ESTIMATING THE EFFECTS OF AMBIENT TEMPERATURE ON … · Guo Y, Barnett AG, Zhang Y, Tong S, Yu W,...
Transcript of ESTIMATING THE EFFECTS OF AMBIENT TEMPERATURE ON … · Guo Y, Barnett AG, Zhang Y, Tong S, Yu W,...
ESTIMATING THE EFFECTS OF AMBIENT TEMPERATURE ON MORTALITY:
METHODOLOGICAL CHALLENGES AND PROPOSED SOLUTIONS
BY
YUMING GUO
Bachelor of Medicine, Master of Medicine
A thesis submitted for the Degree of Doctor of Philosophy
School of Public Health and Social Work
Faculty of Health
Queensland University of Technology
May 2012
ii
ABSTRACT
The health impacts of exposure to ambient temperature have been drawing increasing
attention from the environmental health research community, government, society, industries,
and the public. Case−crossover and time series models are most commonly used to examine
the effects of ambient temperature on mortality. However, some key methodological issues
remain to be addressed. For example, few studies have used spatiotemporal models to assess
the effects of spatial temperatures on mortality. Few studies have used a case−crossover
design to examine the delayed (distributed lag) and non-linear relationship between
temperature and mortality. Also, little evidence is available on the effects of temperature
changes on mortality, and on differences in heat-related mortality over time.
This thesis aimed to address the following research questions:
1. How to combine case−crossover design and distributed lag non-linear models?
2. Is there any significant difference in effect estimates between time series and
spatiotemporal models?
3. How to assess the effects of temperature changes between neighbouring days on
mortality?
4. Is there any change in temperature effects on mortality over time?
To combine the case-crossover design and distributed lag non-linear model, datasets
including deaths, and weather conditions (minimum temperature, mean temperature,
maximum temperature, and relative humidity), and air pollution were acquired from Tianjin
China, for the years 2005 to 2007. I demonstrated how to combine the case−crossover design
with a distributed lag non-linear model. This allows the case−crossover design to estimate the
iii
non-linear and delayed effects of temperature whilst controlling for seasonality. There was
consistent U-shaped relationship between temperature and mortality. Cold effects were
delayed by 3 days, and persisted for 10 days. Hot effects were acute and lasted for three days,
and were followed by mortality displacement for non-accidental, cardiopulmonary, and
cardiovascular deaths. Mean temperature was a better predictor of mortality (based on model
fit) than maximum or minimum temperature.
It is still unclear whether spatiotemporal models using spatial temperature exposure produce
better estimates of mortality risk compared with time series models that use a single site’s
temperature or averaged temperature from a network of sites. Daily mortality data were
obtained from 163 locations across Brisbane city, Australia from 2000 to 2004. Ordinary
kriging was used to interpolate spatial temperatures across the city based on 19 monitoring
sites. A spatiotemporal model was used to examine the impact of spatial temperature on
mortality. A time series model was used to assess the effects of single site’s temperature, and
averaged temperature from 3 monitoring sites on mortality. Squared Pearson scaled residuals
were used to check the model fit. The results of this study show that even though
spatiotemporal models gave a better model fit than time series models, spatiotemporal and
time series models gave similar effect estimates. Time series analyses using temperature
recorded from a single monitoring site or average temperature of multiple sites were equally
good at estimating the association between temperature and mortality as compared with a
spatiotemporal model.
A time series Poisson regression model was used to estimate the association between
temperature change and mortality in summer in Brisbane, Australia during 1996–2004 and
Los Angeles, United States during 1987–2000. Temperature change was calculated by the
iv
current day’s mean temperature minus the previous day’s mean. In Brisbane, a drop of more
than 3 °C in temperature between days was associated with relative risks (RRs) of 1.16 (95%
confidence interval (CI): 1.02, 1.31) for non-external mortality (NEM), 1.19 (95% CI: 1.00,
1.41) for NEM in females, and 1.44 (95% CI: 1.10, 1.89) for NEM aged 65–74 years. An
increase of more than 3 °C was associated with RRs of 1.35 (95% CI: 1.03, 1.77) for
cardiovascular mortality and 1.67 (95% CI: 1.15, 2.43) for people aged < 65 years. In Los
Angeles, only a drop of more than 3 °C was significantly associated with RRs of 1.13
(95% CI: 1.05, 1.22) for total NEM, 1.25 (95% CI: 1.13, 1.39) for cardiovascular mortality,
and 1.25 (95% CI: 1.14, 1.39) for people aged ≥ 75 years. In both cities, there were joint
effects of temperature change and mean temperature on NEM. A change in temperature of
more than 3 °C, whether positive or negative, has an adverse impact on mortality even after
controlling for mean temperature.
I examined the variation in the effects of high temperatures on elderly mortality (age ≥ 75
years) by year, city and region for 83 large US cities between 1987 and 2000. High
temperature days were defined as two or more consecutive days with temperatures above the
90th
percentile for each city during each warm season (May 1 to September 30). The
mortality risk for high temperatures was decomposed into: a “main effect” due to high
temperatures using a distributed lag non-linear function, and an “added effect” due to
consecutive high temperature days. I pooled yearly effects across regions and overall effects
at both regional and national levels. The effects of high temperature (both main and added
effects) on elderly mortality varied greatly by year, city and region. The years with higher
heat-related mortality were often followed by those with relatively lower mortality.
Understanding this variability in the effects of high temperatures is important for the
development of heat-warning systems.
v
In conclusion, this thesis makes contribution in several aspects. Case−crossover design was
combined with distribute lag non-linear model to assess the effects of temperature on
mortality in Tianjin. This makes the case−crossover design flexibly estimate the non-linear
and delayed effects of temperature. Both extreme cold and high temperatures increased the
risk of mortality in Tianjin. Time series model using single site’s temperature or averaged
temperature from some sites can be used to examine the effects of temperature on mortality.
Temperature change (no matter significant temperature drop or great temperature increase)
increases the risk of mortality. The high temperature effect on mortality is highly variable
from year to year.
vi
KEY WORDS
CLIMATE CHANGE
TEMPERATURE
TEMPERATURE CHANGE
UNSTABLE WEATHER
HEAT EFFECT
MORTALITY
TIME SERIES
CASE–CROSSOVER
SPATIOTEMPORAL MODEL
DISTRIBUTED LAG NON-LINEAR MODEL
HEATWAVES WARNING SYSTEM
vii
PUBLICATIONS OR MANUSCRIPTS BY THE CANDIDATE ON MATTERS
RELEVENT TO THE THESIS
Guo Y, Barnett AG, Pan X, Yu W, Tong S. (2011) The impact of temperature on mortality in
Tianjin, China: a case−crossover design with a distributed lag non-linear model.
Environmental Health Perspectives 119:1719–1725.
Guo Y, Barnett AG, Tong S. Spatiotemporal model or time series model for assessing city-
wide temperature effects on mortality? Environmental Research (in press), doi:
10.1016/j.envres.2012.09.001.
Guo Y, Barnett AG, Yu W, Pan X, Ye X, Huang C, Tong S. (2011) A large change in
temperature between neighbouring days increases the risk of mortality. PLoS ONE, 6(2),
e16511.
Guo Y, Barnett AG, Tong S. Associations between high temperatures and elderly mortality
differed by year, city and region in the United States. Scientific Reports (In revision).
viii
PUBLICATIONS OR MANUSCRIPTS BY THE CANDIDATE DURING PHD STUDY
Guo Y, Punnasiri K, Tong S (2012). Effects of Temperature on Mortality in Chