and Monitoring, Division of Policy and Practice, UNICEF; 2001 [cited 2007 Jul 27]. Available from: http://www.unicef.org/statis

44. Ministry of Health. Ministry of Health Annual Report 2002. Paper No.: 44. Suva (FJI): Government of the Republic of Fiji; 2005.

45. Ministry of Health. Ministry of Health Annual Report 2001. Paper No.: 43. Suva (FJI): Government of the Republic of Fiji; 2005.

46. World Health Organization. Country Health Information Profiles. Manila (PHL): WHO, Western Pacific Region; 2005.

47. Fiji Bureau of Statistics. Table 2.2 Summary of Key Demographic Indicators for Fiji by Sex and Ethnicity from the Population Census of 1986 and 1996. In: Key Statistics: March 2006 [Internet]. Suva (FJI): Government of the Republic of Fiji; 2006 [cited 2006 Jul 11]. Available from: http://www.spc.int/prism/country/fj/stats/cens&surveys/Popu_census.htm

48. World Health Organization. Table: Death and Daly estimates by Cause. In: Country Health Information Profiles [internet]. Manila (PHILIPPINES): WHO, Western Pacific Regional Office; 2006 [cited 2006 Jul 11]. Available from: who.int/entity/healthinfo/statistics/bodgddeathdalyestimates.xls

49. World Health Organization. Mortality Country Profile 2006: Fiji [Internet]. Manila (PHL): WHO, Western Pacific Regional Office; 2006 [cited 2007 Jul 27]. Available from: http://www.who.int/whosis/mort/profiles/mort_wpro_fji_fiji.pdf

50. World Health Organization. Country Health Information Profiles. Manila (PHL): WHO, Western Pacific Regional Office; 2007.

51. World Health Organization. Number and Rates of Registered Deaths. In: WHO Statistical Information System [Internet]. Geneva (CHE): WHOSIS; 2006 [cited 2007 Jul 30]. Available from: http://www.who.int/whosis/database/mort/table1_process.cfm

52. United Nations Children’s Fund. Statistical tables: table 1- basic indicators. In: The State of the World’s Children 2002 [Internet]. New York (NY): Statistics and Monitoring, Division of Policy and Practice, UNICEF; 2006 [cited 2007 Jul 30]. p. 12-15. Available from: http://www.childinfo.org/mortality_infantmortality.php

53. United Nations Development Programme. Population and Vital Statistics Report. Statistical Paper: A LVIII(1). New York (NY): United Nations; 2006.

54. Ministry of Health [Internet]. Infant mortality rates. Suva (FJI): Government of the Republic of Fiji; 2008 [cited 2008 Jul 30]. Available from: www.fijihealth.gov.fj

55. Department of Economic and Social Affairs Population Division. World Population Prospects: The 2008 Revision [Internet]. New York (NY): United Nations; 2008 [cited 2008 Jul 30]. Available from: www.un.org/esa/population/publications/wpp2008/wpp2008_text_tables.pdf

56. World Health Organization. Country Health Information Profiles. Manila (PHL): Western Pacific Regional Office, WHO; 2008.

57. World Health Organization. Core Indicators 2005 [Internet]. Manila (PHL): Western Pacific Regional Office, WHO; 2005 [cited 2009 Jul 7]. Available from: www.wpro.who.int/information_sources/databases/core-indicators

58. Economic and Social Commission for Asia and the Pacific. Statistical Yearbook for Asia and the Pacific 2008 [Internet]. Bangkok (THA): United Nations ESCAP; 2009 [cited 2009 Jul 27]. Available from: http://www.unescap.org/stat/data/syb2008

59. Asian Development Bank. Millennium Development Goals [Internet]. Manila (PHL): ADB; 2000 [cited 2009 Jul 27]. Available from: http://www.adb.org/poverty/mdgs.asp

60. Asian Development Bank. PART II Millennium Development Goals – Goal 4. Reduce child mortality. In: Key Indicators for Asia and the Pacific 2008 [internet]. Manila (PHL): ADB; 2008 [cited 27/7/09]. Available from: http://www.adb.org/documents/books/key_indicators/2008/Part-II.asp#goal04

61. Secretariat of the Pacific Community. PRISM Demographic Indicators [Internet]. Noumea (NCL): SPC; 2008 [cited 2009 Jul 27]. Available from: www.spc.int/prism/Social/demog/HealthInd.html

62. Secretariat of the Pacific Community. PRISM MDG Indices [Internet]. Noumea (NCL): SPC; 2008 [cited 2009 Jul 27]. Available from: www.spc.int/prism/MDG/mdg-p/htm

63. World Bank. Key Development Data and Statistics [Internet]. Washington (DC): The World Bank Country Operations Division, East Asia and Pacific Region World Bank; 2009 [cited 2009 Jul 27]. Available from: http://web.worldbank.org/

64. United Nations Children’s Fund. Country Comparison [Internet]. New York (NY): Statistics and Monitoring, Division of Policy and Practice, UNICEF; 2008 [cited 2009 Jul 27]. Available from: www.unicef.org/view_chart/php



65. United Nations Children’s Fund. Child Info Database –Statistics by Area [Internet]. New York (NY): Statistics and Monitoring, Division of Policy and Practice, UNICEF 2009 [cited 27/7/09]. Available from: www.childinfo.org

66. United Nations Development Programme. Fighting climate change: Human solidarity in a divided world. In: Human Development Report – 2007/8 Data [Internet]. New York (NY): UNDP; 2007 [cited 2009 Jul 27]. Available from: http://hdr.undp.org/en/reports/global/hdr2007-8/

67. United Nations Children’s Fund. Statistical tables. In: The State of the World’s Children 2009. New York (NY): Statistics and Monitoring, Division of Policy and Practice, UNICEF; 2009.

68. Economic and Social Commission for Asia and the Pacific. Statistical Indicators for Asia and the Pacific 2004 Compendium, Volume XXXIV [Internet]. Bangkok (THA): United Nations ESCAP; 2004 [cited 2009 Jul 27]. Available from: www.unescap.org/stat/data/statind/pdf.t4_dec04.pdf

69. United Nations Population Fund. State of the World Population 2009. New York (NY): UNFPA; 2009 [cited 2009 Jul 27]. Available from: http://www.unfpa.org/swp/index.html

70. Rowland DT. Demographic Methods and Concepts. New York (NY): Oxford University Press; 2003.

71. Secretariat of the Pacific Community. Population Projections 1996-2005 [unpublished data]. Noumea (NCL): SPC. Personal communication 2008

72. Murray CJ, Rajaratnam JK, Marcus J, Laakso T, Lopez AD. What can we conclude from death registration? Improved methods for evaluating completeness. PLoS Med [Internet]. 2010;7(4):e1000262.

73. Australian Bureau of Statistics. Abridged Life Tables, 2002-2004 [Internet]. Canberra (AUST): ABS; 2008 [cited 2008 Jul 30]. Available from: www.abs.gov

74. Australian Bureau of Statistics. Chapter 2: summary of findings. In: 3302.0 Deaths: Australia 2004 [Internet]. Canberra (AUST): ABS; 2004 [cited 2008 Oct 30]. Available from: http://www.abs.gov.au/AUSSTATS/[email protected]/allprimarymainfeatures/F69F652EDC86B05CCA257235007865FB?opendocument

75. Statistics New Zealand. Tables: deaths. In: Age-specific Death Rates by Sex, Total Population, 1986–2007 [Internet]. Wellington (NZ): Government of New Zealand; 2008 [cited 2008 Oct 30]. Available from: http://www.stats.govt.nz/tables/deaths-tables.htm

76. Statistics New Zealand. Life Tables 2002-2004 [Internet]. Wellington (NZ): Government of New Zealand; 2008 [cited 2008 Jul 30]. Available from: www.stats.govt.nz/tables/abridged-period-life-tables.htm

77. Murray CJL, Ferguson BD, Lopez AD, Guillot M, Salomon JA, Ahmad O. Modified logit life table system: principles, empirical validation, and application. Popul Stud. 2003;57(2):165-82.

78. World Health Organization. International Classification of Diseases. Geneva (CHE): WHO; 2007.

79. National Planning Office. Millennium Development Goals- Fiji National Report. Suva (FJI): National Planning Office; 2004 [cited 2010 Aug 8]. Available from: http://www.spc.int/prism/fjtest/MDGs/Documents/Fiji_MDG%20report.pdf

80. Janssen F, Kunst AE. ICD coding changes and discontinuities in trends in cause-specific mortality in six European countries, 1950–99. Bull World Health Organ. 2004;82(12):891-970.

81. Collins V, Dowse G, Cabealawa S, Ram P, Zimmet P. High Mortality from Cardiovascular Disease and Analysis of Risk Factors in Indian and Melanesian Fijians. Int J Epidemiol. 1996;25(1):59-69.

82. Cameron AJ, Welborn TA, Zimmet PZ, Dunstan DW, Owen N, Salmon J. Overweight and obesity in Australia: the 1999–2000 Australian Diabetes, Obesity and Lifestyle Study (AusDiab). Med J Aust. 2003;178:427-32.

83. Rajaratnam JK, Marcus JR, Levin-Rector A, Chalupka AN, Wang H, Dwyer L, et al. Worldwide mortality in men and women aged 15-59 years from 1970 to 2010: a systematic analysis. Lancet. 2010;375(9727):1704-20.

84. Ministry of Health (Fiji). Fiji Non-communicable Disease (NCD) Steps Survey 2002. Suva (FJI): Government of the Republic of Fiji; 2002.

85. Taylor R, Badcock J, King H, Pargeter K, Zimmet P, Fred T, et al. Dietary intake exercise, obesity and noncommunicable disease in rural and urban populations of three Pacific Island countries. J Am Coll Nutr. 1992;11(3):283-93.

86. Taylor R, Brampton D, Lopez A. Contemporary patterns of Pacific Island mortality. Int J Epidemiol. 2005;(34):207-14.

87. Mathers C, Ma Fat D, Inouie M, Rao C, Lopez A. Counting the dead and what they died from: an assessment of the status of global cause of death data. Bull World Health Organ. 2005;83(3):171-7.


CR

ICO

S N

o. 0

0213

J

QU

HL

059/

R

Do something decisive for your career.

QUT’s Master of Health Management gets straight to the heart of the real issues in public health.As a heavyweight addition to your CV, the degree is one of the most significant boosts you can give your career.It’s already proving so for David Stewart, Director of Queensland Health’s Central Integrated Regional Cancer Service.

“The course is more relevant and practical than I could’ve hoped for,” he said. “Things such as the intensive lecture weekends, where top health sector leaders candidly detail how they’ve dealt with challenges in the health system, are immensely valuable. It’s certainly not just the textbook view – we deconstruct real world examples and debate the merits of different approaches in change management.”

To find out more about the course – and others within QUT’s respected School of Public Health – go to www.qut.edu.au/do-something-decisive And equip yourself to lead and manage decisively in an ever-changing sector.

David Stewart, Director of Queensland Health’s Central Integrated Regional Cancer Service.

QUHL 059 Public Health Press_FPFC PHAA Jrl_FA_m/r.indd 1 2/09/11 9:45 AM

� � �

174

5.5 Capture-recapture analysis

An alternative method of assessing reporting completeness is a capture-recapture

analysis, considered a direct method as completeness is estimated based on

reported deaths from two or more sources, rather than referring to the age

distribution of the deaths. The advantages of this method is that under-reporting is

not assumed to be consistent across all age groups, and estimates of completeness

can be made directly by 5 and 10 year age groups, including for children (228, 323).

As described in Chapter two, the capture-recapture method reconciles different

sources of data to determine the proportion of these records which overlap. The

extent of the overlap in comparison to the overall number of records is used to

estimate the number of deaths not reported by either source (or “missing” records)

and therefore the total number of deaths (233).

Used extensively in the past to assess disease specific prevalence and mortality

(324, 325), for all-cause mortality this method has generally been used to compare

deaths recorded through routine CRVS systems to an independent survey. Use of

capture-recapture methods to assess all-cause mortality from two or more routine

reporting systems without an independent survey is relatively rare, with only one

recent example found in literature from Malawi (although no details of the model

selection procedures were documented) (326).

There are several considerations that must be taken into account when applying this

method to all-cause mortality that are perhaps less of an issue with cause or age-

specific assessments. These are outlined below:

� � �

175

• Independence: due to the structure of CRVS systems, policy and legislation

requiring evidence to be presented to officially register a death or bury/cremate, the

body, routinely collected data sources may not be independent of each other. Two

source analysis relies on the assumption the sources are independent (228), that

reporting the event to one source has no impact on the probability of the death being

reported to the other (233). The choice of model selected in three source

assessments is based on whether or not the sources are independent (and the

combination of these dependencies) (327, 328).

• Catchability: when studying all-cause mortality, the population from which

deaths are captured is much more heterogeneous than would be the case in many

disease specific studies, thus not all deaths in the community may have the same

probability of being captured (228).

• Population: the basic premise is events are drawn from the same population

(228), so only systems with the same coverage can be used as sources of data.

Alternatively, data may need to be separated by geographical area to meet this

criterion.

The following two case studies, for Bohol in the Philippines and Tonga, present

examples of the use of capture-recapture methods for all-cause mortality, and

lessons for how these may be applied in the PICTs. A two source capture-recapture

analysis is also presented for Kiribati, a population that does not have three distinct

data sources available, to set boundaries for plausibility around mortality estimates.

� � �

176

5.5.1 Case study 4: Capture-recapture analysis in Bohol, Philippines – an

application in an “ideal” setting


Ian Riley Population Capture-recapture analysis of all-cause mortality data in Bohol,

Philippines. Population Health Metrics 2011, 9:9 (17 April 2011). doi:10.1186/1478-

7954-9-9

Case study four presents a two and three source analysis for the Bohol province in

the Central Visayas region of the Philippines. Although not a PICT, Bohol shares

many of the physical and social characteristics of the PICTs in regard to small widely

distributed populations. The region is essentially a “best-case” scenario for the

application of capture-recapture methods for assessing all-cause mortality data, as

there are three reporting systems all of reasonable quality, and therefore sufficient

information on each death to allow individuals in the different data collections to be

reconciled. These sources include two government sources, civil registry and MoH

data, and records from the Catholic church, which maintains records of births and

deaths following a standard format across all their parishes. The region is 99%

Catholic (329).

Key findings, as outlined in the paper, show capture-recapture methods as useful for

correcting incomplete data and generating estimates of all-cause mortality. They

may therefore have application in the PICTs, where many countries are likely to have

some, but not complete data. The method is particularly useful as it allows

examination of the data by age group, which is not possible with direct demographic

� � �

177

methods. This is particularly important in the PICTs, as the system assessments in

Chapter three demonstrated a high likelihood of differential reporting completeness

by age across most of the study countries in the absence of direct financial

incentives for reporting (such as in Nauru).

The study also highlights that even in this ideal setting, capture-recapture analysis

must be applied with a thorough understanding of the reporting system structure and

operation from which the data has been generated. This is particularly important in

the three-source analysis where there would otherwise be a significant potential for

inappropriate model selection (327, 330). While there are mathematical approaches

(such as the Aikake Information Criterion (328) and the Bayesian Information

Criterion (328) as described in the paper) to selecting the correct model, the results

show there may be several models that could be described as a “close-fit” based on

these criteria, but which give vastly different estimates of the scale of unreported

deaths. An understanding of the reporting system, such as through a thorough

system assessment, is therefore critical to appropriate model selection and

subsequently generating reliable estimates of mortality level. Three source analyses

are more robust than a two source analysis, as death reporting systems are rarely

completely independent in the real world.

Finally, the study demonstrated that even CRVS systems considered to generate

“good” quality data, as in the Philippines (4) should be assessed for completeness

and measures of mortality corrected accordingly. Findings from the study found

similar results to previous assessments in the Philippines (331, 332), supporting the

use of this method for correcting mortality data.

RESEARCH Open Access

Capture-recapture analysis of all-cause mortalitydata in Bohol, PhilippinesKaren L Carter1*, Gail Williams1, Veronica Tallo2, Diozele Sanvictores2, Hazel Madera2 and Ian Riley1*

Abstract

Background: Despite the importance of mortality data for effective planning and monitoring of health services,official reporting systems rarely capture every death. The completeness of death reporting and the subsequenteffect on mortality estimates were examined in six municipalities of Bohol province in the Philippines using asystem review and capture-recapture analysis.

Methods: Reports of deaths were collected from records at local civil registration offices, health centers andhospitals, and parish churches. Records were reconciled using a specific set of matching criteria, and both a two-source and a three-source capture-recapture analysis was conducted. For the two-source analysis, civil registry andhealth data were combined due to dependence between these sources and analyzed against the church data.

Results: Significant dependence between civil registration and health reporting systems was identified. There were8,075 unique deaths recorded in the study area between 2002 and 2007. We found 5% to 10% of all deaths werenot reported to any source, while government records captured only 77% of all deaths. Life expectancy at birth(averaged for 2002-2007) was estimated at 65.7 years and 73.0 years for males and females, respectively. This wasone to two years lower than life expectancy estimated from reconciled reported deaths from all sources, and fourto five years lower than life expectancy estimated from civil registration data alone. Reporting patterns varied byage and municipality, with childhood deaths more underreported than adult deaths. Infant mortality wasunderreported in civil registration data by 62%.

Conclusions: Deaths are underreported in Bohol, with inconsistent reporting procedures contributing to thissituation. Uncorrected mortality measures would subsequently be misleading if used for health planning andevaluation purposes. These findings highlight the importance of ensuring that official mortality estimates from thePhilippines are derived from data that have been assessed for underreporting and corrected as necessary.

BackgroundEffective planning and monitoring of health servicesrequires accurate measures of mortality by age and sex.Additionally, population health targets such as the Mil-lennium Development Goals [1] require reliable data tomeasure progress. Mortality data in the Philippines arecollected through the civil registration system, the healthsystem, and the parish records of local churches.Despite the importance of mortality data, official

reporting systems rarely capture every death [2]. Thechallenge for public health planners in many countriesis to meaningfully interpret the available data. A global

assessment of mortality data in 2003 listed the Philip-pines as having only “medium” quality death records,with reporting completeness estimated at 77% based ondata supplied to the World Health Organization [2].The completeness of death reporting and its impact onmortality estimates was examined in six municipalitiesof Bohol province in the Philippines using a systemreview and capture-recapture analysis. Variations inreporting patterns by age group, sex, and municipalitywere also examined.Bohol is a province of the Central Visaya region of the

Philippines. More than 95% of the provincial populationand study area is Catholic [3], although there is a smallbut localized Muslim population near the provincialcapital, Tagbilaran City. Major industries are agricultureand fishing [4]. The province has a population of

* Correspondence: [email protected]; [email protected] of Population Health, University of Queensland, Herston (Brisbane),Queensland, AustraliaFull list of author information is available at the end of the article

Carter et al. Population Health Metrics 2011, 9:9http://www.pophealthmetrics.com/content/9/1/9

© 2011 Carter et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative CommonsAttribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction inany medium, provided the original work is properly cited.




approximately 1,227,809 (2007) [3] living in 47 munici-palities and one city.Capture-recapture analysis uses the number of deaths

recorded by each combination of sources to estimatethe number of unreported deaths, and therefore the esti-mated total number of deaths [5]. Capture-recapturemethods have been used extensively to obtain estimatesof disease incidence and to assess coverage of disease-specific surveillance systems and registers [6-10]. Forall-cause mortality, this method provides an additionaltool to quantify and correct underregistration of deaths.Other methods compare reported deaths to expectedpatterns based on age distribution or require repeatedcensuses to evaluate survival [11,12].

MethodsThe study included Baclayon, Balilihan, Cortes, Dauis,and Panglao municipalities, and Tagbilaran city. Thesewere selected as part of a broader study to improveexisting methods for estimating mortality levels fromincomplete data and verbal autopsy methods for assign-ing cause of death under the Gates Grand Challenges inGlobal Health program [13]. The municipalities selectedwere included in this larger study due to their proximityand subsequent access to hospitals in Tagbilaran city.

System reviewA system review was conducted to assess data collectionand management within municipal mortality reportingsystems. The review comprised observation andmapping of processes; examination of forms, policydocuments, and electronic systems; and key informantinterviews held with local civil registrars, health officersand health center staff, hospital staff, church officials,cemetery managers, and staff of the provincial branch ofthe National Statistics Office (NSO).

Data CollectionProject staff visited local civil registration offices, healthcenters and hospitals, and parish churches regularlyduring 2006-2007 and extracted death report data fromlog books and registers using a standard form. Key vari-ables collected were: municipality of report, municipalityof residence, name of agency, nature of agency (localcivil registry [LCR], health center [HC], or church), fullname, sex, date of birth, date and place of death, andage at death. Population by age group, sex, and munici-pality was derived from the 2000 [14] and 2007 [15]censuses using exponential interpolation to provideestimates for intermediate years.

Data reconciliationEach death was assigned a unique number. Recordswere reconciled to create a single list of unique deaths

indicating source(s) in which each was recorded. Projectstaff manually matched deaths by comparing dataextraction forms from all sources. At least three of fivecriteria had to be met for records to be considered amatch: same full name (minor variations in spelling, var-iations in name order or a phonetic match wereallowed), age at death within one year, date of deathwithin five days, same municipality of residence, orsame sex. If more than one record was found as amatch, each record was required to share three criteriawith another record for the individual. A large numberof church records could not be matched because ofincomplete entries, and it was consequently decided thatoutstanding records matching on full name and date ofdeath would also be included. Records were thenentered into an Access database [16] that was analyzedto confirm all matches met stated criteria and nomatches had been missed, to remove duplicate records,and to correct minor data entry errors. Deaths withmissing age for a particular source were redistributedaccording to two imputed age distributions: (a) for thatsource, and (b) for all sources combined. In cases wherematched records did not have identical municipality orage at death, these variables were recorded separately,and sensitivity analyses were used to examine the influ-ence of such mismatches.The following measures of mortality were calculated

using standard methods [12]: age-specific mortality rate;infant mortality rate (IMR) (probability of dying before1 year of age); childhood mortality (probability of dyingbefore 5 years of age); adult mortality (probability ofdying between 15 and 59 years of age inclusive); and lifeexpectancy at birth (LE).

Two-source capture-recapture analysisA key assumption in a two-source analysis is thatsources are independent, i.e., reporting to one sourcedoes not influence the chance of the event beingreported to the other source [17-19]. The system reviewrevealed that LCR and HC were highly interdependentand so they were combined; this combination is hence-forth referred to as the “official” source. Unreporteddeaths were estimated using a two-source capture-recapture analysis of official and church sources. Datawere analyzed using Excel and SAS. Under the assump-tion that official and church sources are independent,the number of missing deaths (x) may be estimated asbc/a, where a, b, c are the numbers of deaths found byboth sources (a), only in the official records (b), andonly in the church records (c) [17]. Confidence intervalswere calculated using the maximum likelihood estima-tor. Unreported deaths were estimated for total deaths,with subgroup analyses for sex, age, and municipality.The estimated percentage of underreporting was


Page 2 of 9

examined by subgroup to identify variation in reportingpatterns. Dependence between official data and churchdata was assessed using the method of Hook and Regal[20], with a value of 1 signifying sources are indepen-dent, and a positive and negative dependency betweensources indicated by values less than and greater than 1,respectively.

Three-source capture-recapture analysisThree-source capture-recapture analysis allows for pair-wise dependence between sources [6-8,17], and usedLCR, HC, and church data separately. Data wereanalyzed using the SAS CATMOD procedure [21] fortotal deaths as well as by municipality and age group.Age was grouped as 0-4 years, 5-14 years, and 15+years. Each three-source analysis resulted in eight possi-ble models (Table 1). The minimal three-source modeluses three parameters, one for each source, and assumesindependence of all sources [7,17]. An extended modelincorporates three additional parameters, which allowsthe estimation of the three possible pair-wise dependen-cies (LCR and HC, LCR and church, and HC andchurch), generating three models with a single pair-wisedependency combination, and three models incorporat-ing two pair-wise dependencies. The “saturated model”has an additional parameter for modification of eachpair-wise dependence by the third source [7,17]. Modelswere selected on the basis of the system evaluation(qualitative assessment) and statistical criteria such asthe significance of associations between sources (usinglikelihood ratio tests) and goodness-of-fit criteria(Bayesian Information Criterion (BIC) and the AkaikeInformation Criterion (AIC)) [17]. Excel and STATAwere used for analysis [22].

Final mortality estimatesFinal estimates of the total number of deaths werederived from both the two- and three-source analyses.Models derived from each of these were assessed for

plausibility and fit based on the statistical techniquesdescribed above and the system assessment. Measures ofmortality, including IMR, childhood mortality, LE, andadult mortality, were then recalculated using the cor-rected estimates.

ResultsSystem AnalysisFilipino legislation [23,24] requires families of thedeceased person, regardless of the cause of death, toobtain a medical certificate of death and to present thisto the Local Civil Registrar to obtain an official deathcertificate prior to burial. Nearly all funerals are over-seen by the local Catholic Church, which also keeps arecord of deaths. Stillbirths are recorded using separateforms and do not appear on the death registers.At the local registry office, the death is recorded in the

register book and a death certificate issued to the family.A copy of the death certificate is sent to the NationalStatistics Office along with an electronic file if the muni-cipality also maintains an electronic database. Healthrecords may be derived from clinic log books, hospitalseparation data, or the medical certificate of death.While any medical practitioner can issue a certificate ofdeath, Filipino law requires that each medical certificateof death is submitted to the local municipal health offi-cer for review and signature [24]. A record of the deathshould be available from the health center even if themedical certificate was not originally issued there.Church records include the name of the deceased, dateof death, age at death, usual place of residence, andplace of death. These records are maintained in a stan-dardized format as set out in log books sold to theparishes through the church hierarchy.The system review demonstrated significant depen-

dence in the reporting of deaths between the LCR andthe HC. Families often approached both health and civilregistry offices independently, but medical certificates ofdeath were also forwarded directly by several municipal

Table 1 Models from three-source capture-recapture analysis (all ages) for Bohol study area, 2002-2007

Model* d.f G2 p-value AIC BIC Estimator (x) Total (N) Lower 95% CI Upper 95% CI

Independent 3 1900.00 <0.0001 1894 1879 147 8222 8208 8236

1-2 2 77.43 <0.0001 73 63 382 8457 8422 8493

1-3 2 1800.00 <0.0001 1796 1786 69 8144 8130 8157

2-3 2 1900.00 <0.0001 1896 1886 148 8223 8206 8240

1-2,1-3 1 69.75 <0.0001 68 63 520 8595 8476 8715

1-2, 2-3 1 16.50 <0.0001 15 9 552 8627 8560 8695

1-3, 2-3 1 1800.00 <0.0001 1798 1793 62 8137 8123 8150

1-2, 1-3, 2-3 0 -0.0002 1.000 0 0 908 8983 8748 9218

Source 1 - LCR data; Source 2 - health data; Source 3 - church data.

* Notation indicates associations between sources taken into account by the model, i.e., the 1-2 model accounts only for an association between source 1 (LCR)and source 2 (health center).


Page 3 of 9

health offices to the LCR. HCs either recorded deathsdirectly from the medical certificate of death at thetime it was issued or reviewed, or else kept no recordof the death at that time and later relied on either amonthly report of deaths from the LCR or on thefamily returning a copy of the official death certificateto the health office after registration. There was noofficial contact between the churches and governmentoffices and no apparent dependence in reportingpractices.In practice, there was considerable variation in report-

ing processes across the six municipalities. Noted varia-tions included whether families were required to firstreport the information to the local registrar or healthcenter, whether there was someone present at the healthcenter who could certify cause of death, and the chargesassociated with reporting. Practices for dealing with thenonavailability of a medical officer to sign the medicalcertificate also varied. Panglao and Dauis had a recipro-cal arrangement between their health centers. In Cortes,the mayor was the alternative authorized person, whilein the other municipalities, the reporting waited untilthe doctor returned.

Observed DeathsCombining all sources, 8,075 unique deaths wererecorded in the study area between 2002 and 2007, giv-ing an observed crude mortality rate of 7.06 deaths per1,000 population. Official sources (LCR and HC)recorded 6,696 (83%) of these deaths for a crude mortal-ity rate of 5.85 deaths per 1,000 population. Of the offi-cially recorded deaths (6,696), 6,297 (94%) wererecorded by the LCR, and 4,999 (75%) by the HC. Thechurch recorded 6,621 deaths.Observed LE at birth for the study area (2000-2007)

was 67.3 years for males and 74.2 years for females.IMR was 19.2 deaths per 1,000 live births for bothmales and females, with childhood mortality 23.2 and23.5 deaths per 1,000 live births for males and females,respectively.

Two-source capture-recapture analysisThe overall two-source analysis of aggregated officialrecords and church records estimated a total of 8,151deaths in the study area for 2002-2007. Aggregation ofsubgroup analyses within strata of age and municipalityprovided estimates that differed little from the overallestimate. Sensitivity analyses revealed that the source ofinformation on age and municipality (in cases of non-match on these) did not affect results, and the LCRrecord was used wherever possible to assign values. Offi-cial data were found to be virtually independent fromchurch data, with dependence coefficient values greaterthan 0.99.

Recording completeness varied by age (Figure 1), withchild deaths less likely to be recorded than adult deaths.Age had the greatest effect on health center records.Both official sources of data showed increasing comple-teness with increasing age at death. Variations in report-ing by age were not consistent across municipalities.Childhood deaths were less likely to be recorded inBaclayon, Dauis, and Panglao than in the othermunicipalities.

Three-source capture-recapture analysisEstimates of total deaths ranged from 8,137 to 8,983based on aggregate data, depending on the modelselected (Table 1). This equates to deaths from allsources being underregistered by 1% to 10%. Estimatesof total deaths varied further when disaggregated by agegroup and totaled to produce an overall figure.The saturated model was selected as the best fit (with

the lowest significant value for the AIC and BIC) fromthe analysis of the aggregate data. This also generatedthe highest estimate of 8,983 total deaths (Table 1). Thenext best fit falls to any of the models incorporating theassociation between the LCR and health data.Pair-wise dependence between sources was, however,

not consistent for all age groups. Table 2 shows mea-sures of association between the three sources and con-firms the dependence identified in the system reviewbetween the LCR and health reporting systems. Theodds ratio is high in the dataset as a whole (14.7) and inall three age groups. It is particularly high in persons 5-14 years (12.39) and in persons 15 years and older(15.86). There is additional evidence, however, of anassociation between the other combinations of sources,although the magnitude of this association as measuredby the odds ratio was much smaller (Table 2). Thisassociation reaches statistical significance in the datasetas a whole and in persons 15 years and older.Taking both the system review and these statistical

tests into account, these results did not clearly identifyany one model as the only reliable solution, but ratherindicated several plausible options depending on thelevel of age disaggregation used.

Final mortality estimatesSix models were identified as plausible and a good fit forthe data from the two- and three-source analyses. Threeyielded low values, and three yielded high values for thetotal number of deaths (Table 3). These models allaccounted for the principal association demonstratedbetween the LCR and health data, and the differences inreporting by age.The three low estimates were derived from those

models that used: a) a two-source analysis (with LCRand health center data combined as a single source


Page 4 of 9

against church data) including unknown age deaths(with age redistributed according to source category) byage; b) a three-source analysis with the model account-ing for association between the LCR and health datausing ages redistributed according to total age distribu-tion; and c) the three-source analysis as defined in (b)with estimates calculated separately for broad agegroups (<5 years, 5-14 years, and 15+ years) andsummed, rather than as a single application for all agescombined.The three high estimates were derived from models

that used: a) a three-source analysis (including unknownage deaths without age distribution) using the saturatedmodel (model accounting for an association betweeneach possible pair of sources); b) a three-source analysiswith the saturated model using ages redistributedaccording to total age distribution (then calculated sepa-rately for <5 years, 5-14 years, and 15+ years); and c) athree-source analysis using ages redistributed according

Table 2 Association between reporting sources byage-group - three-source analysis, Bohol studyarea, 2002-2007

Age group Association OR Phi p

All ages S1 * S2 14.74 (12.79,17.00) 0.52 (0.01) < 0.0001

S1 * S3 1.64 (1.30, 2.08) 0.06 (0.02) < 0.0001

S2 * S3 1.75 (1.53, 1.99) 0.11 (0.01) < 0.0001

0-4 years S1 * S2 4.81 (3.26, 7.10) 0.36 (0.04) < 0.0001

S1 * S3 0.58 (0.31, 1.09) -0.11 (0.06) 0.0726

S2 * S3 0.85 (0.51, 1.42) -0.04 (0.06) 0.5373

5-14 years S1 * S2 12.39 (4.57,33.58) 0.52 (0.08) < 0.0001

S1 * S3 0.94 (0.17, 5.13) -0.01 (0.12) 0.9399

S2 * S3 0.98 (0.36, 2.62) -0.00 (0.10) 0.965

15 + years S1 * S2 15.86 (13.42, 18.74) 0.50 (0.01) < 0.0001

S1 * S3 2.01 (1.54, 2.61) 0.08 (0.02) < 0.0001

S2 * S3 1.78 (1.55, 2.04) 0.11 (0.01) < 0.0001

Source 1 - LCR data, Source 2 - health data, Source 3 - church data.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Reporting completeness

Age Group

Reconciled data

LCR

Health

Church

Figure 1 Proportion of estimated deaths reported, Bohol study area, 2002-2007: two-source analysis.


Page 5 of 9

to total age distribution, calculated separately for <5years (1-2 model), 5-14 years (1-2 model), and 15+ years(saturated model) (Table 3). Low estimates equate to 5%of all deaths in the study area going unreported by anyof the sources, while the high estimates set unreporteddeaths at 10%.Final estimates were derived by selecting the average

of both high and low models and the extreme high andlow values of the 95% confidence intervals for any ofthese models selected. As such, total deaths in the studyarea for 2002-2007 are estimated to be 8,729, with arange of 8,079 to 9,445 deaths.The final estimates for total deaths imply that for the

period 2002-2007, only 77% (95% CI: 71%-83%) ofdeaths in the study were reported to official sources,with 93% (95% CI: 85%-100%) being reported to anysource, including the church. Civil registration was only72% complete (95% CI: 67%-78%). The crude mortalityrate for the study area is therefore estimated at 7.62deaths per 1,000 population (95% CI: 7.19-8.13). Lifeexpectancy at birth is estimated at 65.7 years for malesand 73.0 years for females, one to two years lower thanlife expectancy estimated from reconciled reporteddeaths from all sources, and four to five years lowerthan life expectancy estimated from civil registrationdata alone. The corrected IMR was 22.7 and 22.2 deathsper 1,000 live births for males and females, respectively,indicating that infant deaths were underreported fromall sources by 15% and by 62% from the LCR data(Figure 2). Childhood mortality was 27.3 and 27.1 deathsper 1,000 live births for males and females, respectively,again indicating underreporting by all sources of 13% to

15%. Adult mortality was estimated at 267.2 deaths per1,000 population for males and 137.3 deaths per 1,000population for females, indicating underreporting by theLCR of 24% to 29%.

DiscussionThese findings demonstrate that reporting of deaths inBohol to both official and church sources is incomplete.Estimates suggest that only 77% of deaths are recordedin official LCR or health records, while 5% to 10% of alldeaths are not reported to any source. Civil registrationwas estimated to be only 72% complete for all ages.Although the impact of underreporting of deaths onsummary mortality measures such as LE at birth was inthe order of four to five years, there was a much greaterimpact on age-specific measures of IMR and childhoodmortality, which were 15% to 18% higher than those cal-culated from the unadjusted reconciled data. Infantdeaths were underreported by 62% in the civil registra-tion data. Adult mortality was also 11% higher than theoriginal estimate from the reconciled data.The three-source analysis allows for departure from

the assumption of independence between sources, essen-tial for the two-source analysis. However, the addition ofany further source to the capture-recapture analysisincreases the uncertainty in the estimate and the widthof confidence intervals [17,19]. The additional modelsgenerated for each possible set of associations betweensources as additional sources are added [17] results in agreater onus on researchers to select an appropriatemodel based on their knowledge of the reporting sys-tems. Conducting the three-source analysis is therefore

Table 3 Final estimates of total deaths by method, Bohol study area, 2002-2007

Method/Model Total estimated deaths Crude mortality rate (deathsper 1,000 population)

Low estimates

2-source analysis (combined LCR and health center) including unknown age deaths(with age redistributed by source category) by age

8459 (8226-8692) 7.39 (7.19 - 7.60)

3-source analysis with the 1-2 model using ages redistributed by total age distribution 8457 (8422-8493) 7.39 (7.36 - 7.42)

3-source analysis with the 1-2 model using ages redistributed by total age distributioncalculated separately for <5 yrs, 5-14 yrs, and 15+ years)

8481 (8413 - 8546) 7.41 (7.35 - 7.47)

High estimates

3-source analysis (including unknown age deaths without age distribution) - saturatedmodel

8983 (8748-9218) 7.85 (7.64 - 8.06)

3-source analysis with the saturated model using ages redistributed by total agedistribution (calculated separately for <5 yrs, 5-14 yrs, and 15+ years)

8964 (8676 - 9305) 7.83 (7.58 - 8.13)

3-source analysis calculated separately for <5 yrs (1-2), 5-14 yrs (1-2), and 15+ years(saturated)

9006 (8734 - 9275) 7.87 (7.63 - 8.11)

Averages

Average of all accepted estimates with maximum 95% confidence intervals (ascalculated by any method)

8725 (8354 - 9266) 7.62 (7.19 - 8.13)


Page 6 of 9

only justified if the third source contributes deaths notrecorded by the other two sources to the analysis. Thetwo-source analysis found very weak associationbetween official and church records. Although thethree-source analysis confirmed this, the association wasstatistically significant and thus should not be ignored(Table 2). The two-source analysis is viewed as a lowerlimit for the estimate of total deaths, as the associationindicates that deaths reported to official sources aremore likely to also be reported to the church, resultingin an underestimation of unreported deaths. The three-source results accounting for this dependence thereforerange upward from the two-source estimate. Since onemodel cannot be identified as the only model to fit thedata, the various models are used to set upper andlower limits of the estimate of unreported deaths, andtherefore total deaths, for the study area.Population estimates were derived from the 2000 [14]

and 2007 [15] censuses, using linear interpolation toprovide estimates for intermediate years, and are thusmore likely to have overestimated the population forintermediate years, biasing mortality rates downward.The broad matching criteria used to reconcile the datasources and multiple checks of the matching processminimize the risk of significant errors in the reconcilia-tion of data sources. Further, in this setting, under-matching due to incomplete data is more likely thanovermatching. This would lead to overestimation of the

number of deaths not reported [17], and therefore leadto an inflated estimate of mortality, thus supporting theuse of the higher estimates of total deaths as an upperlimit.The small number of deaths in the study area meant

that while disaggregation by 5-year age group or munici-pality was possible with the two-source analysis, manyof the results became unstable when the data wereeither disaggregated further by year, or when the officialsources were separated for the three-source analysis.The two-source analysis was therefore used to assessvariations in reporting patterns. The important differ-ence in reporting completeness between childhooddeaths and adult deaths was accounted for in the three-source analysis by disaggregating data by broader agegroups (0-4, 5-14, and 15+ years).Capture-recapture analysis requires that: the record

sources capture deaths from the same population, thepopulation is closed, and that every death has the sameprobability of being reported [17,18]. The analysis is alsoaffected by the dependence between the data sources,with independence required for an accurate result froma two-source analysis. These conditions, however, arerarely fully met with human populations, and such ana-lysis therefore requires a solid understanding of report-ing systems in order to assess the magnitude of anydependence and population movements in or out of thearea. In the Bohol study area, while the population is

0 5 10 15 20 25

Civil Registration

All observed deaths

C-RC estimates

Deaths per 1000 live births

Data source

Figure 2 Estimated infant mortality rate by source, Bohol study area, 2002-2007.


Page 7 of 9

not strictly closed, migration both in and out is low,with the latest census reporting an annual growth rateof 1.06 percent for the period 2000 to 2007, althoughmost of this change is accounted for through births anddeaths [3]. Dependence between the LCR and healthreporting systems was accounted for in both the two-and three-source analyses, and the high proportion ofCatholics in the study population meant that for nearlyall deaths, there was a reasonable potential for thatdeath to be reported to the local parish and thus beincluded in this study. As such, while there were minordeviations from the assumed conditions for a capture-recapture analysis, Bohol is an ideal situation in whichto conduct this analysis, and results are highly unlikelyto have been affected by assumptions of the methodbeing violated. The selection of municipalities based onproximity to the urban center of Tagbilaran would likelyresult in the estimates of completeness presented herebeing considered a best-case scenario, with reportingfrom outlying areas likely to be even less complete.Additionally, the gender and age distribution in theseareas is almost identical to that for the broader province[14,15], suggesting that there is some generalizability tothe local area. However, the variation in reporting pat-terns and procedures by municipality identified by thisstudy, despite national legislation and procedural policy,indicates that it is not appropriate to apply these find-ings and assessment to other areas of the Philippineswithout an understanding of reporting processes at alocal level.While there have been very few studies on reporting

completeness in the Philippines, our findings of reportingcompleteness to official sources are consistent with a2003 evaluation of mortality data in the Philippines basedon data obtained by the World Health Organization [2].Completeness of death registration is often assessed aspart of the analyses of census data; however, this has notbeen reported in either of the two most recent censuses,2000 and 2007, in the Philippines [14,15]. Although esti-mates for this study were calculated from only six of the48 municipalities in Bohol, they are significantly higherthan official published estimates that put IMR for Boholprovince at 14.6 in 2000 and 8.53 in 2004 [4]. Publishedestimates of LE at birth for Bohol vary, with estimates for2000-2005 from official sources in the Philippines ran-ging from 65.3 to 68.19 for males and 71.0 to 72.9 forfemales (2000-2005) [3,4]. LE estimates from this studyare consistent with the lower published estimates fromthe National Statistics Office (NSO) for males, but arehigher than the published estimates for females [3]. Asboth sets of published data are derived from civil registra-tion within the province, it is probable that estimatesreleased by the NSO have been corrected for someundercounting in the data.

ConclusionsThis study demonstrates that even though the Philip-pines has national legislation and policies on deathregistration, there is significant underreporting of deathsto official sources in Bohol. While this had only a smallimpact on summary measures of mortality, such as thecrude death rate and LE, the impact of underreportingon age-specific measures such as IMR and childhoodmortality was much greater. Uncorrected mortality mea-sures would subsequently be misleading if used forhealth planning and evaluation purposes. These findingshighlight the importance of ensuring that official mortal-ity estimates from the Philippines are derived from datathat have been assessed for underreporting and cor-rected if necessary.Official LCR and health sources were more incomplete

that church records across nearly all age groups (Figure1), highlighting the potential to improve reporting toofficial sources by developing stronger links betweenlocal civil registrars and parish churches in this setting.Completeness of death reporting within the health sys-tem could be substantially improved by reducing lossesfrom the system through ensuring that a copy of eachmedical certificate of death is either kept or entered intoa logbook rather than relying on the data being returnedfrom the LCR. Variations in local reporting practicesfurther indicate that additional improvements in report-ing completeness may be possible through adaptinglocal practices to align with a standard nationalapproach.Finally, capture-recapture methods are an important

tool for assessing and correcting mortality data forreporting completeness, and have been demonstratedhere to provide improved understanding of mortalityestimates derived from a system generally recognized tobe of reasonable quality. While not an entirely newapplication of capture-recapture analysis, the methodhas potential to significantly improve our understandingof mortality in settings where multiple incomplete data-sets are available, particularly in small areas (such as thesmall countries of the Pacific Islands or provinces of thelarger countries of Asia) where it is possible forresearchers to develop a strong understanding of systemprocesses in order to make sensible decisions aboutboth model selection and matching criteria.

AcknowledgementsThe authors would also like to thank Dr Marilla Lucero, Dr Alvin Tan, and theother staff from the Research Institute of Tropical Medicine, Manila, workingon the Bohol field site for their data collection and guidance during thiswork. Funding for this research was provided under the Population HealthMetrics Research Consortium project of the Gates Grand Challenges inGlobal Health program. The funding body had no role in study design; inthe collection, analysis, and interpretation of data; in the writing of themanuscript; or in the decision to submit the manuscript for publication.


Page 8 of 9

Author details1School of Population Health, University of Queensland, Herston (Brisbane),Queensland, Australia. 2Research Institute of Tropical Medicine, Manila,Philippines.

Authors’ contributionsKC designed the study; conceived and conducted the system review;designed and tested matching criteria; prepared the data for the three-source analysis; performed the statistical analysis for the two-source analysisand final model selection including interpretation of final results; and draftedthe manuscript. GW made substantial contributions to analysis andinterpretation of data, performed the statistical analysis for the three-sourceanalysis and revised the manuscript critically. VT coordinated data collection,collation and matching, and made substantial contributions to theinterpretation of data. DS coordinated initial data cleaning, management,and tabulation of records, and made substantial contributions to theinterpretation of data.HM made substantial contributions to the acquisition and interpretation ofdata. IR conceived and coordinated the study and revised the manuscriptcritically for important intellectual content. All authors read and approvedthe final manuscript.

Competing interestsThe authors declare that they have no competing interests.

Received: 23 December 2010 Accepted: 17 April 2011Published: 17 April 2011

References1. UNESCAP: Achieving the Millennium Development Goals in an Era of

Global Uncertainty: Asia-Pacific Regional Report 2009/10. United NationsEconomic and Social Commission for Asia and the Pacific Statistics, AsianDevelopment Bank, United Nation Development Funds. Bangkok; 2010[http://www.mdgasiapacific.org/regional-report-2009-10].

2. Mathers C, Ma Fat D, Inouie M, Rao C, Lopez A: Counting the dead andwhat they died from: an assessment of the status of global cause ofdeath data. Bulletin of the World Health Organization 2005, 83(3):171-7.

3. Philippines National Statistics Office: Press Release 03: Bohol Recorded aPopulation of 1.2 Million Persons (Results from the 2007 Census of Population)Manila; 2010 [http://www.census.gov.ph/data/pressrelease/2010/pr1003tx.html].

4. Provincial government of Bohol: Official Website. [http://www.bohol.gov.ph/].

5. Kerr CB, Taylor RJ, Heard G, (eds): Handbook of Public Health MethodsRoseville (Australia): McGraw Hill; 1999.

6. International Working Group for Disease Monitoring and Forecasting(IWGDMF): Capture-Recapture and Multiple-Record Systems Estimation 1:History and Theoretical Development. American Journal of Epidemiology1995, 142(10):1047-1057.

7. Wittes JT, Sidel VW: A generalization of the simple capture-recapturemodel with applications to epidemiological research. Journal of ChronicDiseases 1968, 20:287-301.

8. Lewden C, Jougla E, Alioum A, Pavillon G, Lièvre L, Morlat P, Salmon D,May T, Chêne G, Costagliola D, the Mortalité 2000 Study Group: Number ofdeaths among HIV-infected adults in France in 2000, three-sourcecapture-recapture estimation. Epidemiology and Infection 2006,134(6):1345-1352.

9. McCarty DJ, Tull ES, Moy CS, et al: Ascertainment corrected rates:applications of capture-recapture methods. Int J Epidemiol 1993,22:559-65.

10. Clothier HJ, Vu T, Sundararajan V, Andrews RM, Counahan M, Tallis GF,Lambert SB: Invasive pneumococcal disease in Victoria: a bettermeasurement of the true incidence. Epidemiol Infect 2008, 136:225-231.

11. Murray CJL, Rajaratnam JK, Marcus J, Laakso T, Lopez AD: What Can WeConclude from Death Registration? Improved Methods for EvaluatingCompleteness. PLoS Med 2010, 7(4):e1000262.

12. Rowland DT: Demographic Methods and Concepts New York: OxfordUniversity Press; 2003.

13. Grand Challenges in Population Health: Population Health MetricsResearch Consortium Project. [http://www.grandchallenges.org/

MeasureHealthStatus/Challenges/PopulationHealth/Pages/ConsortiumProject.aspx].

14. Philippines National Statistics Office: Table: Household Population by AgeGroup, Sex, and City/Municipality. 2000 Census of Population and HousingFinal Counts by Region, by Province and by Barangay Manila; 2001, [datarequested from NSO]..

15. Philippines National Statistics Office: Table 2: Household Population byAge Group, Sex, and City/Municipality: 2007. National Census Results byProvince. Results from the 2007 Census of Population Manila; 2010 [http://www.census.gov.ph/data/sectordata/2007/Bohol_Table%202.pdf].

16. Microsoft: Access: Computer program 2007.17. Hook E, Regal R: Capture-Recapture Methods in Epidemiology: Methods

and Limitations. Epidemiologic Reviews 1995, 17(2):243-264.18. Johnson J, Omland K: Model Selection in ecology and evolution. Trends in

Ecology and Evolution 2004, 19(2):101-108.19. Sekar C, Deming DW: On a method of estimating birth and death rates

and extent of registration. Journal of American Statistical Association 1949,44:101-115.

20. Hook E, Regal R: Validity of Methods for Model Selection, Weighting forUncertainty and Small Sample Adjustment in Capture-RecaptureEstimation. American Journal of Epidemiology 1997, 145(12):1138-1144.

21. SAS: The CATMOD Procedure , Second[http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#catmod_toc.htm], SAS/STAT(R) 9.2 User’s Guide..

22. Heiden M: RECAP: Stata module to perform capture-recapture analysisfor three sources with Goodness-of-Fit based confidence intervals.Statistical Software Components S456859, Boston College Department ofEconomics 2007.

23. Republic of the Philippines: Act No 3753: Law on Registry of Civil StatusManila; 1930.

24. Republic of the Philippines: Presidential Decree 651: Requiring the registrationof births and deaths in the Philippines which occurred from January 1, 1974and thereafter Manila; 1975.

doi:10.1186/1478-7954-9-9Cite this article as: Carter et al.: Capture-recapture analysis of all-causemortality data in Bohol, Philippines. Population Health Metrics 2011 9:9.










Page 9 of 9

http://www.mdgasiapacific.org/regional-report-2009-10

http://www.ncbi.nlm.nih.gov/pubmed/15798840?dopt=Abstract



http://www.census.gov.ph/data/pressrelease/2010/pr1003tx.html

http://www.census.gov.ph/data/pressrelease/2010/pr1003tx.html

http://www.bohol.gov.ph/

http://www.bohol.gov.ph/













http://www.grandchallenges.org/MeasureHealthStatus/Challenges/PopulationHealth/Pages/ConsortiumProject.aspx



http://www.census.gov.ph/data/sectordata/2007/Bohol_Table%202.pdf

http://www.census.gov.ph/data/sectordata/2007/Bohol_Table%202.pdf







http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#catmod_toc.htm



� � �

187

5.5.2 Case study 5: Capture-recapture analysis in Tonga – application in a

Pacific Island with multiple data sources

Hufanga S, Carter KL, Rao C, Lopez AD, Taylor R. Mortality trends in Tonga: an

assessment based on a synthesis of local data. Population Health Metrics, 2012

Aug 14;10(1):14.

Case study five examines the application of capture-recapture analysis techniques

in Tonga. There are three primary data sources in Tonga: two separate collections

from the MoH (monthly nursing reports and vital registration through the HIS), and

civil registration through the Ministry of Justice. The Prime Minister’s office also

records deaths reported through local community leaders (164) for electoral

purposes, although this collection is considered to be very incomplete, and is

affected by data quality issues that affect the accuracy of reconciliation with other

sources. Out-migration in Tonga is relatively high (-16.6% in 2010 (110)) making

methods such as the Brass Growth Balance method unsuitable, as noted previously.

The analysis was therefore restricted to the main island of Tongatapu (80% of the

population, 2006 Census) where the larger population centre (and possibly the

greater employment and social opportunities) minimise the effect of migration on a

per capita basis.

Differences in collected variables, the quality of data collection and societal

influences, such as the use of different names and titles (including customary titles,

maiden and married names, baptismal names and nicknames) in different settings,

make reconciling the data sources one of the key issues to overcome in a capture-

� � �

188

recapture analysis. In this study, criteria for matching individuals were adapted from

the previous work in the Philippines (333).

Key findings from the paper show the process of reconciling data from multiple

sources is of significant benefit in improving the completeness of death reporting and

subsequently improving the validity of data generated, even if further analyses are

not undertaken. However, data sets such as that held by the Prime Minister’s office,

which have low coverage of the population, should not be used for capture-recapture

analysis, as they break the assumption of equal catchability for each death, and may

introduce significant biases into the results.

As discussed in the paper and seen in Figure 18, findings demonstrate patterns of

reporting completeness vary substantially by age and source, as expected from the

system assessment in Chapter three, and highlight the advantage of capture-

recapture methods over direct demographic methods in the ability to examine age

groups separately.

� � �

189

Figure 18: Reporting completeness of deaths in Tonga by age group, sex and data source (2001-2009)

The reconciled data set provides a minimum value for the mortality estimates derived

through capture-recapture analysis. Published IMR estimates varied significantly

through to the late 1990’s where most estimates converge to a narrower range

between 10 and 20 deaths per 1000 live births. Findings from the empirical data

were consistent with this range, and did not demonstrate any significant trend.

Published estimates of LE from 2000 onwards varied from 65-75 years for males,

and 68-74 years for females, with most clustered around 70-71 for males and 72-73

for females. The reconciled empirical data for 2005-2009 produces an estimate of LE

of 65.2 years (95% CI: 64.6-65.8) for males, and 69.6 years (95% CI: 69.0–70.2) for

females; several years lower than both the MoH and Census estimates. However,

based on the CRVS systems and results of the data analysis presented here, it is

� � �

190

highly likely even reconciled data are under-enumerated and mortality may be even

higher than these reconciled estimates indicate. Final plausible ranges of LE for

2005-2008 from the study are therefore 60.4 - 64.2 years for males and 65.4 - 69

years for females.

Hufanga et al. Population Health Metrics 2012, 10:14http://www.pophealthmetrics.com/content/10/1/14


Mortality trends in Tonga: an assessment basedon a synthesis of local dataSione Hufanga1,2, Karen L Carter2,3*, Chalapati Rao2, Alan D Lopez2 and Richard Taylor2,4

Abstract

Background: Accurate measures of mortality level by age group, gender, and region are critical for health planningand evaluation. These are especially required for a country like Tonga, which has limited resources and worksextensively with international donors. Mortality levels in Tonga were examined through an assessment of availablepublished information and data available from the four routine death reporting systems currently in operation.

Methods: Available published data on infant mortality rate (IMR) and life expectancy (LE) in Tonga were soughtthrough direct contact with the Government of Tonga and relevant international and regional organizations. Datasources were assessed for reliability and plausibility of estimates on the basis of method of estimation, originalsource of data, and data consistency. Unreliable sources were censored from further analysis and remaining dataanalysed for trends.Mortality data for 2001 to 2009 were obtained from both the Health Information System (based on medicalcertificates of death) and the Civil Registry. Data from 2005 to 2009 were also obtained from the ReproductiveHealth System of the Ministry of Health (MoH) (based on community nursing reports), and for 2005–2008, datawere also obtained from the Prime Minister’s office. Records were reconciled to create a single list of unique deathsand IMR and life tables calculated. Completeness of the reconciled data was examined using the Brassgrowth-balance method and capture-recapture analysis using two and three sources.

Results: Published IMR estimates varied significantly through to the late 1990s when most estimates converge to anarrower range between 10 and 20 deaths per 1,000 live births. Findings from reconciled data were consistent withthis range, and did not demonstrate any significant trend over 2001 to 2009.Published estimates of LE from 2000 onwards varied from 65 to 75 years for males and 68 to 74 years for females,with most clustered around 70 to 71 for males and 72 to 73 for females. Reconciled empirical data for 2005 to 2009produce an estimate of LE of 65.2 years (95% confidence interval [CI]: 64.6 - 65.8) for males and 69.6 years (95% CI:69.0 – 70.2) for females, which are several years lower than published MoH and census estimates. Adult mortality(15 to 59 years) is estimated at 26.7% for males and 19.8% for females. Analysis of reporting completeness suggeststhat even reconciled data are under enumerated, and these estimates place the plausible range of LE between 60.4to 64.2 years for males and 65.4 to 69.0 years for females, with adult mortality at 28.6% to 36.3% and 20.9% to27.7%, respectively.

Conclusions: The level of LE at a relatively low IMR and high adult mortality suggests that non-communicablediseases are having a profound limiting effect on health status in Tonga. There has been a sustained history ofincomplete and erroneous mortality estimates for Tonga. The findings highlight the critical need to reconcileexisting data sources and integrate reporting systems more fully to ensure all deaths in Tonga are captured and theimportance of local empirical data in monitoring trends in mortality.

Keywords: Mortality, Cause of death, Noncommunicable diseases, Medical record review, Death certification, Tonga,Pacific Islands

* Correspondence: [email protected] of Population Health, University of Queensland, Brisbane, Australia3Secretariat of the Pacific Community, Noumea, New CaledoniaFull list of author information is available at the end of the article

© 2012 Hufanga et al.; licensee BioMed CentraCommons Attribution License (http://creativecreproduction in any medium, provided the or

l Ltd. This is an Open Access article distributed under the terms of the Creativeommons.org/licenses/by/2.0), which permits unrestricted use, distribution, andiginal work is properly cited.



Hufanga et al. Population Health Metrics 2012, 10:14 Page 2 of 11http://www.pophealthmetrics.com/content/10/1/14

BackgroundMortality level is a key measure of population health,and accurate measures of mortality level by age group,gender, and region are critical for health planning andevaluation. Additionally, population health targets suchas the Millennium Development Goals [1] require reli-able data to measure progress. Reliable health statisticsare crucial for countries like Tonga, which have limitedresources and the international community of partnersand donors that support health programs.Tonga is a small Pacific Island country with 36 inhab-

ited islands. The main island groupings are Tongatapu(including the capital Nuku’alofa) which houses 70% ofthe population, and Ha’apai, Vava’u, Eua, and Niua’s.Tonga has an estimated population of 103,000 (2009)[2], of which 38% are under 15 years of age [2]. Healthservices are provided through a national hospital, threeouter island hospitals, 11 reproductive health centers,and 15 community health centers [3].Despite the importance of mortality data, official

reporting systems rarely capture every death [4]. A glo-bal assessment of mortality data in 2003 listed Tonga ashaving “low” quality death records, with reporting com-pleteness estimated at 86% [4]. Substantial variation inpublished estimates of mortality, particularly for infantand childhood mortality, have been noted [5] and makesassessment of health trends over time difficult.Deaths in Tonga are recorded through three organiza-

tions and four systems. These are: the Health Informa-tion System (HIS) and Reproductive Health SurveillanceSystem (RHS) operated by the Ministry of Health(MoH), the Civil Registry (CR) system managed by theMinistry of Justice, and reporting through the PrimeMinister’s office (Figure 1). Coverage of these systems, inprinciple, is nationwide. Burials for deceased persons in

Town Officer

Civil Registrar’s O(Ministry of Justi

Death Certificate

N

Family

District Officer

Prime Minister’sOffice

Monthly report

Quarterly report

Figure 1 Routine mortality data reporting systems in Tonga.

Tonga occur primarily on community land that is mana-ged by local leaders and do not generally require a for-mal burial permit.Tongan legislation [6] requires families to report

deaths to the Registrar or the nominated Magistrate whoperforms this function for outer islands. The CR officeoften requests that the family produce a medical certifi-cate of death in order to complete the registration; how-ever, this is not required by legislation. Deaths ofTongan citizens that occur overseas may be registered inthe same manner.For deaths in a hospital or health center, the attending

doctor will issue a medical certificate of death. For homeand community deaths, local nurses complete a “noticeof cause of death,” which is used by the certifying doctorto complete a medical certificate of death [7]. A copy ofthis certificate is sent by the hospital to the Health Infor-mation Section, where the data are coded and enteredonto the HIS database. District nurses are also requiredto submit monthly reports, including details of knowndeaths. These forms are forwarded to the reproductivehealth office. Data are aggregated by hand for inclusionin the annual report.Reporting to the Prime Minister’s office is primarily

for the purposes of maintaining the electoral roll, al-though deaths at all ages are collected. Reports are sub-mitted monthly by locally elected community leaders[8].In this study, mortality levels in Tonga were examined

through an assessment of available published informa-tion and data available from the four routine deathreporting systems in operation. Empirical data fromthese systems were reconciled to derive summary mea-sures of mortality, including infant mortality rate (IMR),childhood mortality (probability of dying before the age

fficece)

ational Statistics Office

National Planning Office / Process

HealthSystem

Doctor* Nurse #

Health PublicInformation HealthSystem Nursing

MedicalCertificateof Death

Monthlyreport

NoticeofDeath


of 5, 5q0), average life expectancy at birth (LE), and adultmortality (probability of dying between ages 15 and 59,

45q15). The reconciled dataset was further examined forpotential under-enumeration of deaths using both aBrass analysis [9] and capture-recapture techniques [10],and the plausibility of the derived mortality estimates isdiscussed.

MethodsMeasures included in the analysis and reported in thispaper are from two sources: (1) previously publishedreports [11-60] and (2) empirical data on deaths by agegroup, sex, and period for 2001 to 2009 obtained fromlocal sources. As key informant interviews with theregistrars’ office indicated, most overseas deaths regis-tered are for Tongan residents. All deaths registered inTonga were included.

Published data collection and analysisData collectionAvailable published secondary data [11-60] on mortalityin Tonga were sought through direct contact with thegovernment of Tonga and international and regionalorganizations with an interest in health or mortality inTonga. A literature search was conducted, including aninternet search of websites for United Nations agencies(UNDP, UNICEF, UNFPA, and WHO); regional institu-tions such as the Secretariat of the Pacific Community

Figure 2 Published estimates of Infant Mortality Rate (IMR) by data soMoH = Ministry of Health, SPC = Secretariat of the Pacific Community, WHOrange of best estimates from this study (Table 2) for 2001–2004 and 2005–

(SPC), the World Bank (WB), and the Asian Develop-ment Bank (ADB); and non-governmental organizations.

Data assessmentIMR and LE were chosen as measures of all-cause mor-tality due to availability. IMRs from published reportswere graphed over time by major source of data(Figures 2 and 3). Data sources were assessed for reli-ability and plausibility of estimates on the basis ofmethod of estimation, original source of data, and dataconsistency. Unreliable sources were censored from fur-ther analysis if they met any of the following criteria: (a)data were derived, or were considered likely to havebeen derived, from models assuming a given improve-ment by year, as evidenced by a perfectly linear improve-ment in LE or IMR by year; (b) multiple incompatibleestimates were given by the source for a single year oradjacent years; (c) source data included implausible esti-mates (based on equivalent measures for developedcountries); (d) calculations were based on uncorrectedvital registration data known to be significantly underre-ported or with no assessment of reporting completeness.An exponential trend was fitted to the average IMRs

of each year from data remaining after censoring(Microsoft Excel). A linear trend was fitted to IMRs post1996, excluding the MoH estimates, which were retainedafter censoring but later shown to be incomplete basedon the empirical data analysis. LE from all sources were

urce, 1939–2009. SDT = Statistics Department of Tonga,= World Health Organization, ADB = Asian Development Bank. The

2009 is shown at 2002 and 2007, respectively.

Figure 3 Credible published and empirical estimates of IMR by source and method of adjustment, 1939–2009. SDT = StatisticsDepartment of Tonga, MoH = Ministry of Health, CRC = Capture-recapture analysis y = 200 + 21e-0.023x / y = 0.0025x + 12.16 (fitted to credibleestimates excluding MoH estimates as described in the methods).


graphed over time by major source of data (Figures 4and 5), with unreliable estimates (based on the same cri-teria as described for IMR) censored from further ana-lysis. Insufficient data remained after censoring to fit atrend curve.

Empirical data collection and analysisData collectionData for 2001 to 2009 were obtained from both the HIS(derived from medical certificates of death) and CR.Data from 2005 to 2009 were obtained from the RHSrecords (community nursing reports), and data for 2005to 2008 were obtained from the Prime Minister’s office(Table 1). Additional years from the RHS and Prime

Figure 4 Published estimates of life expectancy at birth by source, m(Table 2) for 2001–2004 and 2005–2009 is shown at 2002 and 2007, respec

Minister’s office were not available at the time of datacollection. Key variables collected were: island of resi-dence, data source, full name, sex, date of birth, dateand place of death, and age at death. Population by agegroup, sex, and island was derived from the 1996[14,15] and 2006 [16,17] censuses using exponentialinterpolation to provide estimates for intermediate andadditional years. Total births for 2005 to 2009 wereobtained from the MoH annual reports for calculationof IMR [21].

Reconciliation of reported deathsEach death was assigned a unique number. Records werereconciled to create a single list of unique deaths

ales, 1930–2009. The range of best estimates from this studytively.

Figure 5 Published estimates of life expectancy at birth by source, females, 1930–2009. The range of best estimates from this study(Table 2) for 2001–2004 and 2005–2009 is shown at 2002 and 2007, respectively.


indicating source(s) in which each was recorded. Investi-gators manually matched deaths through a repeatedsorting of lists by key criteria, and the final list of uniquedeaths was reviewed by two investigators (SH, KC). Cri-teria were established for records to be considered amatch: (1) same surname (minor variations in spelling,variations in name order, or a phonetic match wereallowed), (2) same first name (again, minor variationsallowed), (3) age at death within one year, (4) date ofdeath within the same month, and (5) either same placeof residence or same place of death. Records were con-sidered to match if they met three criteria including sur-name or four criteria excluding surname.Deaths with missing ages were then redistributed

according to imputed age distributions for that source.Deaths only reported through the Prime Minister’sOffice (Table 1) were excluded from the reconciled data,

Table 1 Reported deaths by source, all Tonga: 2001-2009

Year CivilRegistration

HIS (deathcertificates)

RHS (comnursing r

2001 473 567

2002 446 588

2003 434 598

2004 418 543

2005 406 534 133

2006 470 493 146

2007 400 526 144

2008 398 484 543

2009 468 567 393

HIS = Health Information System, RHS = Reproductive Health System.

as the limited information provided and poor data qual-ity meant that it was not possible to determine whetherthese were additional deaths, or whether there was sim-ply insufficient information to identify an existingmatch.The following measures of mortality were calculated

using standard methods [61]: age-specific mortality rate,IMR (probability of dying before 1 year of age), child-hood mortality (probability of dying before 5 years ofage), adult mortality (probability of dying between 15and 59 years of age inclusive), and LE. Childhood(<5 years) mortality was also calculated based on a re-gression of the IMR as a proportion of childhood mor-tality against childhood mortality from internationalestimates for 2006 [62]. Estimated reporting complete-ness by data source was calculated using the total recon-ciled deaths.

munityeports)

Prime Minister’s office(town and districtofficer reports)

Reconciled data(all sources)

605

627

629

574

36 657

44 713

74 735

60 957

836


Assessment of completeness of reconciled dataReconciled data were assessed for under enumeration ofdeaths by (1) the Brass growth-balance method, whichcompares the age distribution of reported deaths withthe expected age distribution of deaths based on theobserved population age-sex structure (assuming aclosed population), applied to ages 25 to 64 years [61,63]and (2) capture-recapture techniques, which use theproportion of reported deaths recorded by each combin-ation of sources to estimate the number of unreporteddeaths [10,63]. Data from the Prime Minister’s officewere excluded from the capture-recapture analysis asvery few deaths were reported (Table 1), and these werefrom a very small area of Tonga, thus violating the as-sumption of equal catch ability for each death [10].A two-source capture-recapture analysis was con-

ducted for HIS data against CR data for 2001 to 2004. Akey assumption in a two-source analysis is that sourcesare independent. Data from the HIS and RHS for 2005to 2009 were therefore reconciled as a single source(henceforth referred to as MoH data) as both systemsare operated by the MoH and therefore demonstratestrong interdependency. MoH data was then assessedagainst CR data for 2005 to 2009. Confidence intervals(CIs) were calculated using the maximum likelihood es-timator [10].Three-source capture-recapture analysis allows for

pair-wise dependence between sources [10,64] and usedHIS, RHS, and CR data separately for 2005 to 2009.Each three-source analysis results in eight possible mod-els [10,64,65]. Models that did not account for depend-ence between the two MoH sources were excludedbased on the evaluation of the death recording systemswith final model selection based on statistical criteriasuch as the significance of associations between sources(p-value) and goodness-of-fit criteria (Bayesian Informa-tion Criterion and the Akaike Information Criterion)[10,65]. HIS data for 2001 to 2004 were adjusted byreporting completeness for the HIS data for 2005 to2009 to generate an adjusted estimate for this periodbased on the three-source analysis. Unreported deathswere estimated for total deaths, with subgroup analysesfor sex, age, and island group for both the two- andthree-source capture-recapture analyses to identify vari-ation in reporting patterns.Reporting completeness for Tongatapu was applied to

all islands to derive adjusted estimates of total deaths, asmost deaths are from Tongatapu and it was anticipatedthat data from the main island group (where data arechecked more regularly) should be of the highest qualityand would therefore be less likely to be under matcheddue to data recording issues. Adjusted measures of LEand IMR were then compared with reconciled data andcredible estimates from the published data.

ResultsAnalysis of published dataThirteen sources of published mortality estimates werefound for Tonga, with 10 of these sources reporting LE.There was very little information on methods of data col-lection and analysis available in the published sources.

Infant mortality ratePublished IMR estimates varied significantly through thelate 1990’s when most estimates converge to between 10and 20 deaths per 1,000 live births. Despite the variation,it is apparent IMR has fallen substantially over theperiod investigated and is now relatively low (Figures 2and 3). Four sources remained after applying the censor-ing criteria, including MoH estimates from reporteddeaths. Significant variation remains in recent estimatesdespite censoring. Based on credible published estimates,IMR for Tonga appears relatively stable for recent years:between 10 and 19 deaths per 1,000 live births.

Life expectancyTen sources of published estimates of LE were found forTonga since 1939. There was a wide variation found inpublished estimates of LE (around 20 years’ differencefor a single year for each sex), as shown in Figures 4 and5. Eight sources were censored from further analysis,leaving only uncorrected MoH data and an earlier studythat used CR data for 1982 to 1986 and 1987 to 1992adjusted for completeness [19]. After censoring, pub-lished estimates of LE were 69.6 years for males and72.9 years for females for 2005 to 2008 (Table 2).

Analyses of reported deathsReconciled dataThere were 3,874 unique deaths recorded for Tonga be-tween 2005 and 2009 (Table 1) and 2,435 deathsrecorded between 2001 and 2004. The MoH recorded3,361 (87%) of total reconciled deaths for 2005 to 2009,while the CR recorded 2,128 deaths (55%). The PrimeMinister’s office recorded 214 deaths for 2005 to 2008,of which only 14 were not able to be matched with an-other source.From the reconciled deaths, IMR was directly calcu-

lated for 2005 to 2009 as 14.7 deaths per 1,000 livebirths, with childhood mortality as 24.5 deaths per 1,000live births for males and 19.5 for females. Imputation ofIMR from these levels of childhood mortality yielded anestimate of 18.2 deaths per 1,000 live births. Adult mor-tality is estimated at 26.7% for males and 19.8% forfemales (Table 2). LE at birth (2005 to 2009) was 65.2(95% CI: 64.6 – 65.8) years for males and 69.6 (95% CI:69.0 - 70.2) years for females. LE is lower than publishedestimates presented above and substantially lower thanpublished estimates that were censored. If overseas

Table 2 Summary estimates of mortality (all Tonga) by method and period

Mortality dataset Period IMR /1,000 IMR /1,000imputed fromchild mortality

Child mortality<5 years /1,000

Adultmortality (%)

Life expectancyat birth (years)

M F M F M F

Crediblepublishedsources

Civil Registry data(adjusted by Brassanalysis) [29]

1982-1986 25.9 - - - - - 66.3 69.5

1987-1992 22.9 - - - - - 68.1 72

Ministry of Health dataonly (unadjusted) [5]

2005-2008 16.4 - - - 18.4 19.1 69.6 72.9

Reconcileddata

2001-2004 8.9 12.6 16.7 11.5 21.8 15.5 67.7 71.7

2005-2009 13.7 18.2 24.5 19.5 26.7 19.8 65.2 69.6

Adjustedreconcileddata

Adjusted by Brassanalysis (>5 years)

2001-2004 NA NA NA NA 21.6 22.1 67.9 68.3

2005-2009 NA NA NA NA 26.7 23.6 65.2 67.8

Adjusted by two-sourceCRC analysis (all ages)

2001-2004 9.1 12.8 17.0 11.8 22.2 15.8 67.5 71.6

2005-2009 14.7 19.4 26.6 20.7 28.6 20.9 64.2 69.0

Adjusted separatelyfor three-source CRCanalysis for 0–4 yearsand remaining deathsby gender

2001-2004† 10.4 21.7 32.1 21.8 39.4 29.8 58.8 64.2

2005-2009 19.1 24.7 34.8 27.6 36.3 27.7 60.4 65.4

Final mortality estimates 2001-2004 9.1 – 21.7 17.0 – 32.1 11.8 – 21.8 22.2 -39.4 15.8 – 29.8 58.8 - 67.5 64.2 - 71.6

2005-2009 14.7 – 24.7 26.6 – 34.8 20.7 – 27.6 28.6 – 36.3 20.9 – 27.7 60.4 - 64.2 65.4 – 69.0

CRC = Capture-recapture analysis.Credible estimates of IMR are shown in Figure 3.†Estimates generated by applying reporting completeness for HIS data from 2005 to 2009 three-source analysis (final disaggregation by 0 to 4 years, andremaining deaths by gender) to HIS data for 2001 to 2004.


deaths are excluded to minimize the possibility of nonre-sidents being included in the analysis, there is less than0.5 years increase in LE for both males and females.There is no change to IMR.Reporting completeness for individual data sources

(2005 to 2008), as determined from a comparison withtotal reconciled deaths, is estimated at 55% for CR, 67%for HIS, 35% for RHS, and 87% for the combined MoHdata. For Tongatapu, reporting completeness was esti-mated at 56% for CR, 70% for HIS, 42% for RHS, and91% for the combined MoH data (Table 3).

Brass analysisBrass analysis of the reconciled data estimates thatreporting completeness for 2005 to 2009 (all sources) is>99% for males and 63% for females; however, the dis-parity between these estimates is not plausible. For Ton-gatapu only, reporting completeness was estimated as>99% for males and 82% for females. The Brass analysisplot of partial births against partial deaths for males (forall deaths and Tongatapu only) did not readily allow astraight line to be fitted regardless of age groupsselected, indicating this estimate of completeness isunreliable.

Capture-recapture analysisThe overall two-source analysis of aggregated healthrecords and civil registration records estimated a total of

2,491 (95% CI: 2465–2517) deaths in Tonga for 2001 to2004 and 4,400 (95% CI: 4295–4506) deaths for 2005 to2009. Compared to these estimates, reconciled data fromboth the CR and health reporting systems were found tobe 98% (95% CI: 97–99) complete for 2001 to 2004 and88% (95% CI: 85–90) for 2005 to 2008 for both sexes.For Tongatapu only, the reconciled data were estimatedas 98% (95%: CI 97–99) complete for both sexes for2001 to 2004 and 93% (95%: CI 89–95) complete for2005 to 2009 (Table 3). Recording completeness variedby age, with child deaths less likely to be recorded thanadult deaths through CR. However, aggregation of sub-group analyses within strata of age and island group pro-vided summary measures of mortality, which differedlittle from the overall aggregated estimates. Reportingcompleteness for individual data sources (2005 to 2008)in Tongatapu, as determined by the two-source analysis,is estimated at 53% for CR data, 66% for HIS data, 40%for RHS data, and 84% for the combined health data(HIS plus RHS).A three-source analysis was performed for HIS, RHS,

and CR data for 2005 to 2009. Estimates of total deathsranged from 4,404 to 6,695 based on aggregate (total)data, depending on the model selected. The modelselected as the best fit was that which included depend-ence between the HIS and RHS and between the HISand CR. This is consistent with the assessment of thedeath reporting system. Based on this model, the

Table 3 Estimated reporting completeness (%) of deaths (all ages, both sexes) by source and method of assessmentfor Tongatapu

Method of assessment Years Completeness (%) and 95% confidence intervals

Reconcileddeath data

Civil registrationof deaths

HIS (deathcertificates)

RHS (communitynursing reports)

MoH(combined)

Comparison against reconciled death data 2001-2004 - 70 96 NA NA

2005-2009 - 56 70 42 91

Brass growth-balance assessment 2001-2004 96 (M) / 63 (F) - - - -

2005-2009 >99 (M) / 82 (F) - - - -

Two-source capture-recapture analysis 2001-2004 98 (97 – 99) 69 (68–70) 94 (93 – 95) NA NA

2005-2009 93 (91–95) 53 (52–54) 66 (65–67) 40 (39–40) 84 (83–86)

Three-source capture-recapture analysis(aggregated ages)

2005-2009 68 (63–73) 37 (34 – 40) 46 (43 – 50) 28 (26 – 30) 63 (58 – 67)

Three-source capture-recapture analysis(final disaggregation*)

2005-2009 69 (57 – 77) 38 (31 – 42) 47 (39 – 52) 28 (23 – 31) 63 (52–70)

95% confidence intervals shown in parentheses where applicable.HIS = Health Information System, RHS = Reproductive Health System, MoH = Ministry of Health (HIS and RHS data combined).* Capture-recapture analysis disaggregated by the following subgroups: 0 to 4 years (both sexes), males 5 years or older, and females 5 years or older.


reconciled data are estimated to be 58% (95% CI: 52 –63) complete. If only Tongatapu is included, the recon-ciled data are estimated to be 70% (95% CI: 63–76)complete for males and 67% (95% CI: 58–74) completefor females. As reporting patterns varied by age (particu-larly for very young deaths) and sex, final estimates ofcompleteness from the three-source analysis werederived from models of best fit for disaggregated datafor children aged 0 to 4 years (both sexes) and theremaining ages by sex. Based on these models, thereconciled data for Tongatapu were found to be 70%(95% CI: 32–91) complete for children aged 0 to 4 years,69% (95% CI: 61–76) complete for males 5 years andover, and 68% (95% CI: 59–75) complete for females5 years and over. Adjusted estimates are shown inTable 2.

Final estimates of mortalityBoth the Brass method analysis and capture-recaptureanalyses indicate that deaths remain under enumeratedfollowing data reconciliation. The Brass method, how-ever, provided implausible results, due most likely to thesensitivity of this method to high levels of migration andsmall numbers [66]. The two-source capture-recaptureanalysis is less sensitive but is known to be affected byprobable positive dependence between the civil registryand health data due to registration practices; therefore, itis expected to overestimate reporting completeness. Assuch, the two-source analysis provides a plausible lowerlimit to mortality levels and subsequent upper limit forLE. Equally, the three-source analysis allows for this de-pendence and can therefore be used to set a plausibleupper limit for mortality levels and hence a lower limitfor LE. Final estimates for 2005 to 2009 place IMR be-tween 14.7 to 24.7 deaths per 1,000 births, with the best

estimate for this period at 19 deaths per 1,000 births, asestimated from childhood mortality from the two-sourcecapture-recapture analysis, direct calculation from thethree-source analysis, and the 2006 census. Final esti-mates place the plausible range of LE at 60.4 to64.2 years for males and 65.4 to 69.0 years for females.Adult mortality is estimated to be between 28.6% to36.3% for males and 20.9% to 27.7% for females.

DiscussionThese findings demonstrate that there has been a sus-tained history of erroneous mortality estimates forTonga, with many published estimates of LE and IMRimplausible and presented without an indication of themethod used. Many of the IMR estimates presentedprior to 2000 (Figure 3) fall below current levels of IMRin Australia (IMR of 5.0 in 2007 [58]) and New Zealand(IMR of 6.0 in 2007 [58]). Credible data sources demon-strate that IMR has declined steadily to below 20 deathsper 1,000 live births, fluctuating in recent years between10 and 19 deaths per 1,000 live births. Although higherthan previous MoH estimates of IMR, reconciled esti-mates of IMR are consistent with this range, while sev-eral adjusted estimates are slightly higher. The range ofplausible estimates suggests that progress toward theMillennium Development Goal targets has been min-imal, with best estimates from the empirical data from2005 to 2009 at very similar levels to census estimatesfor 2006. Although IMR demonstrates no significanttrend since the late 1990s, at these levels it is a minor in-fluence on the overall LE [61].Published estimates of LE from 2000 onwards varied

from 65 to 75 years for males and 68 to 74 years forfemales, with most clustered around 70 to 71 for malesand 72 to 73 for females. Very few of the published


estimates were assessed as reliable, with only MoH dataremaining for recent years. These MoH data estimatedLE for 2005 to 2008 as 69.6 years for males and72.9 years for females. The 2006 census estimates of67.3 years for males and 73.0 years for females were cen-sored from the final results as LE was calculated fromregistration data adjusted using a model with a single in-put parameter (childhood mortality) [16,17].The reconciled empirical data for 2005 to 2009 pro-

duces an estimate of LE of 65.2 years (95% CI: 64.6 -65.8) for males and 69.6 years (95% CI: 69.0 – 70.2) forfemales, which is several years lower than both the MoHand census estimates. These findings demonstrate thatLE is substantially lower than previously reported forTonga. Further, the reconciled data indicate LE mayhave fallen from 2001 to 2004, although this is not evi-dent in the range of plausible published estimates foreach period.Although there remains some uncertainty regarding

the extent of underreporting to all data sources, the datacollection systems and results of the assessments pre-sented here indicate that even reconciled data underenumerated. The Brass analysis, particularly for males,appears significantly affected by migration, which is quitehigh in Tonga (−16.58 per 1,000 population in 2010[67]). However, the capture-recapture assessments pre-sented here allow limits of plausibility to be established.The two-source analysis for 2005 to 2009 accounts forthe major dependency between the HIS and RHS bycombining these data; however, there is also probablepositive dependency between the HIS and CR due to theCivil Registry requesting medical certificates of death. Assuch, this analysis would overestimate reporting com-pleteness and underestimate the total deaths, thus pro-viding a plausible upper limit for LE. The three-sourcecapture-recapture analysis allows these dependencies tobe taken into account, and can therefore be used as alower limit for LE, producing similar results to the Brassanalysis for females. As migration data for Tonga are notcomplete, it is not possible to use changes in populationrecorded through the census as an alternative methodfor assessing reporting completeness for deaths.The low LE is clearly being predominantly driven by

adult mortality. Even prior to adjustment for under-counting, adult mortality based on reconciled data inTonga is roughly three times that seen in Australia,where there is a probability of dying of 8.9% for malesand 5.1% for females in 2002 to 2004 [68], or NewZealand, where there is a probability of dying of 9.7% formales and 6.4% for females in 2002 to 2004 [69]. Thehigh adult mortality is consistent with the limited infor-mation available on noncommunicable disease risk factorprevalence available for Tonga. Provisional results from a2004 survey found that 69% of adults were obese (body

mass index ≥ 30 kg/m2), 23% of adults were affected byhypertension (systolic blood pressure ≥ 140 mmHg and/or diastolic blood pressure ≥ 90 mmHg or currently onmedication), and 18% reported to be diabetic (self-report)[70]. In 2004, Tonga acted to recognize the extensivecommunity burden on noncommunicable diseases bydeveloping the first non-communicable disease strategyin the Pacific region [71]. This pattern of high adult mor-tality limiting LE is also consistent with recent findingselsewhere in the Pacific, such as in Fiji and Nauru[72,73].Difficulties encountered in assessment of published

data were the lack of information concerning primarydata sources, methods of calculation, and assumptions.In order to deal with this, exclusion criteria were speci-fied and unreliable estimates censored from further ana-lyses. The matching criteria used to reconcile the sourcesof death reporting and multiple checks of the matchingprocess minimize the risk of significant errors in the rec-onciliation of the data sources. These criteria were devel-oped and tested to reduce the risk of matching recordsbeing missed, with the matching process checked by twoinvestigators (SH, KC). Removing the Prime Minister’soffice data from the analysis further reduced the potentialfor under matching due to incomplete data, which wouldhave lead to overestimation the number of deaths notreported [10,63]. Completeness estimates for the main is-land group of Tongatapu, as the best quality subset ofdata, were also applied to all data to minimize the poten-tial for overestimating unreported deaths due to mis-alignment in the populations covered by each source orpoorer data quality from the outer islands.The wide range of plausible estimates for the period

2001 to 2004 limit any assessment of trends between thetwo time periods investigated. This uncertainty is largelya result of less data being available for these years. How-ever, as MoH efforts to improve reporting began afterthis period, HIS data for 2001 to 2004 could reasonablybe expected to be less complete than in 2005 to 2009.This is reflected in the increase in the total reporteddeaths per year in the reconciled data between the twoperiods, as shown in the results. The true level of mor-tality is therefore likely to fall closer to the high end ofthe range of plausible mortality estimates for 2001 to2004 (and therefore the lower estimates of LE) asderived by applying reporting completeness for the HISfrom the three-source assessment for 2005 to 2009.While there have been very few studies on reportingcompleteness in Tonga, our findings of reporting com-pleteness for the reconciled data are consistent with a2003 evaluation of mortality data in Tonga based ondata obtained by the World Health Organization. Thisreport found Tonga to have “low” quality mortality andcause of death data, with completeness estimated at 86%


and coverage estimated at 70% [4], compared with anestimated completeness for the reconciled data between68% and 93% for 2005 to 2008, as found in this study.

ConclusionsLE for Tonga for 2005 to 2008 is estimated to be be-tween 60.4 to 64.2 years for males and 65.4 to 69.0 forfemales, well below previously published estimates. Thelow LE, at a relatively low IMR and high prematureadult mortality, suggests that non-communicable dis-eases are having a profound limiting effect on health sta-tus in Tonga.These findings demonstrate that much of the mortality

data that have been previously available to health policymakers in Tonga as well as to international donors andagencies has been misleading, potentially masking theurgency of the public health action need to address adultpremature mortality in Tonga. Additionally, the findingshighlight the critical need both to reconcile existing datasources and integrate reporting systems more fully toensure all deaths in Tonga are captured, as well as theimportance of analysis of local empirical data in moni-toring trends in health status. While census estimatesfor IMR were found to be consistent with best estimatesfrom the routinely reported data, the census was foundto have underestimated adult mortality and thereforeoverestimated LE.

Competing interestsAL is one of the Editors-in-Chief of Population Health Metrics.

Authors’ contributionsSH conducted the system review, supervised the data collection andcollation processes, designed the matching criteria, conducted the datamatching, managed the ethics process and coordination with data collectionagencies, performed the statistical analysis for the two-source analysis of theHIS and PM data for 2005 to 2009, completed the search for publishedestimates, assessed published data sources for reliability, revised themanuscript critically, and made substantial contributions to the interpretationof data. KC oversaw the project, conducted the system review, oversaw thedesign and testing of matching criteria, checked the matching process,assisted with data matching for the civil registration data against othersources, performed the statistical analysis for both the two-source analysisfor 2001 to 2004 and the three-source analysis, including the civil registrydata and final model selection including interpretation of final results andselection of final estimates, supervised the other analysis, and drafted andled the revision of the manuscript. CR revised the manuscript critically forimportant intellectual content. ADL and RT conceived and coordinated thebroader study under which this project was undertaken, reviewed andcontributed significantly to the interpretation of results, and revised themanuscript critically for important intellectual content. RT also generated theregression models to estimate IMR from childhood mortality. All authors readand approved the final manuscript.

AcknowledgmentsThe authors would also like to thank the staff and management of theTongan Ministry of Health, Civil Registry office, and Prime Minister’s office fortheir assistance and cooperation with this research. Particular thanks are dueto Mrs. Sela Paasi, Dr. Toa, Dr. Malakai ‘Ake, Mr. Pita Vuki, and Mrs. ManakoviPahulu. We would also like to thank Claire Mowry for performing the originalpublication search. Funding for this research was provided under the AusAIDAustralian Research Development Awards (HS0024). The funding body hadno role in study design; in the collection, analysis, and interpretation of data;

in the writing of the manuscript; or in the decision to submit the manuscriptfor publication.

Author details1Ministry of Health, Kingdom of Tonga, Tonga. 2School of Population Health,University of Queensland, Brisbane, Australia. 3Secretariat of the PacificCommunity, Noumea, New Caledonia. 4School of Public Health andCommunity Medicine, University of New South Wales, Paddington, Australia.

Received: 7 September 2011 Accepted: 2 August 2012Published: 14 August 2012

References1. UNESCAP: Achieving the Millennium Development Goals in an Era of Global

Uncertainty: Asia-Pacific Regional Report 2009/10. Bangkok: United NationsEconomic and Social Commission for Asia and the Pacific Statistics, AsianDevelopment Bank, United Nation Development Funds; 2010. http://www.mdgasiapacific.org/regional-report-2009-10.

2. Secretariat of the Pacific Community: Pacific Island Populations - Estimatesand projections of demographic indicators for selected years.; 2011. http://www.spc.int/sdp/.

3. Western Pacific Country Health Profiles (CHIPS): 2010 Revision. Manila: WorldHealth Organization; 2011. http://www.wpro.who.int/internet/files/hin/CHIPS2010.pdf.

4. Mathers C, Ma Fat D, Inouie M, Rao C, Lopez A: Counting the dead andwhat they died from: an assessment of the status of global cause ofdeath data. Bull World Health Organ 2005, 83(3):171–177.

5. Hufanga S: Workplace Project Portfolio. In Masters dissertation. Brisbane,Queensland: University of Queensland, School of Population Health; 2010.

6. Kingdom of Tonga: Legislation Births, Deaths and Marriages RegistrationAct of 1988.Adsett N J. In Laws of Tonga (vol. 2). Great Britain: The EasternPress Ltd; 1988.

7. Ministry of Health: Policy and procedures: Notification of a death. Nuku'alofa,Tonga: Kingdom of Tonga; 2007.

8. Kingdom of Tonga: Town Officers Act. Nuku'alofa, Tonga: Government ofTonga; 1973.

9. Brass growth balance method. In Manual X: indirect techniques fordemographic estimation. New York, NY: United Nations; 1983:139–146.

10. Hook E, Regal R: Capture-Recapture Methods in Epidemiology: Methodsand Limitations. Epidemiol Rev 1995, 17(2):243–264.

11. Pongi J: Tonga Census of population and housing. Vol. 2 -Demographicanalysis. Nuku'alofa, Tonga: Department of Statistics, Kingdom of Tonga;1976.

12. Bakker ML: A Demographic Analysis of the Population of Tonga 1977–1975.Noumea: SPC; 1995.

13. Kingdom of Tonga Statistics Department: Demographic Analysis of 1986Census Data: Fertility, Mortality and Population Projections -Volume 2.Nuku'alofa, Tonga: Government of Tonga; 1993.

14. Demmke A: Tonga Population Profile: A Guide for Planners and Policy Makers.Noumea: SPC Population and Demography Programme; 1996.

15. Secretariat of the Pacific Community: Tonga Population Profile Based on 1996Census: A Guide for Planners and Policy Makers. Noumea: SPC; 1999.

16. Kingdom of Tonga Statistics Department: Population Census 2006 AnalyticalReport Volume 2. Nuku'alofa, Tonga: Government of Tonga; 2006.

17. Kingdom of Tonga Statistics Department: Statistical Bulletin on Vital StatisticsReports & Population Census Reports; 2006. http://www.spc.int/prism/country/to/stats/Social/Health/health.htm.

18. The Government of Tonga and United Nations Children's Fund: A SituationAnalysis of Children and Women in Tonga 1996. Suva, Fiji: UNICEF Pacific;1996.

19. Kingdom of Tonga Statistics Department: Vital Statistics Bulletin; 2002. http://www.spc.int/prism/country/to/stats/MDG/mdgh.htm.

20. Kingdom of Tonga Statistics Department: Social Indicators; 2006. http://www.spc.int/prism/country/to/stats/Social/socio-index.htm.

21. Ministry of Health: Report of the Minister for Health for the year. Nuku'alofa,Tonga: Kingdom of Tonga; 2009.







http://www.wpro.who.int/internet/files/hin/CHIPS2010.pdf


http://www.spc.int/prism/country/to/stats/Social/Health/health.htm




25. Ministry of Health: Tonga Health 2000; 2001. http://www.spc.int/prism/country/to/stats/MDG/mdgh.htm.


27. Ministry of Health: Health Problems in the Kingdom of Tonga in the Years1956–1982. Nuku'alofa, Tonga: Kingdom of Tonga; 1984.

28. McArthur N: Island Populations of the Pacific. Canberra: Australian NationalUniversity Press; 1967.

29. Kupu AS: Mortality Analysis for Tonga 1982–1992. Sydney University: MastersDissertation; 1993.

30. Secretariat of the Pacific Community: Demographic Analysis of Tonga.Noumea: SPC; 1995.

31. Secretariat of the Pacific Community: Statistical Bulletin of the South PacificNo. 42. Noumea: SPC; 1995.

32. Secretariat of the Pacific Community: Pacific Island Populations. Noumea:SPC; 1998.

33. Secretariat of the Pacific Community: Pacific Island Populations 2004.Noumea: SPC; 2004.

34. Secretariat of the Pacific Community: Social Indicators. In PRISM PacificRegional Information System Noumea: SPC; 2004. http://www.spc.int/prism/social/health.html.

35. Lopez A, Salomon J, Ahmad O, Murray C, Mafat D: Life Tables for 191Countries: Data, Methods and Results. GPE Discussion Paper Series: No.9. EIP/GPE/EBD. Geneva: World Health Organisation; 2000. http://www.who.int/health-systems-performance/docs/levelofhealth_docs.htm.

36. World Health Organisation: World Health Statistics 2006 and World HealthReport. In World Health Organisation Statistical Information System. Geneva:WHO; 2007. http://www3.who.int/whosis/core/core_select.cfm.

37. World Health Organisation: Tables for Member States: Tonga 2004. Geneva:WHO; 2007. http://www3.who.int/whosis/life/life_tables/life_tables_process.cfm?country=ton&language=en.

38. World Health Organization: Indicators for all countries: Tonga 2006. Geneva:WHO; 2008. http://apps.who.int/whosis/data/Search.jsp?countries=%5bLocation%5d.Members.

39. World Health Organization: World Health Statistics 2010. Geneva: WHO; 2011.http://www.un.org/esa/population.

40. The World Bank Group: WDI (World Development Indicators) Online. WorldBank; 2010. http://devdata.worldbank.org/dataonline/, http://data.worldbank.org/country/tonga#wbboxes-source-SP-DYN-LE00-IN.

41. Asian Development Bank: Tonga: Economic Performance and SelectedDevelopment Issues. ADB, Office of Pacific Operations; 1996.

42. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2000, Volume XXXI. ADB; 2000. http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

43. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2000: Growth and Change in Asian and the Pacific. ADB; 2001.http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

44. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2002: Population and Human Resource trends and challenges. ADB;2002. http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

45. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2003: Special Chapter on Education for Global Participation. ADB;2003. http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

46. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2004. ADB; 2004. http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

47. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2005. ADB; 2005. http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

48. Asian Development Bank: Key Indicators of Developing Asian and PacificCountries 2006: Measuring Policy Effectiveness in Health and Education. ADB;2006. http://www.adb.org/Documents/Books/Key_Indicators/default.asp.

49. United Nations Population Division: Annual Indicators. In DemographicYearbook: Historical Supplement.: United Nations Economic and SocialCommission for Asia and the Pacific, Statistics Division; 1997. www.unescap.org/stat/data/goalIndicatorArea.aspx.

50. United Nations Population Division: Demographic Yearbook. United Nations;2003.

51. United Nations Development Programme: Global MDG Database; 2010.http://mdgs.un.org/unsd/mdg/Data.aspx.

52. United Nations Statistics Division: Population and Vital Statistics Report; 2006.www.unstats.un.org/unsd/demographic/products/vitstats/Sets/SeriesA_%20Jan2006_complete.pdf.

53. United Nations Population Fund Country Support Team: Indicator ofMortality. In Office for the South Pacific. UN: Mission Report No. 94/01: YearEnd Review and Programme Planning; 2010. http://www.unfpa.org/swp/2008/en/09_technical_notes.html.

54. United Nations Economic and Social Commission for Asia and the Pacific:Statistical Indicators for Asia and the Pacific 2005 Compendium Volume XXXV;2005. http://www.unescap.org/stat/data/statind/pdf/index.asp.

55. United Nations Economic and Social Commission for Asia and the Pacific:Statistical Indicators for Asia and the Pacific 2004 Compendium, Volume XXXIV.UN; 2004. http://www.unescap.org/stat/data/statind/pdf/index.asp.

56. United Nations Economic and Social Commission for Asia and the Pacific:Asia Pacific Estimates 2009. In Statistical Yearbook for Asia and the Pacific.UN: Statistics Division; 2009. http://www.unescap.org/stat/data/syb2009/4-Child-health.asp.

57. United Nations Economic and Social Commission for Asia and the PacificStatistics, Asian Development Bank, and United Nation Development Funds:Achieving of Millennium Development Goals in an ear of Global Uncertainty.United Nations; 2010.

58. UNICEF: State of the world’s children;. http://www.unicef.org/sowc2011/pdfs/Table-1-Basic-Indicators_02092011.pdf.

59. UNICEF: State of the world’s children; 2003. http://www.unicef.org/sowc05/english/index.html.

60. UNICEF: State of the world’s children; 1998. http://www.unicef.org/sowc98/tab1a.htm.

61. Rowland DT: Demographic Methods and Concepts. New York: OxfordUniversity Press; 2003.

62. World Health Organization: Statistical Annex, Table 1: Basic Indicators forall WHO member states. In World Health Report. Geneva: WHO; 2005.

63. Wittes JT, Sidel VW: A generalization of the simple capture-recapturemodel with applications to epidemiological research. J Chronic Dis 1968,20:287–301.

64. International Working Group for Disease Monitoring and Forecasting(IWGDMF): Capture-Recapture and Multiple-Record Systems Estimation 1:History and Theoretical Development. Am J Epidemiol 1995,142(10):1047–1057.

65. Hook E, Regal R: Validity of Methods for Model Selection, Weighting forUncertainty and Small Sample Adjustment in Capture-RecaptureEstimation. Am J Epidemiol 1997, 145(12):1138–1144.

66. Murray CJL, Rajaratnam JK, Marcus J, Laakso T, Lopez AD: What Can WeConclude from Death Registration? Improved Methods for EvaluatingCompleteness. PLoS Med 2010, 7(4):e1000262. doi:10.1371/journal.pmed.1000262.

67. Secretariat of the Pacific Community: Migration Statistics. In PRISM; 2010.http://www.spc.int/prism.

68. Australian Bureau of Statistics: Chapter 4: Differentials in Mortality. InDeaths. Australia; 2004. http://www.abs.gov.au/AUSSTATS/[email protected]/allprimarymainfeatures/F69F652EDC86B05CCA257235007865FB?opendocument.

69. Statistics New Zealand: Tables: Deaths 2008, Age-specific death rates by sex,total population; 1986–2007. http://www.stats.govt.nz/tables/deaths-tables.htm.

70. Secretariat of the Pacific Community: Pacific NCD Statistics; 2011. http://www.spc.int/hpl/index.php?option=com_content&task=blogcategory&id=23&Itemid=42.

71. Ministry of Health: Non Communicable Diseases Response Plan. Nuku'alofa,Tonga: Kingdom of Tonga; 2004.

72. Carter KL, Cornelius M, Taylor R, Ali SS, Rao C, Lopez AD, Lewai V, GoundarR, Mowry C: Mortality Trends in Fiji. ANZJPH, . In Press.

73. Carter KL, Soakai TS, Taylor R, Gadabu I, Rao C, Thoma K, Lopez AD:Mortality Trends and the Epidemiological Transition in Nauru. Asia Pac JPublic Health 2011, 23:10.

doi:10.1186/1478-7954-10-14Cite this article as: Hufanga et al.: Mortality trends in Tonga: anassessment based on a synthesis of local data. Population Health Metrics2012 10:14.

http://www.spc.int/prism/country/to/stats/MDG/mdgh.htm

http://www.spc.int/prism/country/to/stats/MDG/mdgh.htm

http://www.spc.int/prism/social/health.html

http://www.spc.int/prism/social/health.html

http://www.who.int/health-systems-performance/docs/levelofhealth_docs.htm

http://www.who.int/health-systems-performance/docs/levelofhealth_docs.htm

http://www3.who.int/whosis/life/life_tables/life_tables_process.cfm?country=ton&language=en

http://www3.who.int/whosis/life/life_tables/life_tables_process.cfm?country=ton&language=en

http://apps.who.int/whosis/data/Search.jsp?countries=%5bLocation%5d.Members

http://apps.who.int/whosis/data/Search.jsp?countries=%5bLocation%5d.Members

http://devdata.worldbank.org/dataonline/

http://data.worldbank.org/country/tonga#wbboxes-source-SP-DYN-LE00-IN

http://data.worldbank.org/country/tonga#wbboxes-source-SP-DYN-LE00-IN

http://www.adb.org/Documents/Books/Key_Indicators/default.asp






www.unescap.org/stat/data/goalIndicatorArea.aspx

www.unescap.org/stat/data/goalIndicatorArea.aspx

www.unstats.un.org/unsd/demographic/products/vitstats/Sets/SeriesA_%20Jan2006_complete.pdf

www.unstats.un.org/unsd/demographic/products/vitstats/Sets/SeriesA_%20Jan2006_complete.pdf

http://www.unfpa.org/swp/2008/en/09_technical_notes.html

http://www.unfpa.org/swp/2008/en/09_technical_notes.html

http://www.unicef.org/sowc2011/pdfs/Table-1-Basic-Indicators_02092011.pdf

http://www.unicef.org/sowc2011/pdfs/Table-1-Basic-Indicators_02092011.pdf

http://www.unicef.org/sowc05/english/index.html

http://www.unicef.org/sowc05/english/index.html

http://www.unicef.org/sowc98/tab1a.htm

http://www.unicef.org/sowc98/tab1a.htm

http://dx.doi.org/10.1371/journal.pmed.1000262

http://dx.doi.org/10.1371/journal.pmed.1000262

http://www.abs.gov.au/AUSSTATS/[email protected]/allprimarymainfeatures/F69F652EDC86B05CCA257235007865FB?opendocument



http://www.stats.govt.nz/tables/deaths-tables.htm

http://www.stats.govt.nz/tables/deaths-tables.htm

http://www.spc.int/hpl/index.php?option=com_content&task=blogcategory&id=23&Itemid=42



� � �

202

5.5.3 Using Capture-recapture analysis to set plausible mortality limits in

Kiribati

As demonstrated in case study five, even when data sources are largely

incomplete, it is possible to use capture-recapture analysis to quantify upper and

lower boundaries to define the range of plausible estimates for mortality. In Kiribati,

previous estimates of mortality level have all been obtained through surveys and

census analyses based mainly on indirect methods; with age-specific mortality

patterns, LE and adult mortality derived through single input model life tables (50,

152). There has been no assessment of mortality patterns based on empirical data

as available sources were considered too incomplete (50, 152). As discussed in

Chapter two, single input parameter model life tables may underestimate adult

mortality across all age groups (but especially adult ages when U5M is low) and

subsequently overestimate LE. Although Kiribati is still relatively early in the

epidemiological transition (334) and therefore may more closely reflect standard

model life tables, the existence of a double burden of disease, as indicated by the

high prevalence of diabetes and other NCDs (335) demonstrates a need to evaluate

the plausibility of these findings.

Only two sources of mortality data are available in Kiribati, civil registration and the

HIS, however the system assessment in Chapter three indicates these operate

largely independently, with no direct data sharing and limited provisions to require a

medical certificate prior to registration of a death. While three source capture-

recapture methods are more robust, the limited interaction between these data

sources indicates a two source analysis could be used. Capture-recapture analysis

� � �

203

therefore offered an opportunity to use the empirical data from these sources for the

first time to provide mortality estimates.

Methods

Methods used in the study closely reflected those employed in Bohol and Tonga.

Recorded deaths were sought from the MoH and Civil Registration Office for 2000-

2009. MoH data were first reconciled between the HIS death registry, hospital

records and nurses reports. Duplicate records were identified based on the same

criteria as used for matching the records between sources (outlined below). Where

duplicates were found, the data from each record was combined, and any

differences in key variables noted in the combined record.

Criteria for matching records within the two sources were based on the previous

work in Tonga and Bohol, and the variables available. These were developed in

conjunction with the Civil Registration Office and MoH staff. Data trials were

conducted to identify a set of criteria that would minimise the potential for over

matching (falsely matching records together that do not refer to the same individual),

and therefore underestimating the deaths not recorded in either source; and equally

minimise potential for under-matching, which would overestimate the missing

number of deaths.

The final criteria for records to be considered to be related to the same individual

were:

� � �

204

• Same surname (minor spelling variations allowed) plus at least two of: first

name (minor spelling variations allowed, or English versions of the same name),

and/or island (place of death), and/or date of death (within 5 days); or

• Same first name, and island, and date of death, and age and gender; or

• Either the same surname or first name, within same month of death, and all of

the following: island, age, gender, and cause (by broad category).

For cause to be considered a match, only specific causes were included (such as

stroke, diabetes, suicide etc) with ill-defined causes such as “old” or “sick” not

included. Cause was also considered a match if the records contained the same

cause in English and I-Kiribati.

These criteria were designed to account for the following issues:

• People may have different first names due to various usage of given versus

baptismal names, major variations in spelling, and interchangeable I-Kiribati and

English spelling (for example “Tiaon” and “John”).

• Adoption (including informal adoption) is common practice in Kiribati. People

may use either (or both) the name of their birth father, or the name of their adoptive

father.

• Surnames in Kiribati are usually the fathers’ first name. People may informally

use different surnames depending upon the head of the household where they

currently reside, marital status, and “nicknames” or name variations commonly used

� � �

205

by their father. In a significant proportion of records, this means no surname match is

possible.

• Minor spelling variations are common both with surnames and first names.

• Date of death was not routinely used for the MoH data until 2003. Prior to this,

the date was recorded only as month of report from the health centre. This was not a

reliable means of verifying potential matches.

The flexible matching criteria required all matching to be carried out manually, by

sorting the data sets by each of the matching criteria successively to identify

potential matches for review. Matches were identified, moved to the final data set

and re-sorted by place of death (by island), then by first name, then by surname.

This process was repeated for each matching criteria (i.e. sorting remaining records

first by place, date of death, name etc. in varying combinations). Each potential

match was then reviewed to ensure the matching criteria were met, and only those

that met these criteria were counted as the same deceased individual. Where

variables about an individual differed, the variable value as recorded in the civil

registration data set was considered to be the true value.

The analysis of data followed Hook and Regal’s (230, 336) method for two sources

as outlined in the two previous case studies. As the two systems operate quite

separately to each other and do not require documentation from the other to register

a death, it can be reasonably assumed the systems are largely independent. This

assumption was tested using the “r” and “B(r)” measures of dependence as outlined

by Hook and Regal (230, 336). Confidence intervals (95%) were also calculated

based on variance using the formula outlined by Sekar Deming (231).

� � �

206

It was noted in discussion with both MoH and civil registration staff that neonatal

deaths are frequently recorded as stillbirths in both reporting systems. Therefore, two

separate tabulations were generated and analysed, one including stillbirths and one

excluding them.

Population by age group, and sex was derived from the 2005 census (as the mid-

point of the study period). Population data by time period for 2000-2004 and 2005-

2009 were derived from the 2001(50) and 2005 censuses (50) using exponential

interpolation/projection to provide estimates for other years.

As the capture-recapture results for deaths in children (0-4 years) appear implausibly

high, final estimates of total deaths by age group and gender were derived using

reconciled data (including stillbirths) for children (0-4 years), and adjusted figures

based on the capture-recapture analysis by age group for all remaining ages.

Summary measures of mortality were then re-calculated based on this adjusted data

for 2000-2009 combined and by 5 year period. As the number of live births was not

known (and could not be verified), IMR was estimated from U5M calculated from the

life tables in each method, using a regression equation based on data from 1990 and

2006 for all countries that equates IMR (x) with U5M (y) mortality derived by Prof

Taylor as follows:

� � �

207

Results

There were 8,728 unique deaths reported (through all sources) in 2000-2009 in

Kiribati, an average of 873 deaths per year. If stillbirths are included in the data set,

this figure was 8,840 deaths in total or 894 deaths per year (Table 19). Including

stillbirths, 59% of these deaths were recorded through civil registration and 63%

through MoH. There were no significant changes in the number of reported deaths

by year.

Table 19: Reported deaths by source, Kiribati: 2000-2009

Age group

Reported deaths by source Civil Registration

Ministry of Health

Reconciled TOTAL

0-4 437 1254 1530 5-14 166 198 294 15-24 327 282 509 25-34 419 355 633 35-44 704 524 953 45-54 851 681 1183 55-64 685 618 1029 65-74 687 650 1073 75-84 473 359 687 85+ 130 81 183 Unknown 313 606 766 Total 5192 5608 8840

NB/ Table includes deaths reported as stillbirths, as noted previously.

There were 22 deaths in overseas residents recorded in the civil registration data

which were excluded from the study, as they were not from the resident population of

Kiribati. The age range of these cases was 22-74 years, with cases residents of Fiji

(8), Nauru (4), USA (2), Germany (2), or others (6). More than 80% of these deaths

� � �

208

were due to drowning, boating or diving accidents. A further 24 deaths (20 from civil

registration and four from MoH data) were excluded as there were no names

recorded.

A large number of duplicate records were removed from the final data set. While

some of these resulted from the multiple systems in use at MoH, many were the

result of duplicate reporting through the local island registrars. Up to 14 entries were

found per death for deaths occurring in Nanouti, with up to 19 entries for a single

death that occurred in Tarawa.

As there was little evidence of secular trends over the study period (Table 20),

summary measures are presented for 2001-2009 inclusive. U5M (both sexes

combined) was estimated at 67.3 deaths per 1,000 live births, excluding stillbirths;

and 73.6 deaths per 1,000 live births, if stillbirths are included. This equates to an

IMR of 48.9 and 52.9 deaths per 1,000 live births respectively using a regression

analysis (developed by Prof R Taylor) of U5M against IMR based on data from the

WHO databank (presented in case study three).

Adult mortality (45q15) was estimated at 46.6-47.1% for males and 27.8-27.9% for

females, depending on whether stillbirths were included in the data set (thereby

adjusting the distribution of deaths of unknown ages). LE for 2001-2009 (total) is

estimated at 54.6-55.3 years for males and 63.4-64 years for females.

Summary measures of mortality from the analyses for 2001-2009 are presented in

the following table by period. Confidence intervals for estimates from the reconciled

� � �

209

data were derived from the life tables (using the Chiang method) with confidence

intervals for the capture-recapture analysis based on estimated reporting

completeness of the source data. Uncertainty for best estimates was derived by

applying the greater of these two values.

��

�

210

Tabl

e 20

: Sum

mar

y m

easu

res

of m

orta

lity

for K

iriba

ti by

dat

a so

urce

and

per

iod,

200

0-20

09

Dat

a so

urce

IM

R

(dea

ths

per

1000

live

bi

rths)

U5M

(pro

babi

lity

of d

ying

(p

er 1

000

pop)

from

age

s 0-

4 in

clus

ive)

(9

5% C

I)

Adu

lt m

orta

lity

(pro

babi

lity

(%) o

f dy

ing

from

age

s 15

-59

incl

usiv

e)

Life

Exp

ecta

ncy

at b

irth

(yea

rs)

(9

5% C

I)

(Bot

h se

xes)

(B

oth

sexe

s)

Mal

es

Fem

ales

M

ales

Fe

mal

es

2001

-20

04

2005

-20

09

2001

-200

4 20

05-2

009

2001

-20

04

2005

-20

09

2001

-20

04

2005

-20

09

2001

-20

04

2005

-20

09

2001

-20

04

2005

-20

09

Cen

sus

(200

5)^

52

69

58.9

63.1

Rec

onci

led

data

In

cl.

Stil

lbirt

hs

52.0

51

.6

72.1

(6

7.5-

76.

7)

71.5

(6

6.8-

76.4

)

48.1

46

.8

29.1

27

.1

54.2

(5

3.8-

54

.6)

55.1

(5

4.6-

55.4

)

63.1

(6

2.6-

63.6

)

64.2

(6

3.7-

64

.6)

Exc

l. st

illbi

rths

49.0

41

.8

67.5

(6

3.0-

71.

9)

56.4

(5

2.3-

60.9

)

48.2

40

.5

29.1

22

.6

54.4

(5

4.0-

54.8

)

58.9

(5

8.4-

59

.3)

63.4

(6

3.0-

63.9

)

68.1

(6

7.7-

68

.6)

Cap

ture

-re

capt

ure

anal

ysis

Incl

. S

tillb

irths

10

6.8

92.7

16

4.8

(1

58.3

- 17

1.0)

139.

6

(133

.6-

146.

2)

61.2

60

.0

44.3

37

.4

43.7

(4

3.2-

44.1

)

45.0

(4

4.5-

45

.4)

49.8

(4

9.2-

53.0

)

54.0

(5

4.4

– 55

.5)

Exc

l. S

tillb

irths

95

.7

71.3

14

4.9

(1

38.7

-15

0.9)

103.

2

(97.

8-

109.

0)

61.3

52

.9

44.3

31

.8

44.4

(4

3.9-

44

.8)

49.7

(4

9.2-

52

.0)

51.2

(5

0.7-

51

.8)

59.2

(5

8.7-

59

.7)

Bes

t Est

imat

e*

52.0

51

.6

72.1

(6

7.5-

76.

7)

71.5

(6

6.8-

76.4

)

61.2

60

.0

44.3

37

.4

47.9

(4

7.5-

48

.2)

48.6

(4

7.2-

48

.3)

55.7

(5

5.3-

56

.1)

57.9

(5

5.0-

56

.3)

^ C

ensu

s an

alys

is b

ased

on

mod

el li

fe ta

ble

from

sin

ge in

put p

aram

eter

– c

hild

hood

mor

talit

y de

rived

from

CE

BC

S m

etho

d. N

o co

nfid

ence

inte

rval

s av

aila

ble

* B

ased

on

capt

ure-

reca

ptur

e an

alys

is fo

r old

er a

ge g

roup

s an

d re

conc

iled

data

for c

hild

ren

(incl

udin

g st

illbirt

hs)

+ C

onfid

ence

inte

rval

s ba

sed

on re

porti

ng c

ompl

eten

ess

for c

aptu

re-re

capt

ure

anal

ysis

and

Chi

ang

life

tabl

e m

etho

ds fo

r re

conc

iled

data

N

ote:

Adu

lt m

orta

lity

varie

s as

add

ing

stillb

irths

affe

cts

all a

ges

thro

ugh

age

redi

strib

utio

n.

� � �

211

Although there were substantial differences in assigned age between the civil

registration and MoH data, the largest differences were in adjacent age groups and

balanced each other out, therefore having negligible effect on summary measures of

mortality (Figure 19).

Figure 19: Differences in total number of reported deaths by age group between Civil registry data and MoH data for Kiribati, 2000-2009

Capture-recapture analysis

The capture-recapture analysis suggests overall the total reported deaths (reconciled

data) is 59% (95% CI: 58-60%) complete. This is plausible in light of what is known

regarding the CRVS systems in Kiribati, particularly as there is little social incentive

for registration of deaths and burial occurs at home. Assessed as a combined total

(not disaggregated by age group), civil registration data were estimated to be 35%

(95% CI: 34-36%) complete and MoH data 38% (95% CI: 37-39%) complete. There

has been no sustained improvement in completeness over the study period for either

�30

�20

�10

0

10

20

30

F

M

� � �

212

reporting system. Estimates from the capture-recapture analysis place completeness

for the civil registration data between 32 and 38%, and the MoH data between 30

and 45%.

Based on the capture-recapture analysis, reporting patterns varied significantly by

age; with significant under-reporting in younger ages, particularly in the CRVS

system, and lower reporting rates for females than males in the older age groups

(Figure 20).

Figure 20: Estimated completeness of reported deaths from 2-source capture-recapture analysis by source, sex and age group for Kiribati, 2000-2009

Sources were found to be largely independent (with “B(r)”, a measure of dependence

where 1 = independent (230)) estimated to equal 0.99). Based on the system design,

particularly the Civil Registrar’s requirement that evidence such as a medical

certificate be presented to complete the registration, any dependence that did affect

the calculations would be anticipated to be positive, that is deaths reported to one

0.00

0.10

0.20

0.30

0.40

0.50

0.60

<�5 5�14 15�24 25�34 35�44 45�54 55�64 65�74 75�84 85+

Reporting�Completeness

Age�Group

Civil�Registry�� MalesCivil�Registry�� FemalesHealth�� MalesHealth�� Females

� � �

213

source would also be more likely to occur in the other. This is supported by the “r”

value as used by Hook and Regal (336), which is estimated at 0.21 in this analysis

(as r > 0 indicates positive dependence), and would result in an overestimation of

reporting completeness. On the other hand, poor data quality may have resulted in

some cases which should have been matched between both sources being missed,

which would lead to an underestimation of reporting completeness.

Based on the capture-recapture analysis, U5M (both sexes combined) for 2001-2009

was estimated at 135.6 deaths per 1,000 live births, excluding stillbirths, and 156.2

deaths per 1,000 live births, if stillbirths are included in the data set. Using regression

methods, this equates to an IMR of 90.4 and 102.1 deaths per 1,000 live births

respectively. Adult mortality was estimated at 60.0% for males and 40.3% for

females depending on whether stillbirths were included in the data set (thereby

adjusting the distribution of deaths of unknown ages). LE is estimated at 44.1- 45.2

years for males and 51.5 - 53.2 years for females.

Best estimates of summary measures of mortality

As noted in the methods, the capture-recapture analysis produced improbably high

estimates for total deaths in the 0-4 year age group, but appeared plausible for other

age groups (Figure 21). This is plausible as there is likely to be greater dependency

between the MoH and Civil Registration Offices given local emphasis on child health

programs, which would encourage reporting of known childhood deaths. Names and

other identifiable data are also poorly recorded in this age group (with names

changed as parents settle on a name for their child), making it more likely that under-

matching would occur. This is apparent from the number of estimated deaths by age

� � �

214

group as shown in Figure 21. This issue with data matching and dependence is

largely confined to the under 5 age group, and as such, the capture-recapture

estimates derived by age and sex were used to adjust all other age groups. Thus

final estimates of mortality were derived from reconciled data (including stillbirths) for

children (0-4 years) and adjusted figures based on the capture-recapture analysis by

sex and age group for all remaining ages. This results in an estimate of 1,307 deaths

per year, or a crude death rate of 14 deaths per 1,000 population.

Figure 21: Estimated average deaths per year (2000-2009) for Kiribati from census data and routinely collected data

The best available estimate of IMR (2001-2009) is from the reconciled data as

described earlier, and is 52.9 deaths per 1,000 live births, with a plausible range

from 48.9 - 113.7 deaths per 1,000 live births based on the analysis of

� � �

215

completeness. U5M is estimated at 73.6 deaths per 1,000 live births, with a plausible

range from 67.3 - 117.3 deaths per 1,000 live births.

Adult mortality for 2001-2009 is significantly higher in males than females, and LE is

significantly lower. Adult mortality is estimated at 60.0% for males (plausible range

46.6 = 63.3%) and 40.3 for females (plausible range 27.8 - 44.0%). LE is estimated

as 48.5 years for males (plausible range 44.1 - 55.3 years), and 56.9 years for

females (plausible range 48.6 - 64.0 years).

There was no significant change in any of the summary measures of mortality

between 2000-2004 and 2005-2009. The only indication of possible improvement

over this period was a slight reduction in estimated adult mortality for females and a

subsequent improvement in best estimates of LE from 55.7 years to 57.9 years

(Table 20).

��

�

216

Tabl

e 21

: Est

imat

ed a

vera

ge d

eath

s pe

r yea

r by

data

sou

rce

and

met

hod

of e

stim

atio

n: K

iriba

ti, 2

000-

2009

by

age

grou

p an

d se

x

M

ale

Fem

ale

Tota

l

Age

gr

oup

Cen

sus

Rec

onci

led

data

(no

adju

stm

ent)

Cap

ture

re

capt

ure

estim

ate

Cen

sus

Rec

onci

led

data

(no

adju

stm

ent)

Cap

ture

re

capt

ure

estim

ate

Cen

sus

Rec

onci

led

data

(no

adju

stm

ent)

Cap

ture

re

capt

ure

estim

ate

< 5

8888

19

580

8118

816

816

938

25-

14

1619

31

1114

2126

3352

15-2

4 23

41

8119

1624

4256

103

25-3

4 22

46

7722

2440

4270

118

35-4

4 38

69

9432

3756

7010

615

045

-54

5284

12

040

4765

9313

118

555

-64

6569

93

4945

8811

411

417

365

-74

7264

96

7055

9514

211

919

075

-84

3834

57

5142

7488

7613

185

+ 6

7 11

1413

3620

2042

Incl

udes

stil

l birt

h an

d re

-dis

tribu

ted

for u

nkno

wn

� � �

217

Discussion

The reconciled data demonstrate LE for males over 2001-2009 was at most 54

years, several years lower than the 2005 census estimate of 59 years. For females,

LE from the reconciled data was estimated to be 63 years, matching findings from

the 2005 census. However, the findings presented here, along with knowledge of the

CRVS systems in Kiribati, suggest it is unlikely that even the reconciled data are fully

enumerated. As described in the results, best estimates place LE for 2001-2009 at

48.5 years for males and 56.9 years for females, however this could be as low as 44

and 49 years respectively. IMR from the reconciled data was estimated at 52.9

deaths per 1,000 live births, almost identical to the 2005 census findings, although

capture-recapture analysis suggests this figure could be higher.

There were no apparent secular trends in any of the summary measures from 2000-

2004 to 2005-2009, in either reconciled data or adjusted data based on the capture-

recapture analysis, with the possible exception of a slight improvement in female

adult mortality. This indicates that while Kiribati is not undergoing any significant

decline in health status, there were also no discernible improvements for most

sectors of the population, despite significant attention from international donors to

improving public health programs over the last several years. There has been little

change in IMR or U5M and Kiribati is highly unlikely to meet its MDG targets (49) in

this area.

The total estimated deaths by age group and sex per year from the census analysis

(using model life tables derived from U5M from CEBCS methods) (50) was very

� � �

218

similar to the total deaths reported (reconciled data). This would indicate either

nearly all deaths in Kiribati are reported to at least one of the routine reporting

systems, or the census has substantially underestimated mortality, as would be

expected from the use of a single input parameter model. The capture-recapture

analysis indicates that overall, the total reported deaths (reconciled data) is 59%

(95% CI: 58-60%) complete. Reporting completeness varied much more significantly

by age group, with the Civil Registration Office recording a much lower proportion of

childhood deaths than MoH. In older age groups, reporting completeness was lower

for females than for males, particularly in the MoH data.

It is possible that unreported deaths may have been overestimated through the

capture-recapture analysis due to poor data quality, resulting in deaths reported in

both the Civil Registration and MoH data not being recognised as a match. The

variation in names used to identify an individual in Kiribati increased the potential for

deaths remaining unmatched if other variables were recorded poorly or absent.

Despite these issues, it is unlikely that missed matches account for the bulk of the

difference between the census estimates and reconciled data, and the estimates

from the capture-recapture analysis due to the matching criteria employed. Matching

criteria were developed and trialled on a sample data set taken from the records,

with multiple variations reviewed by staff from both Civil Registration and MoH to

ensure the least number of missed matches. The final criteria selected are very

broad and allow names to be excluded from the requirements for a match to account

for the problems outlined earlier. The matched data set (including unmatched

deaths) was reviewed independently by both MoH and Civil Registration staff to

minimise possible errors. Missed matches are more likely to have occurred amongst

� � �

219

young children, as this is the age group in which baptism and adoptions are usually

carried out, and individuals are less likely to have settled on their own name. While

the inclusion of stillbirths in the data set may also have led to some overestimation,

the decision to include these in the data set was based upon discussion with a range

of Civil Registration and MoH staff. These discussions produced a general

consensus that early neonatal deaths (up to about 3 days old) are routinely recorded

as stillbirths in both the Civil Registration and MoH (community nursing) data, as this

is more socially acceptable for the family.

Capture-recapture analysis requires: the record sources capture deaths from the

same population, the population is closed, and every death has the same probability

of being reported (229, 230). There is very little external migration to or from Kiribati,

indicating that while individual islands should be treated as open populations, the

population is largely closed at a national level. While capture-recapture analysis is

able to account for dependence between the data sources if three or more data

sources are available, a two source analysis assumes complete independence. The

system assessment in Kiribati identified positive dependence between the Civil

Registration and MoH data, which could result in overestimation of reporting

completeness (and subsequently underestimation of mortality), although the

magnitude of this effect was very small with sources found to be close to

independent. Further, the findings of the capture-recapture analysis are consistent

with both an understanding of how the system operates (164) as well as previous

reports (including the 2005 census (4) which have suggested routinely collected data

were too incomplete for use). Differences to the census estimates may have resulted

from the use of models with a single input parameter (from U5M) to derive them (50).

� � �

220

Assessments from other PICTs, including Fiji (93) and Tonga (111), have

demonstrated these models overestimate LE where adult mortality is high, as

reported here in both the reconciled and adjusted data. For females, where

estimates of adult mortality were lower, the single parameter model may be more

suitable and account for the similarity of findings.

The findings here show LE in Kiribati is lower than previously reported (108), and

may be significantly lower than estimates from the reconciled data. Estimates of LE

in Kiribati from the reconciled data are higher than recent estimates from

neighbouring Nauru (LE 2002-2007 estimated at 52.8 years for males and 57.5

years for females). Nauru also has high adult mortality, but does not have the

overcrowding, malnutrition and sanitation problems noted in Kiribati and would

reasonably be expected to have a higher LE than Kiribati, despite a long history of

NCDs.

The findings presented here demonstrate two critical issues. Firstly, this analysis

demonstrates the importance of using empirical data, despite its flaws, to examine

mortality and health issues at a country level. These findings show the health

situation in Kiribati is worse than recognised through census analysis, highlighting

the problems associated with relying on census estimates that impute adult mortality

from childhood mortality. There is clearly a need to improve routine data collections

through standardising the data collected in key fields, reconciling data between

sources (particularly within MoH), managing duplicates in reported data and

educating data collectors in standard definitions (such as for stillbirths). However, the

� � �

221

need to improve data collection and reporting should not be seen as a reason not to

examine and report on trends using available collected data.

Secondly, regardless of the reconciled data, adjusted data from the capture-

recapture analysis or best estimates from a combination of both, show the health

situation in Kiribati needs urgent attention. IMR and U5M are higher than most other

PICTs (108), (Table 5) with the exception of PNG, with little progress made towards

improving these figures under the MDGs. There is also significant premature male

mortality, which has led to some of the lowest estimates of LE for the region (Table

6). A preliminary review of data from the Civil Registration Office suggests this is at

least in part driven by a very high incidence of suicide, and further investigation into

the causes of death contributing to these low LE’s is urgently needed.

5.6 Mortality Estimation in the absence of routine death

reporting

The system assessment in Chapter three demonstrated that although routine

reporting systems for deaths exist in Vanuatu, these had limited coverage (therefore

may not be representative at a national level) and were largely incomplete, rendering

the data gathered through these systems insufficient to generate empirical estimates

of mortality. National estimates are reliant on estimates from census and survey

data. If all criteria used earlier in this chapter to assess published estimates were

applied, only U5M and IMR from the census and a few surveys remain. Given the

absence of other reliable data, it is necessary to rely on census estimates for LE,

� � �

222

adult mortality and age-specific mortality rates, which were calculated based on

single input parameter (5q0) model life tables.

Published estimates from census and survey sources should still be considered in

light of their plausibility. As demonstrated in the material included in this chapter for

both Fiji and Kiribati, single input parameter model life tables based on U5M alone

are likely to underestimate adult mortality, and therefore overestimate LE. To

demonstrate this more clearly, empirical estimates of mortality for Fiji were input into

both one and two parameter models in Modmatch (235), the results of this are

shown in the figures below.

Figure 22: Empirical age-specific mortality compared with predictions from one and two parameter models for Fiji, Males: 2002-2004

Model: WHO Modmatch (235)

� � �

223

Figure 23: Empirical age-specific mortality compared with predictions from one and two parameter models for Fiji, Females: 2002-2004

Model: WHO Modmatch (235)

Both Figures 22 and 23 show substantially lower death rates across all age groups

from models calculated using a single input parameter of 5q0 when compared with

the empirical rates or estimates derived with two input parameters models (5q0 and

45q15). Similar results occur for Tonga if U5M (24.5 deaths per 1,000 live births for

2005-2008) is inputted into a single parameter model life table system. Modmatch

estimates LE at 69 and 73.4 years for males and females respectively, several years

higher than what has been demonstrated to be plausible in this research (case

study five).

� � �

224

While it could be argued that this effect may be more pronounced in Fiji, given Fiji

has moved through the epidemiological transition with NCDs now the leading causes

of death , results in Kiribati, which is earlier in the transition (and therefore not

dissimilar to Vanuatu), also demonstrated that census LE estimates based on single

parameter inputs were potentially significantly overestimated. Census estimates LE

and adult mortality for Vanuatu, should be considered more as a lower limit of

mortality across age groups 5 years and above.

U5M from the Vanuatu census (2009) was 24 per 1,000 live births. Adult mortality

(45q15) was estimated based on model life tables as 172 deaths per 1,000

population for males and 121 deaths per 1,000 population for females, with LE for

males and females estimated as 69.9 and 73.5 years respectively. These would

represent substantially lower mortality than all other study countries in this work,

which is unlikely given the level of infrastructure in Vanuatu; and as such should be

treated as a lower limit rather than a true reflection of mortality levels.

Adult mortality estimates for Vanuatu derived by Rajaratnam et. al. (41) using

multivariate models including economic and a vast range of international mortality

data, are likely to be more representative, with estimates of adult mortality in 2010 of

around 275 deaths per 1,000 population for males and 170 deaths per 1,000

population for females. These models however, show a continued improvement in

mortality rates over time, which is inconsistent with the patterns seen in neighbouring

countries (as presented in this research) which show a clear stagnation or worsening

in adult mortality, and should be further investigated in Kiribati. Computing life tables

from two parameter inputs including this adult mortality (using Modmatch (235))

� � �

225

gives a LE of approximately 65 years for males, and 70 for females; figures which

are more plausible given findings presented through the rest of the chapter, but

which would still put Vanuatu amongst the highest life expectancies amongst the

study countries. While Vanuatu is yet to move far into the epidemiological transition,

as some other PICTs, such as Nauru and Tonga have; the most recent Vanuatu

STEPS survey (337) certainly indicates NCDs are likely to be an issue due to the

high prevalence of diabetes and risk factors such as hypertension. Further data

collection on adult mortality in Vanuatu is therefore critical to determine whether

Vanuatu has in fact managed to so far avoid the high premature adult mortality from

NCDs seen in the rest of the PICTs, and if so, to examine how best to learn from the

experiences of the rest of the PICTs with NCDs to avoid following the same path.

5.7 Summary of method selection and application

Although not all PICTs have complete reporting of deaths, it is possible to generate

reliable measures of mortality level from routinely reported data in many of them by

evaluating the data and correcting for undercount using the methods described in

this chapter. Even where routinely collected data is insufficient to generate measures

directly, they may be useful in establishing plausible ranges of mortality in order to

help interpret data from other sources, such as censuses and surveys. In spite of

these findings, countries such as Vanuatu, with very under-developed reporting

systems, do not collect sufficient data to accurately estimate mortality levels either

directly or with correction. In this setting, improving CRVS systems must be a key

focus for improving the availability and reliability of mortality data for planning and

evaluation.

� � �

226

As noted at the beginning of this chapter, PICTs fall into three broad categories,

those which have complete or nearly complete reporting, those where CRVS

systems are mostly functional but the level of reporting (or patterns of under-

reporting) is largely unknown, and those which do not collect sufficient information

for analysis. Methods for deriving mortality measures for each of these categories,

as documented in Figure 11, were applied to each of the study countries to

demonstrate the potential for deriving a greater understanding of current mortality

levels and trends from routinely collected data than has previously been available.

In countries where collected data require correction, initial corrections could be made

by reconciling data from different sources of death recording. Where this was

insufficient, either capture-recapture methods, or evaluation and correction using the

Brass Growth Balance Method could be used to correct the data or provide plausible

ranges. Both methods must be applied in the presence of strong system

understanding in order to accurately interpret results, fit the age groups in the Brass

Growth Balance Method, or to select an appropriate model using capture-recapture

methods.

A summary of the data sources and methods used to generate measures of mortality

for each of the study countries is shown in the table below.

� � �

227

Table 22: Summary of data sources and methods used in assessment of mortality level for the Pacific Islands

Country Data Source(s) Used in Final Calculations

AssessmentsApplied*

Adjustmentfor under-enumerationmade to reporteddeaths

Years of dataavailable

Fiji MoH Assessment of Published Data Brass Analysis

No 2002-2004

Kiribati MoH and Civil registry

Reconciliation of sources 2 source capture-recapture analysis

Yes 2000-2009

Nauru MoH/Civil registry (shared data)

Mortality level calculated directly from empirical data

No 2002-2007

Palau MoH (Epidemiology spreadsheet)

Mortality level calculated directly from empirical data

No 2004-2009 (excl 2007)

Tonga MoH (HIS and Community nursing data) and Civil registry

Reconciliation of sources 3 source capture-recapture analysis

Yes 2001-2009

Vanuatu Census and survey estimates

Plausibility Assessment of Published Data

Secondary data only available

2007-2011

* A system assessment and trend analysis was conducted for all countries in

addition to the assessments shown in the table.

Age groups for each country were selected on the basis of population size and

disaggregated population data available. Where there were sufficient deaths to do

so, infant mortality was examined separately. Ten year age groupings were used in

Nauru and Palau, where total population size is very small. Statistical confidence

� � �

228

intervals for life tables for all countries were obtained using the Chiang method (296).

A summary of findings across all countries is given in Chapter seven.

� � �

229

6 Assessing cause of death distributions

While mortality level at various ages is important for measuring the health of a

population and identifying key disparities between populations, it is robust data on

CoD distributions that allows targeted interventions to be shaped to affect these

mortality outcomes. In this chapter, the data available from each of the selected

study countries is explored in more detail in relation to the availability and quality of

the CoD data, in order to generate CoD distributions by age where possible.

The system assessment in Chapter three identified that there is significantly more

data on CoD available from the study countries than is routinely collated or reported,

although the quality of these data are likely to vary significantly (Table 14). Medical

certificates, considered the best source of CoD data when completed correctly, are

� � �

230

required for at least some deaths in all of the study countries. In Nauru, Palau,

Tonga and Fiji, medical certificates are required either by policy or law for all deaths,

although not all of these data are routinely coded or collated; while in Vanuatu and

Kiribati, certificates are completed only for those deaths that occur in hospital.

Additional reporting systems, such as the community health reporting in Vanuatu,

Tonga and Palau, were also identified.

The extent and quality of CoD information available for each country is a function of

which deaths are medically certified, and the procedures and practices at three key

stages of the collection and collation process: (1) designation of a diagnosis for

CoD, and completion of the medical certificate by the medical practitioner: (2) coding

of these data and selection of underlying CoD according to the ICD rules (68); and

(3) tabulation and presentation of the data on deaths by sex, age and cause group.

As with data on mortality level, all data on causes of death should be reviewed for

plausibility. Both the framework for evaluating CoD statistics developed by Rao et. al.

(63, 64) and the more recent framework published by the UQ HIS Hub -

(subsequently developed into the ANACOD tool published by WHO) (320), both

provide useful guidance on determining whether national CoD data are plausible.

Both include aspects such as coverage, completeness and timeliness, which are

equally applicable for all-cause and cause-specific mortality data. They also include

aspects such as the use of ill-defined causes, age and sex dependency, consistency

over time and consistency with general mortality levels.

� � �

231

Ill-defined categories includes those cases which have no known cause, those

attributed to signs and symptoms (such as fever or cough) rather than specific

conditions or diseases, conditions that do not explain the death (such as old age and

senility) and those for which only a “mode” of death (such as stopped breathing) is

recorded. The GBD (103), and the UQ HIS Hub framework advocate redistributing

into specific conditions the so-called “garbage” codes, that is, deaths attributed to

causes which tend to be used as “catch all” categories when more specific

information is not available - such as “cardiac failure” or “heart attack”. This

redistribution requires the data set to be adjusted has been coded accurately, and

assumes a redistribution pattern based on broad global experience (103). What is

considered to be ill-defined may vary based on the detail one is trying to obtain from

the data. As Naghavi et. al. point out, many ICD chapters contain a ‘generic’

category, such as neoplasms of unspecified site (265). While correction for these

causes is not required to look at trends across broad causes, these categories do

not allow detailed analysis of specific diseases within the chapter without further

redistribution.

Cause dependency by sex refers to ensuring that causes of death which are sex-

specific have not been wrongly attributed (such as prostate cancer recorded for a

female, or maternal causes recorded for a male). Age dependency refers to

plausibility of the age distributions by cause. For example, as Rao et. al.

demonstrates for China (63), deaths from maternal haemorrhage occur between the

ages of 15 and 40years, peaking between 20-25 years, with no deaths recorded

outside these age groups. The data are therefore consistent with the ages at which

women are in their childbearing years and potentially at risk of maternal

� � �

232

haemorrhage. Similarly, deaths from lower respiratory illnesses are highest in infants

and children, with very few recorded deaths from this cause in the following age

groups until late adulthood, when numbers again begin to rise. This is consistent with

known distributions of infectious disease mortality and demonstrates internal

consistency within the data. The UQ HIS Hub framework (213) demonstrates this

using the GBD categories (rather than specific disease categories), with Group 1

diseases (primarily infectious and respiratory diseases) following a similar pattern in

Venezuela to the age distribution of respiratory deaths demonstrated by Rao et.al for

China (63).

Both the Rao et. al. (63) and UQ HIS Hub (320) frameworks provide a useful

approach to assessment of the plausibility of CoD data collections, once analysed

and reported. In addition to the above aspects, Rao et.al.’s framework (63) also

describes the concept of content validity, which is a subjective assessment of the

process used to obtain and collate the data, and the impact this may have on data

quality. For example, as the China Disease Surveillance Point System examined by

Rao (63) uses interviews with family members as the primary source of data on CoD,

Rao et. al. (63) noted the collection may have limited content validity. For collections

where the system assessment (as described in Chapter three) has identified

potential content validity concerns due to procedures around the collection and

tabulation of data, further investigation of the data quality through a medical record

or certificate review as described in Chapter two is warranted.

Figure 24 shows the processes involved in collecting CoD data for health monitoring

and evaluation; and the assessments which can be applied to determine the validity

� � �

233

of the data generated through these processes. The ability to conduct these

assessments is largely governed by access to the original data.

Figure 24: Collection and suggested review process for cause of death data

As CoD data from all study countries was obtained as either unit record uncoded raw

text (which was subsequently coded according to ICD rules), or as ICD coded data

without the original text, an audit of coding practices (as described in Chapter two)

for the study countries could not be undertaken. There are several good examples of

coding audits, such as undertaken in Australia and England, recorded in the

literature which demonstrate how this could be undertaken (338-340).

Of the four countries where medical certificates of death are required for all deaths,

the system assessment identified CoD data from Palau as likely to have a high

content validity, indicating estimates could be generated directly from the collected

� � �

234

data and assessed for plausibility. For Tonga, Fiji and Nauru, the system

assessments raised queries regarding the content validity that warrant further review

and potential correction of the data. A medical record review was conducted in

Tonga (case study six) and Nauru. A medical certificate review and multiple CoD

analysis was conducted in Fiji where the same access to medical records was not

available.

For Vanuatu and Kiribati, where medical certificates are only completed for deaths in

hospital, alternative data sources are explored. Community nursing reports are

contrasted against hospital and medical certificate data from Vanuatu. In Kiribati,

selected causes of death of particular public health importance are reviewed through

community reporting to the civil registry office. While these alternative data sources

are not an adequate replacement for reliable CoD data on community deaths

through medical certification or verbal autopsy, these reviews do provide an

indication of the impact of key causes of death, such as external causes (injuries and

suicide), malnutrition and maternal causes, which has not been adequately captured

in country reporting to date.

6.1 Direct tabulation of cause of death from collected data

6.1.1 Analysis of cause of death in Palau

Aim and Objectives

The system assessment in Chapter three identified a robust quality control process

to ensure medical certificates in Palau are completed according to the best available

information, and in accordance with the ICD standards. Each certificate is formally

� � �

235

reviewed and countersigned by the medical director, who checks both the plausibility

of the assigned cause and the sequence on the certificate. Certificates which do not

meet these standards are referred back to the treating doctor to review. All deaths

are also reviewed periodically through a mortality review meeting of hospital doctors.

As a result, Palau has a very low percentage of deaths reported as “Ill-defined” or

“garbage” codes and is expected to have a high content validity, allowing direct

analysis of the data without correction through a medical record or medical certificate

review process.

Palau does not routinely code or report their CoD data, other than as leading causes

of death for all ages (in the regional CHIPs reports). Causal distributions of deaths by

age and sex for 1999-2003 and 2004-2009 were therefore generated by direct

analysis of the tabulated data from the medical certificates.

Methods

Data from medical certificates was obtained from the MoH for 2004-2009 as an

Excel file, capturing each field on the medical certificate as uncoded text exactly as

written on the original certificate. Underlying CoD was then identified [by KC]

according to ICDv10 rules. Due to the small number of deaths, underlying cause was

tabulated directly to the ICDv10 General Mortality List One (103 causes) (68) for

analysis by broad age group and sex. Deaths were initially tabulated in ten year age

groups so the broad groupings used were 0-4 years, 5-14 years, 15-64 years and

�65 years. Confidence intervals for proportional mortality were calculated using the

Poisson distribution due to the small number of events. Confidence intervals for

cause-specific rates were subsequently derived from the uncertainty of the

� � �

236

proportional mortality, without adjustment for uncertainty in the overall mortality

envelope.

As noted in the previous chapter, an unusually low number of deaths were reported

in 2007, so this year was excluded from the analysis as being unrepresentative.

Results

For children aged 0-4 years, congenital issues (1-093) and perinatal conditions (1-

092) were approximately 70% of all deaths, although there was some shift between

these categories across the two periods (Figure 25). The cause-specific death rate

for perinatal causes was 133 per 100,000 population (95% CI: 0-376.6 deaths per

100,000 population) for 2004-2009. Only one death was attributed to unknown

causes. The remainder were spread over a number of cause categories. The overall

number of deaths, even when aggregated over five years of data is very low, and as

such significant uncertainty remains in the cause-specific calculations as shown in

Table 23.

� � �

237

Figure 25: Proportional mortality by period for children aged 0-4 years both sexes: Palau, 1999-2009 (excl 2007)

n=20 n=25

Key 1-001 Infectious diseases 1-026 Neoplasms 1-048 Diseases of the blood 1-051 Endocrine, nutritional and other metabolic disorders 1-058 Diseases of the nervous system 1-064 Diseases of the circulatory system 1-072 Diseases of the respiratory system 1-078 Diseases of the digestive tract

1-082 Diseases of the skin 1-084 Diseases of the genito-urinary system 1-092 Perinatal conditions 1-093 Congenital conditions 1-094 Signs, symptoms and other ill-defined causes 1-095 External causes (injuries)

1�0585% 1�064

5%

1�07810%

1�09251%

1�09319%

1�09510%

1999�2003

1�0484%

1�0584% 1�064

4%

1�0728%

1�0844%

1�09264%

1�0934%

1�0944%

1�0954%

2004�2009�(excl.�2007)�

��

�

238

Tabl

e 23

: Dea

ths

by c

ause

, and

pro

port

iona

l mor

talit

y fo

r chi

ldre

n 0-

4 ye

ars,

by

perio

d fo

r Pal

au; 1

999-

2009

(exc

ludi

ng

2007

)

� � �

239

There were nine deaths in 5-14 year olds certified from 1999-2003 and six from

2004-2009 (excluding 2007). The proportion of deaths attributed to external causes

fell from 44% in the earlier period to 16% (one death) in the latter. Although this trend

was not significant when confidence intervals were calculated, this is largely a

function of the small number of events, and the decrease is certainly relevant at a

policy level. Remaining deaths were spread across other cause categories with only

single deaths appearing in each, making further interpretation of cause patterns in

this age group difficult.

Examining proportional mortality for certified deaths shows very different patterns of

mortality between males and females. As shown in Figure 26, while cancers (1-026)

account for 20-30% of deaths in both men and women, this is substantially higher for

women than men, despite the decline between the two periods (Table 24 and 25).

Breast cancer has risen from 3% of all deaths in women in this age group, with a rate

of 6.8 deaths per 100,000 population (95% CI: 0.8–24.5) in 1999-2003 to 4% of

deaths in 2004-2009, with a cause-specific rate of 15.7 deaths per 100,000

population (95% CI: 5.1–36.7). The proportion of deaths due to cervical cancer has

fallen from 7% in 1999-2003 to 3% in 2004-2009 with a slight (but non-significant) fall

in the cause-specific rate from 13.6 deaths per 100,000 population (95% CI: 3.7 -

34.8) to 12.6 deaths per 100,000 population (95% CI: 3.4-32.3). The leading cancer

deaths in males were from liver and lung cancer, both of which rose slightly over the

two periods reviewed, from 5% and 4% respectively in 1999-2003, to 7% and 5%

respectively in 2004-2009, although these changes were not significant and may

reflect stochastic variation. Lung cancer rates are very low, consistent with the

historically low smoking rates in Palau. In 1998, Palau was the only PICT to report

� � �

240

smoking prevalence lower than Australia, with 22% of men and 8% of women over

20 years of age self classifying as smokers.

Figure 26: Proportional mortality by cause and period for Palauan males: Adults 15-64 years for 1999-2009 (excluding 2007)



0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

1�00

11�

026

1�04

81�

051

1�05

51�

058

1�06

41�

072

1�07

81�

082

1�08

31�

084

1�09

21�

093

1�09

41�

095

M�1999�2003

M�2004�2009

� � �

241

Figure 27: Proportional mortality by cause and period for Palauan females: Adults 15-64 years for 1999-2009 (excluding 2007)



The cause-specific mortality rate from external causes in 2004-2009 was 37.8

deaths per 100,000 population in females (95% CI: 19.5-66.0) and 170.8 deaths per

100,000 population in males (95%CI: 132.4-217.0), indicating a statistically

significant difference between sexes, although there has been a notable decline

across both sexes between the two periods. In 2004-2009, deaths from external

causes accounted for 10% of deaths in females aged 15-64 and 27% of deaths in

males in this age group.

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

1�00

11�

026

1�04

81�

051

1�05

51�

058

1�06

41�

072

1�07

81�

082

1�08

31�

084

1�09

21�

093

1�09

41�

095

F�1999�2003

F�2004�2009

� � �

242

This high mortality from external causes in males is clearly of public health interest

and there is a need to understand the specific causes contributing to this category of

deaths. As shown in Figure 28, 58% of deaths from external causes in males aged

15-64 from 2004-2009 are attributed to “other external causes” when classified

strictly according to the ICD rules. This reflects a tendency of certifying doctors to

record the injuries sustained rather than how the injury occurred and indicates a

need for further training, despite the extensive review processes in place. Data on

external causes is therefore currently of limited use for health planning, although a

medical record or medical certificate review may allow greater disaggregation of

deaths within this causal group. Of the deaths for which a specific cause could be

identified, there were 39 deaths in 2004-2009 in this age group attributed to suicide,

accounting for a rate of at least 22.9 deaths per 100,000 population (95% CI: 10.5 -

43.6 unadjusted for ill-defined external causes).

� � �

243

Figure 28: Proportion of deaths due to external causes, by type of external cause for males aged 15-64 years: Palau, 2004-2009 (excluding 2007)

Figures 26 and 27 also show the high proportion of deaths attributed to heart

diseases (1-064) in both males and females; which has increased for both sexes

over the two periods. While the proportion of deaths attributed to diabetes (1-051)

has fallen for females and risen only slightly in males, the proportion of deaths

attributed to genitourinary diseases (1-084) has almost doubled for both sexes and

may indicate diabetes is not being recorded appropriately on the medical certificates.

Rates for causes of death by ICD chapter are shown in Tables 24 and 25, with rates

for specific causes (according to ICD General Mortality List One) attached in

Appendix four.

1�096�Transport�accidents

13% 1�098�Falls5%1�099�Drowning�

2%

1�100�Accidental�poisoning

2%1�101�

Intentional�Self�Harm13%

1�102�Assault7%

1�103�Other�external�causes

58%

� � �

244

Table 24: Cause-specific mortality by ICD chapter and period for adult males 15-64 years: Palau, 1999-2009 (excluding 2007)

1999-2003 2004-2009 Cause of death Deaths Rate (per

100,000) Lower 95% CI

Upper 95% CI

Deaths Rate (per 100,000)

Lower 95% CI

Upper 95% CI

1-001 Infectious diseases

6 18.9 6.9 41.1 20 51.0 31.2 78.8

1-026 Neoplasms

35 110.2 76.8 153.3 49 125.4 92.4 165.2

1-048 Diseases of the blood

3 9.4 1.9 27.6 1 2.5 0.1 14.2

1-051 Endocrine, nutritional and other metabolic disorders

14 44.1 24.1 74.0 19 48.4 29.2 75.7

1-055 Mental and behavioural disorders

0 0.0 0.0 11.6 4 10.2 2.8 26.1

1-058 Diseases of the nervous system

1 3.1 0.1 17.5 3 7.6 1.6 22.4

1-064 Diseases of the circulatory system

30 94.5 63.7 134.9 55 140.2 105.7 182.5

1-072 Diseases of the respiratory system

4 12.6 3.4 32.3 9 22.9 10.5 43.6

1-078 Diseases of the digestive tract

4 12.6 3.4 32.3 9 22.9 10.5 43.6

1-082 Diseases of the skin

0 0.0 0.0 11.6 0 0.0 0.0 9.4

1-083 Diseases of the musculoskeletal system and connective tissue

4 12.6 3.4 32.3 1 2.5 0.1 14.2

1-084 Diseases of the genito-urinary system

10 31.5 15.1 57.9 12 30.6 15.8 53.5

1-092 Perinatal conditions

0 0.0 0.0 11.6 0 0.0 0.0 9.4

1-093 Congenital conditions

0 0.0 0.0 11.6 0 0.0 0.0 9.4

1-094 Signs, symptoms and other ill-defined causes

1 3.1 0.1 17.5 2 5.1 0.6 18.4

1-095 External causes (injuries)

12 37.8 19.5 66.0 67 170.8 132.4 217.0

�

� � �

245

Table 25: Cause specific mortality by ICD chapter and period for adult females 15-64 years: Palau, 1999-2009 (excluding 2007)

Cause of death Deaths Rate (per 100,000)

Lower 95% CI

Upper 95% CI

Deaths Rate (per 100,000)

Lower 95% CI

Upper 95% CI

1-001 Infectious diseases

6 20.4 7.5 44.4 6 18.9 6.9 41.1

1-026 Neoplasms

22 74.7 46.8 113.1 35 110.2 76.8 153.3

1-048 Diseases of the blood

0 0.0 0.0 12.5 3 9.4 1.9 27.6

1-051 Endocrine, nutritional and other metabolic disorders

9 30.6 14.0 58.0 14 44.1 24.1 74.0

1-055 Mental and behavioural disorders

0 0.0 0.0 12.5 0 0.0 0.0 11.6

1-058 Diseases of the nervous system

0 0.0 0.0 12.5 1 3.1 0.1 17.5


12 41.2 21.1 71.2 30 94.5 63.7 134.9


2 6.8 0.8 24.5 4 12.6 3.4 32.3

1-078 Diseases of the digestive tract

1 3.4 0.1 18.9 4 12.6 3.4 32.3

1-082 Diseases of the skin

0 0.0 0.0 12.5 0 0.0 0.0 11.6

1-083 Diseases of the musculoskeletal system and connective tissue

2 6.8 0.8 24.5 4 12.6 3.4 32.3

1-084 Diseases of the genito-urinary system

3 10.2 2.1 29.8 10 31.5 15.1 57.9

1-092 Perinatal conditions

0 0.0 0.0 12.5 0 0.0 0.0 11.6

1-093 Congenital conditions

0 0.0 0.0 12.5 0 0.0 0.0 11.6

1-094 Signs, symptoms and other ill-defined causes

0 0.0 0.0 12.5 1 3.1 0.1 17.5

1-095 External causes (injuries)

11 37.4 18.6 66.8 12 37.8 19.5 66.0

Discussion

Despite the limited change in deaths attributed to diabetes, both proportional

mortality and age-specific mortality (adults aged 15-59 years) from NCDs including

heart disease, cancer and genitourinary diseases have increased across the two

� � �

246

periods examined. Although these changes were not statistically significant, they

provide strong supportive evidence for calls for urgent public health action to address

premature mortality from NCDs.

6.2 Medical Record Review

While medical certification practices in Palau are monitored and have quality review

mechanisms in place, this level of scrutiny is not present (nor necessarily

practicable) in most PICTs. Indeed, as shown in the example from Palau, where

uncertainty was identified in both capture of external causes and diabetes-related

deaths, even a system with known quality control procedures may show indications

that further review of the data quality (and indeed training to improve collection in the

first place) may be warranted.

Where there is sufficient access to the original data, a medical record review allows

comparison of the underlying CoD as derived from the medical certificate, with the

available information from cases’ symptoms and history preceding death, to

determine whether the recorded sequence is plausible, and whether it is the most

reasonable diagnosis that could have been made. Medical record reviews were

conducted in both Tonga (case study six) and Nauru to assess the quality of CoD

data derived from certification, with data adjusted for systematic reporting biases

based on study findings and applied to the overall mortality envelope (all-cause

mortality rates) calculated in Chapter five to derive cause-specific mortality rates.

� � �

247

The methodology for medical record reviews has been developed over a number of

investigations and case studies in the Philippines, Thailand and Australia (269, 341-

344). Reported methods had to be substantially modified for application in the Pacific

Islands (for both the Tonga and Nauru case studies) for several reasons.

Previous medical record reviews have primarily relied on a doctor (or panel of

doctors) independent from the system being reviewed; in that they have not worked

within the system under review, treated patients or family members of patients who

may appear in the cases to review or otherwise have an interest in running the

health services that may have serviced these patients. It is however helpful if

reviewing doctors are familiar with local terms and practices that may appear in the

notes. In a larger country, it is possible to identify independent doctors who are

familiar with local practices. In the PICTs, the small population and close medical

community mean a medical practitioner available on island could not be reasonably

expected to be independent from the system being reviewed.

In order to limit biases in the study, rather than reviewing the original medical chart

directly, key details from each case were extracted by a trained nurse or medical

records officer using a standardised form and de-identified. The form was then

reviewed by a Pacific doctor in Australia who was familiar with the context in which

the cases occurred, but who had been absent from the local islands health setting for

a number of years, and was therefore unlikely to have had contact with any of the

cases under review. Although significantly adapted, the extraction form was

modelled after a similar concept employed for the verbal autopsy study in the

Philippines which used a standard form to extract information from the medical

� � �

248

record to compare with the verbal autopsy information (341). Data extractors

received training in completing the form based on information provided in the

medical record, without adding their own interpretations to these data, with regular

reviews of the completed data extraction forms conducted throughout the studies. An

example of the data form and instructions for interviewers developed for the Tonga

review is provided in Appendix five.

Previous studies have also started from the assumption that the medical records are

substantially correct and well kept; they are in fact the “standard” against which the

diagnosis on the medical certificate should be judged (269, 341-344). However, it is

widely known that many of the health systems in the PICTs have poorly kept medical

records, with notes and results frequently missing or poorly recorded in the patients

chart, or charts missing altogether. The medical records in this study were seen as a

separate source of CoD data which could be compared to the death certification

data, but they could not automatically be assumed to be more accurate than the

certificate Results were therefore seen as a measure more of the uncertainty and

possible variation in the proportion of deaths attributed to each major cause, rather

than a straightforward measurement of error.

� � �

249

6.2.1 Case study 6: Medical Record Review in Tonga

Carter K, Hufanga S, Rao C, Akauola S, Lopez AD, Rampatige R, Taylor R. Causes

of death in Tonga: quality of certification and implications for statistics. Population

Health Metrics, 2012 Mar 5;10(1):4.

The system assessment in Tonga identified contributory causes of death may be

moved into the causal sequence of Part I of the medical certificate during data entry

into the HIS; and CoD tabulations, when performed, are generated from the

immediate CoD (Line one, Part I). There is also a large proportion of deaths coded

as ‘unknown’ or ‘ill-defined’ causes. This paper demonstrates the impact of poor data

management procedures on the final CoD tabulations, leading to potentially

misleading information on CoD distributions being provided to health planners.

Unit record data from the medical certificate was obtained as coded data only,

therefore neither a certificate review or coding audit could be completed. A review of

the plausibility of the causal sequence (as entered in the database) was undertaken,

and a medical record review conducted to examine the accuracy of the diagnosis

and certification.

As contributory causes of death could not be adequately separated from the CoD

sequence, cause-specific rates were calculated to provide a range of estimates

based on both the inclusion and exclusion of these causes from Part I of the medical

certificate. This sensitivity analysis primarily affected the distribution between

diabetes and cardiovascular related deaths, however both showed a significant

� � �

250

increase over 2001-2004 to 2005-2008, regardless of any effect from these

tabulation practices.

As noted in the paper, and in Figure 29, NCDs (group II diseases based on GBD

categories) are the leading CoD in adults over 35 years of age. Table 4 in the paper

shows age-standardised mortality by cause, clearly demonstrating a significant

increase in a range of NCDs between 2001-2004 and 2005-2009, independent of

any changes in age structure, therefore reflecting a genuine increase in premature

mortality from these causes.

Figure 29: Age-distribution of deaths in Tonga by global burden of disease* category; 2005-2009

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0�4 5�14 15�2425�3435�4445�5455�6465�7475�84 85+

Proportion�of�deaths�(%)

Age�Group

I

II

III


Causes of death in Tonga: quality of certificationand implications for statisticsKaren Carter1*, Sione Hufanga2, Chalapati Rao1, Sione Akauola3, Alan D Lopez1, Rasika Rampatige1 andRichard Taylor1,4

Abstract

Background: Detailed cause of death data by age group and sex are critical to identify key public health issuesand target interventions appropriately. In this study the quality of local routinely collected cause of death datafrom medical certification is reviewed, and a cause of death profile for Tonga based on amended data ispresented.

Methods: Medical certificates of death for all deaths in Tonga for 2001 to 2008 and medical records for all deathsin the main island Tongatapu for 2008 were sought from the national hospital. Cause of death data for 2008 werereviewed for quality through (a) a review of current tabulation procedures and (b) a medical record review. Datafrom each medical record were extracted and provided to an independent medical doctor to assign cause ofdeath, with underlying cause from the medical record tabulated against underlying cause from the medicalcertificate. Significant associations in reporting patterns were evaluated and final cause of death for each case in2008 was assigned based on the best quality information from the medical certificate or medical record. Cause ofdeath data from 2001 to 2007 were revised based on findings from the evaluation of certification of the 2008 dataand added to the dataset. Proportional mortality was calculated and applied to age- and sex-specific mortality forall causes from 2001 to 2008. Cause of death was tabulated by age group and sex, and age-standardized (all ages)mortality rates for each sex by cause were calculated.

Results: Reported tabulations of cause of death in Tonga are of immediate cause, with ischemic heart disease anddiabetes underrepresented. In the majority of cases the reported (immediate) cause fell within the same broadcategory as the underlying cause of death from the medical certificate. Underlying cause of death from themedical certificate, attributed to neoplasms, diabetes, and cardiovascular disease were assigned to other underlyingcauses by the medical record review in 70% to 77% of deaths. Of the 28 (6.5%) deaths attributed to nonspecific orunknown causes on the medical certificate, 17 were able to be attributed elsewhere following review of themedical record. Final cause of death tabulations for 2001 to 2008 demonstrate that noncommunicable diseases areleading adult mortality, and age-standardized rates for cardiovascular diseases, neoplasms, and diabetes increasedsignificantly between 2001 to 2004 and 2005 to 2008. Cause of death data for 2001 to 2008 show increasingcause-specific mortality (deaths per 100,000) from 2001-2004 to 2005-2008 from cardiovascular (194-382 to 423-644in 2005-2008 for males and 108-227 to 194-321 for females) and other noncommunicable diseases that cannot beaccounted for by changes in the age structure of the population. Mortality from diabetes for 2005 to 2008 isestimated at 94 to 222 deaths per 100,000 population for males and 98 to 190 for females (based on the range ofplausible all-cause mortality estimates) compared with 2008 estimates from the global burden of disease study of40 (males) and 53 (females) deaths per 100,000 population.

Discussion: Certification of death was generally found to be the most reliable data on cause of death in Tongaavailable for Tonga, with 93% of the final assigned causes following review of the 2008 data matching those listed

* Correspondence: [email protected] of Population Health, University of Queensland, Herston (Brisbane),Queensland, AustraliaFull list of author information is available at the end of the article


© 2012 Carter et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative CommonsAttribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction inany medium, provided the original work is properly cited.



on the medical certificate of death. Cause of death data available in Tonga can be improved by routinelytabulating data by underlying cause and ensuring contributory causes are not recorded in Part I of the certificateduring data entry to the database. There is significantly more data on cause of death available in Tonga than areroutinely reported or known to international agencies.

Keywords: Mortality, Cause of death, Noncommunicable Diseases, Medical record review, Death Certification,Tonga, Pacific Islands

IntroductionCause of death data by age group and sex is critical infor-mation both to identify key public health issues contribut-ing to premature mortality and to design interventionsthat are appropriately targeted. Several studies have pre-viously noted that there is an important gap in availabilityof data on mortality and cause of death patterns from thePacific Islands [1,2], and the World Health Organization(WHO) mortality database currently lists Tonga as havingno cause of death data available [3].Tonga is a Pacific Island country with an estimated

population of 103,000 (2009) [4], of which 38% are under15 years of age [4]. Seventy percent of the populationreside on the main island of Tongatapu [4]. Recent (2005to 2009) estimates of adult mortality (probability of dyingbetween ages 15 and 59 years inclusive) based on recon-ciled empirical data from the Ministry of Health (MoH)and Civil Registration office, are nearly three times thatfor Australia and New Zealand [5]. Estimated life expec-tancy (LE) from these data is between 60.4 to 64.2 yearsfor males and 65.4 to 69.0 years for females [6] whenassessed for underenumeration, several years lower thanpreviously reported [5-7]. Given the high level of prema-ture adult mortality and the impact on LE, it is criticalthat health planners have access to accurate informationon causes of death.Health services in Tonga are provided through a

national hospital, three outerisland hospitals, 11 repro-ductive health centers, and 15 community health centers[8]. Prior to 2007, medical certification of death was con-ducted by the attending doctor for deaths occurring athealth facilities or by the treating doctor for deaths incommunity if requested by the family. In January 2007,the Tongan MoH introduced a new policy on reportingdeaths as part of a comprehensive effort to improve vitalregistration [9]. This policy requires all deaths to be certi-fied by a medical practitioner. Tonga uses a medical cer-tificate of death (hereafter referred to as the medicalcertificate) that is consistent with the international stan-dard medical certificate of death from the 10th version ofthe International Classification of Diseases (ICD-10) [10].This lists the cause of death sequence starting with theimmediate cause of death (the condition that directlyresulted in the death) and working backwards in Part I ofthe certificate and contributory causes (conditions

present at the time of death, but which did not directlycause the death) in Part II (Additional file 1).In accordance with ICD-10, underlying cause of death

(the condition that started the chain of events thatresulted in the death) is generally selected by workingbackwards from the immediate cause to the cause on thelowest line in Part I, which is able to directly cause thecondition above it. If the cause listed on the lower line isnot a direct cause of the condition listed above (forexample if cancer was listed below ischemic heart dis-ease), the lower line is not included in the causalsequence and the line above is selected as the underlyingcause (ischemic heart disease in the previous example).Modification and selection rules also apply for certainconditions [10]. In Tonga, data from the medical certifi-cates are currently coded line by line (without selectingunderlying cause) by medical records staff who havecompleted a recognized training course in ICD codingand collated in an electronic database. Tabulations aregenerated from this database as required, and until thisstudy, were based on the immediate cause of death (line1 of Part I of the certificate). The Country Health Infor-mation Profile, developed from local data and publishedby the Western Pacific Regional Office of WHO, listsonly five leading causes of death for Tonga (based on2007 data) reported for all ages and both sexes combined.These are: 1) Diseases of the circulatory system; 2) Neo-plasms; 3) Symptoms, signs, and ill-defined conditions;4) Diseases of the respiratory system; and 5) Certaininfectious and parasitic diseases [8].This paper examines the accuracy of the locally gener-

ated cause of death data for Tonga for 2008 as collectedthrough the medical certificates. Cause of death asrecorded on the certificate is compared to informationcontained in the patients’ medical record to assessreported patterns of cause of death and to identify (andcorrect) any systematic misclassifications that may affectselection of underlying cause of death or analysis oftrends. Findings from this assessment were applied tocause of death data from the medical certificates of deathfor 2001 to 2007, which were then added to the dataset.Proportional mortality by age and sex was calculated andapplied to age-specific all-cause mortality for 2001 to2008 [5] to produce age-, sex- and cause-specific mortal-ity for Tonga and assess trends over time.


Page 2 of 15

MethodsAn overview of the study approach is given in Figure 1.

Quality review of cause of death data for 2008Data from the medical certificates were sought from theMoH for all reported deaths in Tonga for 2008 (n = 945).Cause of death data from the certificate (both Part I andPart II) were entered into the database by MoH medicalrecord staff as the ICD-10 code only [10]. During dataentry for some deaths, contributory causes listed in Part IIof the original medical certificate were entered onto thefirst available line in the database sequence in Part I. Thiscould erroneously lead to their consideration within thecausal sequence recorded in the database, resulting in thepossibility that such a contributory cause may be selectedas the underlying cause (according to the General Princi-ple for mortality coding in ICD-10), if it could plausiblylead to the condition listed above it.The quality of population cause of death data obtained

through medical certification is directly affected by threestages in the data collection process: original diagnosis andcertification of the death, coding of the data from the cer-tificate, and the data entry and tabulation practices used to

produce aggregate data. As data from the certificates ofdeath were obtained as an extraction from the MoH data-base (with cause of death available only as ICD codes), cer-tification and coding could not be examined separately.Deaths for 2008 were therefore examined using: (a) areview of the data entry and tabulation practices; (b) amedical record review; and (c) assignment of a final under-lying cause from either the medical certificate of death ormedical record based on the best available information.Review of tabulation practicesAn underlying cause of death was assigned for each deathbased on ICD-10 [10]. When the unit record listed morethan one cause, the underlying cause of death wasassigned assuming no movement of contributory causesinto Part I of the certificate. Assigned underlying causeswere then reviewed to assess whether the selection ofunderlying cause would differ if it was assumed that con-tributory causes had been recorded in empty lines forPart I of the certificate when entered in the database.Coded data for both the immediate and underlying causeof death from the medical certificate of death were aggre-gated to one of the 103 causes listed in the GeneralMortality Tabulation List 1 of ICD-10 (hereafter referred

Figure 1 Overview of study design.


Page 3 of 15

ICD list 1) [10]. These were crosstabulated to comparethe effect of current MoH tabulation practices (based onimmediate cause) on the cause of death profile (Table 1).Medical record reviewMedical records were sought from the national hospitalfor all deaths that occurred in 2008 in Tongatapu whereproximity to the major hospital suggests that deceasedindividuals were more likely to have received care at thehospital and therefore have a medical record. Data fromthe medical record for each death were extracted using astandardized form adapted from similar work in the Phi-lippines [11,12], de-identified, and assigned a study num-ber. Data extraction was completed by qualified nurseswho underwent initial training in data extraction proce-dures based on the written study protocol. Extractionforms were regularly reviewed throughout the project [byKC].An independent medical doctor was provided with

medical record extracts and requested to complete amedical record review form for each death based on theinternational standard medical certificate of death [13].Doctors were provided with both written and verbalinstruction on completing the medical record reviewform based on ICD-10 guidelines [10]. Where age atdeath was less than 60 years and cause of death could notbe determined, the record extracts were reviewed by asecond independent medical doctor. For each cause ofdeath assigned, the doctor was requested to indicate thelevel of certainty of the diagnosis (definite, strong cer-tainty, some certainty, and limited certainty) based onthe available evidence [12]. Assigned causes were subse-quently grouped into two diagnostic evidence categories:high (definite and strong) and low (some and limited)[12].Underlying cause of death from the medical record

review form was then coded according to ICD-10 [10] andmatched to the medical certificate of death by study num-ber. Underlying cause of death as noted on both the certi-ficate of death and the medical record review form wascrosstabulated using ICD list 1 [10]. Differences betweenthe two underlying causes were classified as either minor(different cause as tabulated using ICD list 1, but sameICD chapter) or major (different ICD chapter), and signifi-cant associations in reporting patterns were examined.Assignment of final underlying cause of deathPrevious medical record review studies [14-16] havestarted from the presumption of a reference standard,where one source (the medical record) is assumed to bemore accurate than the other (the medical certificate).However, such an assumption was not valid in this study.A final underlying cause of death was assigned for eachdeath based on the best available data, whether from themedical certificate or the medical record. In general, the

medical certificate-based underlying cause of death wasselected, except in the following instances:a) where cause on the medical certificate was

unknown or coded as nonspecificb) an injury leading to the death was noted in the

medical record, but not mentioned on the deathcertificatec) the underlying cause identified from the medical

record could have led to the underlying cause on themedical certificate of deathd) for all remaining deaths not due to external causes,

the underlying cause from the medical record was con-sidered to be of high certaintyProportional mortality by cause by age and sex was

then tabulated and 95% confidence intervals were calcu-lated using the Poisson distribution.

Cause of death tabulations, 2001 to 2008Based on the assessment of deaths in 2008, it was deter-mined that medical records added little to our understand-ing of underlying cause of death at a population level,particularly for adult deaths. For 2008, 93% of the finalassigned causes of death were consistent with the medicalcertificate of death (see results) following the medicalrecord review. Based on this finding, de-identified datafrom the medical certificates for all deaths for 2001 to2007 were added to the 2008 data set. As immediate causediffered from the underlying cause (ICD list 1) in 37% ofcases for 2008 (see results), each death (2001 to 2007) wasreviewed and assigned an underlying cause of death basedon ICD-10 [10], by either assuming no movement of con-tributory causes into Part I of the certificate or assumingthat contributory causes had been recorded in empty linesfor Part I, as completed for the 2008 data. This resulted intwo separate underlying causes, and tabulations by agegroup and sex were generated for both variables. Propor-tional mortality for both tabulations was calculated (withexclusion of ill-defined and unknown deaths) for 2001-2004 and 2005-2008 and applied to estimates of totaldeaths for Tonga to calculate mortality rates by age, sex,and selected causes (ICD list 1). Total deaths for 2001-2004 and 2005-2009 (the closest available period to 2001-2008) were extracted directly from life tables generated bya recent study that established upper and lower mortalityscenarios based on adjusted empirical data reconciledfrom both MoH and Civil Registry sources [5]. Bothunderlying cause of death tabulations described abovewere applied to the upper and lower mortality scenariosfrom this recent study [5], generating four estimates fortotal deaths by cause by age group and sex for each period.The upper and lower values from these four estimateswere used to generate a plausible range of deaths andmortality rates by cause.


Page 4 of 15

Table 1 Comparison of original cause of death tabulation based on immediate cause from the medical certificate against underlying cause of death from themedical certificate†: ICD 103 cause list (selected causes), Tonga, 2008Original tabulation Underlying cause of death from medical certificate

1

1-001-1-025*

1-012 1-019 1-026 1-052 1-059 1-064-1-071+

1-065 1-067 1-069 1-072 1-080 1-082 1-084 1-087 1-092 1-093 1-094 1-095 Other Unknown Total

1-001 - 1-025* 3 4

1-012 1 2 2 10 3 2 4 2 1 3 30

1-019 1 1

1-026 1 55 56

1-052 12 12

1-064 -1-071+

2 9 37 2 5 6 1 1 63

1-067 6 28 1 35

1-069 2 1 15 2 20

1-072 3 4 4 5 1 34 51

1-080 5 2 6 13

1-083 1 1 2

1-084 1 2 7 10

1-092 13 13

1-093 3 3

1-094 2 5 1 1 1 1 10 1 3 25

1-095 0

Other 1 2 15 1 2 10 31

Injury~ 24 24

Unknown 1 1 36 38

Total 10 4 9 67 59 4 47 2 35 17 42 6 6 12 1 15 3 10 29 17 36 431

†Determined according to ICD principles [10], excluding potential impact of contributory causes (Part II) in the causal sequence (Part I)

Single causes from ICD list 1 are recorded in italics, disease groups are shown in plain text.

1-001-1025 Infectious and parasitic diseases (other than septicemia and viral hepatitis)

1-012 Septicemia

1-019 Viral hepatitis

1-026 Neoplasms

1-052 Diabetes

1-059 Meningitis

1-064 - 1071 Diseases of the circulatory system (excluding rheumatic, ischemic, and cerebrovascular heart disease)

1-065 Rheumatic heart diseases

1-067 Ischemic heart disease

1-069 Cerebrovascular heart disease


1-080 Diseases of the liver

1-084 Diseases of the genitourinary system

1-087 Pregnancy, childbirth, and the puerperium

1-092 Certain conditions originating in the perinatal period

1-093 Congenital malformations, deformations, and chromosomal abnormalities

1-094 Symptoms, signs, and abnormal clinical and laboratory findings not elsewhere classified

1-095 Injury, poisoning, and certain other consequences of external causes

Other All other causes

Carter

etal.Population

Health

Metrics

2012,10:4http://w

ww.pophealthm

etrics.com/content/10/1/4

Page5of

15

Population by age group and sex was derived from the1996 [17] and 2006 [7] censuses using exponential inter-polation to provide estimates for intermediate and addi-tional years. Age-standardized mortality rates for allages were then calculated using the world-standardpopulation [18] to examine trends over time (2001-2004and 2005-2008).Final cause of death was also classified to the WHO

Global Burden of Disease (GBD) Study categories: I(infectious, perinatal, and maternal conditions), II (non-communicable diseases), and III (external causes) [19].Proportional mortality by GBD category for males wascompared with that derived from Codmod [20] to assessbroad plausibility of the results. Codmod is a modelingprogram that generates expected cause of death distri-butions [20] from inputs of level of childhood and adultmortality and is based on associations between theseparameters as recorded in over 100 years of data from awide range of countries. This was not performed forfemale deaths as existing Codmod models [20] do notallow input of the high level of adult mortality estimatedin females in Tonga concurrently with the low estimatesof childhood mortality [5].

ResultsReported deathsOf 945 deaths recorded in Tonga for 2008 based on recon-ciled MoH and civil registry data [5], 431 deaths had a cer-tificate recorded in the MoH database; 390 of these werefrom Tongatapu and therefore had a medical recordsought for review. As such, certificates were not availablefor 54% of the reported deaths (reconciled from the CivilRegistry, nursing records, and health information system).There were 2,310 deaths extracted from the MoH data-

base (based on medical certification) for 2001 to 2004, ofwhich 1,902 cases included cause of death data. For 2005to 2008, 2,025 cases were extracted from the database, ofwhich 1,885 included cause of death data.

Quality review of deaths in 2008Review of tabulation practicesExamination of case data for 2008 from the certificatesrevealed that in 63% of deaths the cause of death originallytabulated on immediate cause was the same as the finalunderlying cause of death when aggregated to the ICD list1 [10] (Table 1). However, a substantial proportion ofdeaths originally tabulated as septicemia (1-012) (93%),cardiovascular diseases (1-064 to 1-071) (25%), respiratoryillnesses (1-072) (33%), and liver disease (1-078) (54%)should have been attributed to other causes (Table 1)based on ICD rules [10], reflecting various nonspecificcauses, such as heart failure and respiratory arrest, whichoften appear on the first line of the medical certificate ofdeath. Twelve of the 17 deaths (71%) recorded as “signs,

symptoms or ill-defined conditions” should have beenattributed to more specific causes (Table 1), even with nofurther information available from the medical record.Diabetes (1-052) was notably underrepresented in the

original tabulation, with 47 of 59 (80%) deaths for whichdiabetes was selected as the underlying cause originallyassigned to other causes, notably septicemia and cardio-vascular disease. Neoplasms (1-026) were also underrepre-sented in the original tabulations with 12 (18%) cancerdeaths assigned to other causes. Although they accountedfor only a small number of deaths, some deaths attributedon the medical certificate of death to underlying causes ofparticular public health importance, such as rheumaticheart diseases and maternal causes, were not apparent inthe original tabulation of immediate cause.Upon review, 36 (8%) deaths for 2008 were found to be

potentially affected by the entry of contributory causesfrom Part II of the certificate into the Part I sequence. Inall other deaths application of the ICD rules would resultin the same underlying cause (when collated to ICD list 1)[10]. Only one of these deaths involved a death assigned toInfectious Diseases, which would have otherwise beencoded to pneumonia (Respiratory Disease); all other deathsoccurred amongst those assigned to noncommunicablediseases - both in the original tabulation of immediatecause and selected underlying cause of death. Four ofthese deaths were minor variations within categories ofcardiovascular disease, while 23 cases were diabetes-related deaths that would be assigned to cardiovascular orrespiratory diseases if the diabetes code had been recordedin Part II of the certificate (as a contributory cause).

Medical record reviewMedical records were located for 294 (75%) certifieddeaths that occurred in Tongatapu (Figure 2) followingexclusion of duplicate records and overseas deaths. Causeof death was identified from the medical record for 190cases, 65% of those with a matching medical record. Diag-nosis of underlying cause of death was considered to be ofhigh certainty by the medical reviewer in only 73 (25%) ofthese records.The sex and age distribution of deaths for which records

could not be located was similar to that for the deathswhere a medical record was found, except for older ages(> 64 years), which had a higher proportion of missingrecords. Twenty-eight percent of deaths missing a medicalrecord occurred at the national hospital. Total reporteddeaths followed a similar age and sex distribution as thedeaths for which a record was located.There were 93 major discrepancies (32% of deaths)

and 19 minor discrepancies (6% of deaths) identifiedbetween the medical death certificate and medicalrecord review, following collation to ICD list 1 andtabulation according to underlying cause of death. If the


Page 6 of 15

dataset was limited to deaths with a diagnosis of highcertainty, there were 18 major discrepancies (24% ofdeaths) and four minor discrepancies (5% of deaths). Asdetermined from the medical certificate, 70% of deathsattributed to neoplasms, 77% of deaths attributed to dia-betes, and 70% of deaths attributed to cardiovasculardisease were assigned to other underlying causes by themedical record review (Table 2). The most commonunderlying causes of death found by the medical recordreview to have been originally certified as due to othercauses (Table 2) were infectious diseases, in particularsepticemia (100% of deaths), cerebrovascular diseases(65% of deaths), and respiratory diseases (71% ofdeaths). If redistribution of deaths is limited to deathswith a diagnosis of high certainty, nearly all the deathswould be assigned to the same broad ICD category byboth the certificate and medical record review.Of the 28 deaths attributed to nonspecific or unknown

causes on the medical certificate, 17 (61%) were able tobe attributed elsewhere following review of the medical

record. Eight (29%) of these unknown cases occurred inchildren less than 5 years of age and were assigned toperinatal or congenital causes following the medicalrecord review. Evidence for this attribution was definiteor strong in each of these cases. Even following medicalrecord review, 11 (6% of all deaths) remained assignedto unknown causes.Assignment of final underlying cause of deathFor 208 deaths (71%) for which a medical record wasfound, the underlying cause of death as determinedfrom the medical record review did not match underly-ing cause as certified. The medical certificate of deathwas used in 179 (86%) and the medical record in 29(14%) of these cases to determine final cause. As notedin the methods, the medical record was used ininstances where it added additional information to thecausal sequence indicated on the certificate or where anexternal cause of death was indicated that was notrecorded on the medical certificate. The medical recordswere not considered to be inherently more reliable than

Figure 2 Flowchart of process for medical record review in Tongatapu.


Page 7 of 15

Table 2 Comparison of underlying cause of death from the medical certificate of death† and from the medical record review: ICD 103 cause list (selected causes),Tongatapu, 2008

Underlying cause of death from medical record review

Underlying cause of death from medical certificate 1-001- 1-025* 1-012 1-019 1-026 1-052 1-064- 1-071 1-067 1-069 1-072 1-080 1-082 1-083 1-084 1-092 1-093 1-095 Other Unknown Total

1-001 - 1-025* 5 1 1 7

1-012 1 1 1 3

1-019 2 3 2 1 8

1-026 1 15 1 2 1 1 1 2 4 23 50

1-052 1 1 10 1 1 8 1 1 3 1 1 14 43

1-059 1 1 1 1 4

1-064 - 1-071+ 2 4 4 1 1 2 18 32

1-065 1 1 2

1-067 1 1 6 1 1 10 20

1-069 1 1 6 1 3 12

1-072 2 1 1 5 10 19

1-080 1 1 2

1-082 4 1 5

1-084 1 1 4 6

1-087 1 1

1-092 1 11 1 1 14

1-093 3 3

1-094 1 1 3 5

1-095 1 1 11 6 19

Other 2 1 1 9 2 16

Unknown 1 3 1 7 3 8 23

Total 12 10 4 20 11 7 11 17 17 4 9 2 4 19 8 17 18 104 294

†Determined according to ICD principles [10], excluding potential impact of contributory causes in the causal sequence

Single causes from the ICD list 1 are recorded in italics, disease groups are shown in plain text.

1-001-1025 Infectious and parasitic diseases (other than septicemia and viral hepatitis)

1-012 Septicemia

1-019 Viral hepatitis

1-026 Neoplasms

1-052 Diabetes

1-059 Meningitis

1-064 - 1071 Diseases of the circulatory system (excluding rheumatic, ischemic, and cerebrovascular heart disease)

1-065 Rheumatic heart diseases

1-067 Ischemic heart disease

1-069 Cerebrovascular heart disease


1-080 Diseases of the liver


1-087 Pregnancy, childbirth, and the puerperium

1-092 Certain conditions originating in the perinatal period

1-093 Congenital malformations, deformations, and chromosomal abnormalities

1-094 Symptoms, signs, and abnormal clinical and laboratory findings not elsewhere classified

1-095 Injury, poisoning, and certain other consequences of external causes

Other All other causes

Carter

etal.Population

Health

Metrics

2012,10:4http://w

ww.pophealthm

etrics.com/content/10/1/4

Page8of

15

the death certificate, as has been assumed in other stu-dies of this type, because of the large proportion ofincomplete or missing records, the most recent entry inthe record for many cases preceding the death by manyyears or months, and the poor legibility and completionof records. An overview of the sources of final cause ofdeath data and reasons for selection is shown in Addi-tional file 2; 93% of the final causes assigned matchedthe underlying cause given by the medical certificate.Perinatal conditions were the leading cause of death in

children (Table 3). Neoplasms were the leading cause ofdeath in adult females, with cardiovascular diseases theleading cause in adult males (Table 4). Diabetes was thesecond leading cause of death in adult males (19%) andthe third leading cause of death in adult females (18%)(Table 4) based on ICD-10, where diabetes is taken as anunderlying cause if stated in Part I of the certificate. Half(50%) of all cardiovascular disease was found to be due toischemic heart disease. Viral hepatitis (4.5% of total adultdeaths) was the most frequently reported infectious dis-ease causing mortality in adults.

Causes of death, 2001 to 2008Cause of death data for 2001 to 2008 were compiled fromthe best available data on underlying cause from either themedical certificate of death or medical record for deaths in2008, and de-identified medical certificate data for 2001 to2007. Infectious diseases and neonatal causes were theleading causes of death in 0 to 4 year olds, with injury andexternal causes the leading cause of death for 15 to 24year olds. From 25 years of age onwards, proportionalmortality from injury and external causes decline withchronic diseases leading proportional mortality for all agegroups above 25 years. Age-standardized rates (all ages)for selected diseases are shown in Figures 3 and 4 (Addi-tional file 2), and indicate a significant increase in mortal-ity rates (for males and females) between 2001-2004 and2005-2008 for a range of noncommunicable diseases,

despite the uncertainty in the estimates described in themethods. For 2005 to 2008, the plausible ranges for non-communicable disease death rates (per 100,000) were car-diovascular diseases (males: 423 - 644 and females: 194 -321), diabetes (males: 94 - 222 and females: 98 - 190), andlung cancer (males: 34 - 49 and females: 17 - 23). Therewas a notable increase in age-standardized mortality dueto intentional self harm in males, with the age-standar-dized rate estimated as 9 to 12 deaths per 100,000 in 2005to 2008.The overall cause distribution by GBD category from

the empirical data for Tongan males for both 2001-2004and 2005-2008 was very similar to the modeled distribu-tion produced by Codmod [20], which reflects the antici-pated cause distribution based on mortality levels asderived from a wide range of previous mortality data sets.However, the Tongan empirical data shows a higher pro-portion of deaths from noncommunicable diseases inearly adult ages than the modeled data, which is consis-tent with the low life expectancy for males in Tonga overthis period. As noted in the methods, Codmod distribu-tions could not be derived for female deaths due to thehigh adult mortality.

DiscussionCertificates were not available for 54% of the reporteddeaths in 2008 despite the current MoH policy [9] requir-ing certification of all deaths. Current tabulation prac-tices based on the immediate cause of death havecontributed to a lack of reliable evidence for decision-making in Tonga. Causes of death such as ischemic heartdisease and diabetes were underrepresented in these ori-ginal tabulations. The tabulated cause was found to fallwithin the same cause aggregated to ICD list 1 as theunderlying cause of death in 63% of instancesDespite data entry procedures potentially moving con-

tributory causes into the causal sequence, the overalleffect of this on selection of underlying cause is likely to

Table 3 Underlying causes of death in Children 0-4 years, Tonga 2008

Cause of death* Reported deaths Proportional mortality (%)

1-092 Perinatal conditions 21 36

1-001 Infectious diseases 8 14

1-093 Congenital abnormalities 6 10

1-072 Respiratory diseases 6 10

1-026 Neoplasms 5 9

1-095 External Causes 4 7

1-059 Meningitis 3 5

1-078 Disease of digestive system 2 3

Other 3 5

* Tabulated according to ICD list 1 [10]. “Infectious diseases” in the table refers to those conditions classified according to chapter 1 of the ICD. Meningitis iscoded as a disease of the nervous system under the ICD [10] and is therefore listed separately, while pneumonia and other infectious respiratory illnesses areincluded in the respiratory diseases chapter


Page 9 of 15

be minor, other than for diabetes, if ICD selection rulesare rigorously applied. The general principle of ICDrules require selection of the lowest completed line onPart I of the certificate as the underlying cause of deathonly if the condition “could have given rise to all theconditions entered above it” [10], thereby excluding thecontributory cause in most cases even if recorded in thecausal sequence of Part I of the certificate. However,while the ICD rules allow cardiovascular conditions tobe considered as a consequence of diabetes, the actualrecording of diabetes in Part 1 or Part 2 in such cases isdriven by individual certifier preferences rather than aninternationally recognized causal relationship betweendiabetes and cardiovascular disease. This affects compar-ability of underlying cause statistics between populations

and across time. In this study, the process of generatingtwo tabulations based on different assumptions of howcontributory causes were treated (as described in themethods) in effect also reflects this potential ambiguityin the results [21]. The increases in cause-specific mor-tality from 2001-2004 to 2005-2008 from noncommu-nicable diseases, including cardiovascular diseases anddiabetes, are significant, with the range of estimatesfrom each period not overlapping, indicating thatincreases in cause-specific mortality from diabetes arenot an artifact of changes in coding conventions. Non-communicable diseases are clearly leading the high ratesof adult mortality, although injury and external causesremain the leading cause of death in 15- to 24-year-olds. Current tabulations of data in Tonga code external

Table 4 Underlying causes of death in adults aged 15-64 years, Tonga, 2008

Cause of death* Males Females

Reporteddeaths

Proportionalmortality (%)

Reporteddeaths

Proportionalmortality (%)

1-064 Diseases of the circulatory system 28 28 16 23

1-067 Ischemic heart diseases 18 18 4 6

1-069 Cerebrovascular diseases 1 1 4 6

1-065 Acute rheumatic fever and chronic rheumatic heart diseases 1 1 1 1

1-051 Endocrine, nutritional, and metabolic diseases (all diabetes(1-052))

19 19 13 18

1-095 External causes of morbidity and mortality 18 18 3 4

1-096 Transport accidents 6 6 0 0

1-101 Intentional self-harm 4 4 0 0

1-026 Neoplasms 15 15 20 28

1-036 Malignant neoplasm of breast 0 3 4

1-037 Malignant neoplasm of cervix uteri 0 2 3

1-039 Malignant neoplasm of ovary 0 2 3

1-031 Malignant neoplasm of liver and intrahepatic bile ducts 3 3 1 1

1-034 Malignant neoplasm of trachea, bronchus, and lung 3 3 0

1-029 Malignant neoplasm of stomach 2 2 2 3

1-030 Malignant neoplasm of colon, rectum, and anus 2 2 2 3

1-001 Certain infectious and parasitic diseases 10 10 3 4

1-019 Viral hepatitis 6 6 2 3

1-072 Diseases of the respiratory system 4 4 3 4

1-078 Diseases of the digestive system 3 3 3 4

1-084 Diseases of the genitourinary system 2 2 3 4

1-094 Symptoms, signs, and abnormal clinical and laboratoryfindings, not elsewhere classified

1 1 1 1

1-058 Diseases of the nervous system 0 0 3 4

1-082 Diseases of the skin and subcutaneous tissue 0 0 2 3

1-087 Pregnancy, childbirth, and the puerperium 0 1 1

Other categories 1 1 0 0

TOTAL 101 100 71 100

* Tabulated according to ICD list 1 [10]. The table shows cause categories by ICD chapter in bold, with important individual cause categories from ICD list 1shown in small print under their assigned ICD chapter


Page 10 of 15

cause by injury type rather than cause (such as fall,motor vehicle accident, etc.), and there was insufficientdetail in many cases (both on the certificate or medicalrecord) to assign a more specific cause category. Injurymortality data are thus currently of limited use to targetpublic health interventions, and doctors should beencouraged to record more detail for these events onthe medical certificate of death.The data for 2008 indicate 93% of the final assigned

causes matched those listed on the medical certificate ofdeath (ICD list 1). There were, however, some clear pat-terns of misreporting on the certificate, primarily for chil-dren, where additional information from the record couldhave led to a more specific diagnosis. Inconsistencieswere also noted in the reporting of diabetes and cardio-vascular disease, with these conditions being the mostlikely underlying causes from the medical certificate tobe assigned to other causes by the medical record review.This demonstrates the importance of certifying doctorsreferring to the medical record, particularly in cases dueto chronic disease, or where death is attributed to an ill-defined or unknown cause. There is also a critical needto improve the quality of medical records and availability

of this information to the certifying doctor. Even follow-ing review, 6% of deaths remained assigned to unknowncauses. In part, this may reflect the community atten-dance (or lack thereof) at the hospital and subsequentlylimited opportunity for data collection. However, 28% ofthe deaths for which no medical record could be locateddied at the hospital and therefore should have had a med-ical record available. In part these missing records maybe due to cases that were “dead on arrival” at the hospitaland therefore certified without a record being generated.For the remainder, as the study was conducted somemonths after the conclusion of 2008, these missingrecords cannot be accounted for by delays in transfer ofthe record from the ward to the medical records office.This highlights a critical need to improve data manage-ment procedures to ensure that data are not lost oncecollected and that all records can be accessed centrally.The medical record review of deaths that occurred in

2008 was conducted in accordance with the principles setout by Johanssen et al. [14] to ensure the transparencyand reproducibility of medical record review studies.However, this work varied from previous applications ofthis technique as the small population and health service

Figure 3 Age-standardized mortality for males (selected causes) for Tonga, 2001-2009 (plausible range based on upper and lowermortality scenarios). Age-standardized mortality rates for all ages were calculated using the world-standard population [18].


Page 11 of 15

arrangements meant that an independent review in-country was not possible, thus introducing an additionalstep of extracting data from the medical record ratherthan having a physician review the record directly. Pre-vious studies have also started from the presumption of areference standard, where one source is assumed to bemore accurate than the other (i.e., the certificate whenreviewing verbal autopsy findings or the medical recordwhen reviewing the medical certificate), which was notappropriate in this setting. In order to ensure the integ-rity of the study given these design differences, dataextractions from the medical record were guided by astandardized form and instruction manual [12], extrac-tions were completed by nurses with a good understand-ing of medical terminology, and forms were routinelyreviewed by the project team for quality. Causes of deathwere coded according to ICD rules [10] and collated tothe recommended tabulations in the ICD [10]. Theapproach used in this study allowed the flexibility toensure the most reliable data from either the medicalrecord or medical certificate of death were used to assignunderlying cause. This was an important considerationgiven the unknown quality of cause of death data in boththe medical certificates of death and medical records at

the outset of the study. A complete review of informationin the medical record, rather than the principal diagnosisas recorded on the discharge summary was used for com-parison to the medical certificate, as the principal diagno-sis is the reason for the admission and is frequently amanifestation of an underlying cause that may not berecorded. For example, a patient admitted for congestiveheart failure and who subsequently dies would have thisrecorded as the principal diagnosis rather than theunderlying cardiopathy that led to this condition. Ascause of death data were obtained as ICD codes it wasnot possible to differentiate whether errors were intro-duced during original certification or coding, and afurther coding audit may also prove useful.There is very little information currently published on

causes of death in Tonga. Of the published data available,the WHO Child Health Epidemiology Reference Groupindicates similar estimates of proportional mortality for2008 in children under 5 years of age from external causes(7%) but higher mortality from respiratory diseases (pneu-monia) (17%) and congenital abnormalities (17%) com-pared to this study [22]. In adults, while significantuncertainty remains due to both previous data entry prac-tices and the small number of deaths in the study, the

Figure 4 Age-standardized mortality for females (selected causes) for Tonga, 2001-2009 (plausible range based on upper and lowermortality scenarios). Age-standardized mortality rates for all ages were then calculated using the world-standard population [18].


Page 12 of 15

findings demonstrate that diabetes is a much greater pub-lic health issue in Tonga than previously reported. Age-standardized mortality in Tonga from diabetes for 2005 to2008 is estimated at 94 to 222 deaths per 100,000 formales and 98 to 190 for females. This is significantlyhigher than the estimated age-standardized rates of 16(males) and 11 (females) deaths per 100,000 populationrespectively for New Zealand (2008) [23] and previousGBD estimates for Tonga (2008) of 40 (males) and 53(females) deaths per 100,000 [19], although the latter mayhave been based on tabulations of immediate cause. Thehigh estimates of cause-specific mortality from diabetes inTonga are, however, broadly consistent with previous stu-dies for New Zealand’s Maori, a group that also demon-strates high mortality from diabetes at a lower prevalenceof disease. A 2004 survey estimated prevalence of diabetesin the Tongan population at 18% (based on self-report)[24], while prevalence (by self-report) in the Maori popula-tion was estimated at 8% for males and 12% for females in2002 [25]. Age-standardized mortality from diabetes forMaori males was estimated at 61 deaths per 100,000 and39 deaths per 100,000 for females in 2004 [25]. Whilelower than the estimates of age-standardized mortalityfrom diabetes in Tonga, the Maori figures are based onthe older WHO standard population (the Segi age distri-bution) [18] and would be substantially higher if thenewer WHO world standard population (which has ahigher proportion of the population in the adult agegroups) had been used, as applied in this study.A study of diabetic patients in New Zealand also

found that Tongan patients tended to have poorer com-pliance with treatment and a more fatalistic attitude totheir illness [26]. This may also be a factor in the highmortality rates seen in Tonga and is consistent with thehigh proportion of medical record extracts collected inthis study in which the case’s diabetes was referred to as“uncontrolled” and “unmanaged.”Age-standardized death rates for ischemic heart disease

and neoplasms were substantially higher in males inTonga compared to Australia and New Zealand. For Ton-gan males, age-standardized mortality from ischemic heartdisease is estimated at 139 to 210 deaths per 100,000(2005-2008), compared with 97 in New Zealand [23] and78 in Australia [19] in 2008. Age-standardized mortality inmales from neoplasms is estimated at 189 to 255 deathsper 100,000 (2005-2008) in Tonga compared with 155deaths per 100,000 in New Zealand and 146 in Australiain 2008 [19]. Age-standardized mortality from ischemicheart disease and neoplasms for females (2005-2008) wasvery similar to estimates reported for New Zealand [23]and slightly higher than those for Australia in 2008 [19].The higher age-standardized rates for cardiovascular dis-ease and neoplasms would contribute to the relatively lowlife expectancy for males in Tonga (60.4 to 64.2 years,

2005-2009) [5], highlighting the critical importance ofaddressing chronic diseases in Tonga to improve lifeexpectancy. Age-standardized rates for ischemic heart dis-ease remain below those seen in Australia (472 deaths per100,000 for males and 246 deaths per 100,000 for femalesat the height of the cardiovascular disease epidemic(1968)) [27].The high age-standardized mortality rates from cardio-

vascular diseases are consistent with the limited informa-tion on noncommunicable disease risk factors available forTonga. A 2004 survey found that 67% of adults wereobese (body mass index ≥ 30 kg/m2), with 23% of adultsaffected by hypertension (systolic blood pressure ≥ 140mmHg and/or diastolic blood pressure ≥ 90 mmHg orcurrently on medication) [24]. The higher age-standar-dized rate of lung cancer in males of 34.4 to 49.0 deathsper 100,000 compared with 16.5 to 23.0 for females (2005-2009) reflects significantly higher smoking prevalenceamongst males (64%) than females (14%), as recorded in apopulation-based survey undertaken in 1992 [28]. Theserates of lung cancer are substantially higher than previousestimates from the GBD study (2008) of cause-specificmortality of 15 deaths per 100,000 for males and eightdeaths per 100,000 for females [19].Increases in age-standardized mortality from infectious

diseases were largely accounted for by cause-specific mor-tality from hepatitis B and septicemia and indicate a needfor further consideration for early universal vaccination forhepatitis B and improved reporting of underlying condi-tions that may result in septicemia (such as diabetes) onthe medical certificate.

ConclusionThis paper presents the first published comparative vali-dation study of cause of death data in a Pacific Islandcountry and demonstrates clearly the impact of non-communicable diseases on life expectancy in Tonga.This is supported by the significant increase in age-specific mortality from a range of noncommunicablediseases noted between the two periods, including dia-betes, lung cancer, and cardiovascular disease, althoughspecific subcategories of disease such as ischemic heartdisease are more likely to be affected by quality issues inthe earlier period (prior to the new MoH policy on cer-tification) and therefore will be less reliable than thebroader categories. Although Tonga was the first coun-try in the region to develop a strategy to combat non-communicable diseases [29], more clearly needs to bedone to address this growing health issue.This research also outlines the potential for improving

cause of death data available in Tonga. Changes havealready been made to data entry and tabulation practicesin order to separate contributory causes in Part II of thecertificate from the causal sequence in Part I and to


Page 13 of 15

ensure tabulations from the MoH database are generatedfrom the underlying cause of death. However, despite thepotential for further improvements in cause of deathdata, there are significantly more data on cause of deathavailable from Tonga than are routinely reported orknown to international agencies, and it is imperative thathealth planners are able to access this information.

Additional material

Additional file 1: World Health Organization. International StatisticalClassification of Diseases and Related Health Problems. Tenth Revision.Geneva: 2007 [cited 26/11/2010] http://www.who.int/classifications/icd/en/.

Additional file 2: Source data for selection of final cause-of-death inmatched records, and reason for selection; Tongatapu (2008).

AcknowledgementsThe authors would like to acknowledge the management and staff of theTonga Ministry of Health for their support and involvement in this project.Particular recognition is due to Dr. Toa, Dr. Ake, Trish Ryan and the medicalrecords unit in Tonga, and, of course, our data extractors Hulita Fihaki and‘Elaine Faletau. We would also like to acknowledge Dr. SamanthilakaGamage for coding the forms.

Author details1School of Population Health, University of Queensland, Herston (Brisbane),Queensland, Australia. 2Ministry of Health, Tonga, Australia. 3Western Health,Melbourne, Australia. 4School of Public Health and Community Medicine,University of New South Wales, Randwick (Sydney), NSW, Australia.

Authors’ contributionsKC designed the study, developed the study forms and procedures,conducted initial training of data extractors, reviewed trial extraction forms,arranged coding and collation of data, analyzed the collected data, anddrafted the manuscript; SH coordinated data extraction in Tonga andprovided significant input to the manuscript; CR assisted with reviewing trialextraction forms and revised the manuscript critically for importantintellectual content; SA reviewed the extracted data from both countries toallocate cause of death from the medical record and provided significantinput into the discussion; RR provided the second review of deaths forwhich cause was originally assigned as unknown and critically revised themanuscript. ADL and RT conceived and coordinated the broader studyunder which this project was undertaken, reviewed and contributedsignificantly to the interpretation of results, and revised the manuscriptcritically for important intellectual content. RT also made substantialcontributions to the study design. All authors read and approved the finalmanuscript.

Competing interestsADL is an Editor-in-Chief of Population Health Metrics.

Received: 7 September 2011 Accepted: 5 March 2012Published: 5 March 2012

References1. Taylor R, Bampton D, Lopez AD: Contemporary patterns of Pacific Island

mortality. Int J Epidemiol 2005, 34(1):207-214.2. Mathers C, Ma Fat D, Inouie M, Rao C, Lopez A: Counting the dead and

what they died from: an assessment of the status of global cause ofdeath data. Bull World Health Organ 2005, 83(3):171-177.

3. World Health Organisation: World Health Organisation Statistical InformationSystem Geneva: WHO 2011 [http://www3.who.int/whosis/core/core_select.cfm].

4. SPC: 2009 Pocket Statistical Summary Noumea: Secretariat of the PacificCommunity;[http://www.spc.int/sdp/index.php?option=com_docman&task=cat_view&gid=28&Itemid=42].

5. Ministry of Health: Report of the Minister for Health for the year 2008Nuku’alofa, Tonga: Kingdom of Tonga; 2009.

6. Hufanga S, Carter KL, Rao C, Lopez AD, Taylor R: All-cause mortality data inTonga: Assessment of available data In review 2011.

7. Kingdom of Tonga Statistics Department: In Population Census 2006Analytical Report. Volume 2. Nuku’alofa, Tonga; 2006, [http://www.spc.int/prism/country/to/stats/Social/Health/health.htm].

8. WHO: Western Pacific Country Health Profiles (CHIPS) 2010 Revision WorldHealth Organization Manila; 2011 [http://www.wpro.who.int/internet/files/hin/CHIPS2010.pdf].

9. Ministry of Health: Policy and procedures: Notification of a death Nuku’alofa,Tonga: Kingdom of Tonga; 2007.

10. World Health Organization: International Statistical Classification of Diseasesand Related Health Problems Tenth Revision Geneva; 2007 [http://www.who.int/classifications/icd/en/].

11. Murray C, Lopez AD, Black R, et al: Population Health Metrics ResearchConsortium gold standard verbal autopsy validation study: design,implementation, and development of analysis datasets. Population HealthMetrics 2011, 9:27.

12. Carter KL: Procedure Manual - Tonga Medical Record Review University ofQueensland: Unpublished; 2008.

13. World Health Organization: International form of medical certificate ofdeath. ICD-10 International Statistical Classification of Diseases and RelatedHealth Problems Tenth Revision: Instruction Manual Geneva; 2005.

14. Johansson LA, Westerling R, Rosenberg HM: Methodology of studiesevaluating death certificate accuracy were flawed. J Clin Epidemiol 2006,59(2):125-131.

15. Rao C, Porapakkham Y, Pattaraarchachai J, Polprasert W, Swampunyalert N,Lopez AD: In Verifying causes of death in Thailand: rationale and methods forempirical investigation. Volume 8. Population Health Metrics; 2010:11[http://www.pophealthmetrics.com/content/8/1/11].

16. Hernández B, Ramírez-Villalobos D, Romero M, Gómez S, Atkinson C,Lozano R: Assessing quality of medical death certification: concordancebetween gold standard diagnosis and underlying cause of death inselected Mexican hospitals. Population Health Metrics 2011, 9:38.

17. Statistics Department Tonga: Population Census Report 1996; Nuku’alofaTonga: Kingdom of Tonga; 1996 [http://www.spc.int/prism/country/to/stats/Social/Health/health.htm].

18. Ahmad O, Boschi-Pinto C, Lopez AD, Murray CJL, Lozano R, Inoue M: Agestandardization of rates: a new WHO standard Geneva, World HealthOrganization; 2001 [http://www.who.int/whosis/indicators/compendium/2008/1mst/en/index.html], Salomon JA.

19. World Health Organization: Disease and Injury Country Estimates 2008; GlobalBurden of Disease [http://www.who.int/healthinfo/global_burden_disease/estimates_country/en/index.html].

20. Murray CJL: The epidemiological transition revisited. Compositionalmodels for causes of death by age and sex. Popul Dev Rev 2002,28(2):205-228.

21. Adair T, Rao C: Changes in certification of diabetes with cardiovasculardiseases increased reported diabetes mortality in Australia and theUnited States. J Clin Epidemiol 2010, 63(2):199-204.

22. CHERG: Estimates of childhood mortality. Annex B: Data Sources andMethods: The Global Burden of Disease: 2004 Update [http://www.who.int/healthinfo/global_burden_disease/2004_report_update/en/index.html].

23. NZ Statistics: New Zealand Mortality Statistics: 1950 to 2010 Auckland 2010[http://www.moh.govt.nz/moh.nsf/Files/mortdemo/$file/mort-50to10-prov.pdf].

24. Secretariat of the Pacific Community: NCD Statistics for the Pacific IslandsCountries and Territories 2011 [http://www.spc.int/hpl/index.php?option=com_content&task=blogcategory&id=23&Itemid=42].

25. New Zealand Ministry of Health: Diabetes in New Zealand Models andforecasts 1996-2011 [http://www.moh.govt.nz/moh.nsf/0/16a3945b4714f9fccc256b74000f3551?OpenDocument].

26. Barnes L, Moss-Morris R, Kaufusi M: Illness beliefs and adherence indiabetes mellitus: A comparison between Tongan and Europeanpatients. New Zealand Med J 2004, 117(1188)[http://hdl.handle.net/2292/4667].


Page 14 of 15

http://www.biomedcentral.com/content/supplementary/1478-7954-10-4-S1.DOC



http://www.biomedcentral.com/content/supplementary/1478-7954-10-4-S2.DOC






http://www3.who.int/whosis/core/core_select.cfm

http://www3.who.int/whosis/core/core_select.cfm














http://www.pophealthmetrics.com/content/8/1/11

http://www.pophealthmetrics.com/content/8/1/11






http://www.who.int/whosis/indicators/compendium/2008/1mst/en/index.html

http://www.who.int/whosis/indicators/compendium/2008/1mst/en/index.html

http://www.who.int/healthinfo/global_burden_disease/estimates_country/en/index.html

http://www.who.int/healthinfo/global_burden_disease/estimates_country/en/index.html




http://www.who.int/healthinfo/global_burden_disease/2004_report_update/en/index.html

http://www.who.int/healthinfo/global_burden_disease/2004_report_update/en/index.html

http://www.moh.govt.nz/moh.nsf/Files/mortdemo/$file/mort-50to10-prov.pdf

http://www.moh.govt.nz/moh.nsf/Files/mortdemo/$file/mort-50to10-prov.pdf



http://www.moh.govt.nz/moh.nsf/0/16a3945b4714f9fccc256b74000f3551?OpenDocument

http://www.moh.govt.nz/moh.nsf/0/16a3945b4714f9fccc256b74000f3551?OpenDocument

http://hdl.handle.net/2292/4667

http://hdl.handle.net/2292/4667

27. Taylor R, Page A, Danquah J: The Australian epidemic of cardiovascularmortality 1935-2005: effects of period and birth cohort. J EpidemiolCommunity Health 2011.

28. Woodward A, Newland H, Kinahoi M: Smoking in the Kingdom of Tonga:a report from a national survey. Tob Control 1994, 3(1):41-45.

29. Ministry of Health: Non Communicable Diseases Response Plan Nuku’alofa,Tonga: Kingdom of Tonga; 2004.

doi:10.1186/1478-7954-10-4Cite this article as: Carter et al.: Causes of death in Tonga: quality ofcertification and implications for statistics. Population Health Metrics 201210:4.










Page 15 of 15

� � �

266

6.2.2 Medical Record Review in Nauru

Aim and Objectives

Certification practices are of particular interest in Nauru for two main reasons. Firstly,

Nauru is the only study country to use a medical certificate that is not compliant with

the international standard. In the Nauru certificate there is an additional line labelled

cause of death” above the first line of Part I (the causal sequence). Secondly, there

is a very high occurrence of co-morbidities for chronic NCDs and it is not uncommon

to see medical certificates (or medical records) with three or more of the following

diseases: gout, hepatitis, liver failure, ischemic heart disease, kidney failure,

diabetes, neoplasms, and rheumatic heart disease. This high level of co-morbidity

across all adult age groups may lead to difficulties in selecting a clear causal

sequence leading to death and subsequent completion of the medical certificate.

Methods

The medical record review in Nauru followed similar methods as the study in Tonga.

Medical records were sought for 2005-2009. Prior to 2005, many records were

discarded when the public hospital moved to its present site.

Variations from the methods used in Tonga were that medical certificates had to be

manually coded and entered and the search for medical records included many

found amongst so called “active” patient files. Where a medical record was not

available for an infant who had died in the first few days after birth, the mother’s

record was reviewed for information on the pregnancy and child’s symptoms.

Although a medical record is to be created for all children born, in practice, early

notes are frequently recorded on the mother’s chart, with a separate chart for the

� � �

267

child not established for several days. Similarly, if a file consisted of several parts,

but these could not all be located, the partial file was used. In both cases, the extract

was marked to indicate the variation. Reviewed records are shown in the following

flow diagram.

Figure 30: Medical record review flowchart for Nauruan study

Where two CoD sequences were possible from the medical record review, if one was

consistent with the medical certificate, this was taken as the correct sequence. As

with the study in Tonga, the reviewing doctor was familiar with the country and

medical facilities, but had not actively worked in this setting for a number of years.

Training was conducted with the doctor by reviewing the study protocol (Appendix

five) and completing a series of example cases that were discussed with the

N:�43

� � �

268

researcher. Any cases where CoD could not be determined were reviewed by a

second doctor with experience in medical record review studies. Reviewing doctors

indicated their level of certainty when identifying the causal sequence resulting in

death (outlined in Appendix five).

Selection of underlying CoD from the original medical certificate, for comparison to

the findings from the medical record review, was done twice; both applying ICD rules

to the causal sequence including the non-standard additional line on the Nauruan

certificate, and again, excluding the additional line and using only the causes listed in

Part I and II of the certificate as it complies with the ICD standard.

The age, sex, and causal distributions (by proportion) of deaths for which a medical

record was found, and those for which no record was found, were compared. As

deaths for which no medical record was found were predominated by older age

categories and ‘ill-defined’ causes, only deaths for which a medical record was found

and could be reviewed were used to determine final causal distributions.

The underlying CoD for each reviewed case could then be selected. In each case

the underlying CoD from the certificate was selected except in the following

scenarios, where data was corrected.

a. Where the underlying cause identified by the two sources did not match but

was potentially related, the more specific cause was selected. If the more

specific cause was from the medical record, this was only selected when the

certainty of diagnosis was considered reasonable (categorised as “definite” or

“some certainty” by the reviewing doctor).

� � �

269

b. If the two causes were not related, and the medical certificate indicated an

“external cause”, this was selected.

c. If the underlying cause from the medical certificate was attributed as “signs,

symptoms or ill-defined causes” the underlying cause was taken from the

medical record review.

d. An unrelated underlying cause was identified from the medical record review

and the diagnosis was classified as “certain” by the reviewing doctor.

CoD distributions were than calculated by age and sex for 2005-2009. Confidence

intervals were based on Poisson distribution of uncertainty from the proportional

mortality, not accounting further for uncertainty in the mortality envelope.

Results

Of the 382 deaths identified in Nauru, a medical record was found and reviewed for

288 (75%) cases.

When medical records were reviewed and compared with the medical certificates,

specific underlying CoD defined according to the ICDv10 General Mortality List One

categories (66), was found to match in 48% of cases if the first (non-standard) line of

the certificate was included. Agreement increases to 58% when using broad

categories (by ICD chapter). Table 26 shows the consistency between the medical

record and medical certificate when including all lines on the certificate. Excluding

the non-standard line, using only the section compliant with the ICD standard, there

was 43% agreement between the medical records and medical certificates at by

specific causes and 54% by ICD chapter.

� � �

270

Using only the additional line before Part I of the Nauruan certificate, compared to

the medical records, there is an 18% match with specific causes and 25% by ICD

chapter.

Many cases with discrepancies in the underlying cause between the medical record

and death certificate had potentially related conditions identified. For example, two

cases certified as septicaemia on the medical certificate were identified from the

medical record as due to diabetes and meningitis, respectively. Both plausibly

related diagnoses. Similarly, cases certified as due to tuberculosis were identified

from the medical record as due to lung disease, although there was insufficient detail

for a specific diagnosis; again, plausibly the same condition. This was similar for a

case of hepatic cancer, certified as due to infectious hepatitis, although this

diagnosis was not identified on the medical record.

The greatest discrepancies (Table 26) were found in cases certified as due to

diseases of the genitourinary tract, with only 10% agreement with the medical

records. A further 30% were found to have an underlying CoD as diabetes in the

medical records. Only 50% of cases certified as due to diabetes on had this

identified as the underlying CoD in the medical record, although a further 24% (22

cases) were attributed to potential complications of diabetes without a diagnosis of

diabetes in the medical record. In contrast, known cases of diabetes, where diabetes

was selected as the underlying CoD from the medical record review, were also

poorly certified. Only 65% of these cases had diabetes as an underlying CoD on the

� � �

271

medical certificate. A further 19% (13 cases) were certified as being due to causes

that are potential complications of diabetes.

Six of 21 deaths (29%) certified as due to perinatal conditions on the medical

certificate were found to be stillbirths upon medical record review. As discussed in

Chapter three, government funeral payments to assist families following a death,

while creating a strong incentive for complete reporting, have also led to pressure

(real or perceived) on doctors to certify late term stillbirths as infant deaths to ensure

families are eligible for this assistance.

��

�

272

Tabl

e 26

: Com

paris

on o

f und

erly

ing

caus

e of

dea

th fr

om th

e m

edic

al c

ertif

icat

e of

dea

th (i

nclu

ding

the

non-

stan

dard

firs

t lin

e in

the

asse

ssm

ent o

f cau

sal s

eque

nce)

aga

inst

find

ings

of t

he m

edic

al re

cord

revi

ew: N

auru

, 200

5-20

09

� � �

273

As noted in the methods, the age, sex and causal distributions of cases for which a

medical record was found were compared to those for which no record was found to

identify whether any significant systematic bias would be introduced by restricting the

analysis to those cases for which a medical record could be reviewed.

Figure 31 shows the underlying cause distribution (by proportion) of deaths for which

a medical record could and could not be found. Causal distribution of deaths for

which a medical record was identified was determined based on underlying cause on

the original certificate (assuming inclusion of the first line), and the corrected

underlying CoD from the medical record review (as outlined in the methods).

Figure 31: Comparison of cause of death distributions by broad category for reviewed* and non-reviewed^ deaths: Nauru, 2005-2009

* deaths reviewed against the medical record; and ^non-reviewed deaths for which no record was found

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35 Missing�RecordsCause�from�DC�for�found�recordsAmmended�cause�based�on�review

� � �

274

As seen in Figure 31, corrections had very little impact on the causal distribution of

deaths by ICD chapter for the reviewed cases. There was however a significant

difference between the causal distributions of deaths for which a medical record

could and could not be found. Deaths with missing records were more much more

likely to be reported as due to ‘ill-defined conditions’ and non-specific circulatory

causes (such as ‘heart stopped’ or ‘heart attack’).

Cases for which a medical record could not be found also had a much higher

proportion of older adults 65 years and above (30% compared to 14% of those for

which a record was located) (Table 27).

Table 27: Comparison of age distribution for deaths with a medical certificate for which no medical record was found; and deaths with a certificate which

were reviewed against the medical record. Nauru, 2005-2009

Age Group

Cases Proportion (%) Not found Found Not found Found

<5 years 11 27 0.12 0.10 5-14 years 5 0.00 0.02 15-24 years 3 11 0.03 0.04 25-34 years 5 22 0.05 0.08 35-44 years 10 34 0.11 0.12 45-54 years 14 98 0.15 0.35 55-64 years 23 46 0.24 0.16 65+ years 28 39 0.30 0.14 Total 94 282 100 100

Based on these findings, the overall cause distributions by age and sex were

generated only from the deaths for which a medical record was found and could be

reviewed, rather than simply using the medical record review to correct individual

records as was done in Tonga (where a much smaller proportion of deaths could not

� � �

275

be located in the medical records). This decision may result in some selection bias,

particularly in the oldest age group, although this should be minimal as there is no

reason to expect any systematic “loss” of medical records based on the certified CoD

(such as a doctor keeping the files of patients with a specific illness). Restricting

cases to those for which cause could be validated should minimise reporting bias

that would potentially result in artificial overestimation of cause-specific mortality

from disease categories such as cardiovascular diseases and diabetes, which may

be used when the causal sequence is unknown.

As seen in Table 28, the majority (44%) of deaths in children aged 0-4 years in

Nauru are due to perinatal conditions, although external causes and respiratory

conditions are also important. Both external causes and respiratory causes featured

as important causes of death for 5-14 year olds (Table 29), although the total

number of deaths is very small and rates are therefore unstable, even when

aggregated over multiple years as shown here.

��

�

276

Tabl

e 28

: Dea

ths

by c

ause

(by

ICD

cha

pter

), pr

opor

tiona

l mor

talit

y an

d ca

use-

spec

ific

mor

talit

y ra

tes

for c

hild

ren

0-4

year

s:

Nau

ru, 2

005-

2009

Row

Lab

els

Dea

ths

Prop

ortio

nal

mor

talit

y (%

)

Low

er

95%

CI

(%)

Upp

er95

% C

I (%

)

Tota

l rat

e (d

eath

s pe

r 1,

000)

Low

er 9

5%

CI (

deat

hs

per 1

,000

)

Upp

er 9

5%

CI (

deat

hs

per 1

,000

) 1-

001

Infe

ctio

us d

isea

ses

0 0

0 9

0.0

0.0

1.1

1-02

6 N

eopl

asm

s 1

4 0

14

0.4

0.0

1.7

1-04

8 D

isea

ses

of th

e bl

ood

0 0

0 9

0.0

0.0

1.1

1-05

1 E

ndoc

rine,

nut

ritio

nal a

nd o

ther

m

etab

olic

dis

orde

rs

0 0

0 9

0.0

0.0

1.1

1-05

8 D

isea

ses

of th

e ne

rvou

s sy

stem

1

4 0

14

0.4

0.0

1.7

1-06

4 D

isea

ses

of th

e ci

rcul

ator

y sy

stem

0

0 0

9 0.

0 0.

0 1.

1

1-07

2 D

isea

ses

of th

e re

spira

tory

sy

stem

4

15

6 26

1.

4 0.

4 3.

1

1-07

8 D

isea

ses

of th

e di

gest

ive

tract

0

0 0

9 0.

0 0.

0 1.

1 1-

082

Dis

ease

s of

the

skin

1

4 0

14

0.4

0.0

1.7

1-08

4 D

isea

ses

of th

e ge

nito

-urin

ary

syst

em

2 7

1 18

0.

7 0.

1 2.

2

1-09

2 P

erin

atal

con

ditio

ns

12

44

35

53

4.2

2.5

6.4

1-09

3 C

onge

nita

l con

ditio

ns

2 7

1 18

0.

7 0.

1 2.

2 1-

094

Sig

ns, s

ympt

oms

and

othe

r ill-

defin

ed c

ause

s 1

4 0

14

0.4

0.0

1.7

1-09

5 E

xter

nal c

ause

s (in

jurie

s)

3 11

3

22

1.1

0.3

2.7

Tota

l 27

100

100

100

9.5

7.2

12.0

� � �

277

Table 29: Deaths by cause (by ICD chapter) and proportional mortality for children 5-14 years: Nauru, 2005-2009

Row Labels Deaths lower upperProportional mortality lower upper

1-048 Diseases of the blood 1 0.0 5.6 20% 2% 48%1-072 Diseases of the respiratory system 1 0.0 5.6 20% 2% 48%1-095 External causes (injuries) 3 0.6 8.8 60% 38% 75%Total 5 2 12 100% 100% 100%

For adults aged 15-59 years, endocrine, metabolic and nutritional diseases were

responsible for the highest proportion of deaths for both males (42%) and females

(32 %), consistent with the findings for all-age proportional mortality from the case

study presented in Chapter five. Cancers were responsible for 22% of deaths in adult

females, with leading cancers including breast cancer (4% of total deaths), lung

cancer (4% of total deaths) and cervical cancer (6% of total deaths). Lung cancer

was the most common cancer identified as the underlying CoD for males in this age

group, but was responsible for a significantly lower proportion of deaths (2%).

Proportional mortality and cause-specific rates for specific diseases are shown in

Table 30.

��

�

278

Tabl

e 30

: Dea

ths

by c

ause

(by

ICD

cha

pter

) and

pro

port

iona

l mor

talit

y fo

r adu

lts 1

5-54

yea

rs: N

auru

200

5-20

09

Cause�of�death�

Female�

Male�

Deaths�

Prop

or�

tiona

l�Mortality�

(%)�

Upp

er�

95%�

CI�

Lower�

95%�

CI�

Rate�

(deaths�

per�

100,000�

peop

le)�

Lower�

95%�

CI�

Upp

er�

95%�

CI�

Deaths

Prop

or�

tiona

l�Mortality�

(%)��

Upp

er�

95%�

CI�

Lower�

95%�

CI�

Rate�

(deaths�

per�

100,000�

peop

le)�

Lower�

95%�

CI�

Upp

er�

95%�

CI�

1�00

1�In

fect

ious

�di

seas

es�

6�6.

4�2.

911

.448

.221

.985

.75

4.5

1.8

8.8

43.8

17.5

�85

.7�

1�02

6�N

eopl

asm

s�21

�22

.3�

17.1

27.9

168.

712

9.2

210.

711

9.9

6.1

14.8

96.5

59.1

�14

4.5�

1�04

8�Di

seas

es�o

f�the

�bl

ood�

0�0.

0�0.

03.

20.

00.

024

.21

0.9

0.0

4.2

8.8

0.3�

40.9

�1�

051�

Endo

crin

e,�

nutr

ition

al�a

nd�o

ther

�m

etab

olic

�diso

rder

s�30

�31

.9�

26.6

37.2

241.

020

1.2

281.

147

42.3

38.2

47.1

412.

237

1.8�

458.

9�1�

058�

Dise

ases

�of�t

he�

nerv

ous�s

yste

m��

1�1.

1�0.

04.

88.

00.

336

.62

1.8

0.3

5.4

17.5

2.6�

53.0

�1�

064�

Dise

ases

�of�t

he�

circ

ulat

ory�

syst

em��

15�

16.0

�11

.121

.512

0.5

83.4

162.

417

15.3

11.0

20.5

149.

110

6.6�

199.

8�1�

072�

Dise

ases

�of�t

he�

resp

irato

ry�sy

stem

�10

�10

.6�

6.3

16.0

80.3

47.7

120.

78

7.2

3.8

11.9

70.2

37.2

�11

5.7�

1�07

8�Di

seas

es�o

f�the

�di

gest

ive�

trac

t�4�

4.3�

1.4

8.9

32.1

10.8

67.2

32.

70.

76.

626

.36.

7�64

.4�

1�08

4�Di

seas

es�o

f�the

�ge

nito

�urin

ary�

syst

em�

5�5.

3�2.

110

.140

.216

.176

.66

5.4

2.4

9.9

52.6

23.7

�95

.9�

1�09

5�Ex

tern

al�c

ause

s�(in

jurie

s)�

2�2.

1�0.

36.

316

.12.

447

.411

9.9

6.1

14.8

96.5

59.1

�14

4.5�

�Tot

al�

94�

100.

0�10

0.0

100.

0��

��

111

100.

010

0.0

100.

0��

��

��

�

279

Cause�of�death�

Female�

Male�

Deaths

Prop

or�

tiona

l�Mortality�

(%)�

Upp

er�

95%�CI

Lower�

95%�CI

Rate�

(deaths�

per�

100,000�

peop

le)�

Lower�

95%�

CI�

Upp

er�

95%�

CI�

Deaths�

Prop

or�

tiona

l�Mortality�

(%)��

Upp

er�

95%�

CI�

Lower�

95%�

CI�

Rate�

(deaths�

per�

100,000�

peop

le)�

Lower�

95%�

CI�

Upp

er�

95%�

CI�

Specific�Diseases�

(ICD�10

3�list)��

��

��

��

��

��

��

��

1�00

3�1

1.06

�0.

034.

848.

00.

336

.61�

0.9

0.0

4.2

8.8

0.3

40.9

�1�

005�

22.

13�

0.32

6.28

16.1

2.4

47.4

0�0.

00.

02.

80.

00.

027

.1�

1�00

6�1

1.06

�0.

034.

848.

00.

336

.61�

0.9

0.0

4.2

8.8

0.3

40.9

�1�

012�

11.

06�

0.03

4.84

8.0

0.3

36.6

1�0.

90.

04.

28.

80.

340

.9�

1�01

9�1

1.06

�0.

034.

848.

00.

336

.62�

1.8

0.3

5.4

17.5

2.6

53.0

�1�

027�

00.

00�

0.00

3.21

0.0

0.0

24.2

1�0.

90.

04.

28.

80.

340

.9�

1�02

8�0

0.00

�0.

003.

210.

00.

024

.21�

0.9

0.0

4.2

8.8

0.3

40.9

�1�

029�

11.

06�

0.03

4.84

8.0

0.3

36.6

1�0.

90.

04.

28.

80.

340

.9�

1�03

0�0

0.00

�0.

003.

210.

00.

024

.21�

0.9

0.0

4.2

8.8

0.3

40.9

�1�

034�

44.

26�

1.43

8.90

32.1

10.8

67.2

2�1.

80.

35.

417

.52.

653

.0�

1�03

5�0

0.00

�0.

003.

210.

00.

024

.21�

0.9

0.0

4.2

8.8

0.3

40.9

�1�

036�

44.

26�

1.43

8.90

32.1

10.8

67.2

0�0.

00.

02.

80.

00.

027

.1�

1�03

7�6

6.38

�2.

9011

.35

48.2

21.9

85.7

0�0.

00.

02.

80.

00.

027

.1�

1�03

9�1

1.06

�0.

034.

848.

00.

336

.60�

0.0

0.0

2.8

0.0

0.0

27.1

�1�

046�

55.

32�

2.14

10.1

440

.216

.176

.64�

3.6

1.2

7.7

35.1

11.7

75.2

�1�

049�

00.

00�

0.00

3.21

0.0

0.0

24.2

1�0.

90.

04.

28.

80.

340

.9�

1�05

0�0

0.00

�0.

003.

210.

00.

024

.21�

0.9

0.0

4.2

8.8

0.3

40.9

�1�

052�

2829

.79�

24.4

935

.18

224.

918

4.9

265.

646

�41

.437

.346

.340

3.4

362.

645

0.5�

1�05

4�2

2.13

�0.

326.

2816

.12.

447

.40�

0.0

0.0

2.8

0.0

0.0

27.1

�1�

061�

11.

06�

0.03

4.84

8.0

0.3

36.6

2�1.

80.

35.

417

.52.

653

.0�

1�06

5�5

5.32

�2.

1410

.14

40.2

16.1

76.6

2�1.

80.

35.

417

.52.

653

.0�

��

�

280

1�06

7�3

3.19

�0.

817.

6224

.16.

157

.57�

6.3

3.1

10.9

61.4

30.3

105.

9�1�

068�

11.

06�

0.03

4.84

8.0

0.3

36.6

3�2.

70.

76.

626

.36.

764

.4�

1�06

9�5

5.32

�2.

1410

.14

40.2

16.1

76.6

5�4.

51.

88.

843

.817

.585

.7�

1�07

0�1

1.06

�0.

034.

848.

00.

336

.60�

0.0

0.0

2.8

0.0

0.0

27.1

�1�

074�

55.

32�

2.14

10.1

440

.216

.176

.66�

5.4

2.4

9.9

52.6

23.7

95.9

�1�

075�

11.

06�

0.03

4.84

8.0

0.3

36.6

1�0.

90.

04.

28.

80.

340

.9�

1�07

6�2

2.13

�0.

326.

2816

.12.

447

.40�

0.0

0.0

2.8

0.0

0.0

27.1

�1�

077�

22.

13�

0.32

6.28

16.1

2.4

47.4

1�0.

90.

04.

28.

80.

340

.9�

1�08

0�1

1.06

�0.

034.

848.

00.

336

.62�

1.8

0.3

5.4

17.5

2.6

53.0

�1�

081�

33.

19�

0.81

7.62

24.1

6.1

57.5

1�0.

90.

04.

28.

80.

340

.9�

1�08

6�4

4.26

�1.

438.

9032

.110

.867

.26�

5.4

2.4

9.9

52.6

23.7

95.9

�1�

088�

11.

06�

0.03

4.84

8.0

0.3

36.6

0�0.

00.

02.

80.

00.

027

.1�

1�09

6�2

2.13

�0.

326.

2816

.12.

447

.42�

1.8

0.3

5.4

17.5

2.6

53.0

�1�

097�

00.

00�

0.00

3.21

0.0

0.0

24.2

0�0.

00.

02.

80.

00.

027

.1�

1�09

8�0

0.00

�0.

003.

210.

00.

024

.23�

2.7

0.7

6.6

26.3

6.7

64.4

�1�

101�

00.

00�

0.00

3.21

0.0

0.0

24.2

2�1.

80.

35.

417

.52.

653

.0�

1�10

3�0

0.00

�0.

003.

210.

00.

024

.24�

3.6

1.2

7.7

35.1

11.7

75.2

�To

tal�

9410

0.00

�10

0.00

100.

0075

5.0

755.

075

5.0

111�

100.

010

0.0

100.

097

3.4

973.

497

3.4�

Con

fiden

ce in

terv

als

for p

ropo

rtion

al m

orta

lity

deriv

ed u

sing

Poi

sson

dis

tribu

tion

Con

fiden

ce in

terv

als

for c

ause

-spe

cific

rate

s de

rived

from

unc

erta

inty

in th

e pr

opor

tiona

l mor

talit

y

� � �

281

Discussion

The results show high cause-specific mortality rates from NCDs in adults aged 15-59

years. Cause-specific mortality from endocrine and metabolic diseases (nearly all of

which is due to diabetes) is estimated at 241 and 412 deaths per 100,000 population

for females and males respectively. In comparison, age-standardised mortality

(based on Segi’s world population) from endocrine diseases in New Zealand in 1996

was 210 per 100,000 population for males and 190 per 100,000 population for

females (345). The high cause-specific mortality from NCDs, including diabetes,

demonstrates these diseases are having a direct impact on premature mortality in

Nauru, resulting in the stagnation in LE reported in Chapter five (case study two).

The study also demonstrated quality issues in both the medical certificates and the

medical records. Numerous cases, as discussed in the results, were found to be

missing a record of specific tests or diagnoses (such as hepatitis infection, or

diabetes), even when these were later implied in the medical certificate. Equally,

many certificates contained no mention of conditions clearly recorded in the medical

record. As such, there is a need to improve documentation and recording across

both sources. There was also limited agreement between the underlying CoD

identified from the medical record review with that identified on the additional line on

the Nauruan certificate. Agreement between sources was much greater when this

additional line was treated as part of the causal sequence when applying the ICD

rules to selecting CoD. In this way, the Nauruan certificate is treated as though there

are four lines in Part I. The extra line, which currently sits outside the causal

sequence section of the certificate should therefore be removed (and a 4th line added

to Part I if required) to ensure the medical certificate in Nauru conforms to the

international standard.

� � �

282

6.3 Medical Certificate of Death Review A medical record review relies on the ability to access and match data from the

medical certificate and the medical record. Where access to the medical records is

not available, it is still possible to make a determination about the quality of the

certification process based on the plausibility of the sequence, the specificity of the

information provided, and consistency with the cases’ age and sex, by reviewing the

original medical certificate.

6.3.1 Review of medical certification of death in Fiji

Aim and Objectives

In Fiji, MoH policy requires a medical certificate for all deaths (both hospital and

community deaths). These are collated in their HIS (PATIS) and routinely reconciled

with monthly nurses’ reports and hospital discharge data. The system assessment

presented in Chapter three identified previous tabulations of these data were

primarily based on the CoD reported on the certificate that coding staff considered to

be of most importance to public health. It was further noted the sequence of causes

was often altered upon data entry to ensure this “most important” cause was entered

first (and would therefore be included in the tabulations); while contributory causes

were entered onto the first available line in the sequence rather than in the labelled

field, and as such may have been moved into the causal sequence recorded. As

access to the medical records was not possible, a medical certificate review was

conducted to evaluate the impact of these practices on the CoD tabulations and

where possible, correct any systematic biases to generate causal distributions of

death by age and sex.

� � �

283

Methods

CoD data were obtained for all deaths reported to the MoH for 2000-2008. Data were

extracted from the MoH databases (PATIS for 2007 and 2008, and separate Access

databases for each earlier year). Medical certificates are coded and entered into

these databases by MoH staff at the national office. Data is checked against monthly

community nursing reports and hospital separation data to ensure all deaths are

captured. Variables extracted for each case included certificate number, sex,

ethnicity, date of birth or age, date of death, place of death (by region), and causes

of death (recorded as ICD codes). Names were not extracted to ensure data were

not identifiable.

Accuracy of the CoD data as collated in the MoH databases, was evaluated by

comparing a random sample of entered cases against the original medical certificate.

A random number generator was used to assign a number to each case extracted

from the database. These were sorted by assigned number and the first 48 records

for each year were selected (a total of 350 records) to provide a random sample.

Original medical certificates were sought for these cases from paper records kept by

the MoH. Age, sex, and causes of death (by line number) were then entered into an

excel spreadsheet from the certificate. Each line of the certificate and underlying

CoD was coded according to ICD rules by the researcher and checked by a WHO

collaborating centre for coding (in Sri Lanka). These fields were then compared to

data extracted from the MoH databases by certificate number.

The original medical certificate for each case was initially assessed to determine: a)

the causal sequence and contributory causes for each death; b) whether the causal

� � �

284

sequence as recorded on the certificate is plausible; and c) the underlying CoD

according to ICD rules. The medical certificate was then compared to the data

extracted from the MoH databases to identify: d) whether causes entered on the

database match those recorded on the certificate; e) which data fields each cause

line on the certificate has been entered into in the database; f) whether the code

assigned to each CoD in the database is the closest possible match for the data

provided; and g) which field (if any) contains the underlying CoD as determined by

applying the ICD rules to the original certificate.

Patterns of misreporting are then used to correct tabulated data from 2000-2008 to

generate a CoD profile for Fiji by age group and gender. Proportional mortality was

applied to the total mortality envelope by age group and sex previously estimated

and reported in Chapter five, to generate age-specific mortality rates. Confidence

intervals (95% CI’s) were generated based on the uncertainty of the cause

distribution (excluding uncertainty in the mortality envelope itself) using Poisson

distribution due to the small number of deaths by individual cause.

As some uncertainty may remain regarding the final allocation of CoD due to

previous data entry and tabulation procedures, a multiple CoD analysis was also

conducted for specific diseases of public health interest (based on ICDv10 General

Mortality List One (103 cause list) (68)) with age specific and age-standardised

mortality rates calculated according to total mentions on the certificate.

Results

Of the 384 records randomly selected for review, the original certificate was located

for 329 (86%) cases as shown in Table 31.

� � �

285

Table 31: Number of selected medical certificates found by age group: Fiji, 2000-2008

Age group (years) <1 1 to 4 5 to 14 15-59 60+ Total

No certificate found 10 5 10 18 12 55 Certificate found 42 7 54 132 94 329 Total records sought

52 12 64 150 106 384

% found 81 58 84 88 89 86

Of the original medical certificates located, only 51 were found to be of the current

certificate form (introduced in late 2007), with the remainder consisting of two

previous versions (both of which used three lines to report the causal sequence of

death, rather than four lines as per the newer certificate) and three deaths with the

causal sequence reported directly to PATIS.

More than half (59%) of all original certificates were recorded with a single CoD (with

only one line completed), consistent across all age groups (Table 32). Of those with

more than one line completed, 52% were found to have a clearly implausible causal

sequence, however in nearly all cases, the ICD rules would still have identified a

plausible underlying cause.

Table 32: Reviewed medical certificates by single versus multiple lines completed on the certificate: Fiji, 2000-2008

Recording on certificate

Age group (years)<1 1 to 4 5 to 14 15-59 60+ Total

Multiple lines completed on certificate

11 3 24 58 39 135

Single cause of death listed only

31 4 30 74 55 194

Total 42 7 54 132 94 329

� � �

286

Data entry practices were found to be less consistent, with the order of some, but not

all certificates, modified upon entry into the database.

Only 44% of original certificates were found to have the code associated with the text

on the first line also in the first line of the database. In order to allow for coding

errors, this was also reviewed by ICD chapter, where 69% of certificates were found

to have the same ICD chapter associated with the text on the first line also in the first

line of the database. This was limited to 60% if only certificates with more than one

line completed were selected. These findings indicate at least 30% (and up to 66%)

of deaths are either coded incorrectly or have causes entered into the database in a

different order to that appearing on the certificate (as anticipated from the system

assessment). As this selective reordering is not consistent however, it is not possible

to simply tabulate deaths from the first line of the database, as has been done for

previously reported data from Fiji. The proportion of deaths for which the underlying

cause was found in line one of the database is shown by age in the Table 33. As

seen in this table, 36% of the deaths which could be reviewed had the underlying

CoD recorded on the first line of the certificate. This increased to 61% when

evaluated by ICD chapter.

� � �

287

Table 33: Column of database where underlying cause of death from the original certificate (specific disease and by chapter) were found, by age group:

Fiji 2000-2008

Deaths Proportion (%)

Age group (years) <1 1 to

45 to 14

15-59 60+ Total <1 1 to

45 to 14

15-59

60+ Total

FIRST LINE (EXACT CAUSE)

11 1 15 45 47 119 27 14 28 34 50 36

FIRST LINE (ICD CHAPTER)

19 4 31 80 66 200 46 57 57 61 70 61

LAST LINE (EXACT CAUSE)

7 1 4 22 10 44 17 14 7 17 11 13

LAST LINE (ICD CHAPTER)

9 1 4 23 13 50 22 14 7 18 14 15

None 12 2 13 20 9 56 29 29 24 15 10 17

The similarity of these figures with the proportion of deaths for which the cause on

the certificate matched the database by ICD chapter, suggests this may be

accounted for, at least in part, by poor coding rather than a deliberate reordering of

the sequence. Only 13% of cases were found to have the underlying cause selected

according to ICD rules on the last line of the database, as would be expected if

certificates were completed correctly and entered as written.

Most concerning is that 17% of cases were found not to have the underlying cause

from the certificate recorded anywhere in the database sequence, resulting in all of

these cases being attributed to an incorrect category. This cannot be adequately

corrected from a medical certificate review and would require the original hard copy

of each certificate to be re-coded. These findings were relatively consistent across

gender, year, age group and ethnicity.

� � �

288

The proportion of deaths by cause for cases where the underlying cause was not

identified in the database is shown in Table 34.

Table 34: Distribution of deaths by broad cause where underlying cause of death was not recorded in the Ministry of Health database. Fiji 2000-2008

Cause Number Proportion (%) 1-001 Infectious Diseases 7 14 1-026 Neoplasms 5 10 1-048 Diseases of the blood and blood forming organs

1 2

1-051 Endocrine, nutritional and metabolic diseases

6 12

1-058 Diseases of the nervous system 4 8 1-064 Diseases of the circulatory system 6 12 1-072 Respiratory Illnesses 1 2 1-087 Maternal causes 1 2 1-092 Perinatal conditions 4 8 1-093 Congenital conditions 4 8 1-094 Signs, symptoms and ill-defined 2 4 1-095 External causes 10 20 Total 51 100

As seen in the above table, causes were distributed across ICD chapters, although

deaths due to external causes were particularly poorly captured.

Given uncertainty in both the original coding and reordering of the causal sequence

during data entry, it was determined not possible to select one line of the database

as the underlying cause by specific disease, nor would it be possible to re-select

underlying cause from the database for 2000-2008 with any accuracy without

reviewing each original certificate. As the original certificates were not available, a

multiple CoD analysis for selected causes of specific public health importance was

� � �

289

conducted. These included maternal deaths, external causes, diabetes and

genitourinary diseases, malnutrition, cancers, hepatitis, TB and septicaemia.

Underlying cause by ICD chapter was also re-selected from the tabulated data for

2000-2008 (excluding 2007 for which not all data was available), to provide an

indicative picture of CoD across age groups, although some uncertainty will remain

in these estimates as the medical certificate review indicated that some underlying

causes were not represented in the tabulated data.

Table 35: Deaths, proportional mortality and death rate by selected cause (total mentions on the certificate) for children 0-4 years: Fiji 2000-2008

�� Females� Males�Total�deaths�

Proportion�of�deaths�with�cause�mentioned

Rate�(deaths�per�100,000�population)

Lower�95%�CI�

Upper�95%�CI�

Meningitis� 23� 28� 51 2.2 7.9 0� 28.9Pneumonia� 49� 49� 98 4.2 15.2 0� 44.2Malnutrition� 20� 23� 43 1.8 6.7 0� 25.9Septicaemia� 97� 97� 194 8.3 30.0 0� 70.9Rheumatic�fever/�RHD� 3� 5� 8 0.3 1.2 0� 9.5Total� 1196� 1146� 2342 100.0 ��

Excluding deaths from 2007 as not all required variables in the data set were available from the extracted data

� � �

290

Table 36: Deaths and proportional mortality by mentioned cause (total mentions on the certificate) for selected causes for children 5-14 years: Fiji

2000-2008

Cause�of�death��Deaths

Proportion�(excluding�ill�

defined)�

Lower�95%�CI�

Upper�95%�CI�

1�001�Infectious�Diseases� 224� 7.7� 2.8� 14.4�1�026�Neoplasms� 91� 3.1� 0.6� 8.8�1�048�Diseases�of�the�blood�and�blood�forming�organs� 14� 0.5� 0.0� 3.7�

1�051�Endocrine,�nutritional�and�metabolic�diseases� 124� 4.2� 1.1� 10.2�

1�055�Mental�Disorders� 1� 0.0� 0.0� 3.7�1�058�Diseases�of�the�Nervous�System� 103� 3.5� 0.6� 8.8�

1�064�Diseases�of�the�circulatory�system� 236� 8.1� 3.5� 15.8�

1�072�Respiratory�Diseases� 288� 9.8� 4.1� 17.1�1�078�Diseases�of�the�digestive�system� 44� 1.5� 0.0� 5.6�

1�082�Diseases�of�the�skin� 13� 0.4� 0.0� 3.7�1�083�Musculoskeletal�Diseases� 1� 0.0� 0.0� 3.7�1�084�Diseases�of�the�genitourinary�tract� 29� 1.0� 0.0� 3.7�

1�087�Maternal�deaths� 7� 0.2� 0.0� 3.7�1�092�Perinatal�Conditions� 1285� 43.9� 31.1� 57.9�1�093�Congenital�Conditions� 259� 8.8� 3.5� 15.8�1�094�Ill�defined� 96� 3.3� 0.6� 8.8�1�095�External�causes� 208� 7.1� 2.8� 14.4�Total�� 3023�Total�excluding�ill�defined� 2927� 100.0�

��

�

291

Tabl

e 37

: Dea

ths

and

prop

ortio

nal m

orta

lity

by u

nder

lyin

g ca

use

(by

ICD

cha

pter

) for

adu

lts 1

5-59

yea

rs:

Fiji

2005

-200

8

Cause�of�death��

Females�

Males�

Deaths�

Prop

ortio

n�(excluding�

ill�defined

)�Lower�

95%�CI�

Upp

er�

95%�CI�

Deaths�

Prop

ortio

n�(excluding�ill�

defin

ed)��

Lower�

95%�CI�

Upp

er�

95%�CI�

1�00

1�In

fect

ious

�Dise

ases

�18

16.

22.

213

.123

05.

31.

65.

6�1�

026�

Neo

plas

ms�

610

20.9

12.2

30.9

328

7.6

2.8

7.2�

1�04

8�Di

seas

es�o

f�the

�blo

od�a

nd�b

lood

�fo

rmin

g�or

gans

�28

1.0

0.0

3.7

200.

50.

03.

7�1�

051�

Endo

crin

e,�n

utrit

iona

l�and

�m

etab

olic

�dise

ases

�68

623

.514

.634

.575

717

.49.

917

.1�

1�05

5�M

enta

l�Diso

rder

s�3

0.1

0.0

3.7

40.

10.

03.

7�1�

058�

Dise

ases

�of�t

he�N

ervo

us�S

yste

m�

441.

50.

05.

667

1.5

0.0

3.7�

1�06

4�Di

seas

es�o

f�the

�circ

ulat

ory�

syst

em�

677

23.2

14.6

34.5

1679

38.7

26.9

38.1

�1�

072�

Resp

irato

ry�D

iseas

es�

209

7.2

2.8

14.4

316

7.3

2.8

7.2�

1�07

8�Di

seas

es�o

f�the

�dig

estiv

e�sy

stem

�82

2.8

0.2

7.2

145

3.3

0.6

3.7�

1�08

2�Di

seas

es�o

f�the

�skin

�10

0.3

0.0

3.7

80.

20.

03.

7�1�

083�

Mus

culo

skel

etal

�Dise

ases

�23

0.8

0.0

3.7

70.

20.

03.

7�1�

084�

Dise

ases

�of�t

he�g

enito

urin

ary�

trac

t�12

84.

41.

110

.219

54.

51.

15.

6�1�

087�

Mat

erna

l�dea

ths�

200.

70.

03.

7��

0.0

0.0

0.0�

1�09

3�Co

ngen

ital�C

ondi

tions

�9

0.3

0.0

3.7

160.

40.

03.

7�1�

094�

Ill�d

efin

ed�

301.

00.

05.

665

1.5

0.0

3.7�

1�09

5�Ex

tern

al�c

ause

s�20

77.

12.

814

.456

813

.16.

913

.1�

�Tot

al�

2949

��

��44

05��

��

Tota

l�exc

ludi

ng�il

l�def

ined

�29

19��

��

4340

��

��

��

�

292

Tabl

e 38

: Dea

ths

and

prop

ortio

nal m

orta

lity

by m

entio

ned

caus

e (to

tal m

entio

ns o

n th

e ce

rtifi

cate

) for

sel

ecte

d ca

uses

for

adul

ts 1

5-64

yea

rs: F

iji 2

000-

2008

Caus

e�of

�dea

th�

Fem

ale�

Mal

e�

Deat

hs�

Prop

ortio

n�of

�dea

ths�

with

�cau

se�

men

tione

d

Rate

�(d

eath

s�per

�10

0,00

0�po

pula

tion)

Low

er�

95%

�CI�

Upp

er�

95%

�CI�

Deat

hs�

Prop

ortio

n�of

�dea

ths�

with

�cau

se�

men

tione

d

Rate

�(d

eath

s�per

�10

0,00

0�po

pula

tion)

Low

er�

95%

�CI�

Upp

er�

95%

�CI�

Tube

rcul

osis�

47�

0.7

2.4

0.0

7.9�

850.

84.

60.

011

.5�

Hepa

titis�

15�

0.2

0.8

0.0

3.9�

410.

42.

20.

07.

6�Di

abet

es�

1391

�21

.370

.540

.510

0.4�

1534

15.3

83.5

50.9

116.

1�Rh

eum

atic

�Hea

rt�

Dise

ase�

157�

2.4

8.0

0.0

18.0

�13

51.

37.

30.

017

.0�

Isch

aem

ic�H

eart

�Di

seas

e�63

6�9.

732

.212

.052

.5�

3083

30.8

167.

912

1.7

214.

0�

Sept

icae

mia

�85

7�13

.143

.419

.966

.9�

868

8.7

47.3

22.7

71.8

�Ca

ncer

�12

49�

19.1

63.3

34.9

91.6

�67

86.

836

.915

.258

.6�

Tota

l�65

37�

��

��

1001

4�

��

��

� � �

293

6.4 Examining cause of death distributions from alternative data

sources: understanding causal patterns in the absence of

nationally representative data from medical certification

Ideally, health planners require medically certified CoD for the whole population, or a

representative sample. Yet many countries have to rely on hospital data for

reporting and monitoring, since that is the only data available. While this is a useful

proxy to monitor changes in specific CoD, it is important to remember CoD

distributions for hospital data may not be representative of the total population (283).

Deaths which occur suddenly from external causes, stroke and myocardial infarction,

where not all cases make it to hospital for treatment, may be under-represented

(283), while treatment practices and the level of services available may affect the

likelihood of death from NCDs (such as specific types of cancer and liver disease)

occurring in hospital.

Differences in reporting patterns from hospital and community deaths are explored

below in relation to Vanuatu, which has three CoD data collections: hospital

separation data (based on primary diagnosis) medical certificates for hospital deaths,

and community nursing reports.

6.4.1 Comparison of cause of death distributions from hospital and

community data in Vanuatu

Aim and Objectives

As noted in Chapter three, reporting completeness from routine CRVS systems in

Vanuatu is currently too low to enable data to be corrected and used directly to

derive estimates of mortality level (61). The most reliable data on mortality level

� � �

294

comes from survey (such as the MICs described in chapter 2 which examined U5M

and IMR)(157), and census analyses (61, 138, 144). However, neither MICS nor

census analyses collect information on CoD. CoD data, when reported, are usually

derived from hospital separation data based on the primary diagnosis and reported

as leading causes of death for all ages.

As national representative CoD data are not available, hospital data (medical

certification and hospital separation data) and community health reporting data are

reviewed to generate proportional mortality distributions by age and sex. Similarities

and differences in the resultant causal distributions are reviewed to: a) identify key

public health issues contributing to premature mortality by cause; b) quantify at a

broad level the contribution of these disease categories to premature mortality where

possible; c) identify public health issues from the community data that may have

been previously unrecognised in published data from Vanuatu based on hospital

reporting; and d) highlight systematic biases that may occur when CoD profiles are

obtained from hospital data alone.

Methods

De-identified unit record data on deaths for 2001-2007 were obtained during in-

country visits from three sources: hospital separation data, medical certificates for

deaths in hospital and notices of death submitted with monthly nursing reports from

smaller health facilities (area health centres, dispensaries and aid posts).

CoD data are collected for deaths in the two main hospitals: Vila the national referral

hospital at Port Vila on Efate(the main Island), and the secondary district hospital at

Luganville, in Santo province. These are two of only three hospitals in Vanuatu

� � �

295

routinely staffed by doctors (Tanna hospital is staffed by a single doctor and the

remaining two district hospitals are staffed by nurse practitioners).

Hospital separation data were extracted from the HIS. This is an Access database

maintained at the national level, with data entry for hospital deaths completed locally

at Vila Central and Luganville hospitals, and copied to the national office periodically

when staff travel to or from the national office. Local copies of the database were

used as they are considered to be more complete than the copy held at the national

office. CoD is recorded according to principal diagnosis, with up to two additional

fields for other causes, and fields for accidents and external causes, and medical

procedures. These fields appear to have been completed as a causal sequence, as

found on a medical certificate (with immediate cause recorded first, working

backwards to the underlying cause), so underlying cause was selected and coded

according to ICDv10 rules for medical certificates (68). Causes are coded according

by medical record staff at the hospital, with both text and code entered in the

database.

Medical certificates are completed semi-routinely (and particularly upon request by

family members) for hospital deaths. They are not routinely coded or collated for

Luganville, and although collated for Port Vila hospital are not routinely coded,

analysed or reported. Data were extracted from the original medical certificates at

Luganville hospital and from an Excel spreadsheet maintained by medical records

staff at Vila Central. The Excel spreadsheet included all causes listed in Part I of the

certificate as text, identified by line. Contributory causes were not recorded. Medical

certificates are predominantly completed by the attending doctor, although ni-

Vanuatu doctors in Luganville noted they completed most certificates as

� � �

296

internationally trained doctors based at the hospital were not always comfortable

undertaking this task due to language difficulties. Most certified deaths occur in

hospital, however doctors occasionally certify deaths that occur in the community.

Place of death was not well recorded on either the original certificate or the

spreadsheet; and it is therefore not possible to separate certified deaths according to

location. Underlying cause was selected and coded [by KC] according to ICDv10

(68).

Deaths in the community are captured through health system reporting by health

centres, dispensaries, and aid posts. Nurses or nurse aides at each facility are

required to submit a monthly report of activities including any deaths in their area. A

notice of death form is also collected for deaths in the community. This includes

name, age, date of death, place of death and CoD as a single free-text field. The

system is designed so deaths are collated and coded according to ICDv10 before

being entered into the national HIS database at the provincial level, although in

practice this is mostly performed at the national HIS office.

Data were sorted according to health facility code and deaths that occurred in Vila or

Luganville hospitals were removed. For deaths where specific age was not recorded,

deaths recorded as “adults” were attributed to age group 15-59 years and “old” or

“elderly” to age group 60+ years.

The approximate proportion of deaths accounted for by each data set was calculated

based on an average of 1,394 deaths per year (as estimated by the Vanuatu

National Statistics Office) (346), assuming little variation over the period under

investigation. Deaths were tabulated by broad age group (0-4 years, 5-14 years, 15-

� � �

297

59 years, 60+ years) to address the differences in CoD distribution by age, while

allowing for a lack of age detail in some of the data. Foetal deaths were identified

and removed from all data sets. Cases where there was uncertainty whether the

death was a stillbirth or early neonatal were counted as neonatal, as separate

reporting mechanisms exist for stillbirths which mean in practice, they should not be

recorded in these data.

All data sets were aggregated to ICDv10 General Mortality List One (103 causes),

with childhood deaths (0-4 years) aggregated to ICDv10 Condensed List three for

U5M. Tabulations of reported deaths by cause, proportional mortality and

proportional mortality adjusted to exclude unknown and ill-defined deaths were then

generated for each data set. Comparisons by broad category of cause, as derived

from the GBD, are also presented.

Results

a) Hospital Separation Data:

There were 954 deaths reported in the hospital separation data. This is an average

of 136 deaths per year, or nearly 10% of the expected deaths in Vanuatu. Of these,

800 deaths were recorded at Vila Hospital and 154 at Luganville. There were 13

foetal deaths removed from the data (eight from Vila Central and five from

Luganville). A further 14 deaths had no age attributed and were excluded from

further analysis.

b) Medical Certificates of Death:

There were 1790 deaths medically certified between 2001 and 2007. This is an

average of 256 deaths per year, or approximately 18% of the expected deaths in

� � �

298

Vanuatu. Of these, 1466 were recorded at Vila Central Hospital and 324 at

Luganville Hospital. One hundred foetal deaths were also certified at Vila Central

Hospital, and a further 33 at Luganville hospital. These were subsequently removed

from the data set. There were an additional 78 deaths excluded from further analysis

as they had no age recorded.

c) Health Facility records

There were 2,563 deaths recorded by health facilities between 2001 and 2007, an

average of 366 deaths per year, or approximately 26% of the expected deaths in

Vanuatu. Twenty one deaths reported from Vila Hospital were removed from the

data set. All were adult deaths (13 aged 15-59 years and eight aged 60 or older).

There were also 102 foetal deaths removed from the data. The age and gender

distribution of deaths reported through the health facilities is shown in Figure 32.

Forty six deaths had no age details and were excluded from further analysis.

Figure 32: Age distribution of reported deaths by data source and gender; Vanuatu, 2001-2007

MC = Medical certificate data, Hosp = Hospital discharge data, and HF = Health

Facility data. M= Males and F= Females.

� � �

299

The age distribution of deaths reported in the hospital separation data and medical

certificates were quite similar, with children aged 0-4 years accounting for 15% and

22% of reported deaths respectively. There were few deaths reported in the 5-14

year age group (4 % and 2% respectively), with the majority of deaths in the adult

age group of 15-59 years (48% and 41% respectively). Approximately 30% were

people aged 60 years or older at time of death (32% and 31% respectively for

hospital separation data and medical certificates) . The community health facility

data showed similar proportions of childhood deaths (16% for ages 0-4 years and

3% for 5-14 year olds) as other sources, but a lower proportion of adult deaths for

ages 15-59 years (29%) and a significantly higher proportion of deaths in people

aged 60 years or older (51%). There was a low proportion of deaths of unknown

ages (1-4%) in each of the data sources. As these accounted for only a small

number of deaths they were excluded from further analysis.

Deaths attributed to unknown or ill-defined causes (Code 1-064 ‘signs, symptoms or

ill-defined causes) are shown in Table 39.

Table 39: Proportion of deaths (%) attributed to Ill-defined or unknown causes by age group, gender and source: Vanuatu 2001-2007

Age group Gender Hospital Separation

Data

MedicalCertificates

Data

HealthFacility

Data0-4 years Both 1.4 3.9 22.6 5-14 years Both 8.6 2.7 24.7 15-59 years Males 1.8 5.9 18.1 15-59 years Females 1.8 6.3 22.1

� � �

300

For deaths in ages 0-4 years, both neonatal causes amenable to intervention

through maternal health status, antenatal care, and birthing practices, and infectious

diseases, were listed in the leading causes of death once unknown and ill-defined

causes were removed (Table 40).

Malaria was the leading CoD reported in both the hospital separation data and health

facility reporting for 5-14 year olds. It ranked second in the medical certificate data,

with meningitis ranked first. External causes of death, drowning and burns (exposure

to smoke fire or flames), appeared in the leading five causes of death for this age

group reported through the health facilities, but did not rank highly in the hospital

data (Table 41).

��

�

301

Tabl

e 40

: Dea

ths,

pro

port

iona

l mor

talit

y an

d ca

use-

spec

ific

mor

talit

y ra

tes

(by

ICD

v10

Gen

eral

Mor

talit

y Li

st 3

) for

chi

ldre

n 0-

4 ye

ars:

Van

uatu

200

0-20

07

Cau

se o

f Dea

th

Rep

orte

d D

eath

s P

ropo

rtion

al M

orta

lity

(%) e

xclu

ding

un

know

n an

d ill

-def

ined

cas

es

Hos

pita

lSe

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

lSe

para

tion

Dat

a

Med

ical

Cer

tific

ates

H

ealth

Faci

lity

Rep

ortin

g 3-

002

Dia

rrho

ea a

nd g

astro

ente

ritis

of

pres

umed

infe

ctio

us o

rigin

4

823

2.8

2.2

7.4

3-00

4 Tu

berc

ulos

is

1

10.

00.

30.

33-

005

Teta

nus

2

21.

40.

00.

63-

008

Men

ingo

cocc

al in

fect

ion

11

0.

70.

30.

03-

009

Sep

ticae

mia

3

3

2.1

0.8

0.0

3-01

3 O

ther

vira

l dis

ease

s 1

2

0.7

0.0

0.6

3-01

4 M

alar

ia

75

294.

91.

39.

43-

015

Rem

aind

er o

f cer

tain

infe

ctio

us a

nd

para

sitic

dis

ease

s

31

0.0

0.8

0.3

3-01

8 R

emai

nder

of m

alig

nant

neo

plas

ms

12

20.

70.

50.

63-

019

Rem

aind

er o

f neo

plas

ms

10.

00.

00.

33-

021

Ana

emia

s

12

0.0

0.3

0.6

3-02

2 R

emai

nder

of d

isea

ses

of th

e bl

ood

and

bloo

d-fo

rmin

g or

gans

and

cer

tain

di

sord

ers

invo

lvin

g th

e im

mun

e m

echa

nism

11

0.

70.

30.

0

3-02

4 M

alnu

tritio

n an

d ot

her n

utrit

iona

l de

ficie

ncie

s 4

128

2.8

3.2

2.6

3-02

5 R

emai

nder

of e

ndoc

rine,

nut

ritio

nal

and

met

abol

ic d

isea

ses

60.

00.

01.

9

3-02

7 M

enin

gitis

2

26

1.4

0.5

1.9

3-02

8 R

emai

nder

of d

isea

ses

of th

e ne

rvou

s sy

stem

5

31

3.5

0.8

0.3

��

�

302

3-02

9 D

isea

ses

of th

e ea

r and

mas

toid

pr

oces

s1

0.7

0.0

0.0

3-03

0 D

isea

ses

of th

e ci

rcul

ator

y sy

stem

3

86

2.1

2.2

1.9

3-03

2 P

neum

onia

10

1751

6.9

4.6

16.5

3-03

3 O

ther

acu

te re

spira

tory

infe

ctio

ns

12

80.

70.

52.

63-

034

Rem

aind

er o

f dis

ease

s of

the

resp

irato

ry s

yste

m

24

41.

41.

11.

3

3-03

5 D

isea

ses

of th

e di

gest

ive

syst

em

24

41.

41.

11.

33-

036

Dis

ease

s of

the

geni

tour

inar

y sy

stem

1

1

0.7

0.0

0.3

3-03

8 Fe

tus

and

new

born

affe

cted

by

mat

erna

l fac

tors

and

by

com

plic

atio

ns

of p

regn

ancy

, lab

our a

nd d

eliv

ery

341

402.

111

.012

.9

3-03

9 D

isor

ders

rela

ting

to le

ngth

of

gest

atio

n an

d fe

tal g

row

th

3484

2423

.622

.67.

8

3-04

0 B

irth

traum

a

1

0.0

0.0

0.3

3-04

1 In

traut

erin

e hy

poxi

a an

d bi

rth a

sphy

xia

1655

511

.114

.81.

63-

042

Res

pira

tory

dis

tress

of n

ewbo

rn

15

10.

04.

00.

33-

043

Con

geni

tal p

neum

onia

1

0.0

0.3

0.0

3-04

4 O

ther

resp

irato

ry c

ondi

tions

of

new

born

10

78

6.9

1.9

2.6

3-04

5 B

acte

rial s

epsi

s of

new

born

2

194

1.4

5.1

1.3

3-04

7 H

aem

orrh

agic

and

hae

mat

olog

ical

di

sord

ers

of fe

tus

and

new

born

1

44

0.7

1.1

1.3

3-04

8 R

emai

nder

of p

erin

atal

con

ditio

ns

56

283.

51.

69.

13-

050

Con

geni

tal h

ydro

ceph

alus

and

spi

nal

bifid

a 1

72

0.7

1.9

0.6

3-05

1 O

ther

con

geni

tal m

alfo

rmat

ions

of t

he

nerv

ous

syst

em

1

0.

70.

00.

0

3-05

2 C

onge

nita

l mal

form

atio

ns o

f the

hea

rt 3

91

2.1

2.4

0.3

3-05

4 D

own'

s sy

ndro

me

and

othe

r ch

rom

osom

al a

bnor

mal

ities

3

81

2.1

2.2

0.3

��

�

303

3-05

5 O

ther

con

geni

tal m

alfo

rmat

ions

5

1715

3.5

4.6

4.9

3-05

8 O

ther

sym

ptom

s, s

igns

and

abn

orm

al

clin

ical

and

labo

rato

ry fi

ndin

gs, n

ot

else

whe

re c

lass

ified

28

90

3-05

9 A

ll ot

her d

isea

ses

23

21.

40.

80.

63-

061

Tran

spor

t acc

iden

ts

13

10.

70.

80.

33-

062

Acc

iden

tal d

row

ning

and

sub

mer

sion

4

0.0

1.1

0.0

3-06

4 E

xpos

ure

to s

mok

e, fi

re a

nd fl

ames

2

61

1.4

1.6

0.3

3-06

5 A

ccid

enta

l poi

soni

ng b

y an

d ex

posu

re

to n

oxio

us s

ubst

ance

s 1

15

0.7

0.3

1.6

3-06

6 A

ssau

lt

5

0.0

0.0

1.6

3-06

7 A

ll ot

her e

xter

nal c

ause

s 3

53

2.1

1.3

1.0

��

�

304

Tabl

e 41

: Dea

ths,

pro

port

iona

l mor

talit

y an

d ca

use-

spec

ific

mor

talit

y ra

tes

(by

ICD

v10

Gen

eral

Mor

talit

y Li

st O

ne) f

or

child

ren

5-14

yea

rs: V

anua

tu 2

000-

2007

R

epor

ted

Dea

ths

Pro

porti

onal

Mor

talit

y (%

) exc

ludi

ng

unkn

own

and

ill-d

efin

ed c

ases

H

ospi

tal

Sepa

ratio

nD

ata

Med

ical

Cer

tific

ates

H

ealth

Faci

lity

Rep

ortin

g

Hos

pita

lSe

para

tion

Dat

a

Med

ical

Cer

tific

ates

H

ealth

Faci

lity

Rep

ortin

g 1-

003

Dia

rrho

ea a

nd g

astro

ente

ritis

of

pres

umed

infe

ctio

us o

rigin

1

0.0

0.0

1.8

1-00

5 R

espi

rato

ry tu

berc

ulos

is

11

13.

12.

81.

8 1-

006

Oth

er tu

berc

ulos

is

1

3.

10.

00.

0 1-

008

Teta

nus

10.

00.

01.

8 1-

012

Sep

ticae

mia

2

2

6.3

5.6

0.0

1-02

1 M

alar

ia

45

1012

.513

.918

.2

1-04

2 M

alig

nant

neo

plas

m o

f men

inge

s, b

rain

an

d ot

her p

arts

of c

entra

l ner

vous

sys

tem

2

1

6.3

0.0

1.8

1-04

3 N

on-H

odgk

in's

lym

phom

a 1

1

3.1

2.8

0.0

1-04

5 Le

ukae

mia

s

1 1

0.0

2.8

1.8

1-04

6 R

emai

nder

of m

alig

nant

neo

plas

ms

11

43.

12.

87.

3 1-

049

Ana

emia

s 1

3

3.1

0.0

5.5

1-05

0 R

emai

nder

of d

isea

ses

of th

e bl

ood

and

bloo

d-fo

rmin

g or

gans

and

cer

tain

di

sord

ers

invo

lvin

g th

e im

mun

e m

echa

nism

2

0.

05.

60.

0

1-05

3 M

alnu

tritio

n 3

1

9.4

0.0

1.8

1-05

4 R

emai

nder

of e

ndoc

rine,

nut

ritio

nal a

nd

met

abol

ic d

isea

ses

1

3.

10.

00.

0

1-05

9 M

enin

gitis

3

6 1

9.4

16.7

1.8

1-06

1 R

emai

nder

of d

isea

ses

of th

e ne

rvou

s sy

stem

1

1 4

3.1

2.8

7.3

��

�

305

1-06

3 D

isea

ses

of th

e ea

r and

mas

toid

pr

oces

s

1

0.0

2.8

0.0

1-06

7 Is

chae

mic

hea

rt di

seas

es

1

3.

10.

00.

0 1-

068

Oth

er h

eart

dise

ases

2

1 6

6.3

2.8

10.9

1-

069

Cer

ebro

vasc

ular

dis

ease

s 1

3.1

0.0

0.0

1-07

1 R

emai

nder

of d

isea

ses

of th

e ci

rcul

ator

y sy

stem

1

3.1

0.0

0.0

1-07

4 P

neum

onia

1

2

3.1

0.0

3.6

1-07

5 O

ther

acu

te lo

wer

resp

irato

ry in

fect

ions

1

0.0

2.8

0.0

1-07

6 C

hron

ic lo

wer

resp

irato

ry d

isea

ses

1

20.

02.

83.

6 1-

080

Dis

ease

s of

the

liver

1 3

0.0

2.8

5.5

1-08

1 R

emai

nder

of d

isea

ses

of th

e di

gest

ive

syst

em

10.

00.

01.

8

1-08

3 D

isea

ses

of th

e m

uscu

losk

elet

al

syst

em a

nd c

onne

ctiv

e tis

sue

2

0.

05.

60.

0

1-08

6 R

emai

nder

of d

isea

ses

of th

e ge

nito

urin

ary

syst

em

11

3.

12.

80.

0

1-09

3 C

onge

nita

l mal

form

atio

ns,

defo

rmat

ions

and

chr

omos

omal

ab

norm

aliti

es

31

9.

42.

80.

0

1-09

4 Sy

mpt

oms,

sig

ns a

nd a

bnor

mal

cl

inic

al a

nd la

bora

tory

find

ings

, not

el

sew

here

cla

ssifi

ed

31

18

1-09

6 Tr

ansp

ort a

ccid

ents

2

0.0

5.6

0.0

1-09

7 Fa

lls

11

13.

12.

81.

8 1-

098

Acc

iden

tal d

row

ning

and

sub

mer

sion

2 4

0.0

5.6

7.3

1-09

9 E

xpos

ure

to s

mok

e, fi

re a

nd fl

ames

1 4

0.0

2.8

7.3

1-10

2 A

ssau

lt

1

0.0

0.0

1.8

1-10

3 A

ll ot

her e

xter

nal c

ause

s

1 3

0.0

2.8

5.5

TO

TAL

3537

73

� � �

306

The leading causes of deaths for males aged 15-59 were heart diseases, liver

diseases and diabetes. Differences in specific cause distribution between other heart

diseases (ischaemic heart disease and cerebrovascular diseases) (Table 42)

between data sources in this age group possibly reflect both more accurate reporting

by doctors on the medical certificate, and greater likelihood of these deaths being

recorded as non-specific heart disease in health facility reports by a nursing aide

rather than reflecting a difference in underlying disease. For females in this age

group, leading causes of death also included heart diseases, liver disease and

diabetes; with cervical and breast cancer also ranking highly in both the hospital and

health facility data (Table 42). As with ages 15-59, chronic NCDs were the leading

causes of death in older adults (Appendix four).

��

�

307

Tabl

e 42

: Dea

ths

and

prop

ortio

nal m

orta

lity

(exc

ludi

ng il

l-def

ined

cau

ses)

by

caus

e (IC

Dv1

0 G

ener

al M

orta

lity

List

1) f

or

adul

ts 1

5-59

yea

rs: V

anua

tu 2

000-

2007

Caus

e�of

�dea

th�

MAL

ES�

FEM

ALES

�

Repo

rted

�Dea

ths

Prop

ortio

nal�M

orta

lity�

(%)*

��Re

port

ed�D

eath

sPr

opor

tiona

l�Mor

talit

y�(%

)*��

Hosp

MC�

HF�

Hosp

�M

C�HF

�Ho

spM

C�HF

�Ho

sp�

MC�

HF�

1�00

1�Infectious�Diseases�

2141

329.3

9.8�

9.6

2227

2310.1

9.1

8.4�

1�00

3�Di

arrh

oea�

and�

gast

roen

terit

is�of

�pr

esum

ed�in

fect

ious

�orig

in�

��2

30

0.5�

0.9

��1

10

0.3

0.4�

1�00

4�O

ther

�inte

stin

al�in

fect

ious

�dise

ases

1��

��0.

40�

01

11

0.5

0.3

0.4�

1�00

5�Re

spira

tory

�tube

rcul

osis�

34

91.

31�

2.7

15

80.

51.

72.

9�1�

006�

Oth

er�tu

berc

ulos

is�1

31

0.4

0.7�

0.3

43

��1.

81

0�1�

008�

Teta

nus�

��1

��0

0.2�

0��

��

00

0�1�

012�

Sept

icae

mia

�4

71

1.8

1.7�

0.3

77

��3.

22.

40�

1�01

9�Vi

ral�h

epat

itis�

620

72.

74.

8�2.

14

45

1.8

1.4

1.8�

1�02

0�Hu

man

�imm

unod

efic

ienc

y�vi

rus�

[HIV

]�dise

ase�

��1

��0

0.2�

0��

��

00

0�

1�02

1�M

alar

ia�

42

101.

80.

5�3

45

71.

81.

72.

5�

1�02

5�Re

mai

nder

�of�c

erta

in�in

fect

ious

�an

d�pa

rasit

ic�d

iseas

es�

21

10.

90.

2�0.

31

11

0.5

0.3

0.4�

1�02

6�Neo

plasms�

3548

4715.5

11.5�

14.1

6691

7730.3

31.5

28�

1�02

7�M

alig

nant

�neo

plas

m�o

f�lip

,�ora

l�ca

vity

�and

�pha

rynx

��

��1

00�

0.3

1��

10.

50

0.4�

1�02

8�M

alig

nant

�neo

plas

m�o

f�oe

soph

agus

��

1��

00.

2�0

11

10.

50.

30.

4�

1�02

9�M

alig

nant

�neo

plas

m�o

f�sto

mac

h�3

4��

1.3

1�0

12

30.

50.

71.

1�1�

030�

Mal

igna

nt�n

eopl

asm

�of�c

olon

,��

12

00.

2�0.

6��

��2

00

0.7�

��

�

308

rect

um�a

nd�a

nus�

1�03

1�M

alig

nant

�neo

plas

m�o

f�liv

er�a

nd�

intr

ahep

atic

�bile

�duc

ts�

1020

144.

54.

8�4.

25

38

2.3

12.

9�

1�03

2�M

alig

nant

�neo

plas

m�o

f�pan

crea

s�1

��

0.4

0�0

��1

��0

0.3

0�1�

033�

Mal

igna

nt�n

eopl

asm

�of�l

aryn

x��

��2

00�

0.6

��

��0

00�

1�03

4�M

alig

nant

�neo

plas

m�o

f�tra

chea

,�br

onch

us�a

nd�lu

ng�

45

21.

81.

2�0.

62

3��

0.9

10�

1�03

6�M

alig

nant

�neo

plas

m�o

f�bre

ast�

��1

��0

0.2�

09

2710

4.1

9.4

3.6�

1�03

7�M

alig

nant

�neo

plas

m�o

f�cer

vix�

uter

i��

��

00�

013

207

5.9

72.

5�

1�03

8�M

alig

nant

�neo

plas

m�o

f�oth

er�a

nd�

unsp

ecifi

ed�p

arts

�of�u

teru

s��

��

00�

03

49

1.4

1.4

3.3�

1�03

9�M

alig

nant

�neo

plas

m�o

f�ova

ry�

��

30

0�0.

93

24

1.4

0.7

1.5�

1�04

1�M

alig

nant

�neo

plas

m�o

f�bla

dder

��

��

00�

0��

1��

00.

30�

1�04

2�M

alig

nant

�neo

plas

m�o

f�men

inge

s,�

brai

n�an

d�ot

her�p

arts

�of�c

entr

al�

nerv

ous�s

yste

m�

12

��0.

40.

5�0

14

��0.

51.

40�

1�04

3�N

on�H

odgk

in's�

lym

phom

a�1

3��

0.4

0.7�

0��

��

00

0�

1�04

4�M

ultip

le�m

yelo

ma�

and�

mal

igna

nt�

plas

ma�

cell�

neop

lasm

s��

��

00�

0��

1��

00.

30�

1�04

5�Le

ukae

mia

s�3

41

1.3

1�0.

36

83

2.7

2.8

1.1�

1�04

6�Re

mai

nder

�of�m

alig

nant

�ne

opla

sms�

107

224.

51.

7�6.

616

1429

7.3

4.9

10.5

�

1�04

7�Re

mai

nder

�of�n

eopl

asm

s�2

��

0.9

0�0

5��

��2.

30

0�1�04

8�Diseases�of�the

�blood

�and

�blood

�form

ing�

organs�

46

41.7

1.5�

1.2

21

11

0.3

0.4�

1�04

9�An

aem

ias�

32

41.

30.

5�1.

21

��1

0.5

00.

4�

1�05

0�Re

mai

nder

�of�d

iseas

es�o

f�the

�bl

ood�

and�

bloo

d�fo

rmin

g�or

gans

�1

4��

0.4

1�0

11

��0.

50.

30�

��

�

309

and�

cert

ain�

diso

rder

s�inv

olvi

ng�th

e�im

mun

e�m

echa

nism

�1�051�En

docrine,�nutritiona

l�and

�metab

olic�

diseases�

��16

4119

7.1

9.9�

5.7

2928

3113.3

9.8

11.3�

1�05

2�Di

abet

es�m

ellit

us�

1640

187.

19.

7�5.

428

2829

12.8

9.8

10.5

�1�

053�

Mal

nutr

ition

��

11

00.

2�0.

31

��1

0.5

00.

4�

1�05

4�Re

mai

nder

�of�e

ndoc

rine,

�nu

triti

onal

�and

�met

abol

ic�d

iseas

es�

��

��0

0�0

��

10

00.

4�

1�055�Men

tal�D

isorde

rs�

11

00.4

0.2�

01

00

0.5

00�

1�05

6�M

enta

l�and

�beh

avio

ural

�diso

rder

s�du

e�to

�psy

choa

ctiv

e�su

bsta

nce�

use�

1��

��0.

40�

01

��

0.5

00�

1�05

7�Re

mai

nder

�of�m

enta

l�and

�be

havi

oura

l�diso

rder

s��

1��

00.

2�0

��

��0

00�

1�05


�Nervous�System�

511

162.2

2.6�

4.8

36

81.4

22.9�

1�05

9�M

enin

gitis

�1

31

0.4

0.7�

0.3

��3

10

10.

4�1�

060�

Alzh

eim

er's�

dise

ase�

��

10

0�0.

3��

��

00

0�

1�06

1�Re

mai

nder

�of�d

iseas

es�o

f�the

�ne

rvou

s�sys

tem

�4

713

1.8

1.7�

3.9

33

71.

41

2.5�

1�06

2�Di

seas

es�o

f�the

�eye

�and

�adn

exa�

��

10

0�0.

3��

��

00

0�

1�06

3�Di

seas

es�o

f�the

�ear

�and

�mas

toid

�pr

oces

s��

1��

00.

2�0

��

��0

00�

1�064�Diseases�of�the

�circ

ulatory�system

�64

119

8928.5

28.8�

26.6

3446

3915.5

1614

.2�

1�06

5�Ac

ute�

rheu

mat

ic�fe

ver�a

nd�c

hron

ic�

rheu

mat

ic�h

eart

�dise

ases

�3

7��

1.3

1.7�

07

13��

3.2

4.5

0�

1�06

6�Hy

pert

ensiv

e�di

seas

es�

75

23.

11.

2�0.

62

24

0.9

0.7

1.5�

1�06

7�Is

chae

mic

�hea

rt�d

iseas

es�

1551

26.

712

.3�

0.6

47

��1.

82.

40�

1�06

8�O

ther

�hea

rt�d

iseas

es�

2633

7111

.68�

21.2

910

244.

13.

58.

7�

��

�

310

1�06

9�Ce

rebr

ovas

cula

r�dise

ases

�13

2114

5.8

5.1�

4.2

1212

115.

54.

24�

1�07

1�Re

mai

nder

�of�d

iseas

es�o

f�the

�ci

rcul

ator

y�sy

stem

��

2��

00.

5�0

��2

��0

0.7

0�

1�072�Re

spira

tory�Diseases�

1934

248.5

8.2�

7.2

1618

197.3

6.2

7�1�

073�

Influ

enza

��

��

00�

0��

1��

00.

30�

1�07

4�Pn

eum

onia

�7

25

3.1

0.5�

1.5

42

41.

80.

71.

5�

1�07

5�O

ther

�acu

te�lo

wer

�resp

irato

ry�

infe

ctio

ns�

��3

10

0.7�

0.3

2��

10.

90

0.4�

1�07

6�Ch

roni

c�lo

wer

�resp

irato

ry�d

iseas

es�

1021

184.

55.

1�5.

49

1314

4.1

4.5

5.1�

1�07

7�Re

mai

nder

�of�d

iseas

es�o

f�the

�re

spira

tory

�syst

em�

28

��0.

91.

9�0

12

��0.

50.

70�

1�07


�digestiv

e�system

�23

3346

10.3

7.9�

13.7

1015

194.6

5.2

6.9�

1�07

9�Ga

stric

�and

�duo

dena

l�ulc

er�

��1

20

0.2�

0.6

��

��0

00�

1�08

0�Di

seas

es�o

f�the

�live

r�21

2940

9.4

7�11

.97

1016

3.2

3.5

5.8�

1�08

1�Re

mai

nder

�of�d

iseas

es�o

f�the

�di

gest

ive�

syst

em�

23

40.

90.

7�1.

23

53

1.4

1.7

1.1�

1�08


�skin�and

�subcutan

eous�tissue

�2

34

0.9

0.7�

1.2

11

40.5

0.3

1.5�

1�08


�musculoskeletal�

system

�and

�con

nective�tissue�

��4

10

1�0.3

2��

��0.9

00�

1�084�Diseases�of�the

�gen

itourinary�tract�

1421

106.3

5.1�

323

2111

10.5

7.3

4�

1�08

5�Gl

omer

ular

�and

�rena

l�tu

bulo

inte

rstit

ial�d

iseas

es�

��

10

0�0.

3��

12

00.

30.

7�

1�08

6�Re

mai

nder

�of�d

iseas

es�o

f�the

�ge

nito

urin

ary�

syst

em�

1421

96.

35.

1�2.

723

209

10.5

73.

3�

1�087�Materna

l�deaths�

00

00

0�0

713

283.2

4.5

10.2�

1�08

8�Pr

egna

ncy�

with

�abo

rtiv

e�ou

tcom

e��

��

00�

01

31

0.5

10.

4�

��

�

311

1�08

9�O

ther

�dire

ct�o

bste

tric

�dea

ths�

��

��0

0�0

48

181.

82.

86.

5�1�

090�

Indi

rect

�obs

tetr

ic�d

eath

s��

��

00�

02

29

0.9

0.7

3.3�

1�09

3�Co

ngen

ital�m

alform

ations,�

deform

ations�and

�chrom

osom

al�

abno

rmalities�

��

10

0�0.3

��2

��0

0.7

0�

1�09

4�Symptom

s,�signs�a

nd�abn

ormal�

clinical�and

�labo

ratory�find

ings,�

not�e

lsew

here�classified

�4

2274

��

��7

1054

��

��

1�095�External�cau

ses�

2052

428.8

12.6�

12.6

317

151.4

5.8

5.6�

1�09

6�Tr

ansp

ort�a

ccid

ents

�9

83

41.

9�0.

9��

1��

00.

30�

1�09

7�Fa

lls�

42

41.

80.

5�1.

2��

��

00

0�

1�09

8�Ac

cide

ntal

�dro

wni

ng�a

nd�

subm

ersio

n�3

99

1.3

2.2�

2.7

��

30

01.

1�

1�09

9�Ex

posu

re�to

�smok

e,�fi

re�a

nd�fl

ames

��

��0

0�0

��2

10

0.7

0.4�

1�10

0�Ac

cide

ntal

�poi

soni

ng�b

y�an

d�ex

posu

re�to

�nox

ious

�subs

tanc

es�

24

60.

91�

1.8

��5

10

1.7

0.4�

1�10

1�In

tent

iona

l�sel

f�har

m�

��5

40

1.2�

1.2

1��

40.

50

1.5�

1�10

2�As

saul

t�1

87

0.4

1.9�

2.1

24

50.

91.

41.

8�1�

103�

All�o

ther

�ext

erna

l�cau

ses�

116

90.

43.

9�2.

7��

51

01.

70.

4��

Unk

nown�

��4

��

��

��5

��

��

��TO

TAL�

228

440

409

��

��22

630

132

9��

��

312

Discussion

For deaths amongst 0-4 year olds, all sources of data reflect the importance of

neonatal causes of death, demonstrating the importance of antenatal care,

appropriate delivery services and improved maternal health in reducing mortality in

this age group. Infectious diseases (pneumonia, malaria and diarrhoea) represent a

significant proportion of deaths in this age group, as reported by all sources. Further

investment in Integrated Management of Childhood Illnesses (IMCI) programs, as

recommended by the World Health Organisation to target these areas at a

community level, will be necessary if Vanuatu is to show considerable progress

towards the MDGs to reduce infant and childhood mortality by 2015. For 5-14 year

olds, the differences in proportional mortality reflects the importance of external

causes (such as drowning and burns) and the high proportion of non-specific causes

recorded in health facility data

NCDs were the leading causes of death for adults in both the 15-59 year age group

and 60+ age group. For males, these were predominantly cardiovascular diseases,

liver diseases and diabetes; while for females, neoplasms were also important, with

both cervical and breast cancer appearing in the leading causes of death.

As data were collected in 2008, the estimates presented here are only for 2001-

2007. As Vanuatu continues to improve data collection and further data becomes

available, it will be important to repeat this analysis for later years. There has been

significant investment in both reproductive health programs and malaria control in

Vanuatu since these data were collected (62). As a consequence, proportional

mortality from neonatal causes and malaria could reasonably be expected to have

313

reduced, although it is anticipated that they will remain important causes of death for

the foreseeable future.

Hospital separation data would normally have the principal diagnosis recorded first,

but this field was recorded as the mode of death (cardio-respiratory arrest, heart

stopped, stopped breathing etc) in many of the cases reviewed, indicating that

tabulation based on the first field of the database would have been misleading.

Although it is not possible to provide national estimates of cause-specific mortality

rates due to the uncertainty in existing data collections, the data presented in this

paper clearly demonstrates Vanuatu is dealing with a double burden of disease, with

significant proportions of mortality attributed to both infectious diseases and NCDs.

This is further supported by the high proportion of deaths in females 15-59 years

attributed to maternal causes. While the high proportion of NCDs in adults from the

health facility data may in part be attributed to reporting of non-specific causes such

as “heart problem” and “heart-stopped”, that this pattern is reflected in the hospital

separation and medical certificate data adds to the evidence that mortality burden

from NCDs is significant. The high proportion of deaths attributed to NCDs such as

cardiovascular disease and diabetes in adults aged 15-59 is consistent with patterns

seen in other PICTs (such as Fiji, Nauru and Tonga (93, 112, 347)), where NCDs

have contributed to high premature adult mortality and subsequently limited

improvements in LE.

The high proportion of ‘other cancers’ reported in adults is (at least in part)

attributable to a high occurrence of thyroid cancer, consistent with published

incidence data for the 1980’s which suggested Vanuatu had one of the highest

314

occurrences of this type of cancer in the world (348). Previous studies also report a

higher incidence of cervical cancer and high-grade epithelial abnormalities (2.0%) on

screening than Australia (349), which is consistent with findings here that show

cervical cancer as an important CoD in adult females. Screening is however very

limited.

The high proportion of infant and childhood deaths attributed to premature birth or

low gestational weight for age found in this study is somewhat inconsistent with the

relatively low proportions of children born (regardless of outcome) in these

categories reported for 1982-2001 for Vila hospital. Thirty seven out of 1,000

deliveries were reported as premature during this time, with 45 per 1,000 deliveries

reported as small for age (350). This earlier study does report difficulties in

ascertaining gestational age and suggests a great deal of uncertainty in these

figures, with results suggesting either implausibly low capture of premature and low

weight births in hospital records, or high overall mortality rates within these high risk

infants (or both).

These findings highlight the need for more reliable data on CoD, particularly for

deaths that occur in the community, outside the hospitals. Despite a national

reporting system through community health facilities, these account for only 26% of

expected deaths nationally, far lower than the proportion of deaths that would occur

in the community. It is clear from the data presented here that deaths from external

causes are under-represented in the hospital data, and as such may not have been

adequately considered in public health priorities. Similarly, meningitis and malaria

appear more frequently in the community health facility reports than in the hospital

data. This may be due to a greater awareness of the importance of these diseases in

315

the community, resulting in better recognition (and subsequently reporting) than for

other causes of death. However, these data pre-date the large scale malaria

intervention in Vanuatu implemented over the last several years.

Despite data shortcomings, findings presented here provide national level CoD data

that will assist health planners and donor agencies ensure health programs can be

more readily targeted to the greatest areas of need.

6.4.2 Cause of death analysis from hospital, community nursing and civil

registry data in Kiribati

Aim and Objectives

Problems such as malnutrition, injuries and infectious diseases have been identified

as health priorities in Kiribati by the Ministry of Health and international organisations

such as the World Health Organisation (WHO) (108) and UNICEF (351), although

there has to date been limited information available to quantify the extent of these

health problems. As presented in Chapter 5, deaths are substantially under-reported

in Kiribati. Although medical certificates were sought from both medical records and

from each of the hospital ward, very few could be located with the only complete

data available coming from 2008 in Betio hospital (a minor health facility). Since

there were insufficient data for medical certificates of death to be used as a reliable

data source, and they were excluded from further analysis.

Methods

As CoD data could not be reliably extracted from medical certificates, both hospital

and civil registration data were examined for 2000 to 2008. Data were extracted

316

from: the database of hospital admissions administered by the medical records

section of the major hospital; the database of deaths from community health clinics;

and the database maintained by the Civil Registration Office. Stillbirths were

removed from the extracted records.

CoD as recorded in the civil registration data was translated from i-Kiribati to English

with the assistance of staff from the MoH, and coded to the underlying cause using

ICDv10 rules (68). CoD as reported through the Civil Registration Office is of limited

value, as it is derived from family reports and frequently consists of terms such as “e

boo” meaning “sudden death” or possibly “stroke” and “aoraki” meaning “sickness”,

along with English terms such as old age and senility; making attribution of cause

unreliable. The notable exception is for deaths due to external causes such as

suicide, vehicle accidents and drowning, where the cause is more immediate and

requires less interpretation. For injury related deaths, the data were also coded for

intent (intentional, unintentional or unknown) and whether the death was reported as

suicide.

Data were then tabulated by age group, gender and CoD. Similarly, hospital data

were tabulated by age group, ICD code of the principal diagnosis, and gender.

Cause, as proportional mortality by age group, is reported separately from the

hospital data and the civil registration data, as it was not possible to reconcile the

two sources from the de-identified data available. It is anticipated there would be

some overlap between these sources. Estimated age-specific mortality rates have

been calculated by applying proportional mortality from this study to the mortality

envelope (all-cause mortality) established in Chapter five for key causes of death of

public health importance. For deaths from external causes, age-specific mortality

317

rates were estimated as a ‘minimum rate’ and were not adjusted for either under-

reporting or “ill-defined” and “unknown” causes, as deaths from injury are often less

under-reported than those from other causes.

Cause is primarily reported by ICD chapter due to the lack of detail available and

small numbers of deaths given the population size. It is broken into more specific

categories (using ICDv10 General Mortality List One) when sufficient data were

available. Confidence intervals were calculated using Poisson distribution due to the

small number of events.

Results

Civil registration reported 2,639 deaths between 2000 and 2004 and 2,026 deaths

between 2005 and 2008. There were 263 deaths recorded in the hospital data from

2000-2004 and 652 deaths recorded from 2005-2008. A further 1,324 deaths were

recorded by the community health clinics in 2000-2004. For 2005 to 2008, there

were 65 males and 48 females for whom age group was not known, and 24 deaths

for which gender was not specified (one an adult 15-59 years, one older than 60

years and the remainder of unknown age). Although redistributed for analysis of

mortality level, these were excluded from further analysis for cause.

Proportional mortality from 2005 to 2008 by gender and broad age group is shown in

Figures 33-36 for deaths recorded through the MoH (including hospital and health

facility reporting), with both health data and civil registration data shown in Tables 43

and 44. A substantial proportion of reported deaths could not be attributed to a

cause, either due to no cause being reported, the death being coded to ‘signs and

symptoms’ or ‘ill-defined’, or in a small number of cases the condition or terminology

318

listed being indecipherable. The proportion of deaths reported through the civil

registration data not be attributed to a cause in 2005-2008 ranged from 25% for

males aged 15-59 years to 65% for males aged less than 1 year. Deaths reported

through the clinics had less than 20% of deaths fall into this category, with the

exception of females aged 15-59 where 23% could not be attributed to a cause.

Deaths occurring in hospital that could not be attributed to a cause ranged from 0%

up to 7.7% for males aged 5-14 years (Table 43).

Figure 33: Proportional mortality in Kiribati 2005-2008 for children 0-4 years, excluding ill-defined causes, as reported through the Ministry of Health

(hospital and community health facilities)

319

Figure 34: Proportional mortality in Kiribati 2005-2008 for children 5-14 years, excluding ill-defined causes, as reported through the Ministry of Health


Figure 35: Proportional mortality in Kiribati 2005-2008 for males 15-59 years, excluding ill-defined causes, as reported through the Ministry of Health


320

Figure 36: Proportional mortality in Kiribati 2005-2008 for females 15-59 years, excluding ill-defined causes, as reported through the Ministry of Health


321

Table 43: Proportion of ill-defined or unknown causes and adjusted proportional mortality (%) by age group and period: Kiribati males, 2000-2008

Age�Cat.�� Cause�of�Death�

MALES�

2000�2004� 2005�2008�

Civil�Reg.�Combined�

Health�Deaths�

Hospital�Deaths�

Clinic�Deaths� Civil�Reg.�

Combined�Health�Deaths�


Clinic�Deaths�

<�1�Year�Ill�defined�/�Not�stated�or�Unclear� 40.7� 19.4 0.0 22.1 65.2� 13.3� 4.3 18.5

�� Infectious�diseases� 15.2� 10.1 0.0 11.9 0.0� 9.9� 4.4 13.6

��

Endocrine,�nutritional�and�metabolic�diseases� 0.0� 6.3 8.3 6.0 0.0� 11.7� 2.2 18.2

��Diseases�of�the�nervous�system� 12.1� 2.5 8.3 1.5 0.0� 2.7� 4.4 1.5

��Diseases�of�the�circulatory�system� 9.1� 0.0 0.0 0.0 12.5� 0.0� 0.0 0.0

��Diseases�of�the�respiratory�system� 30.3� 24.1 41.7 20.9 25.0� 18.9� 22.2 16.7

��Conditions�originating�in�the�perinatal�period� 21.2� 53.2 33.3 56.7 62.5� 52.3� 62.2 45.5

��Congenital�malformations� 0.0� 0.0 0.0 0.0 0.0� 0.9� 2.2 0.0

�� External�causes� 3.0� 2.5 8.3 1.5 0.0� 2.7� 0.0 4.5�� Other�causes� 6.1� 1.3 0.0 1.5 0.0� 0.9� 2.2 0.0

TOTAL� 100� 100 100 100 100� 100� 100 100

1�4�years�Ill�defined�/�Not�stated�or�Unclear� 40.7� 19.7 0.0 31.6 42.9� 8.7� 0.0 17.9

�� Infectious�diseases� 47.1� 44.9 43.5 46.2 18.8� 33.3� 33.9 32.6�� Neoplasms� 2.9� 0.0 0.0 0.0 6.3� 1.9� 1.7 2.2

��






TOTAL� 100� 100 100 100 100� 100� 100 100

322


MALES�

2000�2004� 2005�2008�


Health�Deaths�





Clinic�Deaths�

5�14�years�

Ill�defined�/�Not�stated�or�Unclear� 20.6� 22.6 0.0 27.3 32.4� 6.9� 7.7 6.3


��Diseases�of�blood�and�blood�forming�organs� 1.9� 7.3 22.2 3.1 8.3� 11.1� 25.0 0.0

��





��Diseases�of�the�digestive�system� 30.8� 19.5 11.1 21.9 8.3� 7.4� 0.0 13.3


TOTAL� 100� 100 100 100 100� 100� 100 10015�59�years�



��




��Diseases�of�the�genitourinary�system� 1.0� 0.9 1.7 0.7 0.7� 1.7� 2.8 1.1

��Pregnancy�and�childbirth� 0.0� 0.0 0.0 0.0 0.0� 0.0� 0.0 0.0


TOTAL� 100� 100 100 100 100� 100� 100 100

323

Table 44: Proportion of ill-defined or unknown causes and adjusted proportional mortality (%) by age group and period: Kiribati females, 2000-2008


FEMALES�

2000�2004� 2005�2008�


Health�Deaths�





Clinic�Deaths�

<�1�Year�Ill�defined�/�Not�stated�or�Unclear� 51.2� 11.4 5.3 13.3 38.9� 13.0� 5.4 16.9

�� Infectious�diseases� 13.6� 12.9 5.6 15.4 0.0� 7.4� 0.0 11.9

��





��Conditions�originating�in�the�perinatal�period� 31.8� 50.0 66.7 44.2 54.5� 47.9� 68.6 35.6

��Congenital�malformations� 0.0� 1.4 0.0 1.9 0.0� 2.1� 0.0 3.4


TOTAL� 100� 100 100 100 100� 100� 100 100

1�4�years�Ill�defined�/�Not�stated�or�Unclear� 51.6� 11.0 0.0 17.0 44.2� 3.8� 2.3 5.7


��






TOTAL� 100� 100 100 100 100� 100� 100 100

324


FEMALES�

2000�2004� 2005�2008�


Health�Deaths�





Clinic�Deaths�

5�14�years�



��Diseases�of�blood�and�blood�forming�organs� 8.3� 0.0 0.0 0.0 0.0� 3.4� 6.7 0.0

��







TOTAL� 100� 100 100 100 100� 100� 100 10015�59�years�



��




��Diseases�of�the�genitourinary�system� 4.5� 6.4 13.6 2.6 3.2� 2.4� 3.8 1.3

��Pregnancy�and�childbirth� 3.6� 5.8 6.8 5.3 0.5� 1.2� 1.9 0.7


TOTAL� 100� 100 100 100 100� 100� 100 100

325

The two major causes of death in children aged less than one from all sources were

conditions arising in the perinatal period and respiratory diseases. In children aged

1-4 years, 34% of the deaths occurring in hospital were attributable to malnutrition as

either the primary diagnosis or a contributing cause, equating to an age-specific

mortality rate of 2.5 per 1,000 in 2000-2004, to 4.5 per 1,000 in 2005-2008. In

children aged 5-14 years of age, external causes account for the greatest proportion

of deaths in both periods, with drowning and vehicle accidents featuring prominently.

A significant proportion of deaths in this age group is also attributed to disease of the

digestive tract

Table 45: Proportion (%) of hospital deaths of children aged 1-4 years where malnutrition was reported as either the primary diagnosis or contributing

factor, Kiribati 2000-2008

Hospital

Gender 2000-2004 2005-2008

Males 47.8 35.6

Females 48.3 32.6

Combined 48.1 34.3

The greatest proportion of deaths in both males and females 15-59 years for 2005 to

2008 was from diseases of the circulatory system (16-29% for females and 17-43%

for males). For males, 21-23% of deaths in both periods (based on civil registration

data and health system data) were attributable to diseases of the digestive system.

Causes in this category include hepatitis, liver disease and bleeding ulcers. For

males, deaths due to external causes were approximately 15% of those reported in

2005-2008, down from 22% in 2000-2004. Suicide was common, with the minimum

age-specific mortality rate for males (without adjusting for completeness or

326

percentage of deaths with cause unknown) calculated as 38.9 deaths per 100,000

population. Endocrine, nutritional and metabolic diseases, including diabetes, made

up only 6-9% of adult male deaths in 2005-2008. These were more important for

females, accounting for 7-13% of deaths in 2000-2004 and 12% of deaths in 2005-

2008. Neoplasms were also higher in females than for males, accounting for 14% of

deaths in 2005-2008. Maternal deaths accounted for 3.6-5.8% of adult female deaths

in 2000-2004, although this had reduced to 0.5-1.2% of deaths in 2005-2008.

Table 46: Minimum suicide rates (per 100,000 people) based on uncorrected data, Kiribati 2000-2008

Gender Intent

Age Group / Years 10-14 years 15-59 years

2000-2004

2005-2008

2000-2004

2005-2008

MaleDefinite suicide

11.4 8.7 51.3 38.9

All possible+ 19.0 21.8 80.3 55.6

FemaleDefinite suicide

4.2 0.0 5.6 1.4

All possible+ 12.5 4.6 6.0 3.2

Combined Definite suicide

7.9 4.5 30.5 20.4

All possible+ 15.8 13.5 44.9 30.4 *Civil registration data - not adjusted for “ill-defined” or “unknown” deaths or completeness of reporting. +Possible suicides are those for which intent was not recorded, including shootings, strangulation which may have been either self inflicted or as the result of assault.

327

Table 47: Estimated Age-specific mortality rate from Diabetes (as reported through civil registration and health system reporting) per 100,000 people

Gender Source Year

2000-2004 2005-2008

MaleCivil

Registration 72.0 81.1

Health 65.5 80.8

FemaleCivil

Registration 39.9 71.3

Health 73.3 59.4

Discussion

Despite the shortcomings of the data available, this study highlights that analysis of

even incomplete data can identify key public issues and provide better evidence for

decision making in health policy development and resource allocation in a setting

such as Kiribati. Although there has been a general acknowledgement amongst the

donor community that issues such as infectious diseases and malnutrition amongst

children are important in Kiribati, the scale of the contribution of malnutrition to

childhood mortality has perhaps been under-recognised, particularly given the

preventable nature of this problem. Additionally, the cause data has highlighted other

key population health issues that have perhaps been overlooked and need greater

attention in order to reduce the excess mortality in Kiribati. These issues include

injury related deaths (and suicide in particular), and digestive tract diseases (possibly

related to Hepatitis B prevalence) in adults.

328

6.5 Summary of method selection and application

As demonstrated in this chapter, both medical record and medical certificate review

provide a useful means of examining content validity of CoD data and correcting for

system errors in certificate completion and tabulation. The medical certificate review

is by necessity more limited, as the researcher has access to less of the original

information captured. As shown for Fiji, while this is useful to understanding data

limitations and interpreting estimates of CoD distributions, the medical certificate

review alone is unable to adequately correct for these concerns where significant

data quality errors are identified that are not accounted for by systemic practices.

Where medical certificates are not available for a representative population, the

studies from Vanuatu and Kiribati have demonstrated that while care must be taken

in generalising findings on CoD, it is possible to obtain a greater understanding of

the broad causes contributing to premature mortality in PICTs through the available

data from hospital, community health, and lay reporting sources. In particular, these

have proven useful in understanding previously under-recognised contribution of

external causes (particularly suicide and traffic accidents) in these communities.

These alternative approaches are however, not a long term substitute for quality

CoD data at a representative population level and cannot provide the detail to guide

planning and evaluation that is possible in Palau, Nauru and Tonga, where nationally

representative data is available. �

329

A summary of the data sources and methods used to assess CoD data and generate

CoD distributions is shown in the table below.

Table 48: Summary of data sources and methods for used in assessment of cause of death patterns for selected Pacific Islands

Country Data Source(s) Used in Final Calculations

Assessments and corrections applied*

Years of dataavailable

Fiji MoH – medical certificates Assessment of trends by broad category / Medical certificate review

2000-2008 (excluding 2004 /2007)

Kiribati MoH and Civil registry – hospital discharge data and community reported CoD

Analysis of broad categories only

2000-2009

Nauru MoH/Civil registry (shared data) derived from medical certificates & hospital records

Assessment of trends by broad category / Medical Record review

2005-2009

Palau MoH (Epidemiology spreadsheet) – tabulated data derived from medical certificates

Direct analysis from medical certificate (including coding and tabulation)

2004-2009 (excl 2007)

Tonga MoH HIS data from medical certificates

Medical record review 2001-2009

Vanuatu Hospital data (medical certificates and separation data) and community health centre reporting

Comparison of hospital and community data for broad categories

2007-2011

A summary of mortality and CoD patterns by age will be presented in the following

chapter for each of the selected countries.

330

7 Mortality patterns in the Pacific Islands

As outlined in the introduction to this thesis, there is a critical need for a greater

understanding of the mortality levels and causes of death contributing to these in the

PICTs. Given the available data, it has been possible to generate estimates of

mortality level for Fiji, Nauru and Palau, and to generate a range of plausible

estimates for Tonga and Kiribati. The plausible ranges presented provide greater

certainty around mortality level for health planning and evaluation. These mortality

estimates are presented in Table 49.

For Vanuatu, although there were insufficient data to use the vital registration

systems from either the MoH or Civil Registry office to obtain plausible estimates, it

has been possible to review existing sources of data such as the census analysis

and multi-variate modelling by Rajaratnam (41) for plausibility based on how these

methods performed in the other study countries. An “upper-limit” for LE (and hence a

331

lower limit for mortality) were generated based on these findings, that provides

greater context for planning than previously available (Chapter 5).

CoD distributions by age and sex were able to be generated for Palau, Nauru and

Tonga based on medically certified deaths. A plausible range for cause-specific

mortality rates was also able to be derived for Vanuatu from MoH data (from hospital

and community reporting sources), although this is based on primary diagnosis

rather than underlying CoD. For these countries, proportional mortality by age and

sex for selected causes from the ICD General Mortality List 1 (103 cause list) (68)

was applied to the mortality level derived in Chapter 5.

For Nauru, Palau and Tonga, confidence intervals for proportional mortality were

derived using a Poisson distribution and applied to the range of mortality estimates

(plausible limits for Tonga as described in Chapter 5, and 95% confidence intervals

based on normal approximation of the binomial for Palau and Nauru). A similar

approach was used in Vanuatu with the range of plausible estimates for proportional

mortality applied to the 95% confidence interval (based on a normal approximation of

the binomial) of the mortality rate.

Data from Kiribati and Fiji were both of insufficient quality to be reliable in deriving

cause-specific mortality rates for most diseases. In Fiji although there is near

universal certification of CoD, coding and tabulation errors are substantial but not

systematic (Chapter 6) and therefore cannot be readily corrected without re-coding

the original medical certificates. For Kiribati, deaths are reported through both the

health and civil registry system; with most causes of death captured through lay

reporting by the family or through nursing reports. For both countries however, it has

332

been possible to provide indicative CoD distributions (proportional mortality) by age

and sex by ICD chapter; with cause-specific rates for specific causes either where

reporting is less subject to diagnostic uncertainty (external causes in Kiribati) or

through multiple-CoD analysis (in Fiji) (Chapter 6).

Estimates of mortality and CoD from the studies in Chapter 5 and 6 are collated in

the remainder of this chapter to provide a picture of mortality and CoD by age group

and by country, with findings compared to previous estimates and assessed for

plausibility.

7.1 Summary measures of mortality Table 49 demonstrates a common pattern of low infant and child mortality and high

adult mortality, resulting in relatively low life expectancies across nearly all of the

study countries. Estimates of LE range from 48.6 years for males in Kiribati to 71.2

years for females in Palau. Males fare substantially worse than females across all

countries with an average LE 4-5 years lower than females, except in Kiribati where

the difference is even greater (even allowing for the range of plausible estimates).

IMR and U5 mortality remain fairly high in Kiribati at 48.9 to 52.9 and 67.3 to 73.6 per

1,000 live births respectively, and in Nauru at 28 and 46 deaths per 1,000 live births

respectively. For all other study countries, IMR was estimated to be below 20 deaths

per 1,000 live births. These figures indicate while there are still gains to be made in

improving child health in the region, infant and child mortality are not the major

influence contributing to the low life expectancies seen across countries.

333

Tabl

e 49

: Sum

mar

y of

key

mea

sure

s of

mor

talit

y fo

r sel

ecte

d Pa

cific

Isla

nds

Cou

ntry

R

efer

ence

Y

ear

IMR

(per

1,

000

live

birth

s)

U5M

(per

1,

000

live

birth

s)

Mal

e ad

ult m

orta

lity

(pro

babi

lity

of d

ying

be

twee

n ag

es 1

5-59

, %)

Fem

ale

adul

t m

orta

lity

(pro

babi

lity

of d

ying

bet

wee

n ag

es 1

5-59

, %)

Mal

e LE

at

birth

(y

ears

)

Fem

ale

LE

at b

irth

(yea

rs)

Fiji

2002

-200

4 19

(CI:

18-

20)

21.3

(CI:

20.1

- 22

.5)

27.6

20

.3

64.3

(CI:

64.0

–

64.6

)

69 (C

I: 68

.8

– 69

.4)

Kiri

bati

2000

-200

9 51

(p

laus

ible

ra

nge

41.8

-71.

3)

71.5

(6

6.8-

76.4

) 60

.0 (p

laus

ible

ra

nge

40.5

-60.

0)

37.4

(pla

usib

le

rang

e 22

.6- 3

7.4)

48

.6

(CI:

47.2

- 48

.3)

57.9

(C

I: 55

.0-

56.3

)

Nau

ru

2002

-200

7 28

(CI:

12-

55)

46.0

51

.0

40.1

52

.8

57.5

Pal

au

2004

-200

9 (e

xcl.

2007

) 16

.7

25

59.6

43

.1

65.8

(CI:

64.4

- 67

.2)

71.2

(CI:

69.9

- 72

.5)

To

nga

2005

-200

8 14

.7 –

24

.7

20.7

– 3

4.8

28.6

– 3

6.3

20.9

- 27.

7 65

.2 (C

I: 64

.6 -

65.8

69

.6 (C

I: 69

.0 –

70.

2)

Van

uatu

(low

er

plau

sibl

e lim

it fo

r mor

talit

y)*

2009

21

24

27

.5 (e

stim

ate)

17

.0 (e

stim

ate)

65

70

95%

Con

fiden

ce in

terv

als

(CI)

give

n in

bra

cket

s w

here

ava

ilabl

e.

* IM

R a

nd U

5M d

eriv

ed fr

om c

ensu

s da

ta (6

1), w

ith a

dult

mor

talit

y fro

m c

o-va

riate

mod

ellin

g by

Raj

arat

nam

et.

al. (

237)

334

The estimates of LE at birth as derived from the work presented in this thesis are

consistently lower than those published circa 2010 by WHO and presented in

Chapter 1 (Table 6). Rather than suggesting a decline in LE over the last few years,

the work presented through the previous chapters provides supporting evidence that

in many cases, previously published estimates, from WHO and other sources (2, 14,

115, 352), have under-estimated adult mortality in the Pacific Islands due to the use

of methods that did not adequately account for observed disparity between child and

adult mortality levels. In comparison, there was much greater similarity with the “best

estimates” identified by Taylor et. al. in 2005 (2) which were vetted for data source

and quality where possible (Table 50).

Table 50: Comparison of ‘Best estimates’ of life expectancy at birth (years) from this research against Taylor et. al. 2005 review (2)

Country Best estimate from 2005 review (2)

Best estimate from work presented in this thesis

Ref Year Male Female Ref Year Male Female Fiji 1996 65 69 2002-2004 64.3 69

Kiribati 1995 59 65 2000-2009 48.6

57.9

Nauru 1991-1993 54 61 2002-2007 52.8 57.5

Palau 1995 64 70 2004-2009 (excl. 2007)

65.8 71.2

Tonga 1997-1999 70 72 2005-2008 65.2 69.6 Vanuatu 1999 66 69 2009-2010 65 70

Despite the difference in time periods, estimates were similar for Fiji, Vanuatu, and

Palau. Estimates of LE for Tonga, Kiribati and Nauru were slightly higher than those

derived from the work in this thesis. In Tonga and Kiribati, the “best estimates”

selected by Taylor et. al. (2) were derived from demographic analysis of the most

recent census; both of which were based on model life tables derived from a single

335

parameter input (child survival) in the absence of any better data sources; and as

such are most likely to be based on an under-estimation of adult mortality (93, 112,

347).

For Nauru, the “best estimates” selected by Taylor et. al (2) were based on a direct

analysis of reported civil registration data that is estimated to be essentially complete

from an earlier time period. As such the difference in estimated LE may represent a

true decline over the last decade, although this is not significant and case study 2

demonstrated there has been no substantial change in LE for males or females in

Nauru over the last 50 years (112).

336

7.2 Mortality and cause of death distributions by age group

7.2.1 Infant and child mortality

As shown in Table 49, infant and childhood mortality varies substantially across the

selected study countries. So too does the CoD patterns associated with the mortality

levels. An overview of causes in this age group is given in the Table 51.

Countries such as Fiji, Palau and Tonga, with low U5M are dominated by perinatal

and congenital causes, indicating mortality is clustered in the first days and weeks of

life, and while some further gains may be possible in association with improvements

in ante and post-natal care; significant resource and technology investment may be

required to realise significant gains at these low levels. Kiribati had a much smaller

proportion of deaths attributed to congenital causes (1%) and perinatal causes

(27%), and more associated with causes such as respiratory illness (19%) and

infection (19%), reflecting continued mortality throughout this age group rather than

being concentrated only in the perinatal period.

Kiribati also had a high proportion (18%) of deaths in children (0-4 years) attributable

to Endocrine, nutritional and metabolic disorders, which in this age groups is

primarily accounted for by malnutrition and ‘electrolyte imbalance’. In comparison, in

Fiji where malnutrition (as part of the ICD chapter ‘Endocrine, nutritional and

metabolic diseases) did not feature as an underlying CoD, the multiple CoD analysis

identified malnutrition was a factor in a small proportion of deaths (<2%), with a

related mortality rate of 6.7 deaths per 100,000 population (95% CI: 0-25.9).

337

Measles, an important CoD in children globally (353), was not found to be a

significant CoD in any of the study countries, reflecting the high immunisation rates

in the region (354). Further investigation is warranted in Kiribati where there was a

high proportion of deaths in children from infectious diseases, but insufficient data

quality to allow disaggregation by cause to this level.

Malaria, which is responsible for 20.8% of deaths in children 1-4 years of age

globally (353) only occurs in Vanuatu of the study countries. Modelled estimates by

Murray and Rosenfeld et. al. in 2010 (355) estimated 1 death (95% CI: 0-3) per year

in children 0-4 years in Vanuatu due to most cases in Vanuatu being due to P.vivax

rather than the more serious P. falciparum. While hospital and community reporting

in Vanuatu (Chapter 6) reported only between 5 and 29 deaths respectively over 8

years (2000-2007), an average of <1 to 4 deaths per year, proportional mortality

ranged from 1.3 to 9.4% of deaths in this age group (depending on source),

indicating Malaria is potentially a greater problem in Vanuatu than previously

identified.

338

Tabl

e 51

: Key

mea

sure

s of

chi

ldho

od m

orta

lity

(0-4

yea

rs),

both

sex

es, f

or s

elec

ted

Paci

fic Is

land

s

IM

R (d

eath

s pe

r 1,0

00

live

birth

s)

U5M

(dea

ths

per 1

,000

live

bi

rths)

Pro

porti

onal

mor

talit

y (0

-4 y

ears

) (%

) R

ef. Y

ear

Infe

ctio

us

Dis

ease

s N

eopl

asm

sD

isea

ses

of th

e ne

rvou

s sy

stem

(p

rimar

ily

Men

ingi

tis)

Per

inat

al

Cau

ses

Con

geni

tal

Cau

ses

Res

pira

tory

Ill

ness

es

Inju

ry a

nd

othe

r ex

tern

al

caus

es

Oth

er

caus

es

Fiji

19 (C

I: 18

-20

) 21

.3 (C

I: 20

.1

- 22.

5)

8 3

4 44

9

10

7 15

20

02-2

004

Nau

ru

51

(pla

usib

le

rang

e 41

.8-

71.3

)

71.5

(C

I: 66

.8-

76.4

)

0 4

4 44

7

15

11

15

2000

-200

9

Kiri

bati

(Hea

lth

data

)

28 (C

I: 12

-55

) 46

.0

19

1 6

27

1 18

4

24

2002

-200

7

Pal

au

16.7

25

0

0 4

64

4 8

4 16

20

04-2

009

(exc

l. 20

07)

Tong

a 14

.7 –

24.

7 20

.7 –

34.

8 14

9

5 36

10

10

7

9 Le

vel:

2005

-20

08.

CO

D:

2001

-200

8

Van

uatu

21

24

6-

18

0.5-

1 1-

5 13

-15

6-11

6-

20

5-6

24-6

3 20

05-2

009

* ad

just

ed to

exc

lude

ill-d

efin

ed a

nd u

nkno

wn

caus

es

95%

Con

fiden

ce in

terv

als

(CI)

show

n

339

Estimates for Oceania from the newly released 2010 Global burden of disease study

(240) show fairly similar point estimates at 2010 as found in this study. U5M was

estimated for Fiji at 30 deaths per 1,000 live births, for Kiribati around 40 deaths per

1,000 live births, and Tonga at 20 deaths per 1,000 live births. Estimates for

Vanuatu, the other study country to be included in the GBD, were substantially

higher than the most recent census at around 32 deaths per 1,000 live births (240),

following closely the findings from the MICs survey which were based on a partial

birth history, and noted the potential for significant responder bias in the sample, with

households where a death had occurred less likely to take part in the survey (157).

The findings from the research in this thesis show significant inequalities in child

survival across the PICTs and indeed across the broader region, with IMR and U5M

in Australia and New Zealand having been stable below 5 deaths per 1,000 live

births for many years. While child mortality is not the primary driving factor behind

the stagnation in LE in the region, there are still significant gains to be made in this

area, and greater attention should be given to addressing those causes of death

which are amenable to intervention.

7.2.2 Adult mortality 15-59 years

Adult mortality was high across all the study countries, as shown in the following

tables, with men having a much greater probability of dying before their 60th birthday

than women. Adult mortality in Fiji and Tonga is nearly three times that in Australia

and New Zealand, with estimates for males in Palau, Kiribati and Nauru nearly

double that again (297, 356), highlighting the profound effect of premature mortality

in the region. As for child mortality, findings from the GBD study (240) are

340

reasonably similar at 2010 to the findings from this thesis (excluding female adult

mortality in Tonga), but show an ongoing decline in adult mortality which is not

consistent with the findings from the work presented in this thesis. The sustained

decline in adult mortality for males shown in Fiji (240) is inconsistent with the

stagnation shown for LE over the last 20 years (93). GBD estimates for female adult

mortality in Fiji show a similar decline over the previous decade although this is now

steady. Again this decline in adult mortality is inconsistent with reliable historical

estimates evaluated in Chapter 5, which show the stagnation in female LE began in

the mid 1980s, with no significant improvement since (93). Trends presented in

Chapter 5 (case study 3) which demonstrate this stagnation in LE are based on

both unadjusted empirical data from the Ministry of Health for 1996-2004 and

previously published estimates censored for reliability to exclude estimates for which

methods could not be verified or which include an assumption of mortality decline in

their methodology (such as estimates from the UN). Estimates of adult mortality

derived from the Ministry of Health data showed very little difference over 3 time

periods, with adult mortality (probability of dying between ages 15-59 inclusive)

estimated at 29.5% and 20.5% for males and females in 1996-1998; 30.5% and

20.8% for males and females in 1991-2001 and 27.6% and 20.3% for males and

females in 2002-2004 (93). This stagnation in LE and high adult mortality is

consistent with risk factor data from Fiji. A national survey of risk factor prevalence in

1980 showed adult prevalence of hypertension as 5.1–9.9% (using �160 mm hg

systolic and/or �95 mm hg diastolic and/or on treatment for hypertension) (357, 358)

while the Fiji STEPS survey in 2002 indicated adult prevalence of hypertension of

19% (using �140 mmhg systolic and /or �90 mmhg diastolic and/or on treatment for

hypertension) (359). Diabetes prevalence has remained at similar levels over this

341

time period, with prevalence reported as 1.1–11.8% (using two hour post 75 g

glucose load blood glucose of �11.1 mm/L and/or on treatment for diabetes) in the

1980 survey, and 12% (using fasting blood glucose of �7.0 mm/l and/or on treatment

for diabetes) in the 2002 steps survey (359). Although smoking prevalence has

declined from 46% of the population smoking in 1992 (NCD book) to 37% in 2002,

as reported in case study 3, cigarette sales in Fiji rose 273% from 1956 to 1984,

three times the population increase (88%) (359).

The modelled decline for Fiji may be a key driver of the more moderate declines in

adult mortality the GBD study portrays in Kiribati, Tonga, and Vanuatu, due to the

model drawing on neighbouring populations for which data was available (41) and

the relative size of Fiji (and therefore the number of deaths) to the other PICTs. For

Tonga, female adult mortality from the GBD study shows adult mortality around

13.5% (range approx: 11.5-17.0) (240) substantially lower than the findings

presented in Table 49 (111). This may potentially be explained by the model

anticipating a greater sex-differential in mortality than is seen in Tonga where adult

mortality is high for both men and women (Table 49), and from errors in the source

data. The input data displayed in the GBD modelling indicated it was derived from

vital registration, without correction for undercount, and shows significant

inconsistencies in level. As discussed in Chapters 3 and 5, the CRVS system in

Tonga is known to have some undercount issues, and there are notable differences

in completeness in the various routine collections from civil registry and the Ministry

of Health (all of which may be termed “vital registration data”) (164).

342

Key causes of death for adult males and females 15-59 years are shown in the

following tables (Tables 52-56). Although countries may be at different stages of the

epidemiological transition, NCDs dominate the causes of death in this age group.

Diabetes rates ranged from 38.2 (Palau) to 489 (Nauru) deaths per 100,000

population for males and 34.6 (Palau) to 268 (Nauru) deaths per 100,000 population

in females. Deaths related to diabetes in Fiji (2005-2008) were estimated from the

multiple CoD analysis in Chapter 6 at 83.5 deaths per 100,000 population for males

and 70.5 deaths per 100,000 population for females. Although rates were not

calculated for Kiribati, proportional mortality for mortality appears to be in a similar

range as the other study countries (excluding Nauru). While the extremely high

diabetes rates in Nauru could in part be a result of local certification preferences

given the low proportion of cardiovascular diseases in comparison, the cause-

specific mortality rates were derived following a medical record review with CoD

assigned from the medical record by a doctor external to the system. The high cause

specific mortality is therefore more likely to reflect the high prevalence of diabetes in

Nauru. In 2010, Nauru was estimated to have the highest prevalence of diabetes in

the world based on WHO figures at 30.9% of the population from 20-79 years of age,

with the second highest prevalence occurring in the United Arab Emirates with

18.7% prevalence; substantially lower than in Nauru (360).

Throughout the records reviewed across the study countries, it was noted deaths

due to diabetes were frequently related to septicaemia following infection of skin

abscesses or amputations. Many of the deaths reviewed noted the diabetes was

“uncontrolled”, and anecdotal discussions with physicians in the region suggest

343

many patients only present for care once the infection has spread or traditional

remedies have failed. Similar issues have previously been documented in the Pacific

population in New Zealand (361). This suggests while prevention of new cases is

important, it may also be possible to reduce mortality related to diabetes through

improvements in management of the disease in existing cases. Further work is

clearly needed to improve mortality data to allow reliable analysis of deaths by

specific sub-category of disease, in this case the specific complications of diabetes

according to the third digit of the ICD code; but also to identify current barriers to and

patterns of access to health care and ways of improving the ongoing care of patients

with chronic diseases.

Diseases of the circulatory system were among the top causes of death in all

countries. Both ischaemic and cerebrovascular diseases were found to be important

causes of death across the study countries. Rates could not be adequately identified

for either in Palau where most deaths were recorded as hypertensive heart disease

or other heart conditions, indicating the need for further review of certification

practices. The overall cause-specific mortality rate of 140.2 and 94.5 deaths per

100,000 people from diseases of the circulatory system in Palau (2005-2009) shows

a similar level of cardiovascular disease overall to Nauru and Vanuatu. While Palau

and Nauru are both recognised as being well into the epidemiological transition, that

diseases of the circulatory system occur at similar rates in Vanuatu (Tables 52 and

55) is perhaps a surprise, but is consistent with the risk factor data available for the

country. Small sample risk factor surveys were conducted by the Vanuatu Ministry of

Health in 1998 (62), 2007 (62) and 2011 (337). These surveys reflect a rapidly

increasing proportion of the population affected by NCD risk factors. Only 3% of the

344

population was reported to have diabetes in 1998 (62), this had increased to 12% in

2007 (62), and by 2011 was 18% (defined as percentage with raised fasting blood

glucose capillary whole blood value � 6.1 mmol/L or currently on medication for

raised blood glucose). Similarly hypertension prevalence increased from 13 to 15%

between 1998 and 2007, while those considered overweight (BMI � 25kgm3)

increased from 33% of the population in 1998 to 38% in 2007 and 51% in 2011 (62).

Proportional mortality from Kiribati also indicates similar levels of cardiovascular

disease. Cause-specific mortality related to Ischaemic heart disease in Fiji was

estimated at 167.9 and 32.2 deaths per 100,000 population for males and females

respectively indicating higher rates of ischaemic heart disease, but similar levels of

all cardiovascular diseases.

Rheumatic heart disease, which has previously been identified as an issue in the

Pacific (362-367) varied markedly across the study countries, and clearly remains a

public health issue in Nauru, and to a lesser extent in Palau, Tonga and Fiji (where

cause specific mortality associate with rheumatic heart disease was 7.3 and 8.0

deaths per 100,000 for males and females respectively in 2005-2008).

345

Tabl

e 52

: Adu

lt m

orta

lity

and

prop

ortio

nal m

orta

lity

(%) f

or k

ey c

ause

s (b

y IC

D c

hapt

er) f

or s

elec

ted

Paci

fic Is

land

s; M

ales

Cou

ntry

Adu

lt m

orta

lity

(pro

babi

lity

of

dyin

g be

twee

n ag

es 1

5-59

, %)

Age

gr

oup

for

caus

e of

de

ath

Ref

. ye

ar

Pro

porti

onal

mor

talit

y, e

xclu

ding

ill-d

efin

ed c

ause

s (%

)

Infe

ctio

us

dise

ases

N

eopl

asm

s

End

ocrin

e,

nutri

tiona

l and

m

etab

olic

di

seas

es

Circ

ulat

ory

dise

ases

R

espi

rato

ry

Illne

sses

Dis

ease

s of

the

dige

stiv

e tra

ct

Gen

ito-

urin

ary

dise

ases

Inju

ry a

nd

othe

r ex

tern

al

caus

es

Fiji

27.6

15

-59

2000

-20

08

5.3

7.6

17.4

38

.7

7.3

3.3

4.5

13.1

Nau

ru

51.0

15

-59

2005

-20

09

4.5

9.9

42.3

15

.3

7.2

2.7

5.4

9.9

Kiri

bati

(hea

lth

repo

rting

)

60.0

(pla

usib

le

rang

e 40

.5-

60.0

) 15

-59

2000

-20

09

4 2

9 31

6

21

1 15

Pal

au

59.6

15

-64

2004

-20

09

8.0

19.6

7.

6 21

.9

3.6

3.6

4.8

26.7

Tong

a 28

.6 –

36.

3 15

-59

2001

-20

09

5 16

8-

14

36-4

1 14

-15

4 2

6

Van

uatu

27

.5

(est

imat

e)

15-5

9 20

07-

2011

9-

10

12-1

6 6-

10

27-2

9 7-

9 8-

14

3-6

9-13

346

Table 53: Cause-specific mortality rates for key causes (by ICD General Mortality List 1) for selected Pacific Islands; Males 15-59 years

Cause� Crude�age�specific�mortality�rate�(deaths�per�100,000�population)�

Listcode Disease Palau 2004-

2009Tonga 2005-

2008Nauru 2002-

2007Vanuatu 2000-

2009

1-001 Certain infectious and parasitic diseases

51.0�(31.2��78.8)� 61.1��84.9� 53.2�(13.5��

141.7)� 60.0��75.6�

1-005 Tuberculosis 5.1�(0.6��18.4)� �*� 10.6�(0.2��112.5)� 6.5��20.8�

1-019 Viral hepatitis 22.9�(10.5��

43.6)� 18.4��24.7� 21.3�(2.0��87.7)� 13.6��37.0�

1-021 Malaria �NA� NA�� NA�� 3.2��23.2�

1-026 Neoplasms �125.4�(92.4��165.2)� 189.4��254.7� �116.9�(45.6��

239.0)� 74.2��119.6�

1-031 Malignant neoplasm of liver and intrahepatic bile ducts

43.3�(25.3��69.4)� 17.9��25.5� *�� 27.1��37.0�

1-034 Malignant neoplasm of trachea, bronchus and lung

30.6�(15.8��53.5)� 34.4��49� 21.3�(2.0��

87.7)� 3.9��13.9�

1-051Endocrine, nutritional and metabolic diseases

48.4�(29.2��75.7)� *�� 499.6�(286.5��

759.0)� 36.8��76.4�

1-052 Diabetes mellitus 38.2�(21.4��

63.1)� 94.2��222� 489.0��(279.4��745.1)� 34.9��74.9�

1-058 Diseases of the nervous system �*� 13.5��17.9� 21.3�(2.0��

87.7)� 14.2��37.0�


140.2�(105.7��182.5)� 423��644.2� 180.7�(82.2��

330.5)� 171.7��222.2�

1-065 Acute rheumatic fever and chronic rheumatic heart diseases

2.5�(0.1��14.2)� 8.7��11.6� 21.3�(2.0��

87.7)� 0.0��13.1�

1-067 Ischaemic heart diseases *�� 139.4��210.2� 74.4�(23.3��

175.1)� 3.9��94.9�

1-069 Cerebrovascular diseases *�� 40.4��66.4� 53.2�(13.5��

141.7)� 27.1��44.8�


22.9�(10.5��43.6)� 168.4��230.9� 85.0�(28.7��

191.4)� 46.5��65.6�

1-074 Pneumonia 2.5�(0.1��

14.2)� 14.8��20.2� 63.8�(18.3��158.6)� 3.2��23.9�

1-076 Chronic lower respiratory diseases

10.2�(2.8��26.1)� 92.7��124.6� 0.0�(0.0��

44.8)� 29.0��41.7�

1-078 Diseases of the digestive system

22.9�(10.5��43.6)� 46.6��63� 31.9�(5.1��

106.5)� 51.0��105.7�

1-080 Diseases of the liver 15.3�(5.6��

33.3)� 17.3��23.1� 21.3�(2.0��87.7)� 45.2��91.8�

347


30.6�(15.8��53.5)� 25.1��36.6� 63.8�(18.3��

158.6)� 19.4��48.6�

1-095External causes of morbidity and mortality

�170.8�(132.4��217.0)� 68.2��91.6� *�� 56.8��97.2�

1-096 Transport accidents 22.9�(10.5��

43.6)� 13.1��17.5� 116.9�(45.6��239.0)� 5.8��30.9�

1-098 Accidental drowning and submersion

2.5��(0.1��14.2)� �*� 21.3�(2.0��

87.7)� 8.4��20.8�

1-101 Intentional self-harm 22.9��(10.5��

43.6)� 8.6��11.6� 31.9�(5.1��106.5)� 0.0��9.3�

1-102 Assault 12.7��(4.1��

29.8)� �*� 21.3�(2.0��87.7)� 2.6��16.2�

*NB/ not all rates available for all countries due to different study designs

Proportional mortality from neoplasms from the Ministry of Health data in Kiribati was

very low in comparison both to the other study countries and global norms; at 2% of

adult deaths for males, 10% for adult females. These figures were slightly higher in

the civil registry data at 4% and 13% respectively. The low mortality rates were

consistent with previously published data from Palafox et.al.(368) who reviewed

reported cancers from Ministry of Health data across Micronesia in the late 1990s,

and found Kiribati to have substantially lower reported incidence of a range of

cancers of public health importance. An extract from this data is shown below.

Table 54: Age-standardised incidence rates of specific cancer types 1980s-1990s (cases per 100,000 population) by country extracted from Palafox et. al

(368)

Cancer site Kiribati Palau Marshall Islands

Nauru

Breast 8.0 17.1 36.0 15.4 Cervix 4.5 37.5 60.5 55.0 Liver 0.5 19.4 10.2 5.7 Lung 4.4 34.6 41.1 42.8 Years 1989-1998 1985-1998 1985-1998 1985-1998

Note: age-adjusted to WHO world standard population, annualized.

Source: Palafox, Cancer in Micronesia (368)

348

As shown in the shaded column, incidence rates for each of these four key types of

cancer were substantially lower than would be expected. This is particularly true for

lung cancer given the high prevalence of smoking in Kiribati (74% of males and 45%

of females reported smoking daily in the 2006 STEPS survey) (335). Palafox et.al

noted in their paper the poor state of the records associated with cancer reporting in

Kiribati (368) and the difficulties of diagnosis given the limited diagnostic facilities

and access to higher level health services. It is likely these factors, are responsible

for both the low reported incidence of cancer as well as the low proportional

mortality, rather than a true absence of the disease in the population. This is

supported by Ou et al. whose review of reported cases in Kiribati for the same period

indicated nearly 18% of those cases who were treated, were not formally diagnosed

or treated for more than a year following the onset of symptoms (369). This lack of

diagnostic capacity may also be a factor in the relatively low proportion of deaths

attributed to respiratory disease (including chronic lower respiratory illness) given the

high smoking prevalence.

A significant increase in age-standardised mortality from Hepatitis B was shown in

Tonga between 2001-2004 and 2005-2008. Hepatitis, which codes to liver disease

(under Diseases of the digestive tract) when a causative virus is not recorded on the

medical certificate (68), is also likely to account for the high proportional mortality

from Diseases of the digestive tract reported for males in Kiribati (21% of adult

deaths), and to a lesser extent the proportional mortality attributed to this category in

the other study countries. As such, Hepatitis B represents a significant cause of

mortality in the region that is amenable to intervention. Universal Hepatitis B

vaccination in Pacific Island countries will take some years to generate a reduction in

349

mortality due to the current prevalence of the disease. A reduction in prevalence

(due to early vaccination) would reasonably be expected to result in reduced

mortality in future generations. All Pacific Island Countries offered universal

vaccination by the late 1990s, however by 2005, coverage in Fiji, Samoa, the

Solomon Islands, Tuvalu and Vanuatu remained below 80% (370).

Injury and external causes are a significant CoD for males in this region and need to

be explored further. CoD for these deaths was, as a whole, poorly recorded across

the region, with most datasets consistently capturing the type of injury but not how it

occurred. The available data suggest suicide is a significant issue, with estimates for

Kiribati (2005-2008) as at least 20 deaths per 100,000 population for males 15-59

(from civil registration data) not accounting for under-count in registration or ill-

defined cases. Deaths attributed to suicide in were also very high for males in Palau

and Nauru with cause specific mortality estimated at 22.9 and 31.9 deaths per

100,000 population respectively (without re-distribution of deaths attributed to ‘other

external causes’), with further work required in Fiji to further disaggregate by cause

the cause-specific mortality from external causes (70.7 deaths per 100,000

population for males 15-59 years in 2005-2008). These findings suggest while

methods of suicide may have changed, there has been very little progress in dealing

with this issue since the mid 1990’s when suicide rates in the Pacific were identified

as being amongst the highest in the world (371).

350

Tabl

e 55

: Adu

lt m

orta

lity

and

prop

ortio

nal m

orta

lity

(%) f

or k

ey c

ause

s (b

y IC

D c

hapt

er) f

or s

elec

ted

Paci

fic Is

land

s;

Fem

ales

Cou

ntry

Adu

lt m

orta

lity

(pro

babi

lity

of

dyin

g be

twee

n ag

es 1

5-59

, %

)

Age

gr

oup

for

caus

e of

de

ath

Ref

. ye

ar

Pro

porti

onal

mor

talit

y, e

xclu

ding

ill-d

efin

ed c

ause

s (%

)

Infe

ctio

us

dise

ases

N

eopl

asm

s

End

ocrin

e,

nutri

tiona

l an

d m

etab

olic

di

seas

es

Circ

ulat

ory

dise

ases

R

espi

rato

ry

Illne

sses

Dis

ease

s of

the

dige

stiv

e tra

ct

Mat

erna

l C

ause

s

Gen

ito-

urin

ary

dise

ases

Inju

ry

and

othe

r ex

tern

al

caus

es

Fiji

20.3

15

-59

2000

-20

08

6.2

20.9

23

.5

23.2

7.

2 2.

8 0.

7 4.

4 7.

1

Nau

ru

37.4

(p

laus

ible

ra

nge

22.6

- 37

.4)

15-5

9

2005

-20

09

6.4

22.3

31

.9

16.0

10

.6

4.3

1.1

5.3

2.1

Kiri

bati

(hea

lth

repo

rting

)

40.1

15

-59

2000

-20

09

4 10

7

29

6 23

1

1 6

Pal

au

43.1

15

-64

2004

-20

09

4.8

28.2

11

.3

24.2

3.

2 3.

2 0

8.1

9.7

Tong

a 20

.9- 2

7.7

15-5

9 20

01-

2009

5

20

14-2

0 28

-34

12

4 0-

1 1

3

Van

uatu

17

.0(e

stim

ate)

15

-59

2007

-20

11

8-10

28

-32

10-1

3 14

-16

6-7

5-7

3-10

4-

11

2-6

351

Table 56: Cause-specific mortality rates for key causes (by ICD General Mortality List 1) for selected Pacific Islands; Females 15-59 years

Cause� Crude�age�specific�mortality�rate�(deaths�per�100,000�population)�

Listcode Disease Palau 2004-

2009Tonga 2005-

2008Nauru 2002-

2007Vanuatu 2000-

2009

1-001 Certain infectious and parasitic diseases

18.9�(6.9��41.1)��

33.8 - 48.4 57.4��(15.4��144.2)� 30.6��46.6�

1-005 Tuberculosis 0.0�(0.0��

11.6)� �*� 28.7�(1.9��141.3)� 1.8��13.4�

1-019 Viral hepatitis 9.4�(1.9��

27.6)�7.3 - 10.1

9.6�(0.2��61.5)� 5.1��8.3�

1-021 Malaria �NA� NA�� NA�� 6.2��11.5�

1-026 Neoplasms 110.2�(76.8��153.3)�

136.4 - 195.5 201.1�(90.7��354.4)� 102.0��145.3�

1-031 Malignant neoplasm of liver and intrahepatic bile ducts

9.4�(1.9��27.6)�

5.9 - 8.1 �*� 3.6��13.4�

1-034 Malignant neoplasm of trachea, bronchus and lung

9.4�(1.9��27.6)� 16.5��23.0� 38.3�(7.6��

113.1)� 0.0��4.6�

1-036 Malignant neoplasm of breast

15.7�(5.1��36.7)� 20.7�30.4� 38.3�(7.6��

113.1)� 13.1��43.4�

1-037 Malignant neoplasm of cervix uteri

12.6�(3.4��32.3)�

8.6 - 12.0 57.4�(15.4��

144.2)� 9.1��32.3�

1-051Endocrine, nutritional and metabolic diseases

44.1�(24.1��74.0)� *�� 287.2�(141.2��

472.8)� 35.7��61.4�

1-052 Diabetes mellitus 34.6�(17.3��

62.0)�98.3 - 190.4

268.1�(129.8��446.8)� 35.7��59.0�

1-058 Diseases of the nervous system *�� 17.9 - 24.8 9.6�(0.2��

61.5)� 5.1��13.4�


94.5�(63.7��134.9)�

194.2 - 321.3 143.6�(58.6��273.1)� 51.7��73.8�

1-065 Acute rheumatic fever and chronic rheumatic heart diseases

15.7�(5.1��36.7)� 10.2��14.3� 47.9�(11.3��

128.8)� 0.0��20.8�

1-067 Ischaemic heart diseases �*� 47.3 - 88.1

28.7�(4.3��96.8)� 0.0��11.1�

1-069 Cerebrovascular diseases *�� 31.3 - 46.7

47.9�(11.3��128.8)� 14.6��25.4�


12.6�(3.4��32.3)�

81.7 - 114.2 95.7�(33.5��203.0)� 22.6��33.7�

1-074 Pneumonia 0.0�(0.0��

11.6)�11.1 - 16.0

47.9�(11.3��128.8)� 2.6��8.3�

1-076 Chronic lower respiratory diseases

9.4�(1.9��27.6)�

23.4 - 33.6 19.1�(1.7��

79.8)� 14.9��23.5�

1-078 Diseases of the digestive system

12.6�(3.4��32.3)�

27.6 - 38.6 38.3�(7.6��113.1)� 16.8��31.8�

352

1-080 Diseases of the liver 6.3�(0.8��

22.8)�9.2 - 12.8

9.6�(0.2��61.5)� 11.7��26.8�


31.5�(15.1��57.9)�

20.9 - 32.1 47.9�(11.3��

128.8)� 14.6��48.4�

1-087 Pregnancy, childbirth and the puerperium �*� 4.8 - 6.7 �*� 11.7��47.1�

1-095External causes of morbidity and mortality

37.8�(19.5��66.0)�

21.4 - 29.5 19.1�(1.7��

79.8)� 5.1��26.8�

1-096 Transport accidents 6.3�(0.8��

22.8)� �*� 19.1�(1.7��79.8)� 0.0��1.4�

1-098 Accidental drowning and submersion

0.0�(0.0��11.6)�

1.7 - 2.3 0.0�(0.0��

40.7)� 0.0��5.1�

1-101 Intentional self-harm 12.6�(3.4��

32.3)� *�� 0.0�(0.0��40.7)� 0.0��6.9�

1-102 Assault 3.1�(0.1��

17.5)� *�� 0� 3.3��8.3�

*NB/ not all rates available for all countries due to different study designs

Maternal mortality was responsible for a substantial proportion of deaths in Vanuatu

(ranging from 3-10 % of all female deaths from 15-59 years of age). This accounts

for between 11.7 and 47.1 deaths per 100,000 women. These estimates were

derived from 7 years of data (2001-2007) to minimise the impact of stochastic

variation. In Fiji, cause-specific mortality was estimated at 2.2 deaths per 100,000

women (2005-2008).

7.2.3 Mortality & Older adults

Understanding premature mortality in adults is of primary importance in improving

the health of Pacific Islanders, however the health of older adults is also of interest to

governments, the health sector and social security and pension funds, and to guide

decisions about the social and physical infrastructure required to support this group.

As noted in the previous section, premature adult mortality across the selected study

countries was high. This trend is continued through into older adulthood, with men

353

and women who reach 65 years of age (above the estimated LE at birth for most of

the study countries) only anticipated to live another 12 years for males and 8-13

years on average for females (Table 57). In Australia and New Zealand, men and

women reaching this age would on average be anticipated to live a further 18-19

years for males and 20-22 years for females (Table 57).

.

Table 57: Life expectancy at 65 years for selected countries

Country Reference Year Males Females Fiji 2002-2004 11.4 13.9 Kiribati 2005-2009 8.2 10.5 Nauru 2002-2007 12.4 7.9 Palau 2004-2009 12.2 13.6

Tonga 2005-2009 (observed / unadjusted)

11.5 13.5

Vanuatu (estimated)

2009-2010 12.2 14.7

Australia (372) 2009-2011 19.1 22.0 New Zealand (356) 2005-2007 17.9 20.6

Causes of death in older adults are typically more poorly recorded than in other

groups due to the high proportion of deaths that occur at home (14) and multiple co-

morbidities, which may make determining a single underlying CoD more difficult

(247). This is reflected in the high proportion of ill-defined causes seen in older

adults across the study countries. Leading causes of deaths across all countries in

this age group were NCDs.

354

7.3 Country summaries An overview of the key findings on mortality level and cause for each of the selected

study countries is provided below and briefly discussed in relation to known risk

factors and previous studies.

7.3.1 Fiji

Case study 3 (Chapter 5) which examined both MoH data from Fiji as well as

previously published estimates clearly demonstrated many past estimates of

mortality for Fiji have significantly underestimated adult mortality. When screened for

plausibility, long term trends demonstrate a long term stagnation in LE since the

early 1980s for both males and females, with no sustained improvements over the

previous 20 to 30 years. Although infant and child mortality estimates were found to

be higher than several previous estimates by the Fiji MoH and other government

sources (14), IMR fell significantly over the 1980s to 1990s and is now stable below

20 deaths per 1,000 live births; a level at which IMR could not be considered to be a

primary influence on the stagnation in LE reported in this work. While migration of

healthy adults from Fiji could potentially contribute to some of the stagnation of LE,

further analysis by ethnicity yet to be published, suggests that adult mortality is

increasing most notably amongst Fijian i-Tauki, especially for women. This is

inconsistent with migration which is known to be higher in the Fijian Indian

population.

Despite almost universal medical certification, coding and tabulation practices

currently limit the value of this data for establishing CoD patterns for specific causes,

with 17% medical certificates reviewed found not to have the underlying CoD

355

represented in the data tabulations. Further work is recommended to re-examine

completed certificates in order to correct these issues.

Despite the data uncertainty (as described in the previous chapters), NCDs are

clearly having a critical impact on premature adult mortality. Plausible published

estimates (case study 3) demonstrate the proportion of deaths from NCDs for all

ages has increased substantially over the previous few decades (93) while deaths

due to infectious disease and external causes have both fallen. The medical

certificate review presented in Chapter 6 demonstrates that this is clearly not a

function of additional deaths occurring from NCDs at older ages as people live

longer, with NCDs the leading CoD in both males and females from ages 15-54

years of age.

Given the uncertainty that remains in the data, tabulations were derived for broad

CoD categories only in Fiji. For adults 15-59 years the tabulations for both men and

women demonstrate the vast majority of deaths are due to NCDs, including diseases

of the circulatory system, endocrine and metabolic disease and neoplasms. The

multiple CoD analysis found 21% of deaths in this age group for females and 15%

for males were associated with diabetes. This equates to a cause-specific mortality

rate of 70.5 (95%CI: 40.5-100.4) and 83.5 (95%CI: 50.1 – 116.1) deaths per 100,000

population respectively. As these rates include all deaths for which diabetes was

listed on the medical certificate they are not comparable with other populations, but

do provide health planners an understanding of the scale of the issue in Fiji.

Ischaemic heart disease was a CoD in 30.8% of males and but only 9.7% of

females, indicating cause specific mortality of 167.9 (95%CI 121.7 0 214.0) and 32.2

(95%CI: 12.0 – 52.5) deaths per 100,000 population respectively. Rheumatic heart

356

disease was not a major CoD for adults, and was attributed as a CoD in only 2.4% of

women and 1.3% of males.

Hepatitis was listed as a CoD in only 0.2% of deaths for adult women and 0.4% of

deaths for adult males indicating the hepatitis vaccination program which has been

running for several decades in Fiji has been effective.

The high cause-specific mortality rates from NCDs are consistent with previous

studies in Fiji which note steady increases in hospital admissions from

cardiovascular diseases and diabetes in the 1980s (357). Changes in risk factors

over the last 20-50 years have also been noted, with trends towards consumption of

imported high energy foods (373, 374), and an increasing proportion of the

population leading a sedentary lifestyle (375). Prevalence of cardiovascular

diseases and diabetes has been documented as higher in populations where these

changes were more ubiquitous than in more traditional rural areas.(74, 376)

7.3.2 Kiribati

Results from Kiribati depict a country still in the early stages of the epidemiological

transition with high rates of infant and child mortality, and a high prevalence of

infectious diseases. The data from this study suggests Kiribati has one of the lowest

life expectancies in the region, estimated to be 48.6 years for males and 57.9 years

for females. This is several years lower than published estimates for Nauru (Table

6), and lower for males than published estimates for PNG (Table 6); countries

previously recognised as having the lowest LE in the region. These low life

expectancies reflect high mortality rates across all age groups.

357

Infant mortality was estimated at 51.6 deaths per 1,000 live births and U5M at 71.5

deaths per 1,000 live births. Malnutrition remains a serious contributor to childhood

mortality in Kiribati, with 23% of deaths in children 0-4 years recorded by the hospital

or community health facilities attributed to ‘Endocrine, metabolic and nutritional

disorders’ as defined in the ICD. If these figures are representative at a broader

community level, this would equate to approximately 12 of every 1,000 children born

dying directly from malnutrition before their 5th birthday. A further 27% of deaths in

this age group were attributed to infectious diseases or respiratory conditions; thus

highlighting the large proportion of preventable deaths.

Adult mortality is relatively high, estimated at 60% for males and 34% for females,

and clearly shows the double burden of disease with both infectious diseases and

NCDs present. For both males and females, Infectious diseases and Diseases of the

digestive tract were among the leading causes of death. Deaths attributed to

diseases of the digestive tract included those recorded as peptic ulcers and liver

disease, while infectious disease deaths included both septicaemia and hepatitis

related deaths. Previous studies have shown high prevalence of Hepatitis B in

Kiribati which would contribute deaths in both of these categories (through those

coded as infectious hepatitis, as well as liver disease), with prevalence of the surface

antigen estimated at 31% in 1987(377). The Kiribati immunisation schedule has

included vaccination for Hepatitis B since 1985 when coverage was estimated as

73%. Estimates of immunisation coverage have fluctuated in the years since (as low

as 35% in 1995) with the most recent available estimate at 86% coverage in 2009.

Although immunisation will not impact mortality rates from this disease for many

years, the data highlights the importance of this program in addressing future

358

mortality rates in Kiribati, and the need to focus on ensuring universal vaccination

coverage.

Maternal deaths account for 3.6 to 5.8% of all female deaths between 15-59 years,

depending on the data source. Civil registry data for maternal deaths is likely to be

more reliable than for other causes as there is a clear event which is recorded

(requiring less interpretation by a lay person determining CoD), and as such

maternal mortality is likely to be toward the higher end of range. This would equate

to an average of 11 deaths per year (95%CI: 5-20 using Poisson distribution) over

2005-2008, or a maternal mortality ratio of 444 (95%CI: 202-808) based on the 2475

births per year estimated by the 2005 census. This is substantially higher than

previous estimates from national MDG reporting and should be further investigated.

Of the cancer deaths reported through the hospital, 55% of male deaths were due to

neoplasms of the lung or oesophagus. While a similar number of deaths were

reported for women, these accounted for only 7% of the cancers, with breast or

cervical cancer most reported (44%). The high proportion of cancer deaths

attributable to lung cancer based on hospital mortality, is consistent with the high

smoking prevalence reported in Kiribati. The 2006 STEPs survey indicated 74% of

males and 45% of females smoked daily (335).

Suicide was highlighted as a major CoD for males from civil registry data, but is

almost entirely absent from the hospital data; thus demonstrating the importance of

not relying on hospital data sources in isolation from other information. Even without

adjusting these rates for undercount, the minimum level of mortality attributable to

this cause is 55.6 deaths per 100,000 population for males, on par with countries

359

such as Russia. While the rates appear to have fallen, they are still extremely high

and warrant further investigation urgently.

7.3.3 Nauru

LE in Nauru is low at an estimated 52.8 and 57.5 years for males and females

respectively as shown in case study 2, and has changed little over the last 50 years.

This stagnation is largely due to the extremely high premature adult mortality,

present in both males and females. Endocrine, nutritional and metabolic diseases

(nearly all diabetes) is the leading CoD for both males and females. This group of

causes accounted for 42% of deaths in males and 32% of deaths in females 15-59

years; a mortality rate of 412 and 241.deaths per 100,000 population respectively.

Although extremely high, this is consistent with previous studies that found a high

prevalence of diabetes in Nauru (24 % in 1982) (Zimmet 1998 (378); and a higher

risk of diabetes in Nauruans amongst multi-ethnic studies, independent of other risk

factors including physically inactive lifestyle (379). A 2004 survey found 90% of

Nauruans have a poor diet (defined by low fruit and vegetable intake), and nearly

50% have very low rates of exercise (380). Cardiovascular diseases and cancers

were also a significant CoD in adults 15-59 years for both men and women.

Unusually, respiratory illnesses accounted for a substantial proportion of deaths in

adults 15-59 years, 7% of deaths in males and 11% of deaths in females. In 2004,

45% of males and 53% of females smoked daily (381), which may have contributed

to these levels.

IMR and U5M are still above 30 deaths per 1,000 live births, with U5M having

changed very little over the last 50 years. From the medical record review 44% of

360

deaths for children aged 0-4 years were found to be due to perinatal causes, with

only 7% due to congenital causes. As perinatal causes are more amenable to

improvement through intervention, through improved ante-natal and post-natal care,

it is feasible further gains could be made in this area. These findings are consistent

with the 2007 DHS which found significant problems with both ante-natal and post-

natal care provision in Nauru, despite nearly all births being attended by a medical

professional. Only 40% of women had the recommended number of ante-natal visits

with 75% of women attending their first visit in the second trimester of pregnancy.

One in six did not receive a postpartum check-up (156)

7.3.4 Palau

LE in Palau is estimated at 65.8 years (95%CI: 64.4 – 67.2) for males and 71.2 years

(95% CI: 69.9 – 72.5) for females in 2004-2008, consistent with the census

estimates from 2005. Improvements in LE for both males and females have

stagnated in Palau as in Fiji and Nauru. Although this stagnation happened at slightly

higher LE’s (for females at least), comparison with census data shows no

improvement at all over 20 years, and for females may actually indicate a slight

decline (as there are no confidence intervals for the census data this could not be

demonstrated to be significant). Adult mortality in Palau is extremely high at 59.6%

and 43.1% for males and females respectively (2005-2008), and dominated by

NCDs including diabetes and cancer.

Previously published estimates of IMR and U5M for Palau (14) vary significantly and

have limited value as they are based on single year estimates which are highly

unstable in a population of this small size. This study found IMR for 2004-2008 to be

16.7 deaths per 1,000 live births when aggregated over the 4 years of data available

361

(excluding 2007 due to a problem with the records for that year) and 25 deaths per

1,000 live births for children 0-4 years. At these relatively low levels, U5M is not a

major influence on the stagnation in LE described above.

7.3.5 Tonga

Mortality and CoD patterns in Tonga have been discussed extensively in case

studies 5 and 6. As noted in these studies, data from the MoH and Civil Registry

were reconciled and subjected to a capture-recapture analysis. Estimates of

mortality from this process were also compared to published data sources screened

for plausibility. A medical record review was then conducted to assess certification

and establish CoD patterns.

Published estimates of IMR converge to a range between 10 and 20 deaths per

1,000 live births from the late 1990s. Findings from reconciled data were consistent

with this range, and did not demonstrate any significant trend over 2001 to 2009.

LE in Tonga is substantially lower than has been previously reported, although this

indicates a stagnation in LE rather than a reversal as previous estimates were shown

to have under-estimated adult mortality. Estimates of LE from the reconciled

empirical data for 2005 to 2009 were 65.2�years (95% CI: 64.6 - 65.8) for males and

69.6�years (95% CI: 69.0 – 70.2) for females, several years lower than published

MoH and census estimates. However the capture-recapture analysis of reporting

completeness suggests even reconciled data are under enumerated, placing the

range of plausible estimates as low as 60.4 �years for males and 65.4 �years for

females,

362

Given the reasonably low IMR, these estimates of LE are driven primarily by the high

adult mortality. Adult mortality (15 to 59�years) from reconciled empirical data is

estimated at 26.7% for males and 19.8% for females, while the capture-recapture

analysis placed the range at 28.6% to 36.3% and 20.9% to 27.7% for males and

females respectively. The data shows real age-specific increases in NCDs including

diabetes, cardiovascular diseases and cancer over 2001-2004 to 2005 to 2008 for

both men and women, and demonstrate NCDs are having a critical impact on

premature mortality. Mortality from diabetes for 2005 to 2008 is estimated at 94 to

222 deaths per 100,000 population for males and 98 to 190 for females, and

cardiovascular disease at 423-644 deaths per 100,000 population for males and

108-227 deaths per 100,000 population for females. The high mortality rates from

NCDs are consistent with the limited information on risk factors available for Tonga.

A 2004 survey found 67% of adults were obese (body mass index � 30 kg/m2), with

23% of adults affected by hypertension (systolic blood pressure � 140 mmHg and/or

diastolic blood pressure � 90 mmHg or currently on medication) (380).

7.3.6 Vanuatu

The system assessment identified there was very limited data in Vanuatu on

mortality and CoD, and the collected data was unlikely to be representative at a

national level. The best estimates of mortality level must still therefore come from

other sources such as censuses and surveys. As demonstrated in the work in this

thesis however, estimates of adult mortality and LE based on single parameter

model life tables tend to under-estimate mortality levels and are therefore not

suitable for application in the region. Covariate modelling methods such as those

developed by Ratnarjaram et al which draw on all available estimates as well as

patterns from neighbouring countries where more empirical data is available are

363

therefore more suited for use, although their applicability in the region needs to be

explored further and may be limited by the lack of neighbouring countries (other than

Australia and New Zealand which are significantly different in terms of size,

development and society) with empirical data which can be inputted into the models.

Best estimates of mortality for infant and child mortality in Vanuatu were therefore

derived from the census data, with age specific mortality for other ages (and LE)

derived by inputting both child mortality from the census and adult mortality from

Rajaratnam ‘s work (41) into a two source life table (using modmatch) (235). These

provide a plausible lower limit for mortality in Vanuatu (and therefore an upper limit

for LE) given the recall issues associated with collecting data through surveys or

censuses and the reporting bias likely in Vanuatu as a culture in which discussing

people who have died is particularly difficult. While LE from this approach is at the

higher end of the range of estimates for the region at 65 years for males and 70

years for females, these are much more plausible than the census estimates at 69.9

and 73.5 years respectively given the level of development and infrastructure in

Vanuatu.

As discussed in Chapter 6, the leading causes of death for adults in both the 15-59

year age groups and older age groups were NCDs. For males these were

predominantly cardiovascular diseases, liver diseases and diabetes; while for

females neoplasms were also important, with both cervical and breast cancer

appearing in the leading causes of death. External causes were under-represented

in the hospital data and therefore would not appear as a public health concern on the

leading CoD list compiled from this source for the CHIPs report. Maternal deaths

were equally under-represented in the health data accounting for 2% of deaths in

364

females 15-59 years, but representing 10% of deaths females in this age group in

the community facility data.

7.4 Mortality in Transition

As demonstrated in Table 50, there has been no sustained improvement in LE

across the study countries since mid 1990s. Although Table 50 indicates LE has

actually fallen in Tonga, the case study presented in Chapter 5 demonstrated this is

not actually the case, and that earlier estimates of LE were likely to be an over-

estimation (111).

For countries where longer trend data was available, including Tonga, Fiji and

Nauru, LE has been stagnant for several decades; indicating that these countries

“missed out” on the anticipated increases in LE associated with the decline in

infectious disease mortality seen in the classical pattern of the epidemiological

transition as described by Omran (382) based primarily on experiences in Europe.

Instead, gains in mortality from declining infectious diseases have, at a population

level, been almost entirely offset by increases in premature adult mortality due to

NCDs. This variation in the standard pattern of epidemiological transition has been

described previously for Nauru (112), but could equally apply to Tonga and Fiji, and

although data is more scarce, is consistent with LE and summary measures reported

for Kiribati and Vanuatu. Only Palau appears to have seen a moderate improvement

in LE over the last decade, despite one of the highest adult mortality rates.

There is a potential argument that the stagnation in LE across countries may in part

be being driven by the “healthy migrant effect” where healthy adults are more likely

365

to migrate away from the country for work, thus artificially increasing adult mortality

in some countries such as Fiji and Tonga. Migration from Kiribati, Nauru and

Vanuatu is however essentially negligible, and cannot account for the high adult

mortality in these countries. Additionally, research on the Pacific population in New

Zealand indicates that the mortality of Pacific people on migration to New Zealand

does not change significantly with duration in New Zealand, suggesting no marked

healthy migrant effect (318, 383). As such, migration is unlikely to have a significant

impact on these patterns of mortality. Mortality rates for “Pacific” people in New

Zealand however do appear to be lower than the mortality rates reported in this

paper. This may be due to the high proportion of Pacific Islanders in New Zealand

coming from countries such as Niue and the Cook Islands where life expectancy

(based on complete or near complete civil registration data) is comparatively high

(43), although further investigation and comparison by specific country of origin

would be useful.

Of particular interest is the consistency of this variant pattern of the epidemiological

transition with high adult mortality across the study countries, despite extensive

differences in societal, environmental and economic conditions and countries being

at vastly different stages in the transition. Vanuatu and Kiribati are at an early stage

of the transition, with relatively high infant and child mortality and a high proportion of

deaths attributed to infectious diseases. These early transition countries could still

realise the LE gains anticipated under the classical pattern of transition as

improvements are made in the areas of child and maternal health, and infectious

diseases. However the double burden of disease faced by these countries as they

concurrently deal with increasing premature mortality from NCDs and external

causes, as described in Tables 52 to 56, indicate it is likely these countries are

366

following the variant pattern of epidemiological transition already seen in their

neighbours.

367

8 Conclusion and recommendations

8.1 SynopsisAccurate data on mortality levels and causes of death is critical to assist

governments to improve the health of their populations; by identifying priority areas

for intervention, leveraging resources for such interventions, and monitoring the

impact and effectiveness of health programs. However, there is substantial disparity

in previous estimates of mortality, leaving decision makers without clear information

from which to work. Methods such as the use of single parameter input model life

tables, commonly employed in the region to obtain estimates of LE and adult

mortality from childhood mortality estimates derived from census and survey data

are inappropriate, and have been shown to lead to a substantial under-estimation of

adult mortality (Chapters 2 and 5). CoD data currently available is frequently limited

to a list of leading causes for all age groups (108), and as demonstrated in Chapter

368

3, often tabulated by immediate rather than underlying cause. In this environment of

limited and potentially misleading data, PICTs are facing a significant public health

issue as NCDs impact on premature adult mortality and are facing increased

pressure to report against MDGs on maternal and childhood mortality. More current,

reliable estimates of mortality and CoD by age and sex, based on empirical data

from within the countries, are therefore critically needed in order to inform

governments and health agencies in responding to their population health needs.

The research presented in this thesis demonstrates it is possible to use existing

routine data collections to significantly improve our knowledge of mortality and CoD

patterns in the PICTs. Although gaps remain, the lack of complete data should not

justify failure to critically analyse empirical data on mortality level and CoD.

There is significantly more mortality data available in the countries reviewed than is

currently used for reporting or decision making purposes. While the data are flawed,

as long as these flaws are understood by the researcher or health planner

appropriate judgements can be made about which methods of analysis are suitable

and what biases may affect the final estimates. The framework presented in Chapter

three is a flexible approach for evaluating CRVS systems to understanding content

validity of data and the methods that may be suitable for interrogating the data

further. While the utility of some data sets is limited due to quality concerns, as

outlined in relation to mortality level analysis for Vanuatu and CoD data for Fiji, for all

countries the routinely collected data was able to add to the existing knowledge of

mortality and CoD patterns.

369

Use of local data allows measurement of what is really happening in each country,

rather than estimates that reflect the assumptions of the models they are based on.

Work in Kiribati, Tonga and the Solomon Islands highlights the importance of

reconciling data sources where multiple reporting systems exist, prior to estimation

of mortality level. Both indirect methods of assessing completeness such as the

Brass Growth Balance method, and direct methods such as capture-recapture

analysis, can provide useful tools to correct incomplete mortality data in the region,

provided they are applied with a thorough understanding of the CRVS systems as

described in Chapter three. The methods outlined in Chapters five and six provide a

useful set of tools able to make the best possible use of the data available. In

particular, the framework presented in Chapter five is useful in identifying an

appropriate analytical approach to the data.

The analyses of mortality level presented in Chapter five demonstrates LE across

the study countries has remained steady in the low 60’s or less (with male LE for

both Nauru and Kiribati estimated in the 50’s). These estimates are lower than many

of those previously reported (including those from WHO and UN agencies) (108). At

the same time, infant and child mortality rates were reasonably low for most

countries, and do not appear to be the principal driving factor behind the low

estimates of LE.

Despite the reasonably low IMR and U5M, there remains scope for improvement;

particularly in Kiribati, the Solomon Islands and Vanuatu. Although specific CoD data

could not be obtained for the Solomon Islands, malnutrition (specifically protein

malnutrition) and infectious diseases accounted for a significant proportion of deaths

in children less than five years in Kiribati and Vanuatu.

370

Estimates of IMR appear to have changed little over the previous ten years for Fiji,

Nauru, Palau, and Tonga, suggesting that these countries will have difficulties

reaching the MDG targets to reduce infant and child mortality by two thirds of 1990

levels by 2015. However, estimates for all four countries place IMR at less than 20

deaths per 1000 live births for the mid 2000’s. As child mortality in these countries is

already reasonably low, many of the “easier” gains possible through improved

maternal and child health services and improved hygiene and sanitation have

already been made. Mortality is now clustered in the neonatal age group, and as

such, further improvements in IMR will require more focused interventions and

investment in specialised medical technology and services. Designing these

interventions will subsequently require more detailed CoD data. It is also important

that baseline data used to set MDG targets is assessed for validity to ensure that

countries are working towards plausible goals.

Premature adult mortality (deaths in the 15-59 year age group) was found to be high

for all study countries. Adult mortality in Fiji (2005-2008), Tonga (2005-2008) and

Nauru (2002-2007) was at least three times that for Australia and New Zealand over

a similar period (2002-2004). Although an aging population would be expected to

lead to an increase in proportional mortality from NCDs, proportional mortality from

these causes is increasing across the countries reviewed without a concurrent

increase in LE. This trend suggests NCDs (such as ischaemic heart disease, cancer,

diabetes and liver disease) are responsible for the high premature adult mortality. In

Tonga, where medical certification was assessed as high quality and sufficient data

was available to assess trends in cause specific mortality, it has been possible to

demonstrate significant increases in age-standardised mortality from the early 2000’s

371

to present in ischaemic heart disease, diabetes and lung cancer. These findings

provide supporting evidence NCDs are having a significant impact on premature

adult mortality in other PICTs that demonstrate similar patterns of IMR, LE and

proportional mortality.

The increase in premature adult mortality from NCDs is consistent with the data on

NCD prevalence and risk factors in the region. Hypertension has increased from the

1980’s across several countries. These include Fiji where prevalence rose from 5.1–

9.9% in 1980 (defined as �160 mm hg systolic and/or �95 mm hg diastolic and/or on

treatment for hypertension) to 19% (using �140 mmhg systolic and /or �90 mmhg

diastolic and/or on treatment for hypertension) in 2002 (359). Vanuatu similarly

reported an increase from 13% to 28.6% over 1998 to 2011. In Tonga 23% of adults

were affected by hypertension (systolic blood pressure � 140 mmHg and/or diastolic

blood pressure � 90 mmHg or currently on medication) in 2004, while Kiribati

reported a prevalence of 17.3% in 2009 (335).

Obesity is also recognised as an issue across the region. By 2000, 75% of adults in

seven PICTs were either overweight or obese (384). Tonga reported 67% of adults

as obese (BMI � 30 kg/m2) in 2004. While increases in the prevalence of people

considered overweight (BMI � 25kgm3) in Vanuatu from 33% in 1998 to 51% in 2011

demonstrate the rapid changes underway in the early transition countries.

Other risk factors such as poor diet and a lack of physical activity are also notable

across the region. It is estimated 90% of Nauruans have a poor diet (defined by low

fruit and vegetable intake), and nearly 50% have very low rates of exercise (381).

372

Smoking too remains a significant risk factor in Kiribati, with 74% of males and 45%

of females smoking daily in 2006 (335).

The use of local data allows countries to manage health responses more confidently,

and to have more control in priority setting and resource decisions. Through analysis

of local data in Nauru, it was demonstrated U5M had not increased significantly over

the 1990’s and 2000’s; contrary to some of the modelled estimates. Health planners

in Nauru were subsequently able to use this information in negotiations around child

health interventions with donor agencies. In Tonga, local data analysis demonstrated

U5M is not getting worse (as reported through some modelled data), while adult

mortality from NCDs is significantly higher than previously thought. This information

is being used by the Tongan MoH in planning and budget decisions, and by regional

agencies and donor partners to improve the appropriateness of assistance offered to

PICTs.

Despite the similarities in broad patterns of mortality level and CoD demonstrated

between countries in this thesis, the work has also shown significant differences,

highlighting the diversity in this region and need to tailor responses at a national (or

even sub-national) level. Countries such as Palau and Tonga have clearly moved

through the epidemiological transition, with adult deaths predominately due to NCDs.

Yet amongst these the patterns are very different with cause-specific mortality rates

from diabetes being nearly three times higher in Tonga despite similar proportional

mortality, to cancers playing a more important role in Palau. Kiribati, at the other end

of the spectrum early in the epidemiological transition, still faces substantial public

health issues related to infant and child mortality and infectious diseases across all

ages; yet is also experiencing rapid increases in the prevalence of risk factors for

373

NCDs and the subsequent double burden of disease in adults. Nauru, in comparison,

illustrates an unusual pathway through the epidemiological transition; with high

premature mortality due to NCDs and very few deaths in adults attributed to

infectious diseases, yet almost no progress over 50 years in reducing infant and

child mortality (112).

As mortality from infectious diseases falls, the early transition countries of Vanuatu

and Kiribati (and most likely their fellow Melanesian countries of PNG and the

Solomon Islands) present a real opportunity to divert these populations from

experiencing the same stagnation in LE as seen in the late transition countries. This

would require an urgent and significant investment in NCD prevention programs.

Additionally, the similarities across the study countries suggest may be possible to

develop a regional set of life tables that better reflect the experiences of the PICTs,

which could then be used to help fill gaps for those countries that do not yet have a

reliable CRVS system.

Recommendations from the work presented in this thesis, both in relation to

assessment and improvement of routine data collection and reporting systems and in

terms of public health priorities, are outlined in the following section.

8.2 Recommendations

Most importantly this work highlights the importance of using the available empirical

data, despite its shortcomings, rather than waiting until systems are perfect and

making decisions solely on modelled data.

374

8.2.1 Recommendations for system improvement of CRVS systems

CRVS systems across the region require strengthening to improve the completeness

and quality of data available. As demonstrated in the work included in this thesis,

much can be made of incomplete data from current systems and an imperfect

system should not be ignored as a data source. The data correction and review

techniques which have been presented are however no substitute for functional high

quality data collection and reporting from vital registration covering a representative

population with medical certification, or verbal autopsy where certification is not

possible.

Key recommendations for system improvement based on the strengths and

weaknesses common across the study countries were outlined in case study one.

These included: improving data sharing between departments, minimising

duplication in data collection and entry, improving resourcing of CRVS systems, and

allocating clear roles and responsibilities for data collection, analysis and reporting.

The importance of human elements, including responsibility and authority to

undertake assigned responsibilities, and ownership to ensure improvements are

sustained (164), should not be under-valued. Close interaction between health staff

and local communities provides a solid foundation for improvement in death

reporting.

Although there are common priorities for system improvement across the region, it

must be noted there is a great deal of diversity across PICTs in terms of

infrastructure, social context and existing reporting system capacity. As such, there

375

is no generic solution for these issues and any action to address these priorities will

need to be locally appropriate. There is significant work underway in the region

under the Pacific Vital Statistics Action Plan 2011-2014, endorsed by Pacific

governments under the Ten Year Pacific Statistics Strategy and supported by SPC,

the UQ HIS Hub, WHO, UNICEF and UNFPA and among others. The plan

encourages and assists countries to undertake a national assessment of their CRVS

system and develop a coordinated national improvement plan that can then be

supported by donor partners. Each of the study countries have commenced work on

national improvement plans (319).

A key priority for governments in addressing the current system gaps should be a

commitment to count every death, and that CoD reporting will be implemented in line

with the recommendations of the ICD, including medical certification and coding

practices. This requires adequate government financial and political commitment to

begin to address recommendations in a way that builds sustainable systems.

Despite current CRVS system flaws, there is a significant amount of data currently

collected that cannot be adequately accessed, interrogated or used for health

planning. There is a critical need to improve this situation to ensure available data is

used at a national level, and it is made available regionally and internationally

through reporting mechanisms such as the WHO CHIPs. It is this data that will give

the Pacific Islands a strong voice in international forums such as discussions for the

post 2015 development agenda (385), and it is imperative to establish a user

demand that PICTs should not be represented by modelled estimates when

plausible empirical data exists.

376

International and regional agencies must act to support countries in making data

more visible and accessible. This will encourage ongoing use of the data in health

planning and decision making, regional and international comparisons, and

monitoring and accountability (of both countries and donor agencies) towards

development goals; and in doing so, facilitate the ongoing improvement of data.

8.2.2 Recommendations for further research

The research presented in this thesis highlights a critical need for additional data on

the impact of migration on mortality patterns in the PICTs. To date, this has been

limited by the poor quality of migration data (132), and the difficulty of obtaining this

data given population mobility and large proportion of people that maintain multiple

homes in different countries. Given the recent release of the 2010 GBD (353), there

is also scope to examine the modelling methods used in that research, and update

these with empirical findings presented in this thesis to continue to provide more

accurate options for modelling data where vital registration is not yet fully reliable.

It is further recommended a system assessment is undertaken for those countries

that have not already done so, both to inform system improvements and identify and

understand existing data sources. The Health Minsters’ declaration of NCDs as an

emergency in the region (5), along with moves to incorporate NCDs into the post

2015 MDG agenda (385), mean governments will increasingly need to accurately

monitor the impacts of premature adult mortality from specific causes of death; not

just as a one off exercise when external research support is available, but through

the routine collection, analysis and reporting of robust CoD data. If such monitoring

and target setting is to be meaningful, there is also a need for all PICTs to establish

377

current baselines of adult mortality by cause, as done for selected countries within

this research.

The need for specific action to obtain more detailed data on specific causes of death

has also been discussed in this thesis. Particular areas of need are further

disaggregated data on deaths from external causes in light of the poor data

recording noted in this area, and the high mortality from suicide which has been

noted in Kiribati. The extent of this problem has largely gone unrecognised as

suicide related deaths are unlikely occur in hospital and are therefore not reported in

the hospital data.

Mortality level and CoD findings from Nauru and Kiribati further identified much of the

continued infant and child mortality is from causes that should be amenable to

intervention (386), such as malnutrition. Further investigation to inform an

intervention strategy should be urgently undertaken.

Finally, although limited to selected study countries, the research presented in this

thesis demonstrates serious disparities in health across the region, with LE having

remained steady at very low levels. Further monitoring of these disparities, and

raising greater awareness of these concerns with governments, funding agencies,

and civil society, is critical if there is to be tangible improvements in health status in

the PICTs in the foreseeable future. Targeted, resourced, and sustained support to

health agencies and governments will be required to realise substantial health

improvements and ‘kick start’ LE improvements in the region.

378

9 REFERENCES

1.� WHO.�Health�Metrics�Network.�Geneva:�World�Health�Organization;�2012�[2/5/2012];�Available�from:�http://www.who.int/healthmetrics/en/.�2.� Taylor�R,�Bampton�D,�Lopez�AD.�Contemporary�patterns�of�Pacific�Island�mortality.�International�journal�of�epidemiology.�2005;34(1):207�14.�3.� Taylor�R,�Lewis�ND,�Levy�S.�Societies�in�transition:�mortality�patterns�in�Pacific�Island�populations.�International�journal�of�epidemiology.�1989;18(3):634�46.�4.� Mathers�CD,�Fat�DM,�Inoue�M,�Rao�C,�Lopez�AD.�Counting�the�dead�and�what�they�died�from:�an�assessment�of�the�global�status�of�cause�of�death�data.�Bulletin�of�the�World�Health�Organization.�2005;83(3):171�7.�5.� Pacific��Island�Countries�Health�Ministers,�editor.�Honiara�Communique.�Pacific��Island�Countries�Health�Ministers�Meeting;�2011;�Solomon�Islands:�WHO�and�SPC.�6.� Secretariat�PIF.�2011�Pacific�Plan�Annual�Progress�Report�PIFS,�2011.�7.� AbouZahr�C,�Cleland�J,�Coullare�F,�Macfarlane�SB,�Notzon�FC,�Setel�P,�et�al.�The�way�forward.�Lancet.�2007;370(9601):1791�9.�Epub�2007/11/22.�8.� AbouZahr�C,�Boerma�T.�Health�information�systems:�the�foundations�of�public�health.�Bulletin�of�the�World�Health�Organization.�2005;83(8):578�83.�Epub�2005/09/27.�9.� Mathers�CD,�Murray�CJ,�Ezzati�M,�Gakidou�E,�Salomon�JA,�Stein�C.�Population�health�metrics:�crucial�inputs�to�the�development�of�evidence�for�health�policy.�Population�health�metrics.�2003;1(1):6.�10.� Swartout�HO,�Webster�RG.�To�what�degree�are�mortality�statistics�dependable.�American�journal�of�public�health.�1940;30:811�5.�11.� Byass�P.�Who�needs�cause�of�death�data?�PLoS�medicine.�2007;4(11):1715.�12.� Detmer�DE.�Building�the�national�health�information�infrastructure�is�a�personal�health,�health�care�services,�public�health�and�research.�BMC�medical�Informatics�and�decision�making.�2003;3(1).�13.� Murray�CJ.�A�critical�review�of�international�mortality�data.�Social�science�&�medicine�(1982).�1987;25(7):773�81.�14.� Darefsky�AS,�Dubrow�R.�International�variation�in�the�incidence�of�adult�primary�malignant�neoplasms�of�the�brain�and�central�nervous�system.�Cancer�Causes�Control.�2009;20(9):1593�604.�Epub�2009/07/28.�15.� Murray�CJ,�Lopez�AD,�Feehan�DM,�Peter�ST,�Yang�G.�Validation�of�the�symptom�pattern�method�for�analyzing�verbal�autopsy�data.�PLoS�medicine.�2007;4(11):e327.�16.� Setel�PW,�Macfarlane�SB,�Szreter�S,�Mikkelsen�L,�Jha�P,�Stout�S,�et�al.�A�scandal�of�invisibility:�making�everyone�count�by�counting�everyone.�Lancet.�2007;370(9598):1569�77.�Epub�2007/11/10.�17.� World�Health�Organization.�History�of�the�development�of�the�ICD.�Geneva:�World�Health�Organization;��[2/1/2013];�Available�from:�http://www.who.int/classifications/icd/en/HistoryOfICD.pdf.�18.� Plummer�IC.�Educating�medical�students�and�other�groups�concerned�with�registration�of�vital�statistics�data.�American�journal�of�public�health.�1940;30:270�2.�19.� Bhopal�R.�Fallout�from�the�Shipman�case.�Death�registers�in�general�practice�would�be�a�means�of�preventing�malpractice�and�murder.�BMJ�(Clinical�research�ed.�2000;320(7244):1272.�Epub�2000/06/10.�20.� Berry�C.�Death�certification�and�Dr.�Shipman.�Med�Sci�Law.�2003;43(3):193�4.�21.� Memon�MI,�Memon�MA,�Horner�JS,�McCann�MH.�Medicolegal�aspect�of�death:�medical�referee's�input�in�the�aftermath�of�Shipman.�J�Eval�Clin�Pract.�2001;7(1):81�3.�

379

22.� Dyer�C.�Shipman�inquiry�calls�for�major�changes�in�death�certification.�BMJ�(Clinical�research�ed.�2002;325(7370):919.�23.� Holden�J,�O'Donnell�S.�Shipman�proposals�will�alter�general�practice�profoundly.�BMJ�(Clinical�research�ed.�2003;326(7383):280.�24.� Smith�DJ.�Certifying�and�investigating�deaths:�the�need�for�reform.�1.�The�shipman�inquiry�death�certification.�Med�Sci�Law.�2004;44(4):280�7.�25.� Sinfield�P,�Wilson�A.�Death�certification:�issues�from�a�pilot�of�the�Shipman�Inquiry's�interim�proposals.�Health�Stat�Q.�2005(26):23�32.�26.� Editorial:�Revising�death�certification�in�the�UK.�The�Lancet.�2007;369:716.�27.� Pacific�Islands�Regional�Millenium�Development�Goals�Report�2004.�Noumea:�Secretariat�of�the�Pacific�Community�in�cooperation�with�the�United�Nations,�2004.�28.� Macfarlane�SB,�Thomson�M,�AbouZahr�CL.�Millennium:�invest�in�country�statistical�systems.�Nature.�2007;446(7139):974.�Epub�2007/04/27.�29.� Foundation�TBaMG.�The�Global�Fund�to�fight�AIDS,�Tuberculosis�and�Malaria.��[2/5/2012];�Available�from:�http://www.theglobalfund.org/en/about/donors/private/gatesfoundation/.�30.� Government�U.�About�PEPFAR.�2012�[1/5/2012];�Available�from:�http://www.pepfar.gov/about/index.htm.�31.� .�The�Marrakech�Action�Plan�for�Statistics.�Second�International�Roundtable�on�Managing�for�Development�Results;�2004;�Marrakech,�Morocco:�Multilateral�Development�Banks�(African�Development�Bank,�European�Bank�for�Reconstruction�and�Development,�Inter�American�Development�Bank,�and�World�Bank).�32.� Murray�CJ,�Laakso�T,�Shibuya�K,�Hill�K,�Lopez�AD.�Can�we�achieve�Millennium�Development�Goal�4?�New�analysis�of�country�trends�and�forecasts�of�under�5�mortality�to�2015.�Lancet.�2007;370(9592):1040�54.�Epub�2007/09/25.�33.� Lopez�AD,�AbouZahr�C,�Shibuya�K,�Gollogly�L.�Keeping�count:�births,�deaths,�and�causes�of�death.�Lancet.�2007;370(9601):1744�6.�Epub�2007/11/22.�34.� Mahapatra�P,�Shibuya�K,�Lopez�AD,�Coullare�F,�Notzon�FC,�Rao�C,�et�al.�Civil�registration�systems�and�vital�statistics:�successes�and�missed�opportunities.�Lancet.�2007.�35.� Setel�PW,�Macfarlane�SB,�Szreter�S,�Mikkelsen�L,�Jha�P,�Stout�S,�et�al.�A�scandal�of�invisibility:�making�everyone�count�by�counting�everyone.�Lancet.�2007.�36.� Tukuitonga�C.�Background�Paper:�Population�Health�in�the�Pacific�Region��A�situation�analysis.��WHO/SPC�Meeting�of�Pacific�Health�Ministers2007.�37.� World�Bank.�World�Bank�List�of�Economies�2012.�New�York2012;�Available�from:�siteresources.worldbank.org/DATASTATISTICS/.../CLASS.XLS.�38.� Haberkorn�G.�Vital�registration�and�health�programme�monitoring�in�Pacific�Island�countries–some�myths�and�realities.�Public�Health�Surveillance�in�the�Pacific�Noumea:�Secretariat�of�the�Pacific�Community.�1998:91�104.�39.� Khaleghian�P.�Noncommunicable�diseases�in�Pacific�Island�Countries:�Disease�burden,�economic�costs�and�policy�options.�Noumea:�Secretariat�of�the�Pacific�Community�and�World�bank,�2003.�40.� Coyne�T.�Lifestyle�Diseases�In�Pacific�Communities.�Hughes�R,�Langi�S,�editors.�Noumea:�Secretariat�of�the�Pacfic�Community;�2000.�41.� Rajaratnam�JK,�Marcus�JR,�Levin�Rector�A,�Chalupka�AN,�Wang�H,�Dwyer�L,�et�al.�Worldwide�mortality�in�men�and�women�aged�15�59�years�from�1970�to�2010:�a�systematic�analysis.�Lancet.�2010;375(9727):1704�20.�Epub�2010/05/04.�42.� 50�years�in�the�Western�Pacific��Part�1.�Manila:�World�Health�Organisation;�2010�[cited�2010�28/7/2010];�Available�from:�http://www.wpro.who.int/internet/resources.ashx/RCM/RC49_02.pdf.�43.� PRISM��Pacific�Regional�Information�System.�Noumea:�Secretariat�of�the�Pacific�Community;�2010�[cited�2010�26/7/2010];�Available�from:�http://www.spc.int/prism/.�44.� Secretariat�of�the�Pacific�Community.�2011�Population�and�demographic�indicators.�Noumea:�SPC,�2012.�

380

45.� ABC�Radio�Australia.�Charting�the�Pacific.�Australian�Broadcasting�Corporation;�2012�[2/1/2013];�Available�from:�www.abc.net.au/ra/pacific/places.�46.� Secretariat�of�the�Pacific�Community.�Economic�Regional�Data:�GDP�summary.�Noumea:�SPC;�2012�[1/9/2012];�Available�from:�www.spc.int/prism/data.�47.� Pacific�Islands�Forum�Secretariat.�Pacific�Islands�Forum�Secretariat.��[2/1/2013];�Available�from:�http://www.forumsec.org/pages.cfm.�48.� Fiji�Bureau�of�Statistics.�2007�Census�Report.�Suva:�2007.�49.� Millenium�Development�Goals.�Tarawa:�Republic�of�Kiribati,�2007.�50.� Statistics�and�Demography�Program�SPC,�Kiribati�Statistics�Office.�Kiribati�2005�Census.�Noumea:�2007.�51.� United�Nations.�World�Population�Prospects:�The�2010�Revision,�New�York,�2011�(comprehensive�Excel�tables).�New�York:�United�Nations,�Department�of�Economic�and�Social�Affairs,�Population�Division,,�2011.�52.� Statistics�of�the�Human�Development�Report.�UNDP;��[20/8/2010];�Available�from:�http://hdr.undp.org/en/statistics/.�53.� Klugman�J.�Human�Development�Report�2011.�Sustainability�and�Equity:�A�Better.�2011.�54.� Gadabu�I.�Nauru�Household�Income�and�Economic�Survey.�Nauru:�2006.�55.� Republic�of�Nauru.�National�Sustainable�Development�Strategy�2005�2025,�2009�Revision.�.�Nauru:�2009.�56.� Department�of�Foreign�Affairs�and�Trade.�Nauru�Country�Brief.�Australian�Government;�2012�[2/1/2013];�Available�from:�http://www.dfat.gov.au/geo/nauru/nauru_brief.html.�57.� Palau�Office�of�Statistics,�US�Census�Bureau.�2005�Census�of�Population�and�Housing��Monograph�Final�Report.�Palau:�2005.�58.� Gerundio�Dizon�A.�US�officials�reiterate�support�for�Palau�Compact�Approval.�Island�Times��in�the�Pacific�Islands�Report.�2012�31/10/2012.�59.� Tonga�Department�of�Statisitcs.�Population�Census�2011�(Preliminary�count).�Tonga:�Government�of�Tonga;�2012�[2/1/2013];�Available�from:�www.spc.int/prism/tonga/index.php/population�censuses.�60.� Trade�ADoFAa.�Kingdom�of�Tonga�Country�Brief.�Australian�Government;��[2/1/2013].�61.� Vanuatu�National�Statistics�Office.�Vanuatu�Census�2009:�Summary�Report.�Port�Vila:�VNSO,�2011.�62.� Government�of�Vanuatu,�Programme�UND.�Republic�of�Vanuatu�Millennium�Development�Goals�Report.�2005.�63.� Rao�C,�Yang�G,�Hu�J,�Ma�J,�Xia�W,�Lopez�AD.�Validation�of�cause�of�death�statistics�in�urban�China.�International�journal�of�epidemiology.�2007;36(3):642�51.�64.� Rao�C,�Bradshaw�D,�Mathers�CD.�Improving�death�registration�and�statistics�in�developing�countries:�lessons�from�Sub�Saharan�Africa.�Southern�African�Journal�of�Demography.�2004:81�99.�65.� Preston�SH.�The�Contours�of�Demography:�Estimates�and�Projections.�Demography.�1993;30(4):593�606.�66.� Rowland�DT.�Demographic�Methods�and�Concepts.�New�York:�Oxford�University�Press;�2003.�67.� OECD.�OECD�Glossary�of�Statistical�Terms�2002.�68.� World�Health�Organization.�International�Classification�of�Diseases.�2011;10.�69.� Anderson�RN,�Minino�AM,�Hoyert�DL,�Rosenberg�HM.�Comparability�of�cause�of�death�between�ICD�9�and�ICD�10:�preliminary�estimates.�Natl�Vital�Stat�Rep.�2001;49(2):1�32.�70.� National�Centre�for�Classification�in�Health�(NCCH).�The�International�Statistical�Classification�of�Diseases�and�Related�Health�Problems,�Tenth�Revision,�Australian�Modification�(ICD�10�AM)�–�Fifth�Edition��Tabular�list�of�diseases�and�Alphabetic�index�of�diseases.�.�Sydney:�NCCH,�Faculty�of�Health�Sciences,�The�University�of�Sydney.;�2006.�71.� Adair�T,�Rao�C.�Changes�in�certification�of�diabetes�with�cardiovascular�diseases�increased�reported�diabetes�mortality�in�Australia�and�the�United�States.�Journal�of�Clinical�Epidemiology.�2010;63(2):199�204.�

381

72.� Taylor�R.�History�of�Public�Health�in�Pacific�Island�Countries.�In:�Lewis�MJ,�K.L;�M,�editors.�Public�Health�in�Asia�and�the�Pacific:�Historical�and�comparative�perspectives.�London:�Routledge;�2008.�p.�277�307.�73.� Pirie�P.�The�effects�of�Treponematosis�and�Gonorrhoea�on�the�Populations�of�the�Pacific�Islands.�Human�Biology�in�Oceana.�1972;1(3):187�206.�74.� Taylor�R,�Zimmet�P,�Tuomilehto�J,�Ram�P,�Hunt�D,�Sloman�G.�Blood�pressure�changes�with�age�in�two�ethnic�groups�in�Fiji.�Journal�of�the�American�College�of�Nutrition.�1989;8(4):335�46.�75.� Hezel�FX.�Disease�in�Micronesia:�a�historical�survey.�Pacific�health�dialog.16(1):11�25.�76.� Allen�BJ.�A�bomb�or�a�bullet�or�the�bloody�flux?�Population�change�in�the�Aitape�inland,�Papua�New�Guinea,�1941�1945.�The�Journal�of�Pacific�history.�1983;18(4):218�35.�77.� Underwood�JH.�Population�history�of�Nauru:�a�cautionary�tale.�Micronesica.�1989;22(1):3�22.�78.� Ulijaszek�S.�Modernisation,�migration�and�nutritional�health�of�Pacific�Island�populations.�Environ�Sci.�2005;12(3):167�76.�79.� Hezel�FX.�Disease�in�Micronesia:�a�historical�survey.�Pacific�health�dialog.�2010;16(1):11�25.�Epub�2010/10/26.�80.� Terrell�Perica�S.�Background�Paper�5.�Food�and�lifestyle�related�diseases�in�the�Pacific:�The�size�of�the�problem�and�possible�Stratgies.��Population,�Food�and�Development�USP/SPC/UNFPA�Regional�Symposium;�Suva:�Population�Studies�Program,�University�of�the�South�Pacific�1990.�81.� Expanded�programme�on�Immunisation.�World�Health�Organization;��[28/7/2010];�Available�from:�http://www.wpro.who.int/sites/epi/overview.htm.�82.� Sladden�T.�Twenty�years�of�HIV�surveillance�in�the�Pacific��what�do�the�data�tell�us�and�what�do�we�still�need�to�know?�Pacific�Health�Dialog:�Pacific�Health�Surveillance�and�Response.�2005;12(2):23��37.�83.� AIDS�U.�UNAIDS�Terminology.�Revision�October�2011.�2011.�84.� Hayes�GR.�Microcomputers�and�minipopulations:�the�1981�Cook�Islands�census.�Asian�and�Pacific�census�forum�/�East�West�Population�Institute.�1983;10(2):1,�2,�11�6.�85.� Schooneveldt�M,�Songer�T,�Zimmet�P,�Thoma�K.�Changing�mortality�patterns�in�Nauruans:�an�example�of�epidemiological�transition.�Journal�of�epidemiology�and�community�health.�1988;42(1):89�95.�86.� Lewis�ND,�Rapaport�M.�In�a�sea�of�change:�health�transitions�in�the�Pacific.�Health�and�Place.�1995;1(4):211�26.�87.� Finau�SA,�Wainiqolo�IL,�Cuboni�GG.�Health�transition�among�Pacificans:�unpacking�imperialism.�Pacific�health�dialog.�2002;9(2):254�62.�88.� Jones�RM.�Health�transitions�in�the�Pacific.�Pacific�health�dialog.�2002;9(2):176.�89.� Taylor�R.�The�Double�Disease�Burden�in�Pacific�Island�States�(except�New�Guinea),�.�In:�Lewis�M,�MacPherson�K,�editors.�Health�Transitions�and�the�Double�Disease�Burden�in�Asia�and�the�Pacific�Histories�of�Responses�to�Non�Communicable�and�Communicable�Diseases:�Routledge;�2012.�90.� Taylor�R,�Henderson�B,�Levy�S,�Kolonel�L,�Lewis�N.�Cancer�in�Pacific�Islands�Countries.�Noumea:�South�Pacific�Commission,�1985.�91.� Health�Status�Report��Fiji.�Suva:�Ministry�of�Health,�Government�of�Fiji,�1996.�92.� Solomon�Islands�1986�Population�Census:�Report�2B��Data�analysis.�Bureau�of�Statistics,�Government�of�the�Solomon�Islands,�1986.�93.� Carter�K,�Cornelius�M,�Taylor�R,�Ali�SS,�Rao�C,�Lopez�AD,�et�al.�Mortality�trends�in�Fiji.�Aust�N�Z�J�Public�Health.�2011;35(5):412�20.�Epub�2011/10/07.�94.� Samo�M,�Elymore�A.�Mortality�analysis�of�registered�deaths�in�the�FSM�from�1990�2003.�Pacific�health�dialog.�2010;16(1):115�22.�Epub�2010/10/26.�95.� Report�of�the�Minister�of�Health�for�the�year�2005.�Nuku'�Alofa:�Ministry�of�Health,�Government�of�Tonga,�2005.�96.� Report�of�the�Minister�of�Health�for�the�year�2007.�Nuku'�Alofa:�Ministry�of�Health,�Government�of�Tonga,�2007.�97.� Marshall�Islands�Statistical�Abstract�1995.�Majuro:�Office�of�Planning�and�Statistics,�Republic�of�the�Marshall�Islands,�1995.�

382

98.� Demmke�A,�Haberkorn�G,�Rakaseta�VL,�Lepers�C.�Federated�States�of�Micronesia�Population�Profile.�Noumea:�South�Pacific�Commission,�1997.�99.� Lopez�AD.�Commentary:�Estimating�the�causes�of�child�deaths.�International�journal�of�epidemiology.�2003;32(6):1052�3.�100.� AbouZahr�C,�Cleland�J,�Coullare�F,�Macfarlane�SB,�Notzon�FC,�Setel�P,�et�al.�The�way�forward.�Lancet.�2007;370(9601):1791�9.�Epub�2007/11/22.�101.� Murray�CJ,�Lopez�AD.�Alternative�projections�of�mortality�and�disability�by�cause�1990�2020:�Global�Burden�of�Disease�Study.�Lancet.�1997;349(9064):1498�504.�102.� Murray�CJ,�Lopez�AD.�Global�mortality,�disability,�and�the�contribution�of�risk�factors:�Global�Burden�of�Disease�Study.�Lancet.�1997;349(9063):1436�42.�103.� Murray�CJ,�Lopez�AD.�Mortality�by�cause�for�eight�regions�of�the�world:�Global�Burden�of�Disease�Study.�Lancet.�1997;349(9061):1269�76.�104.� Murray�C,�Lopez�A.�The�Global�Burden�of�Disease:�a�comprehensive�assessment�of�mortality�and�disability�from�diseases,�injuries�and�risk�factors�in�1990�and�projected�to�2020�.�Cambridge:�Harvard�School�of�Public�Health,�1996��Contract�No.:�Vol.�1.�105.� Lopez�AD,�Mathers�CD.�Measuring�the�global�burden�of�disease�and�epidemiological�transitions:�2002�2030.�Annals�of�tropical�medicine�and�parasitology.�2006;100(5�6):481�99.�106.� Jenkins�M,�Mera�J,�Hanerg�K,�James�A,�Mendoza�O,�Nasiru�I,�et�al.�Marshall�Islands�Vital�and�Health�Statistics�Abstract�1989�1993.�Majuro:�Ministry�of�Health�and�Environment,�Republic�of�the�Marshall�Islands,�1993.�107.� Tonga�Ministry�of�Health.�Tonga�Ministry�of�Health�Annual�Report.�Tonga:�2007.�108.� Western�Pacific�Regional�Office�WHO.�Country�Health�Information�Profiles�2010.�Manila:�World�Health�Organisation;�2010.�109.� Division�UNS.�UN�Statistical�yearbook:�United�Nations�Publications;�2010.�110.� Secretariat�of�the�Pacific�Community.�PRISM�Regional�Tables:�Births�Deaths�and�Migration�Data.�2011;�Available�from:�http://www.spc.int/prism/births�deaths�and�migration.�111.� Hufanga�S,�Carter�KL,�Rao�C,�Lopez�AD,�Taylor�R.�Mortality�trends�in�Tonga:�an�assessment�based�on�a�synthesis�of�local�data.�Population�health�metrics.�2012;10(1):14.�Epub�2012/08/16.�112.� Carter�K,�Soakai�TS,�Taylor�R,�Gadabu�I,�Rao�C,�Thoma�K,�et�al.�Mortality�trends�and�the�epidemiological�transition�in�Nauru.�Asia�Pacific�journal�of�public�health�/�Asia�Pacific�Academic�Consortium�for�Public�Health.�2011;23(1):10�23.�Epub�2010/12/21.�113.� Fiji�Millennium�Development�Goals�2nd�Report.�In:�Fiji�Go,�editor.�Suva2009.�114.� Pacific�Islands�Forum�Secretariat.�Pacific�Regional�MDGs�Tracking�Report.�Suva,�Fiji:�PIFs;�2004.�115.� Taylor�R,�Lewis�ND,�Sladden�T.�Mortality�in�Pacific�island�countries�around�1980:�geopolitical,�socioeconomic,�demographic�and�health�service�factors.�Australian�journal�of�public�health.�1991;15(3):207�21.�116.� Taylor�R.�Mortality�Trends�and�Differentials�in�Fiji.�1989.�117.� Division�UNS.�Principles�and�recommendations�for�a�vital�statistics�system:�United�Nations,�Department�of�Economic�and�Social�Affairs,�Statistics�Division;�2001.�118.� Shlomowitz�R.�Infant�mortality�and�Fiji's�Indian�migrants,�1879�1919.�The�Indian�economic�and�social�history�review.�1986;23(3):289�302.�119.� Government�of�Japan.�Japanese�Census�of�the�Micronesian�Islands,�1930�and�1935.�1935.�120.� Islands�HCotTTotP.�Census�Report��Trust�Territories�of�the�Pacific�Islands.�1958.�121.� Carrier�NH,�Farrag�AM.�The�Reduction�of�Errors�in�Census�Populations�for�Statistically�Underdeveloped�Countries.�Population�Studies.�1959;12(3):240�85.�122.� Office�VNS.�Vanuatu�Population�Census�1967.�Port�Vila:�Bureau�of�Statistics,�Government�of�Vanuatu,�1967.�123.� Groenewegen�K.�Censuses�in�the�South�Pacific,�1976�1977.�Asian�and�Pacific�census�forum�/�East�West�Population�Institute.�1979;5(3):1�2,�4�8.�

383

124.� Murray�CJ,�Rajaratnam�JK,�Marcus�J,�Laakso�T,�Lopez�AD.�What�can�we�conclude�from�death�registration?�Improved�methods�for�evaluating�completeness.�PLoS�medicine.�2010;7(4):e1000262.�Epub�2010/04/21.�125.� Hill�K.�Estimating�census�and�death�registration�completeness.�Asian�&�Pacific�Population�Forum.�1987;1(3):8�13.�126.� Preston�SH,�Bennett�NG.�A�Census�based�Method�for�Estimating�Adult�Mortality.�Population�Studies.�1983;37(1):91�104.�127.� Nations�U.�Manual�X:�Indirect�Techniques�for�Demographic�Estimation�New�York:�United�Nations;�1983.�128.� Timaeus�IM.�Measurement�of�adult�mortality�in�less�developed�countries:�a�comparative�review.�Population�index.�1991;57(4):552�68.�129.� Garenne�M,�Friedberg�F.�Accuracy�of�indirect�estimates�of�maternal�mortality:�a�simulation�model.�Studies�in�family�planning.�1997;28(2):132�42.�130.� Fernando�C.�Indirect�estimation�of�infant�mortality�trends:�simulation�tests�on�the�Feeney�method.�Genus.�1985;41(3�4):65�88.�Epub�1985/07/01.�131.� Brass�W.�Demographic�Data�Analysis�in�Less�Developed�Countries:�1946�1996.�Population�Studies.�1996;50(3):451�67.�132.� Tomas�P,�Summers�L,�Clemens�M.�Migrants�Count.�Five�steps�toward�better�migration�data.�Washington:�Centre�for�Global�Development;�2009.�133.� Haberkorn�G.�A�sea�of�islands—a�myriad�of�indicators:�on�the�interface�between�demography�and�planning�in�the�Pacific�Islands.�Asia�Pacific�Viewpoint.�1997;38(3):219�36.�134.� Nauru�Bureau�of�Statistics,�Statistics�and�Demography�Programmme�SPC.�Nauru�2011�Census�of�Population�and�Housing��Preliminary�Report.�2011.�135.� Office�VNS.�Vanuatu�Population�and�Housing�Census�2009.�Port�Vila:�Bureau�of�Statistics,�Government�of�Vanuatu,�2009.�136.� Statistics�TDo.�Tonga�2006�Analytical�Census�Report.�2006.�137.� Nauru�Bureau�of�Statistics,�Statistics�and�Demography�Programmme�SPC.�Nauru�2002�Census�of�Population�and�Housing.�2002.�138.� 1999�National�Population�Census:�Demographic�Analysis�Report.�In:�Office�NS,�editor.:�Republic�of�Vanuatu;�2001.�139.� Demographic�Profile�of�New�Caledonia:�based�on�the�1996�Census.�Noumea:�Secretariat�of�the�Pacific�Community,�2001.�140.� Palau�Office�of�Statistics.�2000�Census�of�Population�and�Housing�of�the�Republic�of�Palau.�Koror:�Office�of�Planning�and�Statistics,�Republic�of�Palau,�2000.�141.� 1999�Census�of�Population�and�Housing:�Final�Report.�Majuro:�Office�of�Planning�and�Statistics,�Republic�of�the�Marshall�Islnads,�1999.�142.� Groenewold�G,�Van�Ginneken�J.�Census�Analysis��Confidential�report��Solomon�Islands.�1999.�143.� Statisitcs�TDo.�Tonga�1996�Census�Analytical�Report.�Tonga:�1996.�144.� National�Population�Census�1989:�Demographic�and�Migration�Analysis.�In:�Office�NS,�editor.:�Government�of�Vanuatu;�1993.�145.� May�JF.�Census�taking�in�Niue,�South�Pacific.�Population�today.�1992;20(2):5.�146.� Kaitu�H,�Telupe�T.�1991�Population�and�Housing�Census,�Volume�1:�Basic�Information.�Funafuti:�Central�Statistics�Division,�Government�of�Tuvalu,�1991.�147.� Fiji�Census.�Bureau�of�Statistics,�Government�of�Fiji,�1989.�148.� Office�VNS.�Vanuatu�Population�Census�1979.�Port�Vila:�Bureau�of�Statistics,�Government�of�Vanuatu,�1979.�149.� Vanuatu�Population�Census�1989.�Port�Vila:�Bureau�of�Statistics,�Government�of�Vanuatu,�1993.�150.� Meredith�I,�Sarfati�D,�Ikeda�T,�Atkinson�J,�Blakely�T.�High�rates�of�endometrial�cancer�among�Pacific�women�in�New�Zealand:�the�role�of�diabetes,�physical�inactivity,�and�obesity.�Cancer�Causes�Control.�2012;23(6):875�85.�Epub�2012/04/25.�

384

151.� National�Census�Report:�1994�Census�of�Population�and�Housing.�Phonepei:�Office�of�Planning�and�Statistics,�Federated�States�of�Micronesia,�1996.�152.� Booth�H.�The�demography�of�Kiribati:�estimates�from�the�1985�census.�Asia�Pacific�population�journal�/�United�Nations.�1994;9(3):55�72.�Epub�1994/09/01.�153.� Huy�TQ,�Long�NH,�Hoa�DP,�Byass�P,�Ericksson�B.�Validity�and�completeness�of�death�reporting�and�registration�in�a�rural�district�of�Vietnam.�Scandinavian�journal�of�public�health.�2003;62:12�8.�154.� Lilienfeld�AM.�Practical�limitations�of�epidemiologic�methods.�Environmental�health�perspectives.�1983;52:3�8.�Epub�1983/10/01.�155.� Obermeyer�Z,�Rajaratnam�JK,�Park�CH,�Gakidou�E,�Hogan�MC,�Lopez�AD,�et�al.�Measuring�adult�mortality�using�sibling�survival:�a�new�analytical�method�and�new�results�for�44�countries,�1974�2006.�PLoS�medicine.7(4):e1000260.�156.� Nauru�Bureau�of�Statistics,�Statistics�and�Demography�Programmme�SPC.�Nauru�Demographic�and�Health�Survey.�Noumea:�2007.�157.� Vanuatu�Ministry�of�Health,�UNICEF.�Vanuatu�Multiple�Indicator�Cluster�Survey�2007.�Port�Vila:�Government�of�Vanuatu,�2008.�158.� Stanton�C,�Abderrahim�N,�Hill�K.�An�assessment�of�DHS�maternal�mortality�indicators.�Studies�in�family�planning.�2000;31(2):111�23.�Epub�2000/07/25.�159.� Bairagi�R,�Becker�S,�Kantner�A,�Allen�KB,�Datta�A,�Purvis�K.�An�evaluation�of�the�1993�94�Bangladesh�Demographic�and�Health�Survey�within�the�Matlab�area.�Asia�Pac�Popul�Res�Abstr.�1997(11):1�2.�160.� Palloni�A.�A�review�of�infant�mortality�trends�in�selected�underdeveloped�countries:�Some�new�estimates.�Population�Studies.�1981;35(1):100�19.�161.� United�Nations.�Handbook�of�vital�statistics�systems�and�methods.�Volume�II:�Review�of�national�practices.�New�York:�1985.�162.� AbouZahr�C,�Adjei�S,�Kanchanachitra�C.�From�data�to�policy:�good�practices�and�cautionary�tales.�Lancet.�2007;369(9566):1039�46.�Epub�2007/03/27.�163.� Abouzahr�C,�Gollogly�L,�Stevens�G.�Better�data�needed:�everyone�agrees,�but�no�one�wants�to�pay.�Lancet.�2010;375(9715):619�21.�Epub�2009/01/20.�164.� Carter�KL,�Rao�C,�Lopez�AD,�Taylor�R.�Mortality�and�cause�of�death�reporting�and�analysis�systems�in�seven�pacific�island�countries.�BMC�public�health.�2012;12(1):436.�165.� Isotalo�A.�Medico�Legal�Aspects�of�Medical�Certification�of�Cause�of�Death.�Acta�Med�Leg�Soc�(Liege).�1963;16:77�81.�166.� Campos�Outcalt�D.�Cause�of�death�certification��not�as�easy�as�it�seems.�J�Fam�Pract.�2005;54(2):134�8.�167.� McAllum�C,�St�George�I,�White�G.�Death�certification�and�doctors'�dilemmas:�a�qualitative�study�of�GPs'�perspectives.�Br�J�Gen�Pract.�2005;55(518):677�83.�168.� Aung�E,�Rao�C,�Walker�S.�Teaching�cause�of�death�certification:�lessons�from�international�experience.�Postgraduate�medical�journal.�2010;86(1013):143�52.�169.� Bloor�M.�A�minor�office:�the�variable�and�socially�constructed�character�of�death�certification�in�a�Scottish�city.�J�Health�Soc�Behav.�1991;32(3):273�87.�170.� Diehl�AK,�Gau�DW.�Death�certification�by�British�doctors:�a�demographic�analysis.�Journal�of�epidemiology�and�community�health.�1982;36(2):146�9.�171.� Dobson�AJ,�Gibberd�RW,�Leeder�SR.�Death�certification�and�coding�for�ischemic�heart�disease�in�Australia.�American�journal�of�epidemiology.�1983;117(4):397�405.�172.� Jougla�E,�Papoz�L,�Balkau�B,�Maguin�P,�Hatton�F.�Death�certificate�coding�practices�related�to�diabetes�in�European�countries��the�'EURODIAB�Subarea�C'�Study.�International�journal�of�epidemiology.�1992;21(2):343�51.�173.� Ravakhah�K.�Death�certificates�are�not�reliable:�revivification�of�the�autopsy.�Southern�medical�journal.�2006;99(7):728�33.�174.� Ashworth�TG.�Inadequacy�of�death�certification:�proposal�for�change.�J�Clin�Pathol.�1991;44(4):265�8.�

385

175.� Cameron�HM,�McGoogan�E.�A�prospective�study�of�1152�hospital�autopsies:�I.�Inaccuracies�in�death�certification.�J�Pathol.�1981;133(4):273�83.�176.� Wright�SC,�Evans�AE,�Sinnamon�DG,�MacMahon�S.�Asthma�mortality�and�death�certification�in�Northern�Ireland.�Thorax.�1994;49:141�3.�177.� Khosravi�A,�Rao�C,�Naghavi�M,�Taylor�R,�Jafari�N,�Lopez�AD.�Impact�of�misclassification�on�measures�of�cardiovascular�disease�mortality�in�the�Islamic�Republic�of�Iran:�a�cross�sectional�study.�Bulletin�of�the�World�Health�Organization.�2008;86(9):688�96.�178.� Mahapatra�P,�Chalapati�Rao�PV.�Cause�of�death�reporting�systems�in�India:�a�performance�analysis.�Natl�Med�J�India.�2001;14(3):154�62.�179.� Rao�C,�Lopez�AD,�Yang�G,�Begg�S,�Ma�J.�Evaluating�national�cause�of�death�statistics:�principles�and�application�to�the�case�of�China.�Bulletin�of�the�World�Health�Organization.�2005;83(8):618�25.�180.� Yang�G,�Hu�J,�Rao�KQ,�Ma�J,�Rao�C,�Lopez�AD.�Mortality�registration�and�surveillance�in�China:�History,�current�situation�and�challenges.�Population�health�metrics.�2005;3(1):3.�181.� Kumar�V,�Datta�N.�Lay�reporting�and�verbal�autopsy�in�assessment�of�infant�mortality.�Indian�journal�of�pediatrics.�1986;53(6):672�4.�Epub�1986/11/01.�182.� Wardlaw�JM,�Murray�V,�Berge�E,�del�Zoppo�G,�Sandercock�P,�Lindley�RL,�et�al.�Recombinant�tissue�plasminogen�activator�for�acute�ischaemic�stroke:�an�updated�systematic�review�and�meta�analysis.�Lancet.�2012;379(9834):2364�72.�Epub�2012/05/29.�183.� Ahmad�M,�D�A.�Counting�the�dead�and�the�living.�Indian�journal�for�the�Practicing�Doctor.�2008;5(1).�184.� Mozumder�KA,�Koenig�MA,�Phillips�JF,�Murad�S.�The�Sample�Registration�System:�an�innovative�system�for�monitoring�demographic�dynamics.�Asia�Pacific�population�journal�/�United�Nations.�1990;5(3):63�72.�185.� Jha�P,�Gajalakshmi�V,�Gupta�PC,�Kumar�R,�Mony�P,�Dhingra�N,�et�al.�Prospective�study�of�one�million�deaths�in�India:�rationale,�design,�and�validation�results.�PLoS�medicine.�2006;3(2):e18.�Epub�2005/12/16.�186.� Rao�C,�Osterberger�B,�Anh�TD,�MacDonald�M,�Chuc�NT,�Hill�PS.�Compiling�mortality�statistics�from�civil�registration�systems�in�Viet�Nam:�the�long�road�ahead.�Bulletin�of�the�World�Health�Organization.88(1):58�65.�187.� Jha�P,�Jacob�B,�Kumar�R.�Commentary:�Reliable�measurement�of�the�causes�of�mortality�in�developing�countries.�International�journal�of�epidemiology.�2007;36(3):651�3.�188.� Luker�V.�Epidemiological�transition,�public�health�and�the�Pacific.�Public�Health�in�Asia�and�the�Pacific:�Historical�and�Comparative�Perspectives.�2007;11:250.�189.� Taylor�R.�Small�area�population�disease�burden.�Aust�N�Z�J�Public�Health.�2001;25(4):289�93.�Epub�2001/09/01.�190.� Baiden�F,�Bawah�A,�Biai�S,�Binka�F,�Boerma�T,�Byass�P,�et�al.�Setting�international�standards�for�verbal�autopsy.�Bulletin�of�the�World�Health�Organization.�2007;85(8):570�1.�Epub�2007/09/05.�191.� Kakrani�VA,�Pratinidhi�AK,�Gupte�AM.�A�study�of�registration�of�deaths�at�primary�health�centre��with�special�reference�to�"verbal�autopsy"�method.�Indian�J�Med�Sci.�1996;50(6):196�200.�192.� Shrivastava�SP,�Kumar�A,�Kumar�Ojha�A.�Verbal�autopsy�determined�causes�of�neonatal�deaths.�Indian�pediatrics.�2001;38(9):1022�5.�Epub�2001/09/25.�193.� Gajalakshmi�V,�Peto�R,�Kanaka�S,�Balasubramanian�S.�Verbal�autopsy�of�48�000�adult�deaths�attributable�to�medical�causes�in�Chennai�(formerly�Madras),�India.�BMC�Public�Health.�2002;2:7.�194.� Fauveau�V.�Assessing�probable�causes�of�death�without�death�registration�or�certificates:�a�new�science?�Bulletin�of�the�World�Health�Organization.�2006;84(3):246�7.�195.� Fauveau�V,�Donnay�F.�Can�the�process�indicators�for�emergency�obstetric�care�assess�the�progress�of�maternal�mortality�reduction�programs?�An�examination�of�UNFPA�Projects�2000�2004.�International�journal�of�gynaecology�and�obstetrics:�the�official�organ�of�the�International�Federation�of�Gynaecology�and�Obstetrics.�2006;93(3):308�16.�Epub�2006/05/10.�

386

196.� Setel�PW,�Whiting�DR,�Hemed�Y,�Chandramohan�D,�Wolfson�LJ,�Alberti�KG,�et�al.�Validity�of�verbal�autopsy�procedures�for�determining�cause�of�death�in�Tanzania.�Trop�Med�Int�Health.�2006;11(5):681�96.�197.� Riley�I.�Demography�and�the�epidemiology�of�disease�in�Papua�New�Guinea.�Papua�and�New�Guinea�medical�journal.�2009;52(3�4):83�95.�Epub�2009/09/01.�198.� World�Health�Organization.�Verbal�Autopsy�Standards:�Ascertaining�and�attributing�causes�of�death��The�2012�WHO�verbal�autopsy�instrument.�Geneva:�WHO,�2012.�199.� Bennett�NG,�Horiuchi�S.�Mortality�estimation�from�registered�deaths�in�less�developed�countries.�Demography.�1984;21(2):217�33.�200.� Craig�TJ,�Volaski�V,�Kadyszewski�PF,�DiStefano�O,�Alexander�MJ,�Richardson�MA,�et�al.�The�current�systems�assessment�and�the�clinical�care�process.�A�tool�for�clinical�supervision�and�the�design�of�clinical�information�systems.�Journal�of�medical�systems.�1984;8(1�2):7�14.�201.� Rao�C,�Porapakkham�Y,�Pattaraarchachai�J,�Polprasert�W,�Swampunyalert�N,�Lopez�AD.�Verifying�causes�of�death�in�Thailand:�rationale�and�methods�for�empirical�investigation.�Population�health�metrics.�2010;8:11.�Epub�2010/05/21.�202.� Network�HM.�Assessing�the�National�Health�Information�System�An�Assessment�Tool�2008.�203.� Tangcharoensathien�V,�Faramnuayphol�P,�Teokul�W,�Bundhamcharoen�K,�Wibulpholprasert�S.�A�critical�assessment�of�mortality�statistics�in�Thailand:�potential�for�improvements.�Bulletin�of�the�World�Health�Organization.�2006;84(3):233�8.�Epub�2006/04/04.�204.� Macfarlane�SB.�Harmonizing�health�information�systems�with�information�systems�in�other�social�and�economic�sectors.�Bulletin�of�the�World�Health�Organization.�2005;83(8):590�6.�Epub�2005/09/27.�205.� Nand�D.�Health�Information�Systems�Reform:�The�Fiji�way.�Pacific�health�dialog.�2012;18(1):164.�206.� Health�TMo.�HMN�Assessment�of�the�Health�Information�System�in�Tonga.�Nuku'alofa:�2011.�207.� Rahimi�B,�Vimarlund�V.�Methods�to�evaluate�health�information�systems�in�healthcare�settings:�a�literature�review.�Journal�of�Medical�Systems.�2007;31(5):397�432.�208.� Heeks�R.�Health�information�systems:�failure,�success�and�improvisation.�International�journal�of�medical�informatics.�2006;75(2):125�37.�Epub�2005/08/23.�209.� Ruzicka�LT,�Lopez�AD.�The�use�of�cause�of�death�statistics�for�health�situation�assessment:�national�and�international�experiences.�World�health�statistics�quarterly.�1990;43(4):249�58.�Epub�1990/01/01.�210.� Buehler�JW,�Hopkins�RS,�Overhage�JM,�Sosin�DM,�Tong�V.�Framework�for�evaluating�public�health�surveillance�systems�for�early�detection�of�outbreaks:�recommendations�from�the�CDC�Working�Group.�MMWR�Recomm�Rep.�2004;53(RR�5):1�11.�211.� AD�L.�Draft�framework�for�assessing�mortality�information�systems.�2007.�212.� World�Health�Organization,�Health�Information�Systems�Knowledge�Hub.�Rapid�assessment�of�national�civil�registration�and�vital�statistics�systems�2010.�213.� C�A,�L�M,�R�R,�A.D�L.�Mortality�statistics:�a�tool�to�improve�understanding�and�quality.�HIS�Knowledge�Hub,�School�of�Population�Health,�2010.�214.� World�Health�Organization,�Hub�UoQHISK.�Improving�the�quality�of�birth,�death�and�cause�of�death�information:�guidance�for�a�standards�based�review�of�country�practices�Authors:��.�Brisbane�WHO�and�UQ;�2010.�215.� McGrail�KM,�Black�C.�Access�to�data�in�health�information�systems.�Bulletin�of�the�World�Health�Organization.�2005;83(8):563.�Epub�2005/09/27.�216.� Siddiqi�S,�Masud�TI,�Nishtar�S,�Peters�DH,�Sabri�B,�Bile�KM,�et�al.�Framework�for�assessing�governance�of�the�health�system�in�developing�countries:�Gateway�to�good�governance.�Health�policy�(Amsterdam,�Netherlands).�2008.�217.� Travis�P,�Egger�D,�Davies�P,�Mechbal�A.�Towards�better�stewardship:�concepts�and�critical�issues.�Global�Programme�on�Evidence�for�Health�Policy�Discussion�Papers.�2002(48).�

387

218.� Franca�E,�de�Abreu�DX,�Rao�C,�Lopez�AD.�Evaluation�of�cause�of�death�statistics�for�Brazil,�2002�2004.�International�journal�of�epidemiology.�2008;37(4):891�901.�219.� Preston�SH,�Bhat�PNM.�New�Evidence�on�Fertility�and�Mortality�Trends�in�India.�Population�and�Development�Review.�1984;10(3):481�503.�220.� Murray�CJL,�Rajaratnam�JK,�Marcus�J,�Laakso�T,�Lopez�AD.�What�can�we�conclude�from�death�registration?�Improved�methods�for�evaluating�completeness.�PLoS�medicine.�2010;7(4):e1000262.�221.� Preston�SH.�An�integrated�system�for�demographic�estimation�from�two�age�distributions.�Demography.�1983;20(2):213�26.�222.� Brass�W.�A�procedure�for�comparing�mortality�measures�calculated�from�intercensal�survival�with�the�corresponding�estimates�from�registered�deaths.�Asian�and�Pacific�census�forum�/�East�West�Population�Institute.�1979;6(2):5�7.�223.� Brass�W.�Evaluation�of�birth�and�death�registration�using�age�distributions�and�child�survivorship�data.�Asian�and�Pacific�census�forum�/�East�West�Population�Institute.�1979;5(3):9�11,�20.�Epub�1979/01/01.�224.� Bennett�NG,�Horiuchi�S.�Estimating�the�Completeness�of�Death�Registration�in�a�Closed�Population.�Population�index.�1981;47(2):207�21.�225.� Preston�S,�Coale�AJ,�Trussell�J,�Weinstein�M.�Estimating�the�completeness�of�reporting�of�adult�deaths�in�populations�that�are�approximately�stable.�Population�index.�1980;46(2):179�202.�226.� Preston�S,�Hill�K.�Estimating�the�Completeness�of�Death�Registration.�Population�Studies.�1980;34(2):349�66.�227.� Murray�CJ,�Rajaratnam�JK,�Marcus�J,�Laakso�T,�Lopez�AD.�What�can�we�conclude�from�death�registration?�Improved�methods�for�evaluating�completeness.�PLoS�medicine.7(4):e1000262.�228.� McCarty�DJ,�Tull�ES,�Moy�CS,�Kwoh�CK,�LaPorte�RE.�Ascertainment�corrected�rates:�applications�of�capture�recapture�methods.�International�journal�of�epidemiology.�1993;22(3):559�65.�Epub�1993/06/01.�229.� Hook�EB,�Regal�RR.�Capture�recapture�estimation.�Epidemiology�(Cambridge,�Mass.�1995;6(5):569�70.�Epub�1995/09/01.�230.� Hook�EB,�Regal�RR.�Capture�recapture�methods�in�epidemiology:�methods�and�limitations.�Epidemiologic�reviews.�1995;17(2):243�64.�Epub�1995/01/01.�231.� Sekar�CC,�Deming�WE.�On�a�method�of�estimating�birth�and�death�rates�and�the�extent�of�registration.�Journal�of�the�American�Statistical�Association.�1949;44(245):101�15.�232.� Peterson�JT,�Thurow�RF,�Guzevich�JW.�An�evaluation�of�multipass�electrofishing�for�estimating�the�abundance�of�stream�dwelling�salmonids.�Transactions�of�the�American�Fisheries�Society.�2004;133(2):462�75.�233.� Wittes�J,�Sidel�VW.�A�generalization�of�the�simple�capture�recapture�model�with�applications�to�epidemiological�research.�J�Chronic�Dis.�1968;21:287�301.�234.� Coale�A,�Guo�G.�Revised�regional�model�life�tables�at�very�low�levels�of�mortality.�Population�index.�1989;55(4):613�43.�235.� Murray�CJL,�Ferguson�BD,�Lopez�AD,�Guillot�M,�Salomon�JA,�Ahmad�O.�Modified�Logit�Life�Table�System:�Principles,�Empirical�Validation,�and�Application.�Population�Studies.�2003;57(2):165�82.�236.� Rajaratnam�JK,�Tran�LN,�Lopez�AD,�Murray�CJ.�Measuring�under�five�mortality:�validation�of�new�low�cost�methods.�PLoS�medicine.7(4):e1000253.�237.� Rajaratnam�JK,�Marcus�JR,�Flaxman�AD,�Wang�H,�Levin�Rector�A,�Dwyer�L,�et�al.�Neonatal,�postneonatal,�childhood,�and�under�5�mortality�for�187�countries,�1970�2010:�a�systematic�analysis�of�progress�towards�Millennium�Development�Goal�4.�Lancet.375(9730):1988�2008.�238.� Lozano�R,�Wang�H,�Foreman�KJ,�Rajaratnam�JK,�Naghavi�M,�Marcus�JR,�et�al.�Progress�towards�Millennium�Development�Goals�4�and�5�on�maternal�and�child�mortality:�an�updated�systematic�analysis.�Lancet.�2011;378(9797):1139�65.�Epub�2011/09/23.�239.� Murray�CJ,�Lopez�AD,�Black�R,�Mathers�CD,�Shibuya�K,�Ezzati�M,�et�al.�Global�burden�of�disease�2005:�call�for�collaborators.�Lancet.�2007;370(9582):109�10.�Epub�2007/07/17.�

388

240.� Wang�H,�Dwyer�Lindgren�L,�Lofgren�KT,�Rajaratnam�JK,�Marcus�JR,�Levin�Rector�A,�et�al.�Age�specific�and�sex�specific�mortality�in�187�countries,�1970–2010:�a�systematic�analysis�for�the�Global�Burden�of�Disease�Study�2010.�The�Lancet.�2013;380(9859):2071�94.�241.� Brass�W,�Swamy�S.�Measurement�of�death�registration�completeness�using�the�growth�balance�procedure�applied�to�data�from�India.�Asian�and�Pacific�census�forum�/�East�West�Population�Institute.�1980;7(1):5�8.�242.� Sibai�AM.�Mortality�certification�and�cause�of�death�reporting�in�developing�countries.�Bulletin�of�the�World�Health�Organization.�2004;82(2):83.�Epub�2004/03/26.�243.� Morton�L,�Omar�R,�Carroll�S,�Beirne�M,�Halliday�D,�Taylor�KM.�Incomplete�and�inaccurate�death�certification�the�impact�on�research.�Journal�of�Public�Health.�2000;22(2):133�7.�244.� Lu�TH,�Sun�SM,�Huang�SM,�Lin�JJ.�Mind�your�manners:�quality�of�manner�of�death�certification�among�medical�examiners�and�coroners�in�Taiwan.�Am�J�Forensic�Med�Pathol.�2006;27(4):352�4.�245.� Villar�J,�Perez�Mendez�L.�Evaluating�and�educational�intervention�to�improve�the�accuracy�of�death�certification�among�trainees�from�various�specialties.�BMC�health�services�research.�2007;7:183�9.�246.� Selinger�CP,�Ellis�RA,�Harrington�MG.�A�good�death�certificate:�improved�performance�by�simple�educational�measures.�Postgrad�Med�J.�2007;83(978):285�6.�247.� Maudsley�G,�Williams�EM.�Death�certification�by�house�officers�and�general�practitioners�practice�and�performance.�Journal�of�public�health�medicine.�1993;15(2):192�201.�248.� Lu�T,�Shih�T,�Lee�M,�Chou�M,�Lin�C.�Diversity�in�death�certification:�a�case�vignette�approach.�Journal�of�clinical�epidemiology.�2001;54:1086�93.�249.� Maudsley�G,�Williams�L.�Death�certification��a�sad�state�of�affairs.�J�Public�Health�Med.�1994;16(3):370�1.�250.� Rutty�G,�Duerden�R,�Carter�N,�Clark�J.�Are�coroners'�necropsies�necessary?�A�prospective�study�examining�whether�a�“view�and�grant”�system�of�death�certification�could�be�introduced�into�England�and�Wales.�Journal�of�clinical�pathology.�2001;54(4):279�84.�251.� Brinkley�D,�Haybittle�JL,�Alderson�MR.�Death�certification�in�cancer�of�the�breast.�British�medical�journal�(Clinical�research�ed.�1984;289(6443):465�7.�252.� Vellacott�K,�Ferro�M.�Cause�of�death�and�accuracy�of�certification�of�colorectal�cancer.�Journal�of�the�Royal�Society�of�Medicine.�1984;77(1):22.�253.� Papapetropoulos�S,�Villar�JM,�Gonzalez�J,�Mash�DC.�Disparities�in�death�certificates�of�Parkinson's�disease�patients:�a�report�from�a�population�of�brain�donors.�Mov�Disord.�2006;21(10):1791�2;�author�reply�2.�254.� Freedman�AN,�Seminara�D,�Gail�MH,�Hartge�P,�Colditz�GA,�Ballard�Barbash�R,�et�al.�Cancer�risk�prediction�models:�a�workshop�on�development,�evaluation,�and�application.�Journal�of�the�National�Cancer�Institute.�2005;97(10):715�23.�255.� Karwinski�B,�Hartveit�F.�Death�certification:�increased�clinical�confidence�in�diagnosis�and�lack�of�interest�in�confirmation�by�necropsy�is�not�justified.�J�Clin�Pathol.�1989;42(1):13�7.�256.� Newens�A,�Forster�D,�Kay�D.�Death�certification�after�a�diagnosis�of�presenile�dementia.�Journal�of�epidemiology�and�community�health.�1993;47(4):293�7.�257.� Accuracy�of�death�certificates�in�bronchial�asthma.�Accuracy�of�certification�procedures�during�the�confidential�inquiry�by�the�British�Thoracic�Association.�A�subcommittee�of�the�BTA�Research�Committee.�Thorax.�1984;39(7):505�9.�258.� Balkau�B,�Papoz�L.�Certification�of�cause�of�death�in�French�diabetic�patients.�Journal�of�epidemiology�and�community�health.�1992;46:63��5.�259.� Hunt�R,�Barr�P.�Errors�in�the�certification�of�neonatal�death.�Journal�of�paediatrics�and�child�health.�2000;36(5):498�501.�260.� Maxwell�JD,�Knapman�P.�Effect�of�coroners'�rules�on�death�certification�for�alcoholic�liver�disease.�British�medical�journal�(Clinical�research�ed.�1985;291(6497):708.�261.� Bell�G,�Cremona�A.�Alcohol�and�death�certification:�influencing�current�practice�and�attitudes.�Br�J�Addict.�1989;84(12):1523�5.�

389

262.� Riley�I.�The�medical�profession�and�the�health�services.�Papua�and�New�Guinea�medical�journal.�1986;29(3):253�60.�Epub�1986/09/01.�263.� Crowcroft�N,�Majeed�A.�Improving�the�certification�of�death�and�the�usefulness�of�routine�mortality�statistics.�Clinical�medicine.�2001;1(2):122�5.�264.� Robertson�HE.�Problems�of�Cancer�Mortality�Statistics.�Am�J�Public�Health�Nations�Health.�1930;20(3):257�62.�265.� Naghavi�M,�Makela�S,�Foreman�K,�O'Brien�J,�Pourmalek�F,�Lozano�R.�Research�Algorithms�for�enhancing�public�health�utility�of�national�causes�of�death�data.�2010.�266.� Abulfatih�NM,�Hamadeh�RR.�A�study�of�ill�defined�causes�of�death�in�Bahrain.�Determinants�and�health�policy�issues.�Saudi�Med�J.�2010;31(5):545�9.�Epub�2010/05/14.�267.� al�Mahroos�R.�Validity�of�death�certificates�for�coding�coronary�heart�disease�as�the�cause�of�death�in�Bahrain.�Eastern�Mediterranean�health�journal�=�La�revue�de�sante�de�la�Mediterranee�orientale�=�al�Majallah�al�sihhiyah�li�sharq�al�mutawassit.�2000;6(4):661�9.�268.� Dunn�KL,�Reddy�P,�Moulden�A,�Bowes�G.�Medical�record�review�of�deaths,�unexpected�intensive�care�unit�admissions,�and�clinician�referrals:�detection�of�adverse�events�and�insight�into�the�system.�Arch�Dis�Child.�2006;91(2):169�72.�Epub�2005/10/27.�269.� Hardy�AM,�Starcher�ET,�2nd,�Morgan�WM,�Druker�J,�Kristal�A,�Day�JM,�et�al.�Review�of�death�certificates�to�assess�completeness�of�AIDS�case�reporting.�Public�Health�Rep.�1987;102(4):386�91.�Epub�1987/07/01.�270.� Horner�JS,�Horner�JW.�Do�doctors�read�forms?�A�one�year�audit�of�medical�certificates�submitted�to�a�crematorium.�J�R�Soc�Med.�1998;91(7):371�6.�271.� Hunt�LW,�Jr.,�Silverstein�MD,�Reed�CE,�O'Connell�EJ,�O'Fallon�WM,�Yunginger�JW.�Accuracy�of�the�death�certificate�in�a�population�based�study�of�asthmatic�patients.�Jama.�1993;269(15):1947�52.�272.� Swift�B,�West�K.�Death�certification:�an�audit�of�practice�entering�the�21st�century.�J�Clin�Pathol.�2002;55:275�9.�273.� World�Health�Organization,�Health�Information�Systems�Knowledge�Hub.�Improving�the�quality�of�birth,�death�and�cause�of�death�information:�guidance�for�a�standards�based�review�of�country�practices�2010.�274.� Lu�TH,�Hsu�PY,�Bjorkenstam�C,�Anderson�RN.�Certifying�diabetes�related�cause�of�death:�a�comparison�of�inappropriate�certification�statements�in�Sweden,�Taiwan�and�the�USA.�Diabetologia.�2006;49(12):2878�81.�275.� Kelson�M,�Farebrother�M.�The�effect�of�inaccuracies�in�death�certification�and�coding�practices�in�the�European�economic�community�(EEC)�on�international�cancer�mortality�statistics.�International�journal�of�epidemiology.�1987;16(3):411�4.�276.� Mackenbach�J,�Van�Duyne�WM,�Kelson�MC.�Certification�and�coding�of�two�underlying�causes�of�death�in�the�Netherlands�and�other�countries�of�the�European�community.�Journal�of�epidemiology�and�community�health.�1987;41:156�60.�277.� Andriessen�K.�Do�we�need�to�be�cautious�in�evaluating�suicide�statistics?�European�journal�of�public�health.�2006;16(4):445.�278.� Santoso�JT,�Lee�CM,�Aronson�J.�Discrepancy�of�death�diagnosis�in�gynecology�oncology.�Gynecologic�oncology.�2006;101(2):311�4.�279.� Redelings�MD,�Wise�M,�Sorvillo�F.�Using�multiple�cause�of�death�data�to�investigate�associations�and�causality�between�conditions�listed�on�the�death�certificate.�American�journal�of�epidemiology.�2007;166(1):104�8.�280.� Engel�LW,�Strauchen�JA,�Chiazze�L,�Jr.,�Heid�M.�Accuracy�of�death�certification�in�an�autopsied�population�with�specific�attention�to�malignant�neoplasms�and�vascular�diseases.�American�journal�of�epidemiology.�1980;111(1):99�112.�281.� Hansell�AL.�Lies,��damn�lies�and�mortality�statistics?�Thorax.�2006;61:923�4.�282.� Newell�KW.�Primary�health�care�in�the�Pacific.�Social�science�&�medicine�(1982).�1983;17(19):1441�7.�Epub�1983/01/01.�

390

283.� Murray�CJ,�Lopez�AD,�Barofsky�JT,�Bryson�Cahn�C,�Lozano�R.�Estimating�population�cause�specific�mortality�fractions�from�in�hospital�mortality:�validation�of�a�new�method.�PLoS�medicine.�2007;4(11):e326.�284.� Salomon�JA,�Murray�CJL.�The�epidemiologic�transition�revisited:�compositional�models�for�causes�of�death�by�age�and�sex.�Population�and�Development�Review.�2002;28(2):205�28.�285.� Morris�SS,�Black�RE,�Tomaskovic�L.�Predicting�the�distribution�of�under�five�deaths�by�cause�in�countries�without�adequate�vital�registration�systems.�International�journal�of�epidemiology.�2003;32(6):1041�51.�286.� Rao�C,�Adair�T,�Kinfu�Y.�Using�historical�vital�statistics�to�predict�the�distribution�of�under�five�mortality�by�cause.�Clinical�medicine�&�research.�2011;9(2):66�74.�287.� AbouZahr�C,�Wardlaw�T.�Maternal�mortality�at�the�end�of�a�decade:�signs�of�progress?�Bulletin�of�the�World�Health�Organization.�2001;79(6):561�8.�Epub�2001/07/05.�288.� Heeks�R.�Health�information�systems:�failure,�success�and�improvisation.�International�Journal�of�Medical�Informatics.�2006;75(2):125�37.�289.� Kaufman�D,�Roberts�WD,�Merrill�J,�Lai�TY,�Bakken�S.�Applying�an�evaluation�framework�for�health�information�system�design,�development,�and�implementation.�Nursing�Research.�2006;55(2�Suppl):S37�42.�290.� Yusof�MM,�Kuljis�J,�Papazafeiropoulou�A,�Stergioulas�LK.�An�evaluation�framework�for�Health�Information�Systems:�human,�organization�and�technology�fit�factors�(HOT�fit).�International�Journal�of�Medical�Informatics.�2008;77(6):386�98.�291.� Yusof�MM,�Papazafeiropoulou�A,�Paul�RJ,�Stergioulas�LK.�Investigating�evaluation�frameworks�for�health�information�systems.�International�Journal�of�Medical�Informatics.�2008;77(6):377�85.�292.� Tonga�Ministry�of�Health.�Tonga�Ministry�of�Health�Annual�Report�2010.�Tonga:�2011.�293.� Begg�S,�Rao�C.�Design�options�for�sample�based�mortality�surveillance.�294.� Toson�B,�Baker�A.�Life�expectancy�at�birth:�methodological�options�for�small�populations.�National�statistics�methodological�series.�2003;33.�295.� Eayres�D,�Williams�E.�Evaluation�of�methodologies�for�small�area�life�expectancy�estimation.�Journal�of�epidemiology�and�community�health.�2004;58(3):243�9.�296.� Chiang�C.�Variance�and�covariance�of�life�table�functions�estimated�from�a�sample�of�deaths.�Vital�Health�Stat�2.�1967;20:1�8.�297.� Australian�Bureau�of�Statistics.�Chapter�4:�Differentials�in�Mortality.�.��Deaths�Australia;.�Canberra:�ABS;�2004.�298.� Zealand�SN.�New�Zealand�life�tables:�2005–07.�Wellington:�Statistics�New�Zealand.�2009.�299.� Tobias�MI,�Cheung�J.�Monitoring�health�inequalities:�life�expectancy�and�small�area�deprivation�in�New�Zealand.�Population�health�metrics.�2003;1(1):2.�Epub�2003/05/30.�300.� J�P,�Baravilala�W,�C�K.�Telehealth�in�the�Pacific�Islands:�a�perspective�and�update�from�the�Fiji�School�of�Medicine.�Pacific�health�dialog.�2000;7(2):6�11.�301.� Deaths�of�Pacific�Islanders�in�Australia.�In:�Australian�Bureau�of�Statistics,�editor.�Unpublished2010.�302.� Statistics�NZ.�Deaths�of�Pacific�Islanders�in�New�Zealand.�In:�Statistics�NZ,�editor.�Unpublished2010.�303.� Institut�de�la�Statistique�et�des�Études�Économique.�Death�Data.�ISEE�2012;�Available�from:�http://www.isee.nc/.�304.� Governments�UAPI.�Pacific�Regional�Comprehensive�Cancer�Control�Plan,�2007�2012.�Hawaii:�2007.�305.� Haddock�RL,�Naval�CL.�Cancer�on�Guam,�especially�among�Micronesians.�Pacific�health�dialog.�2002;9(2):222�4.�306.� Haddock�RL,�Talon�RJ,�Whippy�HJ.�Ethnic�disparities�in�cancer�mortality�among�residents�of�Guam.�Asian�Pac�J�Cancer�Prev.�2006;7(3):411�4.�307.� DoD's�Tripler,�VA�cooperating�to�share,�expand�health�resources.�US�Med.�1980;16(1):2,�12�4.�

391

308.� S�B.�Tripler�continues�care�of�Pacific�Islanders,�enhances�outreach�program.�Honolulu:�Department�of�Defense;�2012�[1/1/12].�309.� Gerundio�Dizon�A.�Medical�Referrals�from�Palau�to�Philippines�Suspended.�Island�Times.�2012�16/7/2012.�310.� Zealand�RN.�Am,�Samoa�Hospital�considers�sending�referrals�to�Philippines.�2012�18/6/2012.�311.� Asher�J.�Federated�States�of�Micronesia�Health�Services�in�relation�to�medical�services.�Journal�of�South�Pacific�Law.�2002;6.�312.� Person�DA.�Pacific�Island�Health�Care�Project:�early�experiences�with�a�Web�based�consultation�and�referral�network.�Pacific�health�dialog.�2000;7(2):29�35.�Epub�2001/10/09.�313.� Gunawardane�KJ.�Telemedicine�in�Majuro�Hospital,�Marshall�Islands.�Pacific�health�dialog.�2000;7(2):51�3.�Epub�2001/10/09.�314.� Secretariat�of�the�Pacific�Community,�Cook�Islands�Statisitics�Office.�Demographic�Profile�of�the�Cook�Islands�1996�2002.�Noumea,�New�Caledonia:�SPC;�2002.�315.� Cook�Islands�Statisitics�Office.�Social�Statistics.�Rarotonga,�Cook�Islands:�Cook�Islands�Statistics�Office;�2011�[cited�2012�3/12/12];�Available�from:�http://www.spc.int/prism/country/ck/stats/Statistics/Demography/popn_estimate.htm.�316.� Blakely�T,�Ajwani�S,�Robson�B,�Tobias�M,�Bonne�M.�Decades�of�disparity:�widening�ethnic�mortality�gaps�from�1980�to�1999.�The�New�Zealand�medical�journal.�2004;117(1199):U995.�Epub�2004/10/12.�317.� Blakely�T,�Atkinson�JN,�Fawcett�J.�Ethnic�counts�on�mortality�and�census�data�(mostly)�agree�for�2001�2004:�New�Zealand�Census�Mortality�Study�update.�The�New�Zealand�medical�journal.�2008;121(1281):58�62.�Epub�2008/09/18.�318.� Blakely�T,�Richardson�K,�Young�J,�Callister�P,�Didham�R.�Does�mortality�vary�between�pacific�groups�in�New�Zealand?�Estimating�Samoan,�Cook�Island�Maori,�Tongan,�and�Niuean�mortality�rates�using�hierarchical�Bayesian�modelling.�The�New�Zealand�medical�journal.�2009;122(1307):18�29.�Epub�2010/02/12.�319.� Secretariat�of�the�Pacific�Community,�Health�Infomation�Systems�Knowledge�Hub.�The�Brisbane�Accord�Group�and�the�Pacific�vital�statistics�action�plan�2011�2014.�BAG�Agencies�Secretariat�of�the�Pacific�Community�University�of�Queensland,�2012.�320.� AbouZahr�C,�Mikkelsen�L,�Rampatige�R,�Lopez�A.�Mortality�statistics:�a�tool�to�enhance�understanding�and�improve�quality.�Pacific�health�dialog.�2012;18(1):247.�321.� Lopez�AD,�Begg�S,�Bos�E.�Demographic�and�Epidemiological�Characteristics�of�Major�Regions,�1990�2001.�In:�Lopez�AD,�Mathers�CD,�Ezzati�M,�Jamison�DT,�Murray�CJL,�editors.�Global�Burden�of�Disease�and�Risk�Factors.�Washington�(DC)2006.�322.� Taylor�R,�Carter�K,�Wainiqolo�I,�Naidu�S,�Linhart�C,�Jayaratne�N,�et�al.�Divergent�mortality�trends�by�ethnicity�in�Fiji.�Australian�and�New�Zealand�Journal�of�Public�Health��Under�review.�2013.�323.� Duque�Portugal�F,�Martin�AJ,�Taylor�R,�Ross�MW.�Mark�recapture�estimates�of�injecting�drug�users�in�Sydney.�Australian�journal�of�public�health.�1994;18(2):201�4.�324.� Khorasani�Zavareh�D,�Mohammadi�R,�Laflamme�L,�Naghavi�M,�Zarei�A,�Haglund�BJ.�Estimating�road�traffic�mortality�more�accurately:�use�of�the�capture�recapture�method�in�the�West�Azarbaijan�province�of�Iran.�Int�J�Inj�Contr�Saf�Promot.�2008;15(1):9�17.�Epub�2008/03/18.�325.� Lewden�C,�Jougla�E,�Alioum�A,�Pavillon�G,�Lievre�L,�Morlat�P,�et�al.�Number�of�deaths�among�HIV�infected�adults�in�France�in�2000,�three�source�capture�recapture�estimation.�Epidemiology�and�infection.�2006;134(6):1345�52.�326.� Cecchi�F,�Nyasulu�P,�Chandramohan�D,�Roberts�B.�Rates�and�causes�of�death�in�Chiradzulu�district,�Malawi,�2008:�a�key�informant�study.�Tropical�medicine�and�international�health.�2010;16(3):375�8.�327.� McGilchrist�CA.�Model�selection�and�population�size�using�capture�recapture�methods.�Journal�of�Clinical�Epidemiology.�1999;52(10):915;�discussion�29.�

392

328.� Hook�EB,�Regal�RR.�Validity�of�methods�for�model�selection,�weighting�for�model�uncertainty,�and�small�sample�adjustment�in�capture�recapture�estimation.�American�journal�of�epidemiology.�1997;145(12):1138�44.�Epub�1997/06/15.�329.� Office;�PNS.�Press�release�03:�Bohol�recorded�a�population�of�1.2�million�persons�(Results�from�the�2007�Census�of�Population).�Manila:�Philippines�NSO;�2010�[2/3/2011];�Available�from:�www.census.gov.ph/data/pressrelease/2010/pr1003tx.html.�330.� Johnson�JB,�Omland�KS.�Model�selection�in�ecology�and�evolution.�Trends�in�Ecology�&�Evolution.�2004;19(2):101�8.�331.� Williamson�NË.�An�attempt�to�reduce�infant�and�child�mortality�in�Bohol,�Philippines.�Studies�in�family�planning.�1982:106�17.�332.� Hufana�L,�Cajita�J,�Morante�L,�Lopez�J,�Tan�C,�Mikkelsen�L,�et�al.�Assessing�the�production,�quality�and�use�of�national�vital�statistics:�a�case�study�of�the�Philippines.�Queensland:�University�of�Queensland.�2009.�333.� Carter�KL,�Williams�G,�Tallo�V,�Sanvictores�D,�Madera�H,�Riley�I.�Capture�recapture�analysis�of�all�cause�mortality�data�in�Bohol,�Philippines.�Population�health�metrics.�2011;9:9.�Epub�2011/04/19.�334.� Omran�AR.�The�Epidemiologic�Transition:�A�Theory�of�the�Epidemiology�of�Population�Change.�The�Milbank�quarterly.�2005;83(4):731�57.�335.� Kiribati�Ministry�of�Health,�World�Health�Organization.�Kiribati�NCD�Risk�Factors�STEPs�Report�2004�2006.�Manila:�WPRO,�WHO;�2006.�336.� Hook�EB,�Regal�RR.�Internal�validity�analysis:�a�method�for�adjusting�capture�recapture�estimates�of�prevalence.�American�journal�of�epidemiology.�1995;142(9�Suppl):S48�52.�Epub�1995/11/01.�337.� Vanuatu�Ministry�of�Health,�World�health�Organization.�Vanuatu�STEPS�Survey�2011��Fact�Sheet.�Manila:�WPRO,�WHO;�2012�[1/1/2013];�Available�from:�http://www.who.int/chp/steps/2011_Vanuatu_FactSheet.pdf.�338.� Henderson�T,�Shepheard�J,�Sundararajan�V.�Quality�of�diagnosis�and�procedure�coding�in�ICD�10�administrative�data.�Medical�Care.�2006:1011�9.�339.� Cheng�P,�Gilchrist�A,�Robinson�KM,�Paul�L.�The�risk�and�consequences�of�clinical�miscoding�due�to�inadequate�medical�documentation:�a�case�study�of�the�impact�on�health�services�funding.�Health�Information�Management�Journal.�2009;38(1):35�46.�340.� Campbell�SE,�Campbell�MK,�Grimshaw�JM,�Walker�AE.�A�systematic�review�of�discharge�coding�accuracy.�Journal�of�Public�Health.�2001;23(3):205�11.�341.� Murray�CJ,�Lopez�AD,�Black�R,�Ahuja�R,�Ali�SM,�Baqui�A,�et�al.�Population�Health�Metrics�Research�Consortium�gold�standard�verbal�autopsy�validation�study:�design,�implementation,�and�development�of�analysis�datasets.�Population�health�metrics.�2011;9:27.�Epub�2011/08/06.�342.� Rao�C,�Porapakkham�Y,�Pattaraarchachai�J,�Polprasert�W,�Swampunyalert�N,�Lopez�AD.�Verifying�causes�of�death�in�Thailand:�rationale�and�methods�for�empirical�investigation.�Population�health�metrics.8:11.�343.� McKelvie�PA.�Medical�certification�of�causes�of�death�in�an�Australian�metropolitan�hospital.�Comparison�with�autopsy�findings�and�a�critical�review.�The�Medical�journal�of�Australia.�1993;158(12):816�8,�20�1.�344.� Weeramanthri�TS.�A�medical�cause�of�death�validation�study�of�adult�aboriginal�deaths�in�the�Northern�Territory�of�Australia�in�1992.�Public�health.�1997;111(6):429�33.�345.� Health�NZMo.�The�Burden�of�Disease�and�Injury�in�New�Zealand.�Public�Health�Intelligence:�Occaisonal�bulletin�#1.�Wellington:�NZ�Ministry�of�Health;�2001.�346.� Vanuatu�National�Statistics�Office.�Social�Summary.�Port�Vila:�VNSO;�2012�[2/1/2013];�Available�from:�www.vnso.gov.vu.�347.� Carter�K,�Hufanga�S,�Rao�C,�Akauola�S,�Lopez�AD,�Rampatige�R,�et�al.�Causes�of�death�in�Tonga:�quality�of�certification�and�implications�for�statistics.�Population�health�metrics.�2012;10(1):4.�Epub�2012/03/07.�348.� Paksoy�N,�Montaville�B,�McCarthy�SW.�Cancer�occurrence�in�Vanuatu�in�the�South�Pacific,�1980�86.�Tropical�and�geographical�medicine.�1990;42(2):157�61.�

393

349.� Fotinatos�N,�Warmington�A,�Walker�T,�Pilbeam�M.�Estimates�for�cervical�abnormalities�in�Vanuatu.�Aust�N�Z�J�Public�Health.�2007;31(6):571�5.�350.� Grace�RF,�Vurobaravu�TS.�Twenty�three�years�of�labour�floor�records,�Vila�Central�Hospital,�Vanuatu.�The�Australian�&�New�Zealand�journal�of�obstetrics�&�gynaecology.�2003;43(2):157�9.�351.� A�regional�a�summary�of�the�Fiji,�Kiribati,�Solomon�Islands�and�Vanuatu�child�protection�baseline�reports.�UNICEF,�2008.�352.� Abbott�KC,�Trespalacios�FC,�Agodoa�LY,�Taylor�AJ,�Bakris�GL.�beta�Blocker�use�in�long�term�dialysis�patients:�association�with�hospitalized�heart�failure�and�mortality.�Arch�Intern�Med.�2004;164(22):2465�71.�Epub�2004/12/15.�353.� Lozano�R,�Naghavi�M,�Foreman�K,�Lim�S,�Shibuya�K,�Aboyans�V,�et�al.�Global�and�regional�mortality�from�235�causes�of�death�for�20�age�groups�in�1990�and�2010:�a�systematic�analysis�for�the�Global�Burden�of�Disease�Study�2010.�The�Lancet.�2012;380(9859):2095�128.�354.� Pacific�Islands�Forum�Secretariat.�Pacific�Regional�MDGs�Tracking�Report.�Suva,�Fiji:�PIFs;�2012.�355.� Murray�CJ,�Rosenfeld�LC,�Lim�SS,�Andrews�KG,�Foreman�KJ,�Haring�D,�et�al.�Global�malaria�mortality�between�1980�and�2010:�a�systematic�analysis.�Lancet.�2012;379(9814):413�31.�Epub�2012/02/07.�356.� Zealand:�SN.�Tables:�Deaths�2008,�Age�specific�death�rates�by�sex,�total�population.�1986–2007.�.�2008;�Available�from:�http://www.stats.govt.nz/tables/deaths�tables.htm.�357.� Tuomilehto�J,�Ram�P,�Eseroma�R,�Taylor�R,�Zimmet�P.�Cardiovascular�diseases�and�diabetes�mellitus�in�Fiji:�analysis�of�mortality,�morbidity�and�risk�factors.�Bulletin�of�the�World�Health�Organization.�1984;62(1):133�43.�358.� Zimmet�P,�Jackson�L,�Whitehouse�S.�Blood�pressure�studies�in�two�Pacific�populations�with�varying�degrees�of�modernisation.�The�New�Zealand�medical�journal.�1980;91(657):249�52.�359.� Fiji�Ministry�of�Health,�World�Health�Organization.�Fiji�Non�communicable�Diseases�(NCD)�STEPS�Survey�2002.�Geneva,�Switzerland:�:�World�Health�Organization�(WHO).�2002.�360.� Shaw�J,�Sicree�R,�Zimmet�P.�Global�estimates�of�the�prevalence�of�diabetes�for�2010�and�2030.�Diabetes�research�and�clinical�practice.�2010;87(1):4�14.�361.� Simmons�D,�Weblemoe�T,�Voyle�J,�Prichard�A,�Leakehe�L,�Gatland�B.�Personal�barriers�to�diabetes�care:�lessons�from�a�multi�ethnic�community�in�New�Zealand.�Diabetic�Medicine.�2004;15(11):958�64.�362.� Dobson�J,�Steer�AC,�Colquhoun�S,�Kado�J.�Environmental�Factors�and�Rheumatic�Heart�Disease�in�Fiji.�Pediatr�Cardiol.�2011.�Epub�2011/11/08.�363.� Seckeler�MD,�Barton�LL,�Brownstein�R.�The�persistent�challenge�of�rheumatic�fever�in�the�Northern�Mariana�Islands.�Int�J�Infect�Dis.�2010;14(3):e226�9.�Epub�2009/08/04.�364.� Sopoaga�F,�Buckingham�K,�Paul�C.�Causes�of�excess�hospitalizations�among�Pacific�peoples�in�New�Zealand:�implications�for�primary�care.�J�Prim�Health�Care.�2010;2(2):105�10.�Epub�2010/08/10.�365.� Singh�PI,�Carapetis�JR,�Buadromo�EM,�Samberkar�PN,�Steer�AC.�The�high�burden�of�rheumatic�heart�disease�found�on�autopsy�in�Fiji.�Cardiology�in�the�young.�2008;18(1):62�9.�366.� Atatoa�Carr�P,�Bell�A,�Lennon�DR.�Acute�rheumatic�fever�in�the�Waikato�District�Health�Board�region�of�New�Zealand:�1998�2004.�The�New�Zealand�medical�journal.�2008;121(1285):96�105.�Epub�2008/12/17.�367.� Cuboni�HD,�Finau�SA,�Cuboni�G.�Rheumatic�fever�and�rheumatic�heart�diseases�in�Fiji:�a�review�from�the�surveillance�system�(1996��2000).�Pacific�health�dialog.�2006;13(2):39�47.�368.� Palafox�NA,�Yamada�S,�Ou�AC,�Minami�JS,�Johnson�DB,�Katz�AR.�Cancer�in�Micronesia.�Pacific�health�dialog.�2004;11(2):78�83.�369.� Ou�AC,�Kienene�T,�Tenaua�K,�Etheredge�GD,�Palafox�NA.�Epidemiology�of�cancer�in�the�Republic�of�Kiribati.�Pacific�health�dialog.�2004;11(2):88�93.�370.� Rani�M,�Yang�B,�Nesbit�R.�Hepatitis�B�control�by�2012�in�the�WHO�Western�Pacific�Region:�rationale�and�implications.�Bulletin�of�the�World�Health�Organization.�2009;87(9):707�13.�371.� Booth�H.�Pacific�Island�suicide�in�comparative�perspective.�Journal�of�biosocial�science.�1999;31(4):433�48.�Epub�1999/12/03.�

394

372.� Statistics;�ABo.�3302.0.55.001��Life�Tables,�States,�Territories�and�Australia,�2009�2011��Canberra:�ABS;�2012�[2/2/2012];�Available�from:�http://www.abs.gov.au/AUSSTATS/[email protected]/allprimarymainfeatures/97E435FA3B82A89DCA2570A6000573D3?opendocument.�373.� Tomisaka�K,�Lako�J,�Maruyama�C,�Anh�N,�Lien�D,�Khoi�HH,�et�al.�Dietary�patterns�and�risk�factors�for�type�2�diabetes�mellitus�in�Fijian,�Japanese�and�Vietnamese�populations.�Asia�Pacific�journal�of�clinical�nutrition.�2002;11(1):8�12.�374.� Lako�JV.�Dietary�trend�and�diabetes:�its�association�among�indigenous�Fijians�1952�to�1994.�Asia�Pacific�journal�of�clinical�nutrition.�2001;10(3):183�7.�Epub�2001/11/16.�375.� Taylor�R,�Ram�P,�Zimmet�P,�Raper�LR,�Ringrose�H.�Physical�activity�and�prevalence�of�diabetes�in�Melanesian�and�Indian�men�in�Fiji.�Diabetologia.�1984;27(6):578�82.�376.� Zimmet�P,�Taylor�R,�Ram�P,�King�H,�Sloman�G,�Raper�LR,�et�al.�Prevalence�of�diabetes�and�impaired�glucose�tolerance�in�the�biracial�(Melanesian�and�Indian)�population�of�Fiji:�a�rural�urban�comparison.�American�journal�of�epidemiology.�1983;118(5):673�88.�377.� Tibbs�CJ.�Hepatitis�B,�tropical�ulcers,�and�immunisation�strategy�in�Kiribati.�British�medical�journal�(Clinical�research�ed.�1987;294(6571):537�40.�378.� Zimmet�PZ,�Finch�CF,�Schooneveldt�MG,�King�HO,�Thoma�K.�Mortality�from�diabetes�in�Nauru.�Results�of�4�yr�follow�up.�Diabetes�care.�1988;11(4):305�10.�379.� King�H,�Zimmet�P,�Pargeter�K,�Raper�LR,�Collins�V.�Ethnic�differences�in�susceptibility�to�non�insulin�dependent�diabetes.�A�comparative�study�of�two�urbanized�Micronesian�populations.�Diabetes.�1984;33(10):1002�7.�380.� Public�Health�Division�SotPC.�NCD�Statistics�for�the�Pacific�Islands�Countries�and�Territories.�Noumea:�SPC;�2011�[2/1/2013];�Available�from:�http://www.spc.int/hpl/index.php?option=com_content&task=blogcategory&id=23&Itemid=42.�381.� World�Health�Organization.�Nauru�NCD�Risk�Factors:�STEPS�report.�:�WHO;�2007.�382.� Omran�AR.�The�Epidemiologic�Transition:�A�Theory�of�the�Epidemiology�of�Population�Change.�The�Milbank�Memorial�Fund�Quarterly.�1971;49(4):509�38.�383.� Hajat�A,�Blakely�T,�Dayal�S,�Jatrana�S.�Do�New�Zealand's�immigrants�have�a�mortality�advantage?�Evidence�from�the�New�Zealand�Census�Mortality�Study.�Ethn�Health.�2010;15(5):531�47.�Epub�2010/07/31.�384.� Hughes�RG,�Lawrence�MA.�Globalization,�food�and�health�in�Pacific�Island�countries.�Asia�Pacific�journal�of�clinical�nutrition.�2005;14(4):298�306.�385.� United�Nations.�Beyond�2015.�2012�[24/1/2013];�Available�from:�www.un.org/millenniumgoals/beyond2015.shtml.�386.� Poikolainen�K,�Eskola�J.�The�effect�of�health�services�on�mortality:�decline�in�death�rates�from�amenable�and�non�amenable�causes�in�Finland,�1969�81.�The�Lancet.�1986;327(8474):199�202.��

i A

ppen

dix

1: B

rass

Gro

wth

Bal

ance

Ass

essm

ents

for K

iriba

ti R

econ

cile

d da

ta (2

005-

2009

)

Not

e: R

esul

ts a

re c

lear

ly im

plau

sibl

e –

see

disc

ussi

on in

cha

pter

5

Mal

es

ii Fe

mal

es

iii

App

endi

x 2:

Add

ition

al c

ount

ry li

fe ta

bles

L

ife ta

bles

for

Nau

ru (2

002-

2007

)M

ales

Age

x Ye

ars

in Int.

Line

-ar

ity

Adj

ust.

Pop.

D

eath

s M

orta

lity

Rat

e

Prob

-ab

ility

of

D

ying

Indi

vid-

uals

Su

rviv

ing

Dea

ths

inIn

terv

al

x

Year

sLi

ved

in

Inte

rval

x

Cum

ul-

ativ

e Ye

ars

Live

d

Expe

ct-

ancy

of

Life

at

Age

x

var(

qx)

lx2[

(1-

ax)n

1+ex

+1]2

var(

qx)2

sum

(lx2[

(1-

ax)n

1+ex

+1]2

var(

qx)

2)

Var (

ex)

SE

Low

er

95%

C

I

Upp

er95

%

CI

0-4

0 5

0.3

1621

.8

39.5

0.

0243

46

0.11

22

1000

00

1121

7 46

0740

53

1086

9 53

.1

0.00

0283

9.

56E

+09

1.52

E+1

0 1.

5243

1.

2346

50

.7

55.6

5-14

5

10

0.5

6241

.2

6.1

0.00

0973

0.

0097

88

783

860

8835

28

4850

129

54.6

1.

53E

-05

3.03

E+0

8 5.

68E

+09

0.72

10

0.84

91

15-2

4 15

10

0.

5 60

46.9

13

.2

0.00

2177

0.

0215

87

923

1893

86

9763

39

6660

1 45

.1

3.45

E-0

5 4.

48E

+08

5.38

E+0

9 0.

6961

0.

8343

25

-34

25

10

0.5

4594

.8

24.3

0.

0052

88

0.05

15

8603

0 44

32

8381

35

3096

839

36.0

0.

0001

04

8.19

E+0

8 4.

93E

+09

0.66

65

0.81

64

35-4

4 35

10

0.

5 28

96.5

41

.5

0.01

4331

0.

1337

81

597

1091

2 76

1415

22

5870

3 27

.7

0.00

0373

1.

7E+0

9 4.

11E

+09

0.61

79

0.78

61

45-5

4 45

10

0.

5 19

13.9

54

.7

0.02

8565

0.

2500

70

686

1766

8 61

8515

14

9728

9 21

.2

0.00

0857

1.

3E+0

9 2.

41E

+09

0.48

25

0.69

46

55-6

4 55

10

0.

5 90

1.53

36

.4

0.04

0429

0.

3363

53

017

1783

0 44

1024

87

8774

16

.6

0.00

206

1.37

E+0

9 1.

47E

+09

0.52

40

0.72

39

65-7

4 65

10

0.

5 24

7.0

17.2

0.

0696

87

0.51

68

3518

7 18

185

2609

49

4377

50

12.4

0.

0074

98

1.11

E+0

9 1.

11E

+09

0.89

48

0.94

60

75-8

4 75

10

0.

5 13

4.6

9.1

0.06

7707

0.

5058

17

003

8600

12

7023

17

6801

10

.4

0.01

3876

1E

+08

1E+0

8 0.

3469

0.

5890

85

+ 85

9.9

2.0

0.20

37

1.00

00

8402

84

02

4977

8 49

778

5.9

0 0

0 0

0

iv

Fem

ales

Age

x Ye

ars

in

Int.

Line

-ar

ity

Adj

ust.

Pop.

D

eath

sM

orta

lity

Rat

e

Prob

-ab

ility

of

D

ying

Indi

vid-

uals

Su

rviv

ing

Dea

ths

inIn

terv

al

x

Year

sLi

ved

in

Inte

rval

x

Cum

ul-

ativ

e Ye

ars

Live

d

Expe

c-ta

ncy

of

Life

at

Age

x

Var (

qx)

lx2[

(1-

ax)n

1+ex

+1]2

var(

qx)

2

sum

(lx2[

(1-

ax)n

1+ex

+1]2

var(

qx)

2)

Var (

ex)

SE

Low

er

95%

C

I

Upp

er95

%

CI

0-4

0 5

0.3

1537

.9

23.1

0.

0150

0.

0714

10

0000

71

38

4750

19

5747

213

57.5

0.

0002

05

7.44

E+0

9 1.

19E

+10

1.19

20

1.09

18

55.3

59

.6

5-14

5

10

0.5

6028

.0

2.0

0.00

03

0.00

33

9286

2 30

9 92

7080

52

7219

4 56

.8

5.49

E-0

6 1.

28E

+08

4.48

E+0

9 0.

5198

0.

7210

15-2

4 15

10

0.

5 56

64.1

8.

0 0.

0014

0.

0141

92

553

1304

91

9014

43

4511

4 46

.9

2.44

E-0

5 3.

78E

+08

4.35

E+0

9 0.

5083

0.

7130

25-3

4 25

10

0.

5 43

71.1

12

.1

0.00

28

0.02

72

9124

9 24

83

9000

80

3426

100

37.5

5.

97E

-05

5.57

E+0

8 3.

98E

+09

0.47

76

0.69

11

35-4

4 35

10

0.

5 28

98.0

28

.1

0.00

97

0.09

26

8876

7 82

18

8465

78

2526

020

28.5

0.

0002

76

8.75

E+0

8 3.

42E

+09

0.43

41

0.65

89

45-5

4 45

10

0.

5 23

67.1

55

.3

0.02

33

0.20

90

8054

9 16

838

7212

99

1679

441

20.8

0.

0006

25

6.77

E+0

8 2.

22E

+09

0.34

26

0.58

53

55-6

4 55

10

0.

5 10

42.0

26

.1

0.02

51

0.22

28

6371

1 14

192

5661

47

9581

43

15.0

0.

0014

76

1E+0

9 2.

55E

+09

0.62

71

0.79

19

65-7

4 65

10

0.

5 27

3.9

33.2

0.

1211

0.

7541

49

519

3734

4 30

8464

39

1996

7.

9 0.

0042

17

1.45

E+0

9 1.

55E

+09

0.63

04

0.79

40

75-8

4 75

10

0.

5 15

1.8

21.1

0.

1390

0.

8200

12

174

9983

71

828

8353

2 6.

9 0.

0057

36

9091

4360

90

9143

60

0.61

34

0.78

32

85+

85

33.6

5.

0 0.

1497

1.

0000

21

91

2191

11

703

1170

3 5.

3 0

0 0

0 0

v L

ife ta

bles

for

Pala

u (1

999-

2009

) 19

99-2

003

Mal

es

Fem

ales

vi

2004

-200

9 (e

xcl.

2007

)

Mal

es

Fem

ales

vii

Life

tabl

es fo

r T

onga

(200

5-20

09)

Obs

erve

d –

Rec

onci

led

Mal

es

Age

x

Year

s�in

�In

terv

al�

Line

arity

�Ad

just

men

t�R

epor

ted

Popu

latio

nDe

aths

�M

orta

lity

Rate

�

Prob

abili

ty�

of�

Dyin

g�In

divi

dual

sSu

rviv

ing�

Deat

hs�

in�

Inte

rval

�x�

Year

s�Li

ved�

inIn

terv

al�

x�Cu

mul

ativ

eYe

ars�L

ived

Expe

ctan

cy�

of�

Life

�at�A

ge�

x�0-

4 0

5 0.

3 36

,031

179.

824

0.00

4991

0.02

45

1000

00

2453

49

1416

65

2112

7 65

.2

5-9

55

0.5

33,8

91�

19.7

150.

0005

820.

0029

97

547

283

4870

29

6029

711

61.8

10-1

4 10

5 0.

5 32

,086

19.7

150.

0006

140.

0031

97

264

298

4855

75

5542

682

57.0

15-1

9 15

5 0.

5 26

,854

33.2

040.

0012

360.

0062

96

966

598

4833

35

5057

107

52.2

20-2

4 20

5 0.

5 23

,389

43.5

810.

0018

630.

0093

96

368

894

4796

07

4573

772

47.5

25-2

9 25

5 0.

5 18

,070

41.5

050.

0022

970.

0114

95

474

1090

47

4647

40

9416

6 42

.9

30-3

4 30

5 0.

5 15

,828

46.6

940.

0029

50.

0146

94

384

1382

46

8466

36

1951

9 38

.3

35-3

9 35

5 0.

5 16

,123

53.9

570.

0033

470.

0166

93

002

1543

46

1153

31

5105

2 33

.9

40-4

4 40

5 0.

5 13

,300

83.0

110.

0062

410.

0307

91

459

2810

45

0269

26

8989

9 29

.4

45-4

9 45

5 0.

5 9,

810

103.

764

0.01

0578

0.05

15

8864

9 45

68

4318

25

2239

630

25.3

50-5

4 50

5 0.

5 8,

231

111.

556

0.01

3553

0.06

55

8408

1 55

11

4066

28

1807

805

21.5

55-5

9 55

5 0.

5 6,

740

134.

893

0.02

0014

0.09

53

7857

0 74

88

3741

31

1401

177

17.8

60-6

4 60

5 0.

5 5,

800

184.

245

0.03

1769

0.14

72

7108

2 10

460

3292

60

1027

046

14.4

65-6

9 65

5 0.

5 5,

523

228.

909

0.04

1449

0.18

78

6062

2 11

384

2746

50

6977

86

11.5

70-7

4 70

5 0.

5 3,

937

305.

065

0.07

7488

0.32

46

4923

8 15

981

2062

38

4231

36

8.6

75+

850.

5 4,

552

658.

900

0.14

4753

1.00

00

3325

7 33

257

2168

99

2168

99

6.5

viii

Fem

ales

Age

x

Year

s�in

�In

terv

al�

Line

arity

�Ad

just

men

t�R

epor

ted

Popu

latio

nDe

aths

�M

orta

lity

Rate

�

Prob

abili

ty�

of�

Dyin

g�In

divi

dual

sSu

rviv

ing�

Deat

hs�

in�

Inte

rval

�x�

Year

s�Li

ved�

inIn

terv

al�

x�Cu

mul

ativ

eYe

ars�L

ived

Expe

ctan

cy�

of�

Life

�at�A

ge�

x�0-

4 0

5 0.

3 33

,126

130.

684

0.00

3945

0.01

95

1000

00

1946

49

3190

69

6071

0 69

.6

5-9

55

0.5

30,4

94�

14.6

720.

0004

810.

0024

98

054

236

4896

83

6467

520

66.0

10-1

4 10

5 0.

5 29

,497

10.4

800.

0003

550.

0018

97

819

174

4886

60

5977

837

61.1

15-1

9 15

5 0.

5 24

,316

10.4

800.

0004

310.

0022

97

645

210

4877

00

5489

177

56.2

20-2

4 20

5 0.

5 22

,870

15.7

200.

0006

870.

0034

97

435

334

4863

39

5001

477

51.3

25-2

9 25

5 0.

5 18

,283

18.8

640.

0010

320.

0051

97

101

500

4842

54

4515

138

46.5

30-3

4 30

5 0.

5 16

,116

39.8

240.

0024

710.

0123

96

601

1186

48

0039

40

3088

4 41

.7

35-3

9 35

5 0.

5 15

,991

40.8

720.

0025

560.

0127

95

415

1212

47

4045

35

5084

5 37

.2

40-4

4 40

5 0.

5 12

,323

48.2

080.

0039

120.

0194

94

203

1825

46

6454

30

7680

0 32

.7

45-4

9 45

5 0.

5 10

,389

61.8

320.

0059

520.

0293

92

378

2709

45

5120

26

1034

6 28

.3

50-5

4 50

5 0.

5 9,

196

108.

458

0.01

1794

0.05

73

8967

0 51

36

4355

07

2155

226

24.0

55-5

9 55

5 0.

5 7,

447

114.

232

0.01

534

0.07

39

8453

3 62

44

4070

55

1719

719

20.3

60-6

4 60

5 0.

5 6,

525

145.

083

0.02

2236

0.10

53

7828

9 82

46

3708

30

1312

664

16.8

65-6

9 65

5 0.

5 5,

472

150.

276

0.02

7462

0.12

85

7004

3 90

00

3277

17

9418

33

13.4

70-7

4 70

5 0.

5 4,

123

161.

391

0.03

9141

0.17

83

6104

4 10

882

2780

13

6141

17

10.1

75+

850.

5 6,

089

554.

390

0.09

1047

1.00

00

5016

2 50

162

3361

03

3361

03

6.7

ix

Fina

l adj

ustm

ent (

repo

rtin

g co

mpl

eten

ess

from

Ton

gata

pu c

aptu

re-r

ecap

ture

)

Mal

es

Age

x

Year

s�in

�In

terv

al�

Line

arity

�Ad

just

men

t�R

epor

ted

Popu

latio

nDe

aths

�M

orta

lity

Rate

�

Prob

abili

ty�

of�

Dyin

g�In

divi

dual

sSu

rviv

ing�

Deat

hs�

in�

Inte

rval

�x�

Year

s�Li

ved�

inIn

terv

al�

x�Cu

mul

ativ

eYe

ars�L

ived

Expe

ctan

cy�

of�

Life

�at�A

ge�

x�

0-4

05

0.3

36,0

31�

256.

891

0.00

713

0.03

48

1000

00

3478

48

7827

60

4479

6 60

.4

5-9

55

0.5

33,8

9128

.573

0.00

0843

0.00

42

9652

2 40

6 48

1595

55

5696

9 57

.6

10-1

4 10

5 0.

5 32

,086

28.5

730.

0008

910.

0044

96

116

427

4795

12

5075

374

52.8

15-1

9 15

5 0.

5 26

,854

48.1

220.

0017

920.

0089

95

689

854

4763

11

4595

862

48.0

20-2

4 20

5 0.

5 23

,389

63.1

610.

0027

0.01

34

9483

5 12

72

4709

97

4119

551

43.4

25-2

9 25

5 0.

5 18

,070

60.1

530.

0033

290.

0165

93

563

1544

46

3956

36

4855

4 39

.0

30-3

4 30

5 0.

5 15

,828

67.6

720.

0042

750.

0212

92

019

1946

45

5230

31

8459

7 34

.6

35-3

9 35

5 0.

5 16

,123

78.1

990.

0048

50.

0240

90

073

2158

44

4968

27

2936

8 30

.3

40-4

4 40

5 0.

5 13

,300

120.

306

0.00

9045

0.04

42

8791

5 38

88

4298

53

2284

400

26.0

45-4

9 45

5 0.

5 9,

810

150.

382

0.01

533

0.07

38

8402

6 62

03

4046

25

1854

547

22.1

50-5

4 50

5 0.

5 8,

231

161.

675

0.01

9642

0.09

36

7782

4 72

85

3709

05

1449

922

18.6

55-5

9 55

5 0.

5 6,

740

195.

497

0.02

9006

0.13

52

7053

8 95

39

3288

45

1079

017

15.3

60-6

4 60

5 0.

5 5,

800

267.

021

0.04

6041

0.20

64

6100

0 12

593

2735

16

7501

71

12.3

65-6

9 65

5 0.

5 5,

523

331.

752

0.06

0071

0.26

11

4840

7 12

641

2104

31

4766

55

9.8

70-7

4 70

5 0.

5 3,

937

442.

124

0.11

2301

0.43

84

3576

6 15

680

1396

28

2662

24

7.4

75+

85

0.5

4,55

295

4.92

70.

2097

861.

0000

20

085

2008

5 12

6596

12

6596

6.

3

x Fe

mal

es

Age

x

Year

s�in

�In

terv

al�

Line

arity

�Ad

just

men

t�R

epor

ted

Popu

latio

nDe

aths

�M

orta

lity

Rate

�

Prob

abili

ty�

of�

Dyin

g�In

divi

dual

sSu

rviv

ing�

Deat

hs�

in�

Inte

rval

�x�

Year

s�Li

ved�

inIn

terv

al�

x�Cu

mul

ativ

eYe

ars�L

ived

Expe

ctan

cy�

of�

Life

�at�A

ge�

x�

0-4

05

0.3

33,1

26�

186.

691

0.00

5636

0.02

76

1000

00

2763

49

0328

65

4120

3 65

.4

5-9

55

0.5

30,4

9421

.576

0.00

0708

0.00

35

9723

7 34

3 48

5325

60

5087

5 62

.2

10-1

4 10

5 0.

5 29

,497

15.4

120.

0005

220.

0026

96

893

253

4838

34

5565

550

57.4

15-1

9 15

5 0.

5 24

,316

15.4

120.

0006

340.

0032

96

640

306

4824

38

5081

716

52.6

20-2

4 20

5 0.

5 22

,870

23.1

180.

0010

110.

0050

96

335

486

4804

59

4599

278

47.7

25-2

9 25

5 0.

5 18

,283

27.7

410.

0015

170.

0076

95

849

724

4774

34

4118

819

43.0

30-3

4 30

5 0.

5 16

,116

58.5

650.

0036

340.

0180

95

125

1713

47

1341

36

4138

5 38

.3

35-3

9 35

5 0.

5 15

,991

60.1

060.

0037

590.

0186

93

412

1739

46

2711

31

7004

4 33

.9

40-4

4 40

5 0.

5 12

,323

70.8

940.

0057

530.

0284

91

673

2599

45

1864

27

0733

4 29

.5

45-4

9 45

5 0.

5 10

,389

90.9

290.

0087

530.

0428

89

073

3815

43

5829

22

5547

0 25

.3

50-5

4 50

5 0.

5 9,

196

159.

497

0.01

7344

0.08

31

8525

8 70

86

4085

76

1819

641

21.3

55-5

9 55

5 0.

5 7,

447

167.

988

0.02

2559

0.10

68

7817

2 83

47

3699

93

1411

065

18.1

60-6

4 60

5 0.

5 6,

525

213.

357

0.03

270.

1511

69

825

1055

4 32

2742

10

4107

2 14

.9

65-6

9 65

5 0.

5 5,

472

220.

995

0.04

0385

0.18

34

5927

2 10

871

2691

81

7183

30

12.1

70-7

4 70

5 0.

5 4,

123

237.

340

0.05

756

0.25

16

4840

1 12

177

2115

60

4491

49

9.3

75+

85

0.5

6,08

981

5.28

00.

1338

931.

0000

36

223

3622

3 23

7589

23

7589

6.

6

xi

Life

tabl

es fo

r K

irib

ati (

2000

-200

9)

Fina

l est

imat

es in

clud

ing

reco

ncile

d da

ta fo

r chi

ldre

n 0-

4 ye

ars

and

capt

ure-

reca

ptur

e an

alys

is c

orre

ctio

n (d

isag

greg

ated

by

age

and

sex)

for r

emai

ning

age

gro

ups.

Incl

udes

dea

ths

reco

rded

as

stillb

irths

in in

fant

dea

ths

(see

Cha

pter

5 fo

r det

ails

)

2000

-200

4

Mal

es

xii

Fem

ales

2005

-200

9

Mal

es

xiii

Fem

ales

xiv

Life

tabl

es fo

r V

anua

tu 2

009-

2010

Li

fe ta

bles

der

ived

from

Mod

mat

ch (2

inpu

t par

amet

ers)

usi

ng U

5M fr

om th

e 20

09 c

ensu

s an

d 20

10 e

stim

ates

of a

dult

mor

talit

y fro

m

Rat

nara

jam

et.

al.

Mal

es

se

x ag

e lx

a

n M

x qx

dx

Lx

Tx

E

x e0

_5

q0

_45q

15

_20q

60

1 1

1 0.

3 1

0.01

9315

0.01

9058

0.01

9058

0.98

666

65.3

7587

65.3

7587

65.3

7587

0.02

40.

2749

950.

6981

54

1 2

0.98

0943

0.

4 4

0.00

1264

0.00

5039

0.00

4943

3.91

1908

64.3

8921

65.6

4014

1 3

0.97

6 0.

5 5

0.00

0586

0.00

2924

0.00

2854

4.87

2865

60.4

773

61.9

6445

1 4

0.97

3146

0.

5 5

0.00

0555

0.00

2773

0.00

2698

4.85

8984

55.6

0444

57.1

3885

1 5

0.97

0448

0.

5 5

0.00

1495

0.00

7447

0.00

7227

4.83

4171

50.7

4546

52.2

9077

1 6

0.96

3221

0.

5 5

0.00

2286

0.01

1367

0.01

0948

4.78

8732

45.9

1129

47.6

6435

1 7

0.95

2272

0.

5 5

0.00

253

0.01

2569

0.01

1969

4.73

1438

41.1

2255

43.1

8361

1 8

0.94

0303

0.

5 5

0.00

3161

0.01

5681

0.01

4745

4.66

4653

36.3

9112

38.7

0148

1 9

0.92

5558

0.

5 5

0.00

4358

0.02

1555

0.01

995

4.57

7916

31.7

2646

34.2

7819

1 10

0.90

5608

0.

5 5

0.00

6404

0.03

1515

0.02

8541

4.45

6689

27.1

4855

29.9

7825

1 11

0.87

7068

0.

5 5

0.00

9499

0.04

6392

0.04

0689

4.28

3617

22.6

9186

25.8

7242

1 12

0.83

6379

0.

5 5

0.01

4082

0.06

8015

0.05

6886

4.03

968

18.4

0824

22.0

0945

1 13

0.77

9493

0.

5 5

0.02

0475

0.09

7388

0.07

5913

3.70

7683

14.3

6856

18.4

3321

1 14

0.70

358

0.5

5 0.

0304

620.

1415

330.

0995

83.

2689

510

.660

8815

.152

33

1

150.

604

0.5

5 0.

0440

30.

1983

180.

1197

842.

7205

47.

3919

2812

.238

29

1

160.

4842

16

0.5

5 0.

0653

830.

2809

870.

1360

582.

0809

334.

6713

899.

6473

28

1

170.

3481

58

0.5

5 0.

0968

980.

3900

10.

1357

851.

4013

252.

5904

567.

4404

69

1

180.

2123

73

0.5

5 0.

1462

990.

5355

990.

1137

470.

7774

971.

1891

35.

5992

64

1

190.

0986

26

0.5

5 0.

2148

380.

6988

440.

0689

240.

3208

20.

4116

344.

1736

82

1

200.

0297

02

0.4

5 0.

3068

50.

7988

60.

0237

280.

0773

260.

0908

143.

0575

33

1

210.

0059

74

0.3

5 0.

4262

710.

8552

970.

0051

10.

0119

870.

0134

882.

2577

01

1

220.

0008

65

5

0.57

5956

10.

0008

650.

0015

010.

0015

011.

7362

43

xv

� Females

sex

age

Lx

a n

Mx

qx

dx

Lx

Tx

ex

e0

_5q0

_4

5q15

_2

0q60

2

1 1

0.3

10.

0182

260.

0179

970.

0179

970.

9874

02

70.9

2056

70.9

2056

70.9

2056

0.02

40.

1699

950.

5500

05

2 2

0.98

2003

0.4

40.

0015

340.

0061

130.

0060

033.

9136

05

69.9

3316

71.2

148

2 3

0.97

60.

5 5

0.00

0578

0.00

2886

0.00

2816

4.87

296

66.0

1955

67.6

4298

2 4

0.97

3184

0.5

50.

0005

110.

0025

510.

0024

834.

8597

13

61.1

4659

62.8

3149

2 5

0.97

0701

0.5

50.

0009

60.

0047

880.

0046

484.

8418

87

56.2

8688

57.9

858

2 6

0.96

6053

0.5

50.

0012

040.

0060

020.

0057

984.

8157

72

51.4

4553

.252

75

2

7 0.

9602

550.

5 5

0.00

138

0.00

6877

0.00

6603

4.78

4769

46

.629

2248

.559

19

2

8 0.

9536

520.

5 5

0.00

1794

0.00

8932

0.00

8518

4.74

6966

41

.844

4643

.878

12

2

9 0.

9451

340.

5 5

0.00

257

0.01

2767

0.01

2066

4.69

5506

37

.097

4939

.251

02

2

10

0.93

3068

0.5

50.

0037

990.

0188

140.

0175

554.

6214

54

32.4

0198

34.7

2628

2 11

0.

9155

130.

5 5

0.00

5584

0.02

7538

0.02

5211

4.51

4539

27

.780

5330

.344

21

2

12

0.89

0302

0.5

50.

0081

770.

0400

670.

0356

724.

3623

31

23.2

6599

26.1

3269

2 13

0.

8546

30.

5 5

0.01

1791

0.05

7269

0.04

8944

4.15

0793

18

.903

6622

.119

11

2

14

0.80

5687

0.5

50.

0176

910.

0847

090.

0682

493.

8578

11

14.7

5286

18.3

1092

2 15

0.

7374

380.

5 5

0.02

6996

0.12

6447

0.09

3247

3.45

4072

10

.895

0514

.774

2

2

16

0.64

4191

0.5

50.

0435

710.

1964

560.

1265

552.

9045

67

7.44

0981

11.5

5089

2 17

0.

5176

360.

5 5

0.07

0476

0.29

9594

0.15

5081

2.20

0478

4.

5364

148.

7637

16

2

18

0.36

2555

0.5

50.

1160

640.

4498

060.

1630

81.

4050

78

2.33

5936

6.44

2978

2 19

0.

1994

760.

5 5

0.18

2978

0.62

7735

0.12

5218

0.68

4334

0.

9308

584.

6665

22

2

20

0.07

4258

0.4

50.

2761

460.

7551

420.

0560

750.

2030

63

0.24

6524

3.31

9843

2 21

0.

0181

830.

3 5

0.39

8952

0.83

2422

0.01

5136

0.03

7938

0.

0434

612.

3902

43

2

22

0.00

3047

5

0.55

1751

10.

0030

470.

0055

22

0.00

5522

1.81

2414

xvi

Appendix 3: Additional analysis on mortality level in Fiji including life tables

KNOWLEDGE HUBS FOR HEALTHStrengthening health systems through evidence in Asia and the Pacific

An assessment of mortality estimates for Fiji, 1949-2008: findings and life

tables

Karen CarterHealth Information Systems Knowledge Hub, School of Population Health, University of Queensland

Margaret CorneliusFiji Ministry of Health

Richard TaylorSchool of Population Health, University of Queensland and School of Public Health and Community Medicine,

University of New South Wales

Shareen S AliFiji Ministry of Health

Chalapati RaoHealth Information Systems Knowledge Hub, School of Population Health, University of Queensland

Alan LopezHealth Information Systems Knowledge Hub, School of Population Health, University of Queensland

Vasemaca LewaiFiji Islands Bureau of Statistics

Ramneek GoudarFiji School of Medicine

Claire MowrySchool of Population Health, University of Queensland

HEALTH INFORMATION SYSTEMS KNOWLEDGE HUB

DOCUMENTATION NOTE SERIES NUMBER 12| NOVEMBER 2010

School of Population HealthUniversity of Queensland

Health Information Systems

Knowledge Hub


Health Information Systems Hub (HIS) Knowledge Hub

2 DOCUMENTATION NOTE SERIES NUMBER 12 | NOVEMBER 2010

ABOUT THIS SERIES

The objective of the HIS Hub Documentation Series is to document in detail the methods and findings of the HIS Knowledge Hub’s activities in partner countries.

The Documentation Series also serves to describe the state of the health information system in a number of Asia and Pacific Island countries. This series provides a baseline for comparison of HIS between countries. It also provides a preliminary diagnostic analysis for use by countries in determining areas requiring improvement. The target audiences for these mappings are all stakeholders with an interest in the functioning or development of HIS, specifically those working in the Pacific and Asia region. The Series also reports on work in progress, particularly for large, complex or multi-year HIS Hub initiatives, or on specific components or aspects of such projects that may be of more immediate relevance to stakeholders

The opinions or conclusions contained in the Documentation Series are those of the authors and do not necessarily reflect the views of institutions or governments.

3 NUMBER 12 | NOVEMBER 2010 DOCUMENTATION NOTE SERIES



ACKNOWLEDGEMENTS

The authors would like to acknowledge the assistance of the Fiji Ministry of Health and Bureau of Statistics. Funding for this research was provided under the AusAID Australian Research Development Awards (HS0024).




ABSTRACT

Published mortality indices for Fiji and causes of death for 1940-2008 were assessed for quality, and compared with those generated from recent Ministry of Health reporting1. Trends in credible estimates are compared with proportional mortality for cause of death, and life tables generated from recent Ministry of Health data (1996-2004). Infant mortality rate (IMR) declined to below 20 by 2000 (per 1000). Life expectancy (LE) increased, but has been stable at 64 and 69 years since the 1980’s for males and females respectively. Proportional mortality from cardiovascular disease has increased from around 20% in the 1960s to over 45%. Credible estimates indicate stagnation in life expectancy in the context of a declining and relatively low infant mortality rate. Proportional mortality trends suggest this is largely due to the emergence of chronic diseases (especially cardiovascular) in adults. Incompatible variations in published estimates of infant mortality rate and life expectancy were indentified. Difficulties encountered in assessment of published data were the lack of information in many publications concerning the primary data sources, methods used for calculation, and the assumptions made. In order to deal with this lack of information, exclusion criteria were specified and unreliable estimates censored from further analyses. The use of single input parameter models to estimate life expectancy from childhood mortality results in estimates that are similar to many of the published estimates that were censored from the final analysis, and overestimates life expectancy by 4-5 years. The most plausible estimates of mortality are derived from two local sources: census analyses from the Fiji Bureau of Statistics and routine death reporting from the Fiji Ministry of Health, with these data found to be mostly more than 95% complete by the Brass method.




INTRODUCTION

Mortality is an important measure of the health of a population. As countries move through the demographic and epidemiological transitions, declines in under-nutrition and infectious diseases are reflected in improvements in survival (especially child survival) which result in improvements in life expectancy at birth. Improvements in the economic, social, and physical environment, and developments in public health and health services result in declining age-specific mortality rates. However, many societies undergoing modernization also experience real increases in adult mortality from noncommunicable diseases (NCDs) and injury, which, if of sufficient magnitude, can lead to a slowing, plateau or reversal in overall mortality decline2-4. Such a situation requires urgent public health action to reduce premature adult mortality from noncommunicable diseases, such as occurred in Australia and New Zealand5,6.

Fiji is a multi-ethnic society (Melanesian 57%, Indian 38%) of 827,900 people in 20077. Located approximately 1,770 kms north of New Zealand, Fiji has over 106 inhabited volcanic islands. The two largest Vitu Levu and Vanua Levu cover over 87% of the land area and contains 84% of the population6. Mortality measures for Fiji are routinely published by government sources as well as by a range of international agencies.

Available data on mortality level and causes of death, over the period 1940-2008, are examined for plausibility to establish an understanding of recent mortality levels. Death reporting from the Ministry of Health is also examined for completeness and life tables generated for 1996-20041.

KEY FINDINGS

The infant mortality rate has fallen significantly over the period investigated to below 20 deaths per 1000 live births. Empirical estimates for 2002-2004 place the infant mortality rate at 17.4 deaths per 1000 live births for males and 16.0 deaths per 1000 live births for females.

LE rose over the period of observation but has stabilized at around 64 years for males and 69 years for females since the 1980s and there have been no indications of any sustained improvement since this time.

Adult mortality is two to three times higher in Fiji than in Australia or New Zealand. The high levels of premature adult mortality in concert with the significant increase in proportional mortality from diseases of the circulatory system suggest that there is a severe epidemic of cardiovascular disease that is preventing improvement in life expectancy. Many of the previously published estimates of life expectancy for Fiji are implausible. This work demonstrates that the most plausible estimates of mortality are derived from two local sources: census analyses from the Fiji Bureau of Statistics and routine death reporting from the Fiji Ministry of Health.

STUDY METHODS

Data collection

Measures of level of mortality included in this work are from two sources:

1. Published reports by local and international agencies from 1940 (Appendix 1 and 2). 2. Data on reported deaths by age group, sex and period for 1996-2004 obtained from the Fiji Ministry of Health –

previously unpublished as measures of population mortality, except for the infant mortality rate.

Available data from published reports8-69 on all-cause and cause-specific mortality in Fiji from1940-2008 were sought through direct contact with Fiji government departments (including the Ministry of Health and Fiji Bureau of Statistics) and international and regional organizations known to have an interest in health or mortality in Fiji. Internet searches




included websites for United Nations agencies (UNDP1, UNICEF2, UNFPA3, and WHO4), regional institutions such as the Secretariat of the Pacific Community (SPC), The World Bank (WB), and Asian Development Bank (ADB), and non-government organisations. A literature search was also undertaken in PubMED and Medline.

Information on method of data collection and analysis was rarely available, with the notable exceptions of demographic analyses of the Fiji censuses by the Fiji Bureau of Statistics37,38. These census estimates of life expectancy were produced from separate estimates for childhood and adult mortality, with adult mortality calculated using the orphanhood technique, and childhood mortality calculated by a combination of indirect estimation using the children ever-born, children surviving technique, and a direct estimate from questions about the date of birth and survival of the most recent child born (partial birth history)37,38. Data sources and reasons for their exclusion from the final analysis are shown in Appendix 1. Limited data on causes of death have been published for Fiji, many of which included only leading causes of death and could not be adjusted to exclude ill-defined and unknown causes from proportional mortality. Sources used in this analysis are shown in Appendix 2.

Data assessment

All cause mortality

The infant mortality rate and life expectancy were chosen as measures of all-cause mortality because of availability. Levels of the infant mortality rate from Fiji Ministry of Health death registration (1996-2004) and published reports were tabulated and graphed over time by major source of data. Data sources were assessed for reliability and plausibility on the basis of method of estimation, original source of data and data consistency, with implausible or unreliable estimates censored from further analysis (shown in Appendix 1).

Data sources were removed from the analysis if they met any of the following criteria:

a) the source publication noted that the data were derived from models that assume a given improvement by year or data from a source showed a linear or curvilinear improvement in life expectancy or infant mortality rate by year (indicating use of models that incorporate this assumption);

b) multiple incompatible estimates were given by the source for a single year or incompatible estimates were given by the source for adjacent years;

c) source data included estimates that were extremely implausible (based on equivalent measures for developed countries within the region);

d) calculations were documented to be based on uncorrected vital registration data where there was either no assessment of completeness or data was known to be significantly under-reported (Appendix 1) .

An exponential trend was fitted to the average infant mortality rates of each year from data remaining after censoring (Microsoft Excel). Childhood mortality (probability of dying before 5 years of age) was largely unavailable from published reports. Recent measures were estimated by applying the proportion of childhood mortality accounted for by infant mortality (78%) from credible data sources for which both measures were available (analyses from the 198637 and 199638 censuses, and Ministry of Health data 1996-2004), to credible estimates of infant mortality rate for 2006-200854 .

Age-specific mortality rates and life tables71 for 1996-98, 199-2001, and 2002-04 were calculated from the MoH death data using census populations provided by the Secretariat of the Pacific Community (SPC)72. Due to the small population size, data were aggregated by triennia to reduce stochastic variation. Death recording by the Ministry of Health was analysed for completeness using the Brass growth-balance method71 and since registration was assessed to be mostly over 95% complete, no adjustment was made to number of the recorded deaths >1 year. Details of the Brass analyses are in Appendix 3.

Life expectancy and adult mortality (probability dying between the ages of 15 and 59 years inclusive) were then derived

1 United Nations Development Programme2 United Nations Children’s Fund3 United Nations Population Fund4 World Health Organization




from the life tables for each period (Appendix 4). Age-specific mortality for 2002-2004 was compared to measures derived from published life tables for Australia and New Zealand for the same period73-76.

Empirical measures of life expectancy derived from the Ministry of Health data for 1996-2008 were compared with similar measures for Fiji predicted by the WHO Model Life Table system77. The model was used to predict two estimates of life expectancy; the first using an empirical probability of dying before age five years as a single input; and the second using two input parameters; probabilities of childhood mortality and adult mortality (probability of dying between ages 15 and 59 years). The details of these analyses are in Appendix 5.

Calculated life expectancy for the Ministry of Health death recording, and reported life expectancy from previously published sources were tabulated and graphed over time by major source of data, and implausible or unreliable estimates were censored from further analysis. Difficulties encountered in assessment of published data were the lack of information in many publications concerning the primary data sources, methods used for calculation, and the assumptions made. In order to deal with this lack of information, strict exclusion criteria were specified and unreliable estimates censored from further analyses.

Causes of death

Available data on cause-specific mortality were examined using a range of measures (rates, number of deaths and proportions) and levels of disease or cause aggregation (shown in the Appendix 2). To assess trends over time, data are presented as proportional mortality by International Classification of Disease chapter78 since in several instances total deaths are not given and data are frequently under-enumerated. Cause of death was calculated for the total population as there was insufficient information in available tabulations of data to examine by age group, sex, or ethnicity. proportional mortality, excluding unknown and ill-defined cases, was calculated for sources where all deaths were included in the data source (allowing redistribution by proportion), or the information could be extracted directly from the reported data. Trends in proportional mortality, and proportional mortality excluding unknown and ill-defined cases, were graphed over time for each International Classification of Disease chapter. See Appendix 6.

FINDINGS

All cause mortality

Published estimates

There is extensive variation in the published estimates of infant mortality by source, with several sources reporting levels that are below those of developed countries in the region70 (Figure 1).




Figure 1: Estimates of infant mortality rate in Fiji, by source: 1940-2004

0

10

20

30

40

50

60

70

80

90

1940 1950 1960 1970 1980 1990 2000 2010

Infa

nt M

orta

lity

Rate

(pe

r 10

00 b

irths

).

Year






Census (Total) (12,36,37,39,46)





MOH Registration (Total) (9,13-15,19,24,26,28,33,34,42,43,53)

MOH Registration (Fijian) (13-15)

MOH Registration (Indo-Fijian) (13-15)

Brass adjustment of Vital registration (total) (25)

Brass adjustment of Vital registration (Fijian) (25)

Brass adjustment of Vital registration (Indo-Fijian) (25)

WHO - WPRO (45,49,56)

UNDP (total) (35,52,54)

UNICEF (total) (40,42,51,63,64,66)

UNESCAP (68)


SPC (total) (38,61,62)


WHO (48,50,55)

Figure adapted from Carter et al1.

Of 16 sources for infant mortality rate, 12 were censored from further analysis with a further two partially censored (Appendix 1). The overall trend is consistent with the infant mortality rate falling significantly over the last several decades (1940–2006) to a current level below 20 deaths per 1000 live births (Figure 2). The infant mortality rate in 2008 is estimated by projection at 15 deaths per 1000 live births which suggests a childhood (<5 years) mortality of 17 deaths per 1000 live births.




Figure 2: Estimated infant mortality rate in Fiji, sources remaining after censoring: 1940-2004

0

10

20

30

40

50

60

70

80

1940 1950 1960 1970 1980 1990 2000 2010

Infa

nt M

orta

lity

Rat

e (p

er 1

000

birt

hs).

Year

Average of remaining estimates


Published estimates of life expectancy at birth between 1995 and 2000 vary by more than 10 years in males (from 61 to 72 years), and by 8 years in females (from 68 to 76 years) (Figures 3 and 4).

Figure 3: Estimates of male life expectancy at birth in Fiji, by source: 1940-2006


ADB- Asian Development Bank; BoS - Fiji Bureau of Statistics; MoH - Fiji Minstry of Health; SPC - Secretariat of the Pacific Community; UNDP - United Nations Development Program; WHO - World Health Organization; WPRO – WHO Western Pacific Regional Office; VR – Vital registrationNB – Census data co-published by SPC with the Fiji Bureau of Statistics listed under BoS onlyReferences shown in brackets after each source within figure.




Figure 4: Estimates of female life expectancy at birth in Fiji, by source: 1940-2006


ADB- Asian Development Bank; BoS - Fiji Bureau of Statistics; MoH - Fiji Minstry of Health; SPC - Secretariat of the Pacific Community; UNDP - United Nations Development Program; WHO - World Health Organization; WPRO – WHO Western Pacific Regional Office; VR – Vital registrationNB – Census data co-published by SPC with the Fiji Bureau of Statistics listed under BoS onlyReferences shown in brackets after each source within figure.

Of the 16 data sources identified for life expectancy data, 13 were censored from the final analysis (Appendix 1). Once unreliable sources are removed, the estimates suggest that life expectancy began to level out in the 1980’s, and has not shown a significant improvement since (Figures 5 and 6). Life expectancy for 2008 is estimated at 64 years for males and 69 years for females.




Figure 5: Reliable estimates of male life expectancy at birth in Fiji, by source: 1940-2006


Figure 6: Reliable estimates of female life expectancy at birth in Fiji, by source: 1940-2006


Ministry of health death recording

The Ministry of Health death registration data, aggregated by year, sex and age group were obtained for the years 1996 to 2004. Application of the Brass growth-balance method for age groups 35 to 74 years suggests the recorded deaths to be mostly more than 95% complete (except for female deaths for 2002-2004 which was 85% complete) as shown in Figure 7 below and Appendix 3. The Brass analysis of data for 1996-2001 is shown in Appendix 3.

40

45

50

55

60

65

70

75

80

1940 1950 1960 1970 1980 1990 2000 2010

Life

Exp

ecta

ncy

at B

irth

(yea

rs)

Year

MOH Reported Deaths (Brass adjustment) (21,26)BoS Census (13,37,38,40,47)MOH reported deaths (unpublished data)

40

45

50

55

60

65

70

75

80

1940 1950 1960 1970 1980 1990 2000 2010

Life

Exp

ecta

ncy

at B

irth

(yea

rs)

Year

MOH Reported Deaths (Brass adjustment) (21,26)BoS Census (13,37,38,40,47)MOH reported deaths (unpublished data)




Figure 7: Analysis for mortality data completeness by gender: Fiji, 2002-2004

a) Males b) Females

Completeness (c) is estimated by 1/K where K= the slope of the line

Age-specific mortality

As shown in Figure 8, mortality across all age groups is significantly higher than in Australia and New Zealand.

Figure 8: Age specific mortality rates: Fiji, Australia and New Zealand, 2002-2004

Source data: Fiji - calculated from deaths reported to the Ministry of Health (previously unpublished); New Zealand75, 76; Australia73-74

K=0.89C=>100%

K=1.18C=85%




Life expectancy

Analysis of the Ministry of Health reported deaths shows life expectancy between 1996 and 2004 was stable at 64-65 years for males and 69 for females (see life tables in Appendix 4).

Table 1: IMR and life expectancy at birth for Fiji, Australia and New Zealand, by period and gender: 1996-2004

Age group (years) IMR (deaths per 1,000 live births LE at birthSex Males Females Males Females

Fiji1996-1998 17.4 (16.6-18.2) 16.0 (15.2-16.8) 65.0 (64.9-65.2) 69.2 (68.9-69.4)1999-2001 15.7 (14.9-16.5) 13.3 (12.6-14.0) 63.7 (63.4-63.9) 68.7 (68.4-69.0)2002-2004 16.9 (16.1-17.7) 15.1 (14.3-15.9) 64.3 (64.0-64.6) 69.1 (68.8-69.4)

New Zealand 2002-2004 5.7 5.0 77.0 81.3Australia 2002-2004 5.3 4.5 78.1 83.0

Source data: Fiji - calculated deaths reported to the Ministry of Health; New Zealand75-76; Australia73-74

Adult mortality

Adult mortality (probability of dying between 15 and 59 years) also did not change significantly over the period 1996-2004. The adult mortality levels in Fiji for 2002-2004 were three times higher than in Australia or New Zealand. The absolute difference is far greater between ages 35-59 years, than other adult ages. The probabilities of dying (for 2002-2004) between ages 35-59 years in Fiji were 25% for males and 18% for females, compared to 7 - 8% and 4 - 6%, for males and females, respectively, in Australia and New Zealand73-76(Table 2).

Table 2: Probability of dying between ages 15-59 years (%) by age group in Fiji, Australia and New Zealand, males and females, 1996-2004

Age group (years) 15-34 35-59 15-59Sex Males Females Males Females Males Females

Fiji1996-1998 3.6 2.5 26.8 18.5 29.5 20.51999-2001 4.1 2.9 27.5 18.4 30.5 20.82002-2004 3.7 2.9 24.9 17.9 27.6 20.3

New Zealand 2002-2004 2.0 0.9 7.9 5.5 9.7 6.4Australia 2002-2004 1.7 0.7 7.2 4.4 8.9 5.1

Table adapted from Carter et al1. Source data: Fiji- calculated from deaths reported to the Ministry of Health (previously unpublished); New Zealand75-76; Australia73, 74.

Comparison of empircal estimates with modelled estimates of mortality

Life expectancy and adult mortality predictions from model life tables using the empirical childhood mortality (<5 years) and adult mortality (15-59 years) from the Ministry of Health death recording varied less than 1 year from results calculated using empirical age-specific mortality rates (Figure 9 and 10). Modmatch analysis of data from earlier years is shown in Appendix 5 and 6. However, life expectancy predicted from the Ministry of Health childhood mortality alone (Figure 9), produced estimates similar to those from sources considered in this study as erroneous (Appendix 1), and over-estimated life expectancy by 4-5 years.




Figure 9: Empirical age-specific mortality compared with predictions from one and two parameter models for Fiji, males: 2002-2004

Model: WHO Modmatch77

Figure 10: Empirical age-specific mortality compared with predicitions from one and two parameter models for Fiji, females: 2002-2004

Model: WHO Modmatch77

Summary of mortality estimates

Sources remaining after censorship for quality and method demonstrate that infant mortality rate has fallen steadily to below 20 deaths per 1000 live births. Empirical estimates from the Ministry of Health data placed the infant mortality rate for 2002-2004 at 17.4 deaths per 1000 live births for males and 16.0 deaths per 1000 live births for females (Appendix 4). Credible sources indicate the infant mortality rate for 2006-08 was 18 -20 deaths per 1000 live births, suggesting s a childhood mortality rate (<5 years) of 23-26 deaths per 1000 live births (see methods).




Life expectancy rose but has stabilized at around 64 years for males and 69 years for females from the 1980’s. While the infant mortality rate has not fallen during the period of stagnation in life expectancy, it had fallen previously (Figure 2), and at levels since the late 1980s, the infant mortality rate is a minor influence on the overall life expectancy71.

The most reliable all cause mortality estimates are those from local census data analysis and recent Ministry of Health death registration. The estimates of life expectancy produced from the Ministry of Health data were amongst the lower estimates found in the published data. The Ministry of Health data were not adjusted for under-reporting as the data were found to be substantially complete, however if an adjustment had been made this would have resulted in an even lower measure of life expectancy. Level of mortality (and life expectancy) calculated from the Ministry of Health death registration (direct methods)71 correlates closely with the published Census mortality estimates (indirect methods)37,38, thus providing independent verification of these figures, and by implication, the completeness of death reporting by the Ministry of Health in Fiji.

Cause specific mortality

The proportion of deaths coded to ill-defined and unknown causes over the period since 1960 generally fluctuated below 8% of deaths, with substantially higher peaks (up to 18%) in the early part of the 1980’s and again in the early 2000’s.

The proportion of deaths attributable to “Diseases of the Circulatory System” (International Classification of Disease) Chapter IX) shows a significant and ongoing linear increase over 1960-2000, from around 20% of deaths to over 45% of all deaths. This trend is clearly apparent both before (Figure 11) and after the data are refined to censor ill-defined and unknown causes of death (Figure 12), although it is important not to over-interpret reported proportional mortality in all ages and both sexes which provides only an approximate picture of cause of death patterns.

Proportional mortality due to International Classification of Disease Chapter IX is far greater than for any other cause. Early data relating to 1960 were only available as proportional mortality >5 years7; even though this would lead to an over-estimate of all-ages proportional mortality from cardiovascular disease, the observed proportion (15-18%) was the lowest ever observed in Fiji. There were insufficient data available to examine disease-specific contributions within diseases of the circulatory system.The increase in proportional mortality from cardiovascular diseases, and levels of adult mortality in Fiji, are consistent with documented trends since the 1970s in both hospital admissions for cardiovascular disease, and with risk factor exposure for these conditions8, 83-87.

Figure 11: Proportional mortality from diseases of the circulatory system in Fiji: all ages, 1950-20101

Source data: 7-9, 11, 14-17, 20, 25, 26, 29, 35, 44, 51

Figure adapted from Carter et al1

0

10

20

30

40

50

60

70

80

1950 1960 1970 1980 1990 2000 2010

Prop

rotio

n of

dea

ths

(%)

Year

Diseases of the Circulatory System (all ages)

Diseases of the Circulatory System (age >5 years)




Figure 12: Proportional mortality from diseases of circulatory system (excluding ill-defined and unknown causes) in Fiji: all ages, 1950-2010

0

10

20

30

40

50

60

1950 1960 1970 1980 1990 2000

Pro

po

rtio

n o

f dea

ths

(%)

YearDiseases of the Circulatory System

Source data: 7-9, 11, 14-17, 20, 25, 26, 29, 35, 44, 51

Figure adapted from Carter et al1

Proportional mortality due to the infectious diseases showed no significant trend fluctuating between 5-12% of all deaths over the period investigated.

Figure 13: Proportional mortality from infectious and parasitic diseases (excluding ill-definded and unknown causes) in Fiji: all ages, 1950-2010

Source data: 7-9, 11, 14-17, 20, 25, 26, 29, 35, 44, 51




Respiratory diseases fell linearly from 14% to 7%, and injuries varied between 4-10% over this period, with a gradual decline in evidence since the mid 1980’s

Figure 14: Proportional mortality from diseases of the respiratory system (excluding ill-defined and unknown causes) in Fiji: all ages, 1950-2010

Source data: 7-9, 11, 14-17, 25, 26, 29, 35, 44, 51

There were no clear trends for any other International Classification of Diseases disease categories (except perinatal conditions); including neoplasms, endocrine and metabolic diseases or diseases of the digestive tract and genitourinary tract, particularly once data was adjusted to account for ill-defined and unknown causes (Appendix 6). Proportional mortality from these chronic disease categories together (20-30%), excluding diseases of the circulatory system, showed no obvious trend. Although only around 8% at the beginning of the 1970’s, proportional mortality from conditions arising in the perinatal period fell to around 2 %.

DISCUSSION

A difference in life expectancy at birth in 2000 of approximately 8 years is evident between various published estimates for Fiji. Many of the previously published estimates of life expectancy are implausible. Spurious increases in life expectancy imply that the health situation is improving when this is clearly unsupported by more reliable data. It is likely that many published estimates were produced through the use of single parameter models used to predict life tables (using infant or childhood mortality) which underestimate adult mortality in Pacific Island populations, and subsequently over-estimate life expectancy.

Extensive variation in the published estimates of infant mortality by source (Figure 1) was also noted, with several reporting levels that are below those of developed countries in the region such as Australia (IMR of 5.0 in 2007)67 and New Zealand (IMR of 6.0 in 2007)67 and are therefore highly improbable (Figure 1). A previous study found substantial uncertainty about mortality conditions in Pacific Island populations with variations of 10 years or more in life expectancy in the 1990s, depending on the source90. Principal issues include: under-enumerated vital registration data; annual stochastic fluctuations in mortality in small populations; errors in the imputation of adult mortality from infant and childhood rates; implausible results from indirect demographic methods; use of possibly inappropriate model life tables to adjust death




data, or for indirect methods; and inadequately described and implausible projections90.Reconciliation of mortality data in Fiji to reduce uncertainty is urgently needed to ensure that policy and program initiatives to address health issues can be appropriately evaluated.

The quality assessment in this study indicates that the measures provided by Fijian government sources based on indirect methods from censuses and the Ministry of Health death registration data provide the most reliable estimates of mortality and life expectancy at birth in Fiji. As such, these directly observed data should be used by decision makers as key tools in health planning rather than estimates based on theoretical models.

Sources remaining after censorship for quality and method demonstrate that the infant mortality rate has fallen steadily to below 20 deaths per 1000 live births. Life expectancy rose over the period of observation but has stabilized at around 64 years for males and 69 years for females since the 1980s. While the infant mortality rate has not fallen during the period of stagnation in life expectancy, it had fallen previously, and at the levels since the late 1980s, the infant mortality rate is a minor influence on the overall life expectancy71. Proportional mortality in Fiji from cardiovascular disease increased from just below 20% of deaths around 1960 to over 45% of deaths in 2001 (adjusted for unknown and ill defined causes).The proportional mortality from circulatory diseases is substantially higher than that for infectious diseases (as defined in International Classification of Diseases Chapter I) and respiratory illnesses (including infectious respiratory diseases, such as TB). While a similar stagnation in life expectancy may be driven by increasing mortality from HIV/AIDS, Fiji is estimated to have only 0.1% HIV prevalence in the adult population50, and a low mortality from AIDS82.

CONCLUSION

This assessment demonstrates the extensive variation in published estimates of mortality in Fiji; and the neet to evaluate published estimates cricically. Local data from the Ministry of Health routine reporting and analysis of census data were shown to be the most reliable sources, with many others proven implausible.

The stagnation in life expectancy, at relatively low levels of infant mortality, in concert with the epidemiological transition indicated by the levels and trends in cardiovascular disease mortality, morbidity and risk factors, suggests that an epi-demic of chronic non-communicable diseases is having a profound limiting effect on further mortality decline in Fiji. While further research is required, there is an urgent need for Fiji’s health services to place continued and increased emphasis on effective prevention and treatment strategies for cardiovascular disease.




REFERENCES

1. Carter KL, Cornelius M, Taylor R, Ali SS, Rao D, Lopez AD, Lewai V, Goundar R, Mowry C. Mortality Trends in Fiji. Australian and New Zealand Journal of Public Health [in press]

2. Taylor R, Lewis M, Powles J. The Australian Mortality Decline: all-cause mortality 1788 - 1990. Australian and New Zealand Journal of Public Health 1998;22(1): 27-36.

3. Taylor R, Lewis M, Powles J. Australian mortality decline: cause-specific mortality 1907-1990. Australian and New Zealand Journal of Public Health. 22(1):37-44. 1998

4. Thom T, Epstein F, Feldman J, Leaverton P, Wolz M. Total mortality and morbidity from heart disease, cancer and stroke from 1950 to 1987 in 27 countries. NIH publication 92-3088: Washington. National Heart, Lung and Blood Institute, National Institutes of Health; 1992.

5. Taylor R, Dobson A, Mirzaei M. Contribution of changes in risk factors to the decline of coronary heart disease mortality in Australia over three decades. European Journal of Cardiovascular Prevention and Rehabilitation 2006; 13(5): 760-8.

6. Tobias M, Taylor R, Yeh L, Huang K, Mann S, Sharpe N. Did it fall or was it pushed? The contribution of trends in established risk factors to the decline in premature coronary heart disease mortality in New Zealand. Australian and New Zealand Journal of Public Health 2008;32(2): 117.

7. Fiji Bureau of Statistics. Census 2007 Results: Population size, growth, structure and distribution. Statistical News. Suva: Fiji Bureau of Statistics; 2008

8. Tuomeilehto J, Ram P, Eseroma R, Taylor R, Zimmet P. Cardiovascular diseases and diabetes mellitus in Fiji: analysis of mortality, morbidity and risk factors. Bulletin of the World Health Organization 1984, 62(1): 133-143.

9. Savou SF. Magnitude of Cancer Problem in the South Pacific and Possible Aetiologic Factors (Fiji Islands Context) – Draft Document. Undated.

10. Biumaiwai M; Bavadra T; Olakowski T. Health Situation and Trend Analysis in Fiji Based on Hospital Morbidity and Mortality by Causes. World Health Organization, Regional Office for the Western Pacific; Undated.

11. Ministry of Health. The Nation’s Health. Health Information Unit, Ministry of Health. Suva; undated12. McArthur. Annual Report of the Registrar-General 1967. Fiji Bureau of Statistics. Suva; 196713. Zwart, F. Report on the Census of the Population 1976, Volume II: Demographic Characteristics. Suva; 1979.14. Ministry of Health. Annual Tabulation 1982. Government of the Republic of Fiji. Suva; 198215. Ministry of Health. Annual Reports for the years 1979 and 1980. Parliamentary Paper No. 49 of 1982. Parliament of

Fiji. Suva16. Ministry of Health. Annual Report for the year 1981. Parliamentary Paper No. 25 of 1983. Parliament of Fiji. Suva17. Biumaiwai M, Bavadra T, Olakowski T. Materials for WHO Intercountry Seminar on Strengthening of Epidemiological

Surveillance System and Utilisation of Existing Health Board in Pacific Island Countries. World Health Organization: 1984.

18. Taylor R, Henderson B, Levy S, Kolonel L, Lewis N. Table 13: Proportion of Mortality Due to Cancer in Pacific Island Countries. Cancer in Pacific Island Countries. Secretariat of the Pacific Community; New Caledonia; 1985

19. Fiji Bureau of Statistics. Social indicators for Fiji 1986: 5.20. Ministry of Health. Annual Reports for the year 1984. Parliament of Fiji;1986 21. Taylor R, Naroba V. Analysis of Mortality in the Population of Fiji 1981-83, and a Review of Previous Mortality Data.

South Pacific Commission; Noumea;1986. 22. Fiji Bureau of Statistics. Vital Statistics in Fiji (Statistics of Births, Deaths and Marriages) - A Report for the Year 1979-

80. Suva; 1988.23. Ministry of Health and Social Welfare. Table 4 - Five Major Causes of Hosptial Mortality 1980 – 1987. In: AIDAB. AIDAB

Pacific Multicounty Health Initiatives Program: Health Sector Desk Review - Draft Working Paper. Suva; 1989 24. Taylor R, Davis Lewis N, Levy S. Societies in Transition: Mortality Patterns in Pacific Island Populations. International

Journal of Epidemiology 1989. 18(3):634-46.25. Ministry of Health. Fiji Ministry of Health Annual Report 1988 and 1986. Paper No 7 1989. Parliament of Fiji. Suva26. Balkaran S, Taylor R, Naroba V. Mortality trends and differentials in Fiji (Chapter 5). In: Chandra R, Bryant J, editors.

Population of Fiji. Noumea: Secretariat of the Pacific Community; 1990.27. Ministry of Health. Table: 81. Ministry of Health Annual report for the years 1987 and 1988. Paper No 13 of 1991.

Parliament of Fiji. Suva28. Government of the Republic of Fiji. Opportunities for growth: policies and strategies for Fiji in the medium term; 1993.29. Ministry of Health. Appendix IX. Ministry of Health Annual Report 1990. Parliamentary Paper No 49 of 1993. Parliament

of Fiji. Suva




30. Department for Women and Culture. Women in Fiji: a statistical gender profile. Government of the Republic of Fiji;199431. Government of the Republic of Fiji. The Rights of the Child; 1995.32. World Bank. Fiji: restoring growth in a changing global environment. Report No. 13862-FIJ. World Bank. Country

Operations Division, East Asia and Pacific Region; 1995.33. Fiji Bureau of Statistics. Fiji Fertility Survey. Fiji Bureau of Statistics. Suva;199634. Ministry of Health. Figure 1993: 33. In: Chaimberlain K, Kamm D. A Situation Analysis of Children and Women in Fiji

1996. Government of the Republic of Fiji and UNICEF;199635. Ministry of Health. Annual Tabulation 1997. Government of the Republic of Fiji. Suva; 199736. UNDP.Table 9a - Expectation of life at specified ages for each sex: 1948-1997. United Nations: Demographic Yearbook,

Historical Supplement; 199737. Fiji Bureau of Statistics. 1986 Fiji Census of Population and Housing: Analytical Report, Part 1 Demographic

Characteristics. Suva; 199838. Fiji Bureau of Statistics. 1996 Fiji Census of Population and Housing: Analytical Report, Part 1 Demographic

Characteristics. Suva; 199839. SPC. Pacific Island Populations. – Revised edition. International Conference on Population and Development 1994:

Cairo, Egypt. Secretariat of the Pacific Community; New Caledonia. 199840. Rakeseta V. Fiji Islands population profile based on the 1996 census: a guide for planners and policy makers.

Secretariat of the Pacific Community. New Caledonia;199941. UNICEF. Fiji Mid-Term Review 1999. Undated42. Ministry of Information. Fiji Today. Government of the Republic of Fiji ; 200043. UNICEF. Country Statistics. 2001 cited [27/7/07]; Available from: http://www.unicef.org/statis.44. Ministry of Health. Ministry of Health Annual Report 2002. Paper No 44 of 2005. Parliament of Fiji. Suva45. Ministry of Health. Table 11. Ministry of Health Annual Report 2001. Paper No 43 of 2005. Parliament of Fiji. Suva46. WHO – WPRO. 2005. Country Health Information Profiles. WHO. Manilla.47. Fiji Bureau of Statistics. Tbl 2.2 Summary of Key Demographic Indicators for Fiji by Sex and Ethnicity from the

Population Census of 1986 and 1996. Key Statistics: March 2006 [cited 11/7/2006]. Available from: http://www.spc.int/prism/country/fj/stats/cens&surveys/Popu_census.htm

48. WHO. Table:Death and Daly esimates by Cause - From Burden of Disease and DALY estimates at who.int/entity/healthinfo/statistics/bodgddeathdalyestimates.xls in Western Pacific Regional Office. Country Health Information Profiles. WHO. Manilla. 2006

49. WHO. 2006. Mortality Country Profile 2006: Fiji [cited 27/7/07]. Available from: http://www.who.int/whosis/mort/profiles/mort_wpro_fji_fiji.pdf

50. WHO – WPRO. 2007. Country Health Information Profiles. WHO. Manilla.51. WHO. Number and Rates of Registered Deaths. WHOSIS (WHO Statistical Information System). [cited 30/7/07].

Available from: http://www.who.int/whosis/database/mort/table1_process.cfm 52. UNICEF. Statistical Tables: Table 1, Basic Indicators. The State of the World’s Children 2002: 12-15. [cited 30/7/07].

Available from: http://www.childinfo.org/mortality_infantmortality.php 53. UNDP. Population and Vital Statistics Report. Statistical Papers A LVIII(1). New York; 2006 [cited 7/7/08]. 54. Ministry of Health. [cited 30/7/08]. Available from: www.fijihealth.gov.fj. 55. UN Department of Economic and Social Affairs Population Division World Population Prospects: The 2008 revision.

United Nations. 2008. www.un.org/esa/population/publications/wpp2008/wpp2008_text_tables.pdf [cited 30/7/08];56. Western Pacific Regional Office. Country Health Information Profiles. WHO. Manilla. 2008.57. WHO – WPRO. 2005. Core Indicators 2005. WHO. Manilla. www.wpro.who.int/information_sources/databases/core-

indicators [cited 27/7/09]58. ESCAP 2008. ESCAP Statistical Yearbook for Asia and the Pacific; http://www.unescap.org/stat/data/syb2008 [cited

27/7/09]59. Asian Development Bank – Statistics Database – millennium Development Goals www://sdbs.adb.org/sdbs/index.jsp

[cited 27/7/09]60. Asian Development Bank. www.adb.org/Documents/Books/Key indicators/2008/Part II.asp#goal 4 [cited 27/7/09]61. SPC. PRISM Demographic Indicators. www.spc.int/prism/Social/demog/HealthInd.html [cited 27/7/09]62. SPC. MDG Indicies – Prism. www.spc.int/prism/MDG/mdg-p/htm [cited 27/7/09]63. World Bank (database).Key development Data and Statistics. http://web.worldbank.org/ [cited 27/7/09]64. Unicef Monitoring and statistics. Country Comparison. www.unicef.org/view_chart/php [cited 27/7/09]65. Unicef. 2009. Child Info database –Statistics by Area. www.childinfo.org [cited 27/7/09]




66. UNDP. Human Development Report 2007 Data. www.hdrstats.undp.org [cited 27/7/09]67. UNICEF 2009. Statistical Tables. The State of the World’s Children 2009. UN WHO. 2009. Mortality and Burden of

Disease in World Health Statistics. 68. ESCAP. Statistical indices. www.unescap.org/stat/data/statind/pdf.t4_dec04.pdf [cited 27/7/09]69. UNFPA. State of the World Population. Available from: http://www.unfpa.org/swp/index.html [cited 27/7/09]70. WHO. Basic Indicators for all WHO member states. Statistical Annex, Table 1. World Health Report. World Health

Organization; 200571. Rowland, DT. Demographic Methods and Concepts. Oxford University Press. New York; 2003. 72. SPC. Population Projections 1996-2005 (Unpublished data). Secretariat of the Pacific Community. New Caledonia;

200873. Australian Bureau of Statistics. Abridged Life Tables, 2002-2004. 2008 [cited 30/7/08]; Available from: www.abs.gov74. Australian Bureau of Statistics.. Chapter 4: Differentials in Mortality. Deaths: Australia 2004. 3302.0. [cited 30/10/08];

Available from: http://www.abs.gov.au/AUSSTATS/[email protected]/allprimarymainfeatures/F69F652EDC86B05CCA257235007865FB?opendocument

75. Statistics New Zealand. Tables: Deaths. Age-specific death rates by sex, total population, 1986–2007. 2008. [cited 30/10/08] Available from: http://www.stats.govt.nz/tables/deaths-tables.htm

76. Statistics New Zealand. Life tables 2002-2004. 2008 [cited 30/7/08]; Available from: www.stats.govt.nz/tables/abridged-period-life-tables.htm

77. Murray CJL, Ferguson BD, Lopez AD, Guillot M, Salomon JA, Ahmad O. Modified Logit Life Table System: Principles, Empirical Validation, and Application. Population Studies, 2003; 57(2): 165-182; Available from: http://www.jstor.org/stable/3595746

78. WHO. International Classification of Diseases. World Health Organization. Geneva; 200779. Janssen F, and Kunst A.E. ICD coding changes and discontinuities in trends in cause-specific mortality in six European

countries, 1950–99. Bulletin of the World Health Organization 2004; 82(12): 891-97080. Bennet S and Magnus P. Trends in cardiovascular risk factors in Australia. Results for the National Heart Foundation’s

Risk Factor Study, 1980-1989. Medical Journal of Australia 1994; 161(9): 519-27.81. Cameron AJ, Welborn TA, Zimmet PZ, Dunstan DW, Owen N, Salmon J. Overweight and obesity in Australia: the

1999–2000 Australian Diabetes, Obesity and Lifestyle Study (AusDiab). Medical Journal of Australia 2003; 178:427–432.

82. WHO, UNAids, Unicef. Epidemiological Fact Sheet on HIV and AIDS - Fiji. 2008 [cited 15/5/2010]; Available from: http://apps.who.int/globalatlas/predefinedReports/EFS2008/full/EFS2008_FJ.pdf

83. Pathik B, Ram P. Acute myocardial infarction in Fiji: a review of 300 cases. Medical Journal of Australia 1974;2: 922-4.84. Reed D, Feinleib M. Changing Patterns of Cardiovascular Disease in the Pacific Basin: Report of an International

Workshop. Journal of Community Health, Spring 1983; 8 (3). 182-205. 85. Ram P, Collins V, Zimmet P, Taylor R, King H, Sloman G, Hunt D. Cardiovascular disease risk factors in Fiji: the results

of the 1980 survey. Fiji Medical Journal1983; 11(7/8): 88-94.86. Taylor R, Zimmet P, Levy S, Collins V. Group comparisons of blood pressure and indices of obesity and salt intake in

Pacific populations. Medical Journal of Australia 1985, 142: 499-501.87. Taylor R, Badcock J, King H, Pargeter K, Zimmet P, Fred T, Lund M, Ringrose H, Bach F, Wang RL. Dietary intake

exercise, obesity and noncommunicable disease in rural and urban populations of three Pacific Island countries. Journal of the American College of Nutrition 1992. 11(3): 283-93.

88. Collins V, Dowse G, Cabealawa S, Ram P, and Zimmet P. High Mortality from Cardiovascular Disease and Analysis of Risk Factors in Indian and Melanesian Fijians. International Journal of Epidemiology 1996; 25(1): 59 – 69.

89. Taylor R and Lopez A. Differential Mortality among Pacific Island Countries and Territories. Asia-Pacific Population Journal 2007; 22(3): 45 – 58.

90. Taylor R, Brampton D, Lopez A. Contemporary patterns of Pacific Island mortality. International Journal of Epidemiology 2005 (34):207-214. www.unstats.un.org/unsd/demographic/products/vitstats/Sets/SeriesA_%20Jan2006_complete.pdf

91. Mathers C, Ma Fat D, Inouie M, Rao C, Lopez A. Counting the dead and what they died from: an assessment of the status of global cause of death data. Bulletin of the World Health Organization 2005;83(3): 171 - 7.




APPENDIX 1: Data sources and reliability for IMR and LE in Fiji, 1940-2008

SourceYears available

Censored from final estimates Comments and reason for exclusion

IMR LE IMR LE

Govt. - Registrar General

Vital Registration Data 11, 18, 39 46+, 56+, 86+, 96+

46, 56 Yes Yes Calculations were based on uncorrected vital registration data where no assessment of completeness of registration was provided.

Govt. – BoS BoS Census data 12, 36, 37, 39,46 57, 76, 86, 96

57, 66*, 73, 76*, 83, 86, 96

No No Estimates by ethnicity excluded in years where total was available. Total estimate for 1996 and 1976 calculated based on population proportions at the closest census.

Govt. – MoH MoH registration data adjusted for under-enumeration 20, 25

82 82* Partially No Brass adjustment 25 for IMR excluded as this method is not suitable for younger age groups.

LE from unpublished MoH data

Nil 99, 01, 03

NA No Present study

MoH Registration (Total) (9, 13,

14, 15, 19, 24, 26, 28, 33, 34, 42, 43, 53)76-81, 82*, 84, 90, 95-97, 00, 02-04, 06-08

Nil Partially NA 2008 IMR excluded as inconsistent with previous years. Estimates by ethnicity excluded in years where total was available. Total LE estimate for 1982 calculated based on population proportions at 86 census.

Govt. - other Fiji Coordinating Committee on Children 30

90, 93 90‡, 93‡ Yes Yes Source data not identified

Fiji Department for Women and Culture 29

Nil 86 NA Yes Source data not identified - appears to be an MoH estimate from an earlier period.

Fiji Ministry of Information 41 00, 00, Yes Yes

Fiji Fertility Survey (Total) 32 72 Nil No NA

Fiji Government (Total) 27 92 Nil Yes NA Original source not specified

SPC 38, 61, 62 94, 96, 97, 00, 02, 07

95, 97, 01

Yes Yes Incompatible estimates given by the same source for adjacent years

UN UNDP 35, 52, 54 48-88,91, 93-96, 98, 99, 05

53, 58, 63, 68, 73, 76, 83, 88, 93, 05

Yes Yes Source publication noted that the data were derived from estimation models that assume a given improvement in LE or IMR by year,

WHO - World Health Statistics

48, 50, 55, 90, 00, 06, 07

90, 00, 04, 06, 07, 08

Yes Yes Methods vary but not individually identified. These include census data and models derived from the WHO life table system.

WHO - WPRO 45, 49, 56 98, 99, 02, 05, 07, 08

02, 04, 05, 07, 08

Yes Yes Source/method described as local WHO office

UNFPA 68 Nil 02,03 NA Yes Source data not identified

UNESCAP 57, 67 00, 05, 06 00-08 Yes Yes Data showed a perfectly linear improvement by year when graphed indicating use of models that assume a given improvement in LE or IMR; IMR shows incompatible estimates given by the same source for adjacent years

UNICEF (total) 40, 42, 51, 63, 64, 66 97, 00, 01, 05, 07

Nil Yes Yes Estimates derived from other sources by committee consensus

World Bank 31, 62 85, 90, 00, 08

85‡, 90‡ Yes Yes Estimates not available by gender, method not specified

Asian Development Bank (ADB) 58, 59 80, 90, 98, 00, 07

80, 90, 95, 98, 99, 07,

Yes Yes Multiple incompatible estimates given for a single year ; estimates implausible given accepted levels of LE and IMR in other countries

+By gender and thenicity only, *By ethnicity only, ‡Both genders combined. NA: Not available, NB- Census data co-published by SPC with the Fiji

Bureau of Statistics listed under BoS only




APPENDIX 2: Published cause of death data available Fiji, 1960-2008

Data pro-vided

CategoriesAll deaths included

Years Method of collection / analysis References

Deaths ICDNo 77,78

From the register for Cancers (split by ethnicity of the two main groups)

8

Yes71, 75, 79-81, 85-88, 90, 97

Reported Deaths - MoH10, 13-16, 19, 24, 26, 28, 34

Proportion

Broad Disease Groups

No

60Cardiovascular disease only – method not stated

6

02 Burden of Disease estimates 47

80,88 Vital Statistics 21

ICDNo

74, 78-90, 94, 98-01

Deaths reported to the health statistics section

8, 10, 15, 20, 22, 28, 33 43

Yes 82, 94, 99, 02 Reported Deaths - MoH 8, 10, 45, 52

ICD (combined)

No 77, 85 Reported Deaths - MoH 15, 28

Rate ICD No 71-85Hospital discharge records coded to 17 categories based on ICD. Only categories accounting for >5% included.

9




APPENDIX 3: Brass analysis for mortality data completeness by gender, Fiji

Completeness (c) is estimated by 1/K where K= the slope of the line

a) 1999-2001 - Males

b) 1999-2001 - Females

K=0.74C=>100%

K=10..C=100%




c) 1996-1998 - Males

d) 1996-1998 - Females

K=0.77C=>100%

K=0.9156C=>100%




APPENDIX 4: Life tables calculated from Ministry of Health death reporting

Life table for Fiji 2002 - 2004, Males

Age Deaths Reported Population

Mortality Rate

Linearity Adjustment

Years in Interval

Probability of Dying

Individuals Surviving

Deaths in Interval x

Years Lived in Interval x

Cumulative Years Lived

Expectancy of Life at

Age x

x 5Dx 5Nx nMx a n nqx lx ndx nLx Tx Tx

<11-45-9

10-1415-1920-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-80

80+

518139

8997

156222202188283435655807928

1117955

1008694

1226

29309115190143632133136134768123364

94662793157938878910659895221140757295481971011784

60133742

0.01770.00120.00060.00070.00120.00180.00210.00240.00360.00550.00990.01550.02280.03780.04850.08550.11540.3276

0.10.30.50.50.50.50.50.50.50.50.50.50.50.50.50.50.50.5

14555555555555555

0.01740.00480.00310.00360.00580.00900.01060.01180.01770.02720.04840.07440.10770.17270.21610.35240.4479

1

1000009826097788974859713196570957059468993574919218942185091787607027658140455762951716298

1740473302354561865

1016111616532499433063318484

1213612564160591321916298

98434391718488182486539484252480688475986470658463736453355436281409626372589321040259291187734114537101244

6495892639745860057405517558503101945467674066079359009331194352655699220234417660631356436

983847662807403515215781101244

64.9665.1161.4256.6051.8047.0842.4937.9133.3428.8924.6320.7517.2214.0011.40

8.857.316.21

Life table for Fiji 2002-2004, Females


Mortality Rate


Years in Interval







Age x


<11-45-9

10-1415-1920-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-80

80+

447109

7681

117151146159207308385602662878741823595117

0

27617108255135047124737126544116172

88074752127807276747646635352542767324512254514483

8186

6574

0.01620.00100.00060.00060.00090.00130.00170.00210.00270.00400.00600.01120.01550.02710.03290.05680.0727

0.1780

0.10.30.50.50.50.50.50.50.50.50.50.50.50.50.50.50.5

0.5

14555555555555555

0.01600.00400.00280.00320.00460.00650.00830.01050.01320.01990.02930.05470.07450.12670.15190.24880.3075

1

10000098405980099773497417969689634095545945409329591441887598390477652678135751543207

29919

1595395275317449628795

1005124518532682485562529839

102981430913288

29919

98564392512489359487878485963483269479711475211469587461841450502431659403892363664313321251805182813

193751

69153036816738642422659348675446989496102644777573998046352283430532472591406214090417092451305353

941690628369376564

193751

69.1569.2765.5560.7255.9151.1646.4841.8437.2632.7328.3424.1220.3716.8113.8910.93

8.72

6.48




Life table for Fiji 1999- 2001, Males


Mortality Rate


Years in Interval





Cumulative Years lived


Age x


<11-45-9

10-1415-1920-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-80

80+

470196139100156213221239354482637898917997

1031948727

1269

29603117430138977134619137510114383

89752842368440176600597494888836811267321753410654

54683855

0.01590.00170.00100.00070.00110.00190.00250.00280.00420.00630.01070.01840.02490.03730.05880.08900.13290.3292

0.10.40.50.50.50.50.50.50.50.50.50.50.50.50.50.50.50.5

14555555555555555

0.01570.00670.00500.00370.00560.00930.01220.01410.02070.03100.05190.08780.11720.17060.25640.36410.4988

1

1000009843497779972929693396386954939432492994910668824383665763226737755880415512642413244

1566655486359547893

1169133019272823457873438945

1149714328151281318013244

98590392163487677485563483297479697474543468295460151448275429771399968359247308141243578169938

9916981078

6369140627055058783875390710490514744218493942152346761029993152539164209089016611181261151

901904593763350185180247

81078

63.6963.7060.1255.4150.6045.8841.2836.7632.2527.8823.6919.8516.5213.3910.63

8.436.826.12

Life table for Fiji 1999- 2001, Females


Mortality Rate


Years in Interval







Age x

x 5dx 5Nx nMx a n nqx lx ndx nLx Tx Tx

<11-45-9

10-1415-1920-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-80

80+

374163

8792

124156138158230276410544624758770726561

1170

27901110474130369126460129976108066

84175819728255773679595005015138338291941937012833

68866311

0.01340.00150.00070.00070.00100.00140.00160.00190.00280.00370.00690.01080.01630.02600.03980.05660.08150.1854

0.10.40.50.50.50.50.50.50.50.50.50.50.50.50.50.50.50.5

14555555555555555

0.01330.00590.00330.00360.00470.00720.00820.00960.01380.01850.03390.05280.07820.12200.18090.24790.3384

1

1000009867498094977699741296950962539546894552932439151588412837437719867784555254176027630

1326580325357462696786916

130917283103466965459415

12259137651413027630

98806393303489656487952485904483007479303475050469487461894449817430388402354362455308270243211173475177976

68723086773502638019958905425402590491668644336793954377347932730098402547946209812916677411265386

902932594662351451177976

68.7268.6565.0460.2555.4650.7146.0641.4236.8032.2827.8423.7319.9116.3913.3210.71

8.426.44




Life table for Fiji 1996-1998, Males


Mortality Rate


Years in Interval







Age x


<11-45-9

10-1415-1920-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-80

80+

518173117

92172187176218301409708717817852919738676986

30295116570134542141550131306105237

921269026985982693375646444314335972372715910

948754993872

0.01710.00150.00090.00060.00130.00180.00190.00240.00350.00590.01250.01620.02430.03590.05780.07780.12300.2546

0.10.40.50.50.50.50.50.50.50.50.50.50.50.50.50.50.50.5

14555555555555555

0.01690.00590.00430.00320.00650.00890.00950.01200.01730.02910.06080.07780.11470.16480.25240.32570.4704

1

1000009831597735973109699696364955109460093465918468917583755772406838357115427012879415251

1685580425315632854910

113516192671542165148858

1126714414139081354315251

98483391868487613485765483399479685475275470164463280452554432325402488364058313745249541178737110111

94298

6433389633490659430385455425496966044862604006575353130030611362597856214530217129771310489

946432632687383146204409

94298

64.3364.4360.8156.0651.2446.5641.9537.3332.7528.2824.0620.4516.9713.8411.08

8.977.106.18

Life table for Fiji 1996- 1998, Females


Mortality Rate


Years in Interval







Age x


<11-45-9

10-1415-1920-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-80

80+

438141

6672

104122128144178216493469576667644572532

1082

28584109571126125133682125031

9919288738885478254367417567704484035244248681748510576

67065722

0.01530.00130.00050.00050.00080.00120.00140.00160.00220.00320.00870.01050.01630.02680.03680.05410.07930.1892

0.10.40.50.50.50.50.50.50.50.50.50.50.50.50.50.50.50.5

14555555555555555

0.01510.00510.00260.00270.00420.00610.00720.00810.01070.01590.04250.05090.07840.12570.16870.23820.3308

1

1000009848797981977269746597060964659577094992939739248188547840367744367705562854287928694

1513506255261405596695778

101814923934451165929738

11420134061418528694

98639392735489269487979486314483812480585476903472412466135452568431456403698362871309976247910178932185298

69074926808853641611959268505438871495255744687443988159351125630388442572710212014116886851284988

922116612140364230185298

69.0769.1365.4860.6555.8051.0346.3341.6436.9632.3427.8223.9420.0916.5913.6210.88

8.496.46




APPENDIX 5: Empirical age-specific mortality compared with predictions from one and two parameter models for Fiji

1. 1999-2001, Males

2. 1999-2001, Females




3. 1996-1998, Males

4. 1996-1998, Females




APPENDIX 6:

Proportional mortality in Fiji 1970-2004: all ages (unadjusted for ill-defined or unknown causes)

Source data: 7-9, 11, 14-17, 20, 25, 26, 29, 35, 44, 51

1. Infectious and parasitic diseases - unadjusted for ill-defined and unknown cases

2. Diseases of the respiratory system - unadjusted for ill-defined and unknown causes




3. Neoplasms

a) Unadjusted for ill-defined and unknown causes

b) Adjusted for ill-defined and unknown causes




4. Endorcrine, nutritional and metabolic disorders






5. Diseases of the nervous and sensory systems






6. Diseases of the digestive system

a) unadjusted for ill-defined and unknown causes





7. Diseases of the genitourinary system






8. Conditions arising in the perinatal period






9. Congenital abnormalities






10. Injury and poisoning



KNOWLEDGE HUBS FOR HEALTHStrengthening health systems through evidence in Asia and the Pacific

A strategic partnerships initiative funded by the Australian Agency for International Development

The Nossal Institute for Global Health

lvii

Appendix 4: Additional cause of death distributions by age and sex

Palau Children 0-4 years

1999�2003

�103�ICD�code�

Female�deaths�

Male�deaths�

Total�deaths�

Rate�(deaths�per�100,000�pop)�

Upper�95%�CI�

Lower95%��CI�

1�048� 0 0� 0 0.0 0.0 55.2�1�058� 1 0� 1 0.4 0.0 83.4�1�064� 1 0� 1 0.4 0.0 83.4�1�072� 0 0� 0 0.0 0.0 55.2�1�078� 0 2� 2 3.6 0.0 108.1�1�084� 0 0� 0 0.0 0.0 55.2�1�092� 2 8� 10 71.7 0.0 275.1�1�093� 3 1� 4 9.3 0.0 131.2�1�094� 0 0� 0 0.0 0.0 55.2�1�095� 1 1� 2 3.6 0.0 108.1��

2004�2009�(excl.�2007)

�103�ICD�code�

Female�deaths�

Male�deaths�

Total�deaths�


Upper�95%�CI�

Lower95%��CI�

1�048� 0 1� 1 0.4 0.0� 80.8�1�058� 0 1� 1 0.4 0.0� 80.8�1�064� 1 0� 1 0.4 0.0� 80.8�1�072� 1 1� 2 3.5 0.0� 104.7�1�078� 0 0� 0 0.0 0.0� 53.5�1�084� 0 1� 1 0.4 0.0� 80.8�1�092� 5 11� 16 132.6 0.0� 376.6�1�093� 0 1� 1 0.4 0.0� 80.8�1�094� 0 1� 1 0.4 0.0� 80.8�1�095� 0 1� 1 0.4 0.0� 80.8�

lviii

5-14 years

1999�2003

ICD�103�list�code�

Female�deaths�

Male�deaths�

Total�deaths

Proportion�(%)�


Upper�95%�CI�

Lower95%��CI�

1�001� 1�005� 0� 0� 0 0 0 0� 22.462761�012� 0� 0� 0 0 0 0� 22.462761�013� 0� 0� 0 0 0 0� 22.462761�017� 0� 0� 0 0 0 0� 22.462761�019� 0� 0� 0 0 0 0� 22.462761�020� 0� 0� 0 0 0 0� 22.462761�025� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.92857

1�001�Total� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.928571�026� 1�027� 0� 0� 0 0 0 0� 22.46276

1�029� 0� 0� 0 0 0 0� 22.462761�030� 0� 0� 0 0 0 0� 22.462761�031� 0� 0� 0 0 0 0� 22.462761�032� 0� 0� 0 0 0 0� 22.462761�034� 0� 0� 0 0 0 0� 22.462761�036� 0� 0� 0 0 0 0� 22.462761�037� 0� 0� 0 0 0 0� 22.462761�038� 0� 0� 0 0 0 0� 22.462761�039� 0� 0� 0 0 0 0� 22.462761�040� 0� 0� 0 0 0 0� 22.462761�041� 0� 0� 0 0 0 0� 22.462761�042� 0� 0� 0 0 0 0� 22.462761�043� 0� 0� 0 0 0 0� 22.462761�044� 0� 0� 0 0 0 0� 22.462761�046� 0� 0� 0 0 0 0� 22.462761�047� 0� 0� 0 0 0 0� 22.46276

1�026�Total� 0� 0� 0 0 0 0� 22.462761�048� 1�049� 0� 0� 0 0 0 0� 22.462761�048�Total� 0� 0� 0 0 0 0� 22.462761�051� 1�052� 0� 0� 0 0 0 0� 22.46276

1�053� 0� 0� 0 0 0 0� 22.462761�054� 0� 0� 0 0 0 0� 22.46276

1�051�Total� 0� 0� 0 0 0 0� 22.462761�055� 1�056� 0� 0� 0 0 0 0� 22.46276

1�057� 0� 0� 0 0 0 0� 22.462761�055�Total� 0� 0� 0 0 0 0� 22.462761�058� 1�059� 0� 0� 0 0 0 0� 22.46276

1�061� 0� 0� 0 0 0 0� 22.462761�058�Total� 0� 0� 0 0 0 0� 22.46276

lix

1�064� 1�065� 0� 0� 0 0 0 0� 22.462761�066� 0� 0� 0 0 0 0� 22.462761�067� 0� 0� 0 0 0 0� 22.462761�068� 0� 0� 0 0 0 0� 22.462761�069� 0� 0� 0 0 0 0� 22.462761�071� 0� 0� 0 0 0 0� 22.46276

1�064�Total� 0� 0� 0 0 0 0� 22.462761�072� 1�074� 0� 0� 0 0 0 0� 22.46276

1�076� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.928571�077� 2� 0� 2 22.22222 0.000122 1.473567� 43.99388

1�072�Total� 2� 1� 3 33.33333 0.000183 3.769165� 53.383311�078� 1�080� 0� 0� 0 0 0 0� 22.46276

1�081� 1� 0� 1 11.11111 6.09E�05 0.152228� 33.928571�078�Total� 1� 0� 1 11.11111 6.09E�05 0.152228� 33.928571�082� 1�082� 0� 0� 0 0 0 0� 22.462761�082�Total� 0� 0� 0 0 0 0� 22.462761�083� 1�083� 0� 0� 0 0 0 0� 22.462761�083�Total� 0� 0� 0 0 0 0� 22.462761�084� 1�086� 0� 0� 0 0 0 0� 22.462761�084�Total� 0� 0� 0 0 0 0� 22.462761�092� 1�092� 0� 0� 0 0 0 0� 22.462761�092�Total� 0� 0� 0 0 0 0� 22.462761�093� 1�093� 0� 0� 0 0 0 0� 22.462761�093�Total� 0� 0� 0 0 0 0� 22.462761�094� 1�094� 0� 0� 0 0 0 0� 22.462761�094�Total� 0� 0� 0 0 0 0� 22.462761�095� 1�096� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.92857

1�097� 0� 0� 0 0 0 0� 22.462761�098� 0� 0� 0 0 0 0� 22.462761�099� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.928571�101� 0� 0� 0 0 0 0� 22.462761�102� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.928571�103� 0� 1� 1 11.11111 6.09E�05 0.152228� 33.92857

1�095�Total� 0� 4� 4 44.44444 0.000244 6.63714� 62.36476Grand�Total� 3� 6� 9 100 0.000548 25.05672� 104.0326

� �

lx

2004�2009

ICD 103 list code Female deaths

Male deaths

Total deaths

Proportion (%)

Rate (deaths per 100,000 pop)

Upper 95% CI

Lower95% CI

1-001 1-005 0 0 0 0 0 0 21.012151-012 0 0 0 0 0 0 21.012151-017 0 0 0 0 0 0 21.012151-019 0 0 0 0 0 0 21.012151-020 0 0 0 0 0 0 21.012151-025 0 0 0 0 0 0 21.01215

1-001 Total 0 0 0 0 0 0 21.012151-026 1-027 0 0 0 0 0 0 21.01215

1-028 0 0 0 0 0 0 21.012151-029 0 0 0 0 0 0 21.012151-030 0 0 0 0 0 0 21.012151-031 0 0 0 0 0 0 21.012151-032 0 0 0 0 0 0 21.012151-033 0 0 0 0 0 0 21.012151-034 0 0 0 0 0 0 21.012151-036 0 0 0 0 0 0 21.012151-037 0 0 0 0 0 0 21.012151-038 0 0 0 0 0 0 21.012151-039 0 0 0 0 0 0 21.012151-040 0 0 0 0 0 0 21.012151-042 1 0 1 16.66667 5.7E-05 0.142397 31.737521-043 0 0 0 0 0 0 21.012151-044 0 0 0 0 0 0 21.012151-045 0 0 0 0 0 0 21.012151-046 0 0 0 0 0 0 21.012151-047 0 0 0 0 0 0 21.01215

1-026 Total 1 0 1 16.66667 5.7E-05 0.142397 31.737521-048 1-049 0 0 0 0 0 0 21.01215

1-050 1 0 1 16.66667 5.7E-05 0.142397 31.737521-048 Total 1 0 1 16.66667 5.7E-05 0.142397 31.737521-051 1-052 0 0 0 0 0 0 21.01215

1-053 0 0 0 0 0 0 21.012151-054 0 0 0 0 0 0 21.01215

1-051 Total 0 0 0 0 0 0 21.012151-055 1-056 0 0 0 0 0 0 21.01215

1-057 0 0 0 0 0 0 21.012151-055 Total 0 0 0 0 0 0 21.012151-058 1-061 1 0 1 16.66667 5.7E-05 0.142397 31.737521-058 Total 1 0 1 16.66667 5.7E-05 0.142397 31.737521-064 1-065 1 0 1 16.66667 5.7E-05 0.142397 31.73752

1-066 0 0 0 0 0 0 21.01215

lxi

1-067 0 0 0 0 0 0 21.012151-068 0 0 0 0 0 0 21.012151-069 1 0 1 16.66667 5.7E-05 0.142397 31.737521-070 0 0 0 0 0 0 21.012151-071 0 0 0 0 0 0 21.01215

1-064 Total 2 0 2 33.33333 0.000114 1.378406 41.152841-072 1-074 0 0 0 0 0 0 21.01215

1-075 0 0 0 0 0 0 21.012151-076 0 0 0 0 0 0 21.012151-077 0 0 0 0 0 0 21.01215

1-072 Total 0 0 0 0 0 0 21.012151-078 1-080 0 0 0 0 0 0 21.01215

1-081 0 0 0 0 0 0 21.012151-078 Total 0 0 0 0 0 0 21.012151-082 1-082 0 0 0 0 0 0 21.012151-082 Total 0 0 0 0 0 0 21.012151-083 1-083 0 0 0 0 0 0 21.012151-083 Total 0 0 0 0 0 0 21.012151-084 1-085 0 0 0 0 0 0 21.01215

1-086 0 0 0 0 0 0 21.012151-084 Total 0 0 0 0 0 0 21.012151-092 1-092 0 0 0 0 0 0 21.012151-092 Total 0 0 0 0 0 0 21.012151-093 1-093 0 0 0 0 0 0 21.012151-093 Total 0 0 0 0 0 0 21.012151-094 1-094 0 0 0 0 0 0 21.012151-094 Total 0 0 0 0 0 0 21.012151-095 1-096 0 0 0 0 0 0 21.01215

1-098 0 0 0 0 0 0 21.012151-099 0 0 0 0 0 0 21.012151-100 0 0 0 0 0 0 21.012151-101 1 0 1 16.66667 5.7E-05 0.142397 31.737521-102 0 0 0 0 0 0 21.012151-103 0 0 0 0 0 0 21.01215

1-095 Total 1 0 1 16.66667 5.7E-05 0.142397 31.73752Grand Total 6 0 6 100 0.000342 12.54236 74.38268

lxii

15-6

4 ye

ars

1999�2003

Cause�(103�cod

e)�

Females�

Males�

Deaths�

Prop

ortio

nRa

te�(p

er�

100,000)�

Lower�

95%�CI�

Upp

er�

95%�CI�

Deaths�

Prop

ortio

n

Rate�

(per�

100,000)�

Lower�

95%�

CI�

Upp

er�

95%�CI�

1�001�

1�005�

1.0�

1.5

3.4

0.1

18.9�

0.0

0.0

0.0

0.0

9.7

1�012�

1.0�

1.5

3.4

0.1

18.9�

3.0

1.6

7.9

1.6

23.0

1�013�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�017�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�019�

3.0�

4.4

10.2

2.1

29.8�

4.0

2.1

10.5

2.9

26.9

1�020�

0.0�

0.0

0.0

0.0

12.5�

1.0

0.5

2.6

0.1

14.6

1�025�

1.0�

1.5

3.4

0.1

18.9�

2.0

1.0

5.2

0.6

18.9

1�001�Total�

6.0�

8.8

20.4

7.5

44.4�

12.0

6.2

31.5

16.3

55.0

1�026�

1�027�

3.0�

4.4

10.2

2.1

29.8�

3.0

1.6

7.9

1.6

23.0

1�029�

1.0�

1.5

3.4

0.1

18.9�

2.0

1.0

5.2

0.6

18.9

1�030�

1.0�

1.5

3.4

0.1

18.9�

1.0

0.5

2.6

0.1

14.6

1�031�

1.0�

1.5

3.4

0.1

18.9�

10.0

5.2

26.2

12.6

48.2

1�032�

0.0�

0.0

0.0

0.0

12.5�

3.0

1.6

7.9

1.6

23.0

1�034�

2.0�

2.9

6.8

0.8

24.5�

8.0

4.1

21.0

9.1

41.3

1�036�

2.0�

2.9

6.8

0.8

24.5�

0.0

0.0

0.0

0.0

9.7

1�037�

4.0�

5.9

13.6

3.7

34.8�

0.0

0.0

0.0

0.0

9.7

1�038�

1.0�

1.5

3.4

0.1

18.9�

0.0

0.0

0.0

0.0

9.7

1�039�

2.0�

2.9

6.8

0.8

24.5�

0.0

0.0

0.0

0.0

9.7

1�040�

0.0�

0.0

0.0

0.0

12.5�

1.0

0.5

2.6

0.1

14.6

lxiii

1�041�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�042�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�043�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�044�

1.0�

1.5

3.4

0.1

18.9�

0.0

0.0

0.0

0.0

9.7

1�046�

4.0�

5.9

13.6

3.7

34.8�

7.0

3.6

18.4

7.4

37.8

1�047�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�026�Total�

22.0�

32.3

74.7

46.8

113.1�

37.0

19.2

97.0

68.3

133.7

1�048�

1�049�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�048�Total�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�051�

1�052�

8.0�

11.7

27.2

11.7

53.5�

9.0

4.7

23.6

10.8

44.8

1�053�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�054�

1.0�

1.5

3.4

0.1

18.9�

3.0

1.6

7.9

1.6

23.0

1�051�Total�

9.0�

13.2

30.6

14.0

58.0�

12.0

6.2

31.5

16.3

55.0

1�055�

1�056�

0.0�

0.0

0.0

0.0

12.5�

4.0

2.1

10.5

2.9

26.9

1�057�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�055�Total�

0.0�

0.0

0.0

0.0

12.5�

4.0

2.1

10.5

2.9

26.9

1�058�

1�059�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�061�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�058�Total�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�064�

1�065�

3.0�

4.4

10.2

2.1

29.8�

0.0

0.0

0.0

0.0

9.7

1�066�

1.0�

1.5

3.4

0.1

18.9�

1.0

0.5

2.6

0.1

14.6

1�067�

1.0�

1.5

3.4

0.1

18.9�

13.0

6.7

34.1

18.1

58.3

1�068�

2.0�

2.9

6.8

0.8

24.5�

10.0

5.2

26.2

12.6

48.2

1�069�

5.1�

7.5

17.5

5.5

39.6�

11.0

5.7

28.8

14.4

51.6

1�071�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�064�Total�

12.1�

17.8

41.2

21.1

71.2�

35.0

18.1

91.8

63.9

127.6

1�072�

1�074�

0.0�

0.0

0.0

0.0

12.5�

3.0

1.6

7.9

1.6

23.0

lxiv

1�076�

1.0�

1.5

3.4

0.1

18.9�

6.0

3.1

15.7

5.8

34.2

1�077�

1.0�

1.5

3.4

0.1

18.9�

2.0

1.0

5.2

0.6

18.9

1�072�Total�

2.0�

2.9

6.8

0.8

24.5�

11.0

5.7

28.8

14.4

51.6

1�078�

1�080�

0.0�

0.0

0.0

0.0

12.5�

8.0

4.1

21.0

9.1

41.3

1�081�

1.0�

1.5

3.4

0.1

18.9�

1.0

0.5

2.6

0.1

14.6

1�078�Total�

1.0�

1.5

3.4

0.1

18.9�

9.0

4.7

23.6

10.8

44.8

1�082�

1�082�

0.0�

0.0

0.0

0.0

12.5�

1.0

0.5

2.6

0.1

14.6

1�082�Total�

0.0�

0.0

0.0

0.0

12.5�

1.0

0.5

2.6

0.1

14.6

1�083�

1�083�

2.0�

2.9

6.8

0.8

24.5�

1.0

0.5

2.6

0.1

14.6

1�083�Total�

2.0�

2.9

6.8

0.8

24.5�

1.0

0.5

2.6

0.1

14.6

1�084�

1�086�

3.0�

4.4

10.2

2.1

29.8�

4.0

2.1

10.5

2.9

26.9

1�084�Total�

3.0�

4.4

10.2

2.1

29.8�

4.0

2.1

10.5

2.9

26.9

1�092�

1�092�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�092�Total�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�093�

1�093�

0.0�

0.0

0.0

0.0

12.5�

1.2

0.6

3.1

0.1

14.6

1�093�Total�

0.0�

0.0

0.0

0.0

12.5�

1.2

0.6

3.1

0.1

14.6

1�094�

1�094�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�094�Total�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�095�

1�096�

4.0�

5.9

13.6

3.7

34.8�

16.0

8.3

41.9

24.0

68.1

1�097�

0.0�

0.0

0.0

0.0

12.5�

2.0

1.0

5.2

0.6

18.9

1�098�

2.0�

2.9

6.8

0.8

24.5�

6.0

3.1

15.7

5.8

34.2

1�099�

0.0�

0.0

0.0

0.0

12.5�

0.0

0.0

0.0

0.0

9.7

1�101�

2.0�

2.9

6.8

0.8

24.5�

15.0

7.8

39.3

22.0

64.9

1�102�

3.0�

4.4

10.2

2.1

29.8�

6.0

3.1

15.7

5.8

34.2

1�103�

0.0�

0.0

0.0

0.0

12.5�

17.0

8.8

44.6

26.0

71.4

1�095�Total�

11.0�

16.1

37.4

18.6

66.8�

62.0

32.1

162.5

124.6

208.4

Grand�Total�

68.1�

100.0

231.4

179.3

292.8�

193.2

100.0

506.4

213.3

318.9

lxv

2004�2009

Cau

se (1

03 c

ode)

Fem

ales

M

ales

Dea

ths

Pro

porti

on

Rat

e (p

er

100,

000)

Lo

wer

95

% C

I U

pper

95

% C

I D

eath

s P

ropo

rtion

R

ate

(per

10

0,00

0)

Low

er

95%

CI

Upp

er

95%

CI

1-00

1 1-

005

0.0

0.0

0.0

0.0

11.6

2.

00.

85.

10.

618

.4

1-01

2 2.

0 1.

66.

30.

822

.8

5.0

2.0

12.7

4.1

29.8

1-01

7 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

019

3.0

2.4

9.4

1.9

27.6

9.

03.

622

.910

.543

.6

1-02

0 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

025

1.0

0.8

3.1

0.1

17.5

2.

00.

85.

10.

618

.4

1-00

1 To

tal

6.0

4.8

18.9

6.9

41.1

20

.08.

051

.031

.278

.8

1-02

6 1-

027

0.0

0.0

0.0

0.0

11.6

1.

00.

42.

50.

114

.2

1-02

8 1.

0 0.

83.

10.

117

.5

1.0

0.4

2.5

0.1

14.2

1-

029

0.0

0.0

0.0

0.0

11.6

1.

00.

42.

50.

114

.2

1-03

0 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

031

3.0

2.4

9.4

1.9

27.6

17

.06.

843

.325

.369

.4

1-03

2 1.

0 0.

83.

10.

117

.5

1.0

0.4

2.5

0.1

14.2

1-

033

0.0

0.0

0.0

0.0

11.6

1.

00.

42.

50.

114

.2

1-03

4 3.

0 2.

49.

41.

927

.6

12.0

4.8

30.6

15.8

53.5

1-

036

5.0

4.0

15.7

5.1

36.7

0.

00.

00.

00.

09.

4 1-

037

4.0

3.2

12.6

3.4

32.3

0.

00.

00.

00.

09.

4 1-

038

1.0

0.8

3.1

0.1

17.5

1.

00.

42.

50.

114

.2

1-03

9 5.

0 4.

015

.75.

136

.7

0.0

0.0

0.0

0.0

9.4

1-04

0 1.

0 0.

83.

10.

117

.5

3.2

1.3

8.1

1.6

22.4

1-

042

0.0

0.0

0.0

0.0

11.6

0.

00.

00.

00.

09.

4 1-

043

1.0

0.8

3.1

0.1

17.5

1.

00.

42.

50.

114

.2

1-04

4 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

lxvi

1-04

5 1.

0 0.

83.

10.

117

.5

0.0

0.0

0.0

0.0

9.4

1-04

6 7.

0 5.

622

.08.

945

.4

8.0

3.2

20.4

8.8

40.2

1-

047

2.0

1.6

6.3

0.8

22.8

0.

00.

00.

00.

09.

4 1-

026

Tota

l 35

.0

28.2

110.

276

.815

3.3

49.2

19.6

125.

492

.416

5.2

1-04

8 1-

049

1.0

0.8

3.1

0.1

17.5

0.

00.

00.

00.

09.

4 1-

050

2.0

1.6

6.3

0.8

22.8

1.

00.

42.

50.

114

.2

1-04

8 To

tal

3.0

2.4

9.4

1.9

27.6

1.

00.

42.

50.

114

.2

1-05

1 1-

052

11.0

8.

934

.617

.362

.0

15.0

6.0

38.2

21.4

63.1

1-

053

0.0

0.0

0.0

0.0

11.6

0.

00.

00.

00.

09.

4 1-

054

3.0

2.4

9.4

1.9

27.6

4.

01.

610

.22.

826

.1

1-05

1 To

tal

14.0

11

.344

.124

.174

.0

19.0

7.6

48.4

29.2

75.7

1-

055

1-05

6 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

057

0.0

0.0

0.0

0.0

11.6

3.

01.

27.

61.

622

.4

1-05

5 To

tal

0.0

0.0

0.0

0.0

11.6

4.

01.

610

.22.

826

.1

1-05

8 1-

061

1.0

0.8

3.1

0.1

17.5

3.

01.

27.

61.

622

.4

1-05

8 To

tal

1.0

0.8

3.1

0.1

17.5

3.

01.

27.

61.

622

.4

1-06

4 1-

065

5.0

4.0

15.7

5.1

36.7

1.

00.

42.

50.

114

.2

1-06

6 3.

0 2.

49.

41.

927

.6

2.0

0.8

5.1

0.6

18.4

1-

067

7.0

5.6

22.0

8.9

45.4

27

.010

.768

.845

.410

0.2

1-06

8 5.

0 4.

015

.75.

136

.7

6.0

2.4

15.3

5.6

33.3

1-

069

9.0

7.3

28.3

13.0

53.8

14

.05.

635

.719

.559

.9

1-07

0 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

071

1.0

0.8

3.1

0.1

17.5

4.

01.

610

.22.

826

.1

1-06

4 To

tal

30.0

24

.294

.563

.713

4.9

55.0

21.9

140.

210

5.7

182.

5 1-

072

1-07

4 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

075

0.0

0.0

0.0

0.0

11.6

0.

00.

00.

00.

09.

4 1-

076

3.0

2.4

9.4

1.9

27.6

4.

01.

610

.22.

826

.1

1-07

7 1.

0 0.

83.

10.

117

.5

4.0

1.6

10.2

2.8

26.1

1-

072

Tota

l 4.

0 3.

212

.63.

432

.3

9.0

3.6

22.9

10.5

43.6

lxvii

1-07

8 1-

080

2.0

1.6

6.3

0.8

22.8

6.

02.

415

.35.

633

.3

1-08

1 2.

0 1.

66.

30.

822

.8

3.0

1.2

7.6

1.6

22.4

1-

078

Tota

l 4.

0 3.

212

.63.

432

.3

9.0

3.6

22.9

10.5

43.6

1-

082

1-08

2 0.

0 0.

00.

00.

011

.6

0.0

0.0

0.0

0.0

9.4

1-08

2 To

tal

0.0

0.0

0.0

0.0

11.6

0.

00.

00.

00.

09.

4 1-

083

1-08

3 4.

0 3.

212

.63.

432

.3

1.0

0.4

2.5

0.1

14.2

1-

083

Tota

l 4.

0 3.

212

.63.

432

.3

1.0

0.4

2.5

0.1

14.2

1-

084

1-08

5 2.

0 1.

66.

30.

822

.8

1.0

0.4

2.5

0.1

14.2

1-

086

8.0

6.5

25.2

10.9

49.6

11

.04.

428

.014

.050

.2

1-08

4 To

tal

10.0

8.

131

.515

.157

.9

12.0

4.8

30.6

15.8

53.5

1-

092

1-09

2 0.

0 0.

00.

00.

011

.6

0.0

0.0

0.0

0.0

9.4

1-09

2 To

tal

0.0

0.0

0.0

0.0

11.6

0.

00.

00.

00.

09.

4 1-

093

1-09

3 0.

0 0.

00.

00.

011

.6

0.0

0.0

0.0

0.0

9.4

1-09

3 To

tal

0.0

0.0

0.0

0.0

11.6

0.

00.

00.

00.

09.

4 1-

094

1-09

4 1.

0 0.

83.

10.

117

.5

2.0

0.8

5.1

0.6

18.4

1-

094

Tota

l 1.

0 0.

83.

10.

117

.5

2.0

0.8

5.1

0.6

18.4

1-09

5 1-

096

Tran

spor

t ac

cide

nts

2.0

1.6

6.3

0.8

22.8

9.

03.

622

.910

.543

.6

1-09

8 Fa

lls

0.0

0.0

0.0

0.0

11.6

3.

01.

27.

61.

622

.4

1-09

9 D

row

ning

0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

100

Acc

iden

tal

pois

onin

g 0.

0 0.

00.

00.

011

.6

1.0

0.4

2.5

0.1

14.2

1-

101

Inte

ntio

nal

Sel

f Har

m

4.0

3.2

12.6

3.4

32.3

9.

03.

622

.910

.543

.6

1-10

2 A

ssau

lt 1.

0 0.

83.

10.

117

.5

5.0

2.0

12.7

4.1

29.8

1-

103

Oth

er

exte

rnal

cau

ses

5.0

4.0

15.7

5.1

36.7

39

.015

.599

.470

.713

5.9

1-09

5 To

tal

12.0

9.

737

.819

.566

.0

67.0

26.7

170.

813

2.4

217.

0 G

rand

Tot

al

124.

0 10

0.0

390.

525

6.2

383.

0 25

1.2

100.

064

0.4

207.

531

0.1

lxviii

65+

year

s

1999�2003

Fem

ales

M

ales

Dea

ths

Pro

porti

on

Rat

e (p

er

100,

000)

Lo

wer

95

% C

I U

pper

95

% C

I D

eath

s P

ropo

rtion

R

ate

(per

10

0,00

0)

Low

er

95%

CI

Upp

er

95%

CI

1-00

1 1-

005

0 0

00

120.

4028

00

00

166.

2474

1-

012

3 1.

8087

4297

.914

9520

.203

1228

6.14

015

3.67

6471

225.

3285

73.1

4163

525.

8266

1-

013

0 0

00

120.

4028

10.

7352

9445

.065

71.

1266

4325

1.10

61

1-01

7 0

00

012

0.40

281

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

019

1 0.

6029

1432

.638

320.

8159

5818

1.86

070

00

016

6.24

74

1-02

0 0

00

012

0.40

280

00

016

6.24

74

1-02

5 0

00

012

0.40

280

00

016

6.24

74

1-00

1 To

tal

4 2.

4116

5613

0.55

3335

.575

7733

4.28

177

5.14

7059

315.

4599

126.

8149

649.

9826

1-

026

1-02

7 0

00

012

0.40

280

00

016

6.24

74

1-02

9 0

00

012

0.40

280

00

016

6.24

74

1-03

0 0

00

012

0.40

281

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

031

0 0

00

120.

4028

10.

7352

9445

.065

71.

1266

4325

1.10

61

1-03

2 0

00

012

0.40

282

1.47

0588

90.1

314

10.9

059

325.

5997

1-

034

9 5.

4262

2729

3.74

4913

4.30

6755

7.62

576

4.41

1765

270.

3942

99.2

3467

588.

513

1-03

6 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

00

00

166.

2474

1-

037

1 0.

6029

1432

.638

320.

8159

5818

1.86

070

00

016

6.24

74

1-03

8 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

00

00

166.

2474

1-

039

1 0.

6029

1432

.638

320.

8159

5818

1.86

070

00

016

6.24

74

1-04

0 0

00

012

0.40

284

2.94

1176

180.

2628

49.1

2161

461.

5629

1-

041

1 0.

6029

1432

.638

320.

8159

5818

1.86

070

00

016

6.24

74

1-04

2 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

00

00

166.

2474

1-

043

0 0

00

120.

4028

21.

4705

8890

.131

410

.905

932

5.59

97

lxix

1-04

4 0

00

012

0.40

280

00

016

6.24

74

1-04

6 11

6.

6320

5535

9.02

1517

9.21

764

2.38

7411

8.08

8235

495.

7227

247.

4558

886.

9831

1-

047

2 1.

2058

2865

.276

647.

8984

7323

5.81

180

00

016

6.24

74

1-02

6 To

tal

28

16.8

8159

913.

8729

607.

2685

1320

.807

2719

.852

9412

16.7

7480

1.85

417

70.3

61

1-04

8 1-

049

3 1.

8087

4297

.914

9520

.203

1228

6.14

010

00

016

6.24

74

1-04

8 To

tal

3 1.

8087

4297

.914

9520

.203

1228

6.14

010

00

016

6.24

74

1-05

1 1-

052

11

6.63

2055

359.

0215

179.

217

642.

3874

1611

.764

7172

1.05

1241

2.12

5811

70.9

42

1-05

3 0

00

012

0.40

282

1.47

0588

90.1

314

10.9

059

325.

5997

1-

054

3 1.

8087

4297

.914

9520

.203

1228

6.14

011

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

051

Tota

l 14

8.

4407

9745

6.93

6424

9.81

3776

6.67

4119

13.9

7059

856.

2483

515.

5066

1337

.144

1-

055

1-05

6 0

00

012

0.40

280

00

016

6.24

74

1-05

7 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

00

00

166.

2474

1-

055

Tota

l 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

00

00

166.

2474

1-

058

1-05

9 0

00

012

0.40

280

00

016

6.24

74

1-06

1 2

1.20

5828

65.2

7664

7.89

8473

235.

8118

00

00

166.

2474

1-

058

Tota

l 2

1.20

5828

65.2

7664

7.89

8473

235.

8118

00

00

166.

2474

1-

064

1-06

5 4

2.41

1656

130.

5533

35.5

7577

334.

2817

00

00

166.

2474

1-

066

5 3.

0145

716

3.19

1652

.971

9938

0.82

392

1.47

0588

90.1

314

10.9

059

325.

5997

1-

067

16

9.64

6625

522.

2131

298.

4774

848.

0414

2216

.176

4799

1.44

5462

1.32

0815

01.0

48

1-06

8 11

6.

6320

5535

9.02

1517

9.21

764

2.38

7411

8.08

8235

495.

7227

247.

4558

886.

9831

1-

069

31.8

6111

19

.209

5110

39.8

9368

7.46

0914

36.1

5114

10.2

9412

630.

9198

344.

9329

1058

.593

1-

071

1 0.

6029

1432

.638

320.

8159

5818

1.86

071

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

064

Tota

l 68

.861

11

41.5

1733

2247

.511

1723

.466

2813

.619

5036

.764

7122

53.2

8516

72.4

3329

70.6

86

1-07

2 1-

074

3 1.

8087

4297

.914

9520

.203

1228

6.14

011

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

076

10

6.02

9141

326.

3832

156.

5007

600.

2187

75.

1470

5931

5.45

9912

6.81

4964

9.98

26

1-07

7 2

1.20

5828

65.2

7664

7.89

8473

235.

8118

32.

2058

8213

5.19

7127

.895

6739

5.09

1 1-

072

Tota

l 15

9.

0437

1148

9.57

4827

3.99

8780

7.47

211

8.08

8235

495.

7227

247.

4558

886.

9831

1-

078

1-08

0 0

00

012

0.40

284

2.94

1176

180.

2628

49.1

2161

461.

5629

1-

081

3 1.

8087

4297

.914

9520

.203

1228

6.14

013

2.20

5882

135.

1971

27.8

9567

395.

091

lxx

1-07

8 To

tal

3 1.

8087

4297

.914

9520

.203

1228

6.14

017

5.14

7059

315.

4599

126.

8149

649.

9826

1-

082

1-08

2 2

1.20

5828

65.2

7664

7.89

8473

235.

8118

10.

7352

9445

.065

71.

1266

4325

1.10

61

1-08

2 To

tal

2 1.

2058

2865

.276

647.

8984

7323

5.81

181

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

083

1-08

3 0

00

012

0.40

281

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

083

Tota

l 0

00

012

0.40

281

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

084

1-08

6 9

5.42

6227

293.

7449

134.

3067

557.

6257

53.

6764

7122

5.32

8573

.141

6352

5.82

66

1-08

4 To

tal

9 5.

4262

2729

3.74

4913

4.30

6755

7.62

575

3.67

6471

225.

3285

73.1

4163

525.

8266

1-

092

1-09

2 0

00

012

0.40

280

00

016

6.24

74

1-09

2 To

tal

0 0

00

120.

4028

00

00

166.

2474

1-

093

1-09

3 0

00

012

0.40

280

00

016

6.24

74

1-09

3 To

tal

0 0

00

120.

4028

00

00

166.

2474

1-

094

1-09

4 15

9.

0437

1148

9.57

4827

3.99

8780

7.47

26

4.41

1765

270.

3942

99.2

3467

588.

513

1-09

4 To

tal

15

9.04

3711

489.

5748

273.

9987

807.

472

64.

4117

6527

0.39

4299

.234

6758

8.51

3 1-

095

1-09

6 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

00

00

166.

2474

1-

097

0 0

00

120.

4028

10.

7352

9445

.065

71.

1266

4325

1.10

61

1-09

8 0

00

012

0.40

281

0.73

5294

45.0

657

1.12

6643

251.

1061

1-

099

0 0

00

120.

4028

00

00

166.

2474

1-

101

0 0

00

120.

4028

00

00

166.

2474

1-

102

0 0

00

120.

4028

00

00

166.

2474

1-

103

0 0

00

120.

4028

00

00

166.

2474

1-

095

Tota

l 1

0.60

2914

32.6

3832

0.81

5958

181.

8607

21.

4705

8890

.131

410

.905

932

5.59

97

Gra

nd T

otal

16

5.86

11

100

5413

.428

2655

.584

3969

.701

136

100

6128

.935

3666

.726

5481

.206

��

lxxi

2004�2009

Cau

se (1

03 li

st)

Fem

ales

M

ales

Dea

ths

Pro

porti

on

Rat

e (p

er

100,

000)

Lo

wer

95

% C

I U

pper

95

% C

I D

eath

s P

ropo

rtion

R

ate

(per

10

0,00

0)

Low

er

95%

CI

Upp

er

95%

CI

1-00

1 1-

005

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-01

2 13

7.

1038

2537

1.16

4919

7.63

1163

4.69

216

3.51

2195

255.

4556

93.7

5221

555.

9991

1-

017

0 0

00

105.

3252

10.

5853

6642

.575

931.

0643

9823

7.23

31

1-01

9 0

00

010

5.32

521

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

020

0 0

00

105.

3252

00

00

157.

0626

1-

025

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-00

1 To

tal

15

8.19

6721

428.

2672

239.

6869

706.

3554

84.

6829

2734

0.60

7514

7.05

7367

1.12

44

1-02

6 1-

027

2 1.

0928

9657

.102

36.

9093

7820

6.28

211

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

028

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-02

9 1

0.54

6448

28.5

5115

0.71

3779

159.

087

21.

1707

3285

.151

8710

.303

3830

7.61

11

1-03

0 1

0.54

6448

28.5

5115

0.71

3779

159.

087

21.

1707

3285

.151

8710

.303

3830

7.61

11

1-03

1 2

1.09

2896

57.1

023

6.90

9378

206.

2821

42.

3414

6317

0.30

3746

.407

7743

6.06

27

1-03

2 2

1.09

2896

57.1

023

6.90

9378

206.

2821

00

00

157.

0626

1-

033

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-03

4 3

1.63

9344

85.6

5345

17.6

7316

250.

3079

116.

4390

2446

8.33

5323

3.78

4583

7.97

95

1-03

6 4

2.18

5792

114.

2046

31.1

2075

292.

4209

00

00

157.

0626

1-

037

0 0

00

105.

3252

00

00

157.

0626

1-

038

2 1.

0928

9657

.102

36.

9093

7820

6.28

210

00

015

7.06

26

1-03

9 3

1.63

9344

85.6

5345

17.6

7316

250.

3079

00

00

157.

0626

1-

040

0 0

00

105.

3252

15.8

3333

9.26

8293

674.

1189

357.

425

1053

.329

1-

042

0 0

00

105.

3252

00

00

157.

0626

1-

043

1 0.

5464

4828

.551

150.

7137

7915

9.08

72

1.17

0732

85.1

5187

10.3

0338

307.

6111

1-

044

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-04

5 0

00

010

5.32

520

00

015

7.06

26

lxxii

1-04

6 11

6.

0109

2931

4.06

2615

6.77

4456

1.94

377

4.09

7561

298.

0315

119.

8087

614.

0727

1-

047

2 1.

0928

9657

.102

36.

9093

7820

6.28

210

00

015

7.06

26

1-02

6 To

tal

37

20.2

1858

1056

.393

743.

786

1456

.109

44.8

3333

26.2

439

1908

.821

1361

.153

2514

.875

1-

048

1-04

9 3

1.63

9344

85.6

5345

17.6

7316

250.

3079

00

00

157.

0626

1-

050

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-04

8 To

tal

4 2.

1857

9211

4.20

4631

.120

7529

2.42

090

00

015

7.06

26

1-05

1 1-

052

21

11.4

7541

599.

5741

371.

1364

916.

5205

179.

9512

272

3.79

0942

1.62

9511

58.8

74

1-05

3 1

0.54

6448

28.5

5115

0.71

3779

159.

087

00

00

157.

0626

1-

054

6 3.

2786

8917

1.30

6962

.869

6337

2.84

951

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

051

Tota

l 28

15

.300

5579

9.43

2253

1.22

2711

55.4

0818

10.5

3659

766.

3668

454.

2001

1211

.2

1-05

5 1-

056

1 0.

5464

4828

.551

150.

7137

7915

9.08

71

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

057

0 0

00

105.

3252

00

00

157.

0626

1-

055

Tota

l 1

0.54

6448

28.5

5115

0.71

3779

159.

087

10.

5853

6642

.575

931.

0643

9823

7.23

31

1-05

8 1-

061

1 0.

5464

4828

.551

150.

7137

7915

9.08

73

1.75

6098

127.

7278

26.3

545

373.

2632

1-

058

Tota

l 1

0.54

6448

28.5

5115

0.71

3779

159.

087

31.

7560

9812

7.72

7826

.354

537

3.26

32

1-06

4 1-

065

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-06

6 9

4.91

8033

256.

9603

117.

488

487.

7964

42.

3414

6317

0.30

3746

.407

7743

6.06

27

1-06

7 13

7.

1038

2537

1.16

4919

7.63

1163

4.69

2125

14.6

3415

1064

.398

688.

836

1571

.265

1-

068

6 3.

2786

8917

1.30

6962

.869

6337

2.84

957

4.09

7561

298.

0315

119.

8087

614.

0727

1-

069

18

9.83

6066

513.

9207

304.

5837

812.

2231

1911

.121

9580

8.94

2748

7.02

6112

63.2

71

1-07

0 0

00

010

5.32

521

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

071

3 1.

6393

4485

.653

4517

.673

1625

0.30

791

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

064

Tota

l 50

27

.322

414

27.5

5710

59.5

6218

82.0

6357

33.3

6585

2426

.828

1838

.046

3144

.233

1-

072

1-07

4 8

4.37

1585

228.

4092

98.6

1567

450.

0518

10.

5853

6642

.575

931.

0643

9823

7.23

31

1-07

5 0

00

010

5.32

520

00

015

7.06

26

1-07

6 6

3.27

8689

171.

3069

62.8

6963

372.

8495

127.

0243

951

0.91

1226

4.01

3489

2.47

67

1-07

7 4

2.18

5792

114.

2046

31.1

2075

292.

4209

31.

7560

9812

7.72

7826

.354

537

3.26

32

1-07

2 To

tal

18

9.83

6066

513.

9207

304.

5837

812.

2231

169.

3658

5468

1.21

4938

9.35

6911

06.2

5 1-

078

1-08

0 0.

5 0.

2732

2414

.275

570

105.

3252

1.5

0.87

8049

63.8

639

1.06

4398

237.

2331

lxxiii

1-08

1 2

1.09

2896

57.1

023

6.90

9378

206.

2821

21.

1707

3285

.151

8710

.303

3830

7.61

11

1-07

8 To

tal

2.5

1.36

612

71.3

7787

6.90

9378

206.

2821

3.5

2.04

878

149.

0158

26.3

545

373.

2632

1-

082

1-08

2 1

0.54

6448

28.5

5115

0.71

3779

159.

087

10.

5853

6642

.575

931.

0643

9823

7.23

31

1-08

2 To

tal

1 0.

5464

4828

.551

150.

7137

7915

9.08

71

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

083

1-08

3 0

00

010

5.32

520

00

015

7.06

26

1-08

3 To

tal

0 0

00

105.

3252

00

00

157.

0626

1-

084

1-08

5 0

00

010

5.32

520

00

015

7.06

26

1-08

6 9

4.91

8033

256.

9603

117.

488

487.

7964

31.

7560

9812

7.72

7826

.354

537

3.26

32

1-08

4 To

tal

9 4.

9180

3325

6.96

0311

7.48

848

7.79

643

1.75

6098

127.

7278

26.3

545

373.

2632

1-

092

1-09

2 0

00

010

5.32

520

00

015

7.06

26

1-09

2 To

tal

0 0

00

105.

3252

00

00

157.

0626

1-

093

1-09

3 0

00

010

5.32

520

00

015

7.06

26

1-09

3 To

tal

0 0

00

105.

3252

00

00

157.

0626

1-

094

1-09

4 8.

5 4.

6448

0924

2.68

4898

.615

6745

0.05

188.

54.

9756

136

1.89

5414

7.05

7367

1.12

44

1-09

4 To

tal

8.5

4.64

4809

242.

6848

98.6

1567

450.

0518

8.5

4.97

561

361.

8954

147.

0573

671.

1244

1-

095

1-09

6 0

00

010

5.32

521

0.58

5366

42.5

7593

1.06

4398

237.

2331

1-

098

0 0

00

105.

3252

00

00

157.

0626

1-

099

0 0

00

105.

3252

00

00

157.

0626

1-

100

0 0

00

105.

3252

00

00

157.

0626

1-

101

0 0

00

105.

3252

00

00

157.

0626

1-

102

1 0.

5464

4828

.551

150.

7137

7915

9.08

70

00

015

7.06

26

1-10

3 7

3.82

5137

199.

858

80.3

4294

411.

7932

63.

5121

9525

5.45

5693

.752

2155

5.99

91

1-09

5 To

tal

8 4.

3715

8522

8.40

9298

.615

6745

0.05

187

4.09

7561

298.

0315

119.

8087

614.

0727

G

rand

Tot

al

183

100

5224

.86

2323

.036

3472

.591

170.

8333

100

7273

.389

3464

.148

5178

.383

POP

3502

.486

23

48.7

45

lxxiv

Ton

gaA

ge s

tand

ardi

sed

rate

s (s

tand

ardi

sed

to T

onga

pop

ulat

ion

2005

-200

8)

MA

LES

20

01-2

004

2005

-200

8

N

o ad

just

men

t for

co

ntrib

utor

y ca

uses

A

djus

ted

for e

ffect

of

cont

ribut

ory

caus

es

No

adju

stm

ent f

or

cont

ribut

ory

caus

es

Adj

uste

d fo

r effe

ct o

f co

ntrib

utor

y ca

uses

Lo

w

Mor

talit

y E

stim

ates

Hig

h M

orta

lity

Est

imat

es

Low

M

orta

lity

Est

imat

es

Hig

h M

orta

lity

Est

imat

es

Low

M

orta

lity

Est

imat

es

Hig

h M

orta

lity

Est

imat

es

Low

M

orta

lity

Est

imat

es

Hig

h M

orta

lity

Est

imat

es

1-00

1 C

erta

in in

fect

ious

and

pa

rasi

tic d

isea

ses

20.3

40.0

20.3

40.0

61.1

81.4

63.7

84.9

1-01

9 V

iral h

epat

itis

4.0

8.0

4.0

8.0

18.4

24.5

18.5

24.7

1-

026

Neo

plas

ms

66.1

130.

166

.113

0.1

191.

125

4.7

189.

425

2.4

1-02

9 M

alig

nant

neo

plas

m o

f sto

mac

h 5.

210

.15.

210

.115

.420

.515

.620

.8

1-03

1 M

alig

nant

neo

plas

m o

f liv

er a

nd

intra

hepa

tic b

ile d

ucts

9.

618

.99.

618

.919

.225

.517

.923

.9

1-03

4 M

alig

nant

neo

plas

m o

f tra

chea

, br

onch

us a

nd lu

ng

12.5

24.6

12.5

24.6

36.7

49.0

34.4

45.8

1-04

0 M

alig

nant

neo

plas

m o

f pro

stat

e 10

.420

.310

.420

.326

.034

.626

.935

.9

1-05

2 D

iabe

tes

mel

litus

26

.051

.325

.550

.216

6.5

222.

094

.212

5.6

1-05

8 D

isea

ses

of th

e ne

rvou

s sy

stem

5.

210

.05.

210

.013

.517

.913

.517

.9

1-06

4 D

isea

ses

of th

e ci

rcul

ator

y sy

stem

19

3.9

380.

519

4.4

381.

542

3.0

563.

948

3.2

644.

2

1-06

5 A

cute

rheu

mat

ic fe

ver a

nd

chro

nic

rheu

mat

ic h

eart

dise

ases

0.

20.

50.

20.

58.

711

.68.

711

.6

1-06

6 H

yper

tens

ive

dise

ases

11

.522

.412

.023

.525

.033

.427

.136

.1

1-06

7 Is

chae

mic

hea

rt di

seas

es

44.8

88.4

44.8

88.4

139.

418

5.8

157.

721

0.2

1-06

8 O

ther

hea

rt di

seas

es

119.

923

4.9

119.

923

4.9

203.

427

1.1

232.

631

0.1

1-06

9 C

ereb

rova

scul

ar d

isea

ses

17.5

34.2

17.5

34.2

40.4

53.9

49.8

66.4

lxxv

1-

072

Dis

ease

s of

the

resp

irato

ry

syst

em

55.0

107.

655

.010

7.6

168.

422

4.5

173.

323

0.9

1-07

4 P

neum

onia

14

.327

.814

.327

.814

.819

.715

.120

.2

1-07

6 C

hron

ic lo

wer

resp

irato

ry

dise

ases

25

.649

.925

.649

.993

.512

4.6

92.7

123.

5

1-07

8 D

isea

ses

of th

e di

gest

ive

syst

em

22.8

45.2

22.8

45.2

46.6

62.1

47.2

63.0

1-08

0 D

isea

ses

of th

e liv

er

8.6

17.2

8.6

17.2

17.3

23.1

17.3

23.1

1-

084

Dis

ease

s of

the

geni

tour

inar

y sy

stem

14

.328

.414

.328

.425

.133

.527

.536

.6

1-09

2 C

erta

in c

ondi

tions

orig

inat

ing

in th

e pe

rinat

al p

erio

d 1.

83.

41.

83.

411

.314

.911

.314

.9

1-09

5 Ex

tern

al c

ause

s of

mor

bidi

ty

and

mor

talit

y 29

.558

.329

.558

.368

.290

.868

.891

.6

1-09

6 Tr

ansp

ort a

ccid

ents

6.

813

.66.

813

.613

.117

.513

.117

.5

1-10

1 In

tent

iona

l sel

f-har

m

1.6

3.1

1.6

3.1

8.6

11.5

8.7

11.6

lxxvi

FEM

ALE

S 20

01-2

004

2005

-200

8

N

o ad

just

men

t for

co

ntrib

utor

y ca

uses

A

djus

ted

for e

ffect

of

cont

ribut

ory

caus

es

No

adju

stm

ent f

or

cont

ribut

ory

caus

es

Adj

uste

d fo

r effe

ct o

f co

ntrib

utor

y ca

uses

Lo

w

Mor

talit

y E

stim

ates

Hig

h M

orta

lity

Est

imat

es

Low

M

orta

lity

Est

imat

es

Hig

h M

orta

lity

Est

imat

es

Low

M

orta

lity

Est

imat

es

Hig

h M

orta

lity

Est

imat

es

Low

M

orta

lity

Est

imat

es

Hig

h M

orta

lity

Est

imat

es

1-00

1 C

erta

in in

fect

ious

and

pa

rasi

tic d

isea

ses

12.3

25.2

12.3

25.2

33.8

46.6

35.4

48.8

1-01

9 V

iral h

epat

itis

1.3

2.6

1.3

2.6

7.3

10.1

7.4

10.2

1-

026

Neo

plas

ms

57.5

118.

457

.511

8.4

141.

419

5.5

136.

418

8.5

1-02

9 M

alig

nant

neo

plas

m o

f sto

mac

h 3.

87.

93.

87.

97.

310

.16.

38.

7 1-

031

Mal

igna

nt n

eopl

asm

of l

iver

and

in

trahe

patic

bile

duc

ts

3.2

6.6

3.2

6.6

5.9

8.1

5.9

8.1

1-03

4 M

alig

nant

neo

plas

m o

f tra

chea

, br

onch

us a

nd lu

ng

2.5

5.3

2.5

5.3

16.5

22.8

16.6

23.0

1-03

6 M

alig

nant

neo

plas

m o

f bre

ast

14.8

30.1

14.8

30.1

22.0

30.4

20.7

28.7

1-

037

Mal

igna

nt n

eopl

asm

of c

ervi

x ut

eri

5.6

11.6

5.6

11.6

8.6

12.0

8.7

12.0

1-03

8 M

alig

nant

neo

plas

m o

f oth

er a

nd

unsp

ecifi

ed p

arts

of u

teru

s 4.

79.

64.

79.

66.

89.

55.

88.

0

1-03

9 M

alig

nant

neo

plas

m o

f ova

ry

3.0

6.2

3.0

6.2

7.7

10.6

7.7

10.6

1-

052

Dia

bete

s m

ellit

us

35.0

72.4

34.1

70.7

137.

819

0.4

98.3

135.

9 1-

058

Dis

ease

s of

the

nerv

ous

syst

em

2.5

5.1

2.5

5.1

17.9

24.6

18.0

24.8

1-06

4 D

isea

ses

of th

e ci

rcul

ator

y sy

stem

10

8.2

225.

310

9.1

227.

019

4.2

268.

523

2.5

321.

3

lxxvii

1-06

5 A

cute

rheu

mat

ic fe

ver a

nd

chro

nic

rheu

mat

ic h

eart

dise

ases

1.

22.

51.

22.

510

.214

.210

.314

.3

1-06

6 H

yper

tens

ive

dise

ases

7.

215

.07.

215

.012

.317

.016

.122

.3

1-06

7 Is

chae

mic

hea

rt di

seas

es

14.1

29.6

14.5

30.4

47.3

65.3

63.7

88.1

1-

068

Oth

er h

eart

dise

ases

70

.614

6.6

70.6

146.

689

.412

3.6

104.

814

4.9

1-06

9 C

ereb

rova

scul

ar d

isea

ses

15.2

31.7

15.6

32.6

31.3

43.3

33.8

46.7

1-

072

Dis

ease

s of

the

resp

irato

ry

syst

em

26.6

55.2

26.6

55.2

81.7

112.

882

.711

4.2

1-07

4 P

neum

onia

7.

816

.27.

816

.211

.115

.411

.516

.0

1-07

6 C

hron

ic lo

wer

resp

irato

ry

dise

ases

8.

617

.78.

617

.724

.333

.623

.432

.4

1-07

8 D

isea

ses

of th

e di

gest

ive

syst

em

14.3

29.6

14.3

29.6

27.6

38.1

28.0

38.6

1-08

0 D

isea

ses

of th

e liv

er

6.5

13.5

6.5

13.5

9.2

12.7

9.3

12.8

1-

084

Dis

ease

s of

the

geni

tour

inar

y sy

stem

8.

718

.18.

718

.120

.928

.923

.332

.1

1-08

7 Pr

egna

ncy,

chi

ldbi

rth

and

the

puer

periu

m

1.1

2.2

1.1

2.2

4.8

6.7

4.8

6.7

1-09

2 C

erta

in c

ondi

tions

orig

inat

ing

in th

e pe

rinat

al p

erio

d 1.

32.

41.

32.

410

.714

.410

.714

.4

1-09

5 Ex

tern

al c

ause

s of

mor

bidi

ty

and

mor

talit

y 7.

214

.87.

214

.821

.429

.520

.428

.2

1-09

6 Tr

ansp

ort a

ccid

ents

2.

04.

12.

04.

11.

72.

31.

72.

3 1-

101

Inte

ntio

nal s

elf-h

arm

0.

00.

00.

00.

02.

12.

92.

12.

9

lxxviii

Cause-of-death distributions by global burden of disease category (cumulative

%) for Tonga: Empirical data and modelled distribution based on mortality

level

a) Males, 2001-2004

Modelled estimates by GBD category obtained using Codmod models using inputs of

5q0 = 25 deaths per 1000 live births, 45q15 = 30.6% probability of dying (closest

possible model fit to midpoint of plausible range of mortality estimates), and GDP =

$2105 US per capita

Group I

Group

II

Group

III

lxxix

b) Males, 2005-2008

WHO Global Burden of Disease (GBD) Study categories: I (Infectious, perinatal and

maternal conditions), II (NCDs) and III (external causes)

Modelled estimates by GBD category obtained using Codmod models using inputs of

5q0 = 30 deaths per 1000 live births, 45q15 = 31.9% probability of dying (closest

possible model fit to midpoint of plausible range of mortality estimates), and GDP =

$2961 US per capita

�

Group

III

Group

II

Group I

lxxx

Nauru Deaths by cause (by ICD chapter) and proportional mortality (for children 5-14

years: Nauru, 2005-2009

Row Labels Deaths Proportionalmortality

Lower 95% CI

Upper95% CI

1-001 Infectious diseases 0 0% 0% 32% 1-026 Neoplasms 0 0% 0% 32% 1-048 Diseases of the blood 1 20% 2% 48% 1-051 Endocrine, nutritional and other metabolic disorders

0 0% 0% 32%

1-058 Diseases of the nervous system 0 0% 0% 32% 1-064 Diseases of the circulatory system

0 0% 0% 32%


1 20% 2% 48%

1-078 Diseases of the digestive tract 0 0% 0% 32% 1-082 Diseases of the skin 0 0% 0% 32% 1-084 Diseases of the genito-urinary system

0 0% 0% 32%

1-092 Perinatal conditions 0 0% 0% 32% 1-093 Congenital conditions 0 0% 0% 32% 1-094 Signs, symptoms and other ill-defined causes

0 0% 0% 32%

1-095 External causes (injuries) 3 60% 38% 75% Total 5 100% 100% 100%

lxxxi

Deaths in adults aged 65 years and older

Cause Female Male

ICD CHAPTER Deaths Propor-tional

mortality

Lower 95% CI

Upper95% CI Deaths

Propor-tional

mortality Lower 95% CI

Upper95% CI

1-026 Neoplasms 2 10% 2% 23% 3 17% 6% 31% 1-051 Endocrine, nutritional and other metabolic disorders

9 45% 34% 55% 2 11% 2% 25%

1-064 Diseases of the circulatory system 5 25% 13% 38% 5 28% 15% 41%

1-072 Diseases of the circulatory system 1 5% 0% 18% 5 28% 15% 41%

1-078 Diseases of the digestive tract 1 5% 0% 18% 1 6% 0% 20%

1-082 Diseases of the skin 1 5% 0% 18% 0% 0% 13%

1-084 Diseases of the genito-urinary system 1 5% 0% 18% 1 6% 0% 20%

1-095 External causes (injuries) 0% 0% 12% 1 6% 0% 20% Total 20 100% 100% 100% 18 100% 100% 100% Specific Causes (103 list) 1-034 1 11% 1% 33% 1 50% 10% 77% 1-037 1 11% 1% 33% 0% 0% 51% 1-046 0% 0% 22% 2 100% 100% 100% 1-052 9 100% 100% 100% 2 100% 100% 100% 1-067 0% 0% 22% 2 100% 100% 100% 1-068 3 33% 15% 51% 1 50% 10% 77% 1-069 2 22% 6% 42% 1 50% 10% 77% 1-071 0% 0% 22% 1 50% 10% 77% 1-074 0% 0% 22% 1 50% 10% 77% 1-075 0% 0% 22% 1 50% 10% 77% 1-076 1 11% 1% 33% 1 50% 10% 77% 1-077 0% 0% 22% 2 100% 100% 100% 1-081 1 11% 1% 33% 1 50% 10% 77% 1-082 1 11% 1% 33% 0% 0% 51% 1-086 1 11% 1% 33% 1 50% 10% 77% 1-103 0% 0% 22% 1 50% 10% 77% Total 20 222% 297% 181% 18 900% 4404% 394%

lxxxii

FijiDeaths and proportional mortality by mentioned cause (total mentions on the

certificate) for children 5-14 years: Fiji 2000-2008

Cause of death Deaths

Proportion (excluding ill defined)

Lower 95% CI

Upper 95% CI

1-001 Infectious Diseases 76 8.5 8.0 3.454 15.7631-026 Neoplasms 97 10.8 10.0 4.795 18.391-048 Diseases of the blood and blood forming organs 21 2.3 2.0 0.242 7.2251-051 Endocrine, nutritional and metabolic diseases 30 3.4 3.0 0.619 8.7671-055 Mental Disorders 1 0.1 0.0 0 3.6891-058 Diseases of the Nervous System 83 9.3 9.0 4.115 17.0851-064 Diseases of the circulatory system 122 13.6 13.0 6.922 22.231-072 Respiratory Diseases 117 13.1 13.0 6.922 22.231-078 Diseases of the digestive system 30 3.4 3.0 0.619 8.7671-082 Diseases of the skin 7 0.8 0.0 0 3.6891-083 Musculoskeletal Diseases 13 1.5 1.0 0.025 5.5721-084 Diseases of the genitourinary tract 27 3.0 3.0 0.619 8.7671-087 Maternal deaths 1 0.1 0.0 0 3.6891-092 Perinatal Conditions 4 0.4 0.0 0 3.6891-093 Congenital Conditions 29 3.2 3.0 0.619 8.7671-094 Ill-defined 30 3.4 3.0 0.619 8.7671-095 External causes 237 26.5 26.0 16.984 38.096Total 925 0 3.689Total excluding ill-defined 895 100.0 100.0 81.364 121.627

lxxxiii

V

anua

tuPr

opor

tiona

l mor

talit

y by

GB

D c

ateg

orie

s, b

y so

urce

for 2

000-

2007

Chi

ldre

n 0-

4 ye

ars

Cau

se o

f de

ath

(GB

D

Gro

up)

Rep

orte

d D

eath

sP

ropo

rtion

al M

orta

lity

(%)

Pro

porti

onal

Mor

talit

y (%

) exc

ludi

ng u

nkno

wn

and

ill-d

efin

ed c

ases

H

ospi

tal

Sepa

ratio

n D

ata

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Gro

up I

119

327

265

81.5

86.1

66.4

82.6

87.9

85.8

G

roup

II

18

2629

12.3

6.8

7.3

12.5

7.0

9.4

Gro

up II

I 7

1915

4.8

5.0

3.8

4.9

5.1

4.9

Unk

now

n 2

890

Tota

l 14

6 38

039

9

To

tal e

xcl

unkn

own

144

372

309

Chi

ldre

n 5-

14 y

ears

C

ause

of

deat

h (G

BD

G

roup

)

Rep

orte

d D

eath

sP

ropo

rtion

al M

orta

lity

(%)

Pro

porti

onal

Mor

talit

y (%

) exc

ludi

ng u

nkno

wn

and

ill-d

efin

ed c

ases

H

ospi

tal

Sepa

ratio

n D

ata

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Gro

up I

15

1517

42.9

40.5

23

.346

.941

.730

.9

Gro

up II

16

14

2545

.737

.8

34.2

50.0

38.9

45.5

G

roup

III

1 7

132.

918

.9

17.8

3.1

19.4

23.6

U

nkno

wn

3 1

18

To

tal

35

3773

Tota

l exc

l un

know

n 32

36

55

lxxxiv

A

dult

mal

es15

-59

year

s C

ause

of

deat

h (G

BD

G

roup

)

Rep

orte

d D

eath

sP

ropo

rtion

al M

orta

lity

(%)

Pro

porti

onal

Mor

talit

y (%

) exc

ludi

ng u

nkno

wn

and

ill-d

efin

ed c

ases

H

ospi

tal

Sepa

ratio

n D

ata

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Gro

up I

29

5041

12.7

12.0

12

.212

.912

.112

.2

Gro

up II

17

5 31

225

276

.874

.6

75.2

78.1

75.4

75.2

G

roup

III

20

5242

8.8

12.4

12

.58.

912

.612

.5

Unk

now

n 4

40

Tota

l 22

8 41

833

5

To

tal e

xcl

unkn

own

224

414

335

Adu

lt fe

mal

es15

-59

year

s C

ause

of

deat

h (G

BD

G

roup

)

Rep

orte

d D

eath

sP

ropo

rtion

al M

orta

lity

(%)

Pro

porti

onal

Mor

talit

y (%

) exc

ludi

ng u

nkno

wn

and

ill-d

efin

ed c

ases

H

ospi

tal

Sepa

ratio

n D

ata

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Hos

pita

l Se

para

tion

Dat

a

Med

ical

Cer

tific

ates

Hea

lthFa

cilit

y R

epor

ting

Gro

up I

36

4858

16.4

16.8

21

.116

.416

.821

.1

Gro

up II

18

0 22

120

282

.277

.3

73.5

82.2

77.3

73.5

G

roup

III

3 17

151.

45.

9 5.

51.

45.

95.

5 U

nkno

wn

To

tal

219

286

275

Tota

l exc

l un

know

n 21

9 28

627

5

lxxxv

Appendix 5: Medical record review handbook

PROCEDURES MANUAL

TONGA MEDICAL RECORD

REVIEW

UNIVERSITY OF QLD / TONGAN MINISTRY OF HEALTH

V 1: 15 SEPTEMBER 2009

1 | P a g e

INTRODUCTION

The purpose of this study is to review underlying causes of death to provide an overview of the

major health problems causing premature deaths in Tonga. This study will help both resource

allocation and improving data collections.

The data extracted from the medical records file will be reviewed by a doctor who will assign

underlying and contributory causes of death based on this information. Causes will then be

compared to the death certificate and the data used to complement this information to

develop a complete picture of causes of death in Tonga.

PRIVACY ISSUES

The confidential nature of the data must be respected at all times during the study, and all data

must be de-identified before leaving Tonga. De-identification will be done by matching a study

number to each case. The master list will remain with Sione and is to be password protected.

Hard copies of this list will be provided for data extraction in batches of 50-100 names, and

must be returned to Sione once complete. Sione will ensure details of the completed records

are marked against the master sheet and that the hard copy is destroyed.

2 | P a g e

STUDY PROCEDURES

CONSTRUCTION OF STUDY FILES

Creation of the master list will be done by one person in the HIS section of the Tongan Ministry

of Health to protect privacy.

Use the following instructions to create the master list and associated de-identified files:

Pull up copy of HIS records of all deaths for 2008 (Include all fields at this stage)

Keep

- All deaths in Tongatapu

- All deaths in ANY hospital

Remove – all deaths outside hospital on the outer islands

Randomly sort all remaining records and add a new column for STUDY_ID. Assign each death a

3-digit number (you can use the drag and fill function in excel – but not formulas)

Save 2 copies of this file

1. MASTER LIST

Keep only variables: STUDY_ID, PATIENT_NO (Hospital Number), FIRST NAME, SURNAME, AGE,

DATE OF BIRTH and DATE OF DEATH. Empty variables for Sent, Returned, and Complete should

also be added. This file is to be locked and kept only by Sione.

For data extraction – sort by hospital and print list for each hospital for data extraction (list per

week rather than all at once). This should be provided in hard copy only to the Data extractor

and returned once the data has been extracted. Sione is to check that lists have been returned

and to ensure hard copies are destroyed.

2. DEATH CERTIFICATE FILE

Keep only variables: STUDY_ID, Village, Island, Place of death (hospital vs community) and AGE

(using variable names already in the system)

File to be kept by Sione and copy to Karen (this is de-identified data only)

3. RESULTS FILE

Create another empty spreadsheet called MRResults with variables STUDY ID, Study Cause 1

(Text + ICD Code), Study Cause 2 (Text + ICD Code), Study Cause 3 (Text + ICD Code), Study

Contributing Cause 1 (Text + ICD Code), Study Contributing Cause 1 (Text + ICD Code)

3 | P a g e

INSTRUCTIONS FOR MEDICAL RECORD DATA EXTRACTION

Although the task requires data to be extracted as it appears in the record without

interpretation, finding the appropriate information in the record requires a good understanding

of disease processes, medical terminology, laboratory tests and other investigations. This is the

most important part of the project!!

The form has been designed in consultation with Co-investigators, doctors, and HIS staff from

Tonga. However, this is a new process and your feedback on how the form is working (including

whether it is easy to follow or whether there should be any changes) will help improve the

study.

TRIAL

When starting data extraction, each person will be asked to complete a trial of ½ to 1 day (7-10

forms) before continuing with any further data extraction. These forms are to be sent to Karen

for review and feedback prior to further work. The intention of this trial is both to make sure

the data extractor has understood the intent of the questions and that the appropriate level of

detail has been provided, as well as to provide an opportunity to review the form itself before

further work is done.

GENERAL INSTRUCTIONS

- You will be given a list of patients for whom we need to collect data. You must not copy or

lose this file, and it must be returned to Sione when completed.

- You will either need to find the file using name and patient number or ask medical records

to find the file for you. If a file cannot be found please mark this on the list and let Sione

know when the list is returned.

- For each patient you will need to complete a data extraction form. Make sure the study ID

number is recorded on each page. The patient name or patient number should NOT be

marked anywhere on the extraction form.

- When completing the extraction form, please be as accurate as possible. If the doctor has

included abbreviations or local words, please record these as is rather than translating. We

are hoping to have a Tongan doctor review the data.

4 | P a g e

- You will only have access to the medical record. Please do not chase up additional data

from the death certificate or other data. While this may be the correct things to do for

surveillance system, it is not part of this study protocol and could affect interpretation of

the results.

- While much of the information in the first 4 sections may be found in the coversheet/

admission sheet – if additional information is included in the medical records, please

include this as well.

- Once you have completed the extraction form, please make sure you have

o Completed all sections

o Entered your name and date at the end

o Recorded the study ID on each page

o Ticked the individual off your list

o Returned the file to medical records (or other process for returning files as set out by

Sione)

o Put your completed form into the marked box.

- We have estimated that each record extraction should take up to 30-40 mins on average

(once you have done the first few which will of course be slower!). The time for each form

will vary according to the cause of death and information available. We therefore expect

this to be around 12-14 records per day. If you are consistently completing less than this,

please talk to Sione so we can look at why and whether our estimates need to be revised.

SPECIFIC INSTRUCTIONS BY SECTION

Section 1

1.1 Record the study number from the master sheet. Tick the name off on the master sheet

printout and mark that this has been done on the form.

1.2 Record the gender of the case

1.3 Record the age at death/ discharge from hospital (i.e. at the time of the last record). If one

year or older only years needs to be completed. For cases less than one year old at time of

death, record the age as months or days if case was less than one month old at time of death.

Leave the other sections of this question blank.

5 | P a g e

1.4 If case is a stillbirth – please note on form. There is no need to complete any further

information.

1.5 Format for date of birth is DD/MM/YYYY. If the case died outside hospital and there is no

information on the date of birth in the medical record please tick the box to indicate this.

1.6 Format for date of death is DD/MM/YYYY. If the case died outside hospital and there is no

information on the date of death in the medical record please tick the box to indicate this.

1.7 Record the usual residence (prior to admission to hospital) by island and village

Section 2

2.1 Record the hospital where the records were found

2.2 Indicate whether the case was in hospital at the time of death

2.3 If the case died outside hospital, record the date of discharge. Use the format

DD/MM/YYYY. If the case died in hospital indicate NA

Sections 3-5

If a section is not applicable please tick the box marked NA and move to the next section. Please

ensure that all relevant information available in the medical record is included – not just that

marked on the hospital coversheet.

Section 3

For males or females under 10 years of age indicate that the section is not applicable and move

to section four. Complete this section for all females 10 years or older or for whom age is

unknown.

3.1 Indicate whether the case was pregnant at the time of death

3.2 Indicate whether the case had delivered a child or had an abortion (deliberate or non-

deliberate) within 6 weeks of the death.

3.3 Record the details of the pregnancy (including progress and investigations) if either Q3.1 or

Q3.2 apply.

6 | P a g e

Section 4

Complete this section for all deaths where injury was a factor. For deaths that occurred from

other causes indicate that the section is not applicable and move to section five.

4.1 Indicate whether the intention of the injury was known. For example, suicide, homicide,

murder, assault are all intentional. Motor vehicle accidents (MVA’s) would most likely be

unintentional. If intent is not known (for example in a poisoning where it is not known in the

death was accidental or if the person committed suicide) record the intent as “unknown”.

4.2 If the death was known to be intentional in Q4.1, indicate whether the death occurred due

to suicide (i.e. caused by the individual who died) or homicide (i.e. caused by someone other

than the person who died).

4.3 Indicate the mode of the death, that is how the injury occurred (ie car accident, fell of ledge

etc).

4.4 Describe each injury, including what the injury was (i.e. burn, fracture, laceration etc) and

which body part was affected (ie base of skull, right lower arm, etc).

4.5 If any details of why the injury occurred are known, include this information here (i.e.

known alcohol consumption, kava etc)

Section 5

Complete this section for all deaths excluding those that occurred from injuries without

underlying natural causes. For example for a death from an MVA, skip this section unless the

accident was related to an underlying health problem. For injury deaths where this section is

not relevant , indicate that the section is not applicable and move to section six.

5.1 Include all relevant details of the illness or condition preceding death. Timeframe for this

question is not fixed and will vary from case to case based on cause. Please include all of the

most recent admission with details of any relevant events that led up to that visit (such as

concerns / tests / GP visits). Include the relevant sections of the record as originally recorded.

5.2 Tick if any of the tests in the list were performed either during the most recent admission or

in the events leading to this admission.

5.3 Make sure any test mentioned in Q5.1 or Q5.2 where a positive result was recorded, please

record the date of the test and the result as shown in the medical record.

7 | P a g e

5.4 Indicate if any imaging studies were performed (either locally or overseas) either during the

most recent admission/ illness or in the events leading to this admission.

5.5 For any imaging studies indicated in Q 5.4 that had positive or abnormal findings, record the

date and the results as shown. If you are unsure whether the results indicate an abnormal

finding, include them here.

5.6 If any histopathological investigations were performed, record the date of the procedure

and the result as shown.

5.7 Indicate whether the case had a chronic condition or disease. If no chronic condition is

recorded tick ‘No” and skip to 5.12. Common chronic diseases/ conditions may include:

After effects of stroke, Alcoholism, Asthma, Cancer (active or in remission), Cardiovascular

disease, Cirrhosis, COPD, Drug Use (long term), Diabetes, Disability (paraplegia etc),

Emphysema, Hepatitis B, Hepatitis C, Hypertension, Mental Disorders, Rheumatic Fever, etc.

Please note that we ask you to include chronic infectious diseases such as Tuberculosis at this

question as well as non-communicable diseases.

5.8 For cases with a chronic condition noted in the record state which one(s)

5.9 Record the evidence in the file that indicates the case had this condition (these may be

doctors clinical diagnosis, diagnostic tests or investigations). Include when the diagnosis was

made or the tests were done.

5.10 Record the month and year when this condition was first noted. If the case had multiple

chronic conditions indicate this timeframe for each.

5.11 Record details of any complications listed on file (e.g. cellulitis of the left leg from diabetes

etc). Indicate when these complications occurred. If there are no complications noted in the file

please note ‘None’.

5.12 Record whether the case had any surgery or invasive procedures in the four months prior

to death. If no, tick the appropriate box and skip to Q5.14

5.13 Describe any medication / treatment the case was taking in order to treat this illness at the

time of death.

5.14 Record details of any surgery or procedure indicated in Q5.12. Include date, what the

procedure was, where the procedure was performed (ie hospital, clinic, overseas), outcome,

complications, and length of stay if known.

8 | P a g e

5.15 Indicate whether the person has a positive HIV result on file. If there is a positive result,

indicate the date of the test as DD/MM/YYYY.

5.16 Review the of potential AID’s defining conditions for developing countries by WHO and

indicate if any of these conditions are noted on file. For any that are listed in the medical

record, please record the details of the condition from the record in the space provided.

Remember to include the date of findings.

END

Please ensure your name and the date are marked on the first page for follow up if data on

form is not clear. Also check the questionnaire is securely fastened together and the patient

number is clear.

ISSUES / QUESTIONS

Issues should be raised with Sione in the first instance for anything concerning obtaining and

returning records, name lists, missing records or completed forms. Issues related to content

should be discussed with Karen ([email protected])

9 | P a g e

INSTRUCTIONS FOR CAUSE OF DEATH REVIEW

INTRODUCTION

The Data Collection Form contains an extract of the case details as shown on the medical

record (without reference to any other data sources). Data is contained in the following

sections

1/ General/ demographic details

2/ Hospital/ admission details

3/ Details of female deaths (aged > 10 years)

4/ Details of injury-related deaths

5/ Case history including:

o Most recent illness

o Chronic disease history

o Previous surgery

o AIDs defining conditions

Please use the information provided to make a clinical judgment about the sequence of causes

that resulted in the death, as well as whether there were any additional causes that did not

contribute directly to this sequence. You will be asked to note these causes, including your level

of certainty regarding the cause, on the Review and Coding Sheet. Please do not write in the

section of the sheet marked “Coding” and ignore the text in italics (these are variable names to

assist with data entry).

SPECIFIC INSTRUCTIONS

1.1 Record the study ID on the form

1.2 Indicate the IMMEDIATE cause of death. This is the cause that directly resulted in the

death (i.e. “stroke”). Please note that this field should not include text such as “cardio-

respiratory arrest” or “stopped breathing”. Record only one cause per line.

1.3 If there is one, record the cause of death that resulted in the cause listed in Q1.2. In

other words, the cause of death listed in Q1.2 occurred DUE TO the one listed here. This

is the PRECIPITATORY cause of death. Record only one cause per line.

10 | P a g e

1.4 Continuing this sequence, if there was a further underlying cause of death that led

directly to the cause listed in Q1.3, please record it here. This should be there overall

UNDERLYING cause of death. Record only one cause per line.

1.5 Record any conditions that may have contributed to the death but were did not directly

result in the sequence recorded above in Q1.2 – Q1.4. Record only one cause per line.

1.6 For each cause of death listed in Q1.2 to Q1.5, indicate in the space marked how certain

you are of each diagnosis based on the following criteria.

o Definite – Clinical diagnosis is supported by positive tests and investigations in a

combination which could not be indicative of any other disease or sequence of

events (i.e. in the case of injury or external causes, the injuries could not have

occurred in any other way), or autopsy supports findings.

o Strong certainty – Clinical diagnosis is supported by positive tests and

investigations which are diagnostic for this disease, or in the case of injury and

external causes, the case history is consistent with the cause noted and there is

other evidence (such as a police report or file note) to support the diagnosis.

o Some certainty – Clinical diagnosis is supported by positive tests and

investigations which are indicative for this disease, or case history is consistent

with the disease and does not suggest a differential diagnosis, or in the case of

injury and external causes, the case history is consistent with the described

sequence of events.

o Limited certainty – Cause is the most likely interpretation of the data available.

1.7 For any death where the cause cannot be identified, please indicate this at this question

and provide details in the space provided.

1.8 Note your initials and the date you completed the review.

11 | P a g e

INSTRUCTIONS FOR CODING

You will be given a copy of the Review and Coding Sheet for each case. Code each cause of

death using ICD 10-AM, following the sequence as you would for a medical certificate of death.

The code should be written onto the space provided in the form.

Data is to be entered onto the RESULTS file. For each case enter each cause of death and the

associated ICD code onto the spreadsheet against the Study_ID. Variable names are shown on

the form. Once data from the Review and Coding Sheet has been entered into the spreadsheet,

tick the box at the bottom right of the form to indicate this is complete.

ADMINISTRATION ISSUES

WEEKLY TASKS FOR MOH CO_INVESTIGATOR

• Provide a list to the data collectors at the beginning of each week of records to be

reviewed.

• At the end of the week collect the paper lists, check the completed records marked on

the sheet against the completed forms, mark the completed records on the Master

sheet, and destroy the paper copy.

• E-mail Karen to advise number completed.

• At the end of every week collect the completed forms and either send to Karen by

registered post OR photocopy the forms and mail the original by regular post (the copy

should be kept on-site in a dedicated folder)

OTHER TASKS

• For each batch of coversheets returned for coding, mark the sheets returned against

their study ID numbers on the Master Sheet.

• Provide the coversheets to Nauna for coding and entry into the Record Review Results

File.

• Once entered, mark the ID number as complete in the Master Sheet.

• E-mail a copy of the MRResults to Karen as progress is made

• Send signed invoices to UQ for payment of agreed funds as milestones are reached.

12 | P a g e

FORMS

DATA COLLECTION FORM

See attached file

13 | P a g e

REVIEW AND CODING SHEET

Next Page

REVIEW AND CODING SHEET

1.1 PATIENT ID � � �

From the information in the medical record, please

indicate:

1.2 Immediate cause of death (IC_T)

________________________________________

1.3 Precipitatory cause of death (PC_T)

________________________________________

1.4 Underlying cause of death (ie the cause that

started the sequence) (UC_T)

________________________________________

1.5 Contributory Causes (that contributed to death, but

did not directly lead to the death)

(CC1_T)____________________________________

(CC2_T)____________________________________

(CC3_T)____________________________________

1.6 � Cause cannot be determined (XC)

________________________________________________

________________________________________________

1.8 Initials _________________________

Date: ______ / ______ / ________

Definite Strong

certainty

Some

certainty

Limited

certainty

Definite Strong

certainty

Some

certainty

Limited

certainty

Definite Strong

certainty

Some

certainty

Limited

certainty

Definite Strong

certainty

Some

certainty

Limited

certainty

Definite Strong

certainty

Some

certainty

Limited

certainty

Definite Strong

certainty

Some

certainty

Limited

certainty

ICD CODE

Coder’s use only

1.6 Certainty of Diagnosis

Initials :

______________

Date:

____ /____ / ____

Study ID 22 Sep 2009 v2

Page 1 of 6

MEDICAL RECORD DATA ABSTRACTION FORM

PATIENT DEMOGRAPHICS

1.1 Study ID Number: Name checked against Study ID

Master Sheet?

1.2 Sex 1 Male 2 Female

1.3 Age at Death _____ years (if ≥ 1 yr) _____ months (if < 1 yr) _____ days (if < 1 month)

1.4 Is case a stillbirth? �� if yes, stop here (If uncertain, please continue)

1.5 Date of Birth /____/____/______/ No Info

dd mm yyyy

1.6 Date of Death /____/____/______/

dd mm yyyy

1.7 Residence of Deceased (as detailed as possible)

_______________________________________________ Village |____|____|

_______________________________________________ Island |____|____|

HOSPITAL DETAILS

2.1 Hospital Name: ______________________________________

2.2 Case died in Hospital? 1 YES 2 NO

2.3 Date of Final/Most Recent Admission /____/____/______/ Not Applicable

dd mm yyyy

2.4 If case died outside hospital –

Date of Most Recent Discharge /____/____/______/ Not Applicable

dd mm yyyy

Initials:

______________

Date:

____ /____ / ____


Page 2 of 6

FEMALE DEATHS Not Applicable

3.1 Was this person pregnant 1 YES 2 NO

3.2 Did the death occur within 6 weeks of either a delivery or abortion?

1 YES 2 NO

3.3 If yes to either question:

Provide details – (include Date of Delivery, complications, single or multiple birth, child born

alive etc)

INJURY DEATHS Not Applicable

4.1 Intent of the injury

1 Intentional 2 Unintentional 3 Not known / Not stated

For Intentional Injuries:

4.2 Was this death: 1 Suicide 2 Homicide

For All Injuries:

4.3 Mode of death (i.e. fall / poisoning / hanging / stabbing / burn / MVA etc):

4.4 Describe the injuries (include body part and effect)

4.5 Additional Info: (How did the injury occur?)

i) road traffic accident – whether

pedestrian or occupant of vehicle,

type of vehicles involved

ii) accidental poisoning – nature of

poison, circumstances of poisoning

iii) fall – from where, how

iv) drowning – well, lake, river, sea

v) burns – how (stove burst, gas

cylinder, house fire, chemical burns

etc)

vi) bite of venomous animal

vii) other unintentional injuries such as

occupational injuries – nature of

work, mechanism of injury etc.


Page 3 of 6

NON-INJURY RELATED DEATHS / (And injury related deaths where other disease factors were involved)

Not Applicable

HISTORY, EXAMINATION, INVESTIGATION AND PROGRESS

Recent History

5.1 Clinical case summary: (please write a brief description of the presenting illness and

clinical events during hospitalization in chronological sequence, culminating in either death or

discharge of the patient. Include any relevant investigation results and diagnoses, as recorded

by treating physicians in the case record).


Page 4 of 6

5.2 Were any of the following laboratory investigations performed? (locally or overseas)

Test Performed? (Tick if Yes)

Test Performed? (Tick if Yes)

Haematology Biochemistry

Prostate specific antigen

Haemoglobin Infectious disease serology

White cell count Hepatitis B

Differential leucocyte count Hepatitis C

ESR HIV

Urinalysis Cardiovascular diagnostics

Glucose ECG

Albumin Echocardiography

Microscopy Cardiac perfusion scan

Biochemistry Coronary angiography

Biochemistry Glucose metabolism Cardiac enzymes – Ck-MB)

Fasting plasma glucose Troponins

2 hour post prandial plasma

glucose

Pulmonary diagnostics

Ketones Chest x ray

glycosylated haemoglobin Sputum for AFB

Biochemistry Liver function tests AFB culture report

Serum bilirubin Spirometry

ALT FEV1 %

AST Peak expiratory flow rate

Prothrombin time Ventilation perfusion scan

Biochemistry Renal function tests

Blood urea

Serum creatinine

Electrolytes

5.3 Provide details of all positive test results – include dates / test type and results

5.4 Were any of the following imaging studies performed? (locally or overseas)

Study Performed? Test Performed?

Barium contrast radiology Laparoscopy

Ultrasonography Endoscopy

CT Scan Colonoscopy

MRI Colposcopy/hysteroscopy

Other x rays (including

mammography)


Page 5 of 6

5.5 Provide details of all positive imaging results – include dates / study type and results

5.6 Describe any Histopathology results (cancers, liver and renal biopsies)

Not Applicable

Chronic Diseases / Conditions

5.7 Did this person have a chronic disease or condition 1 YES 2 NO

5.8 What disease(s) / condition(s):

5.9 Was there any evidence recorded when this diagnosis was ORIGINALLY made (include

results and dates) (ie. If the case has chronic Hep B is there an antigen result on file etc)

5.10 What is the earliest date in record where this disease/condition was noted?

Month/Year: __________________________________________________________________

5.11 Any noted complications from the chronic condition? (Describe including date and

outcome):


Page 6 of 6

5.12 Was the person taking any medication / treatment to control this disease or condition?

1 YES 2 NO � If yes, please describe below (include name/ dose)

Surgery / Invasive Procedures

5.13 Was any surgery / invasive procedure performed on this person within the last 4 months?

1 YES 2 NO

5.14 If yes � Describe the procedure (include date, outcome and complications if noted)

AIDS Defining Conditions

5.15 Is there a positive HIV test result on file 1 YES � ____ / ____ / ____ 2 NO

5.16 Does the medical record show evidence of AIDS defining conditions?

Karposi sarcoma YES NO

Pulmonary of extrapulmonary tuberculosis YES NO

Meningitis YES NO

Weight lost > 10% of body weight YES NO

Chronic diarrhoea > 1 month YES NO

Prolonged fever > 1 month YES NO

Persistent cough > 1 month YES NO

History of Herpes Zoster YES NO

Oropharyngeal candidiasis YES NO

Chronic or disseminated herpes simplex YES NO

Generalised lymphadenopathy YES NO

Persistent dermatitis YES NO

Neurological impairment (not related to known accident)

YES NO

Recurrent pneumonia YES NO

Invasive cervical cancer YES NO

END OF QUESTIONNAIRE – Remember to sign and date the front

Mortality and Cause of Death in the Pacific309197/s41545625_phd_fin… · mortality was high across...

Documents

Transcript of Mortality and Cause of Death in the Pacific309197/s41545625_phd_fin… · mortality was high across...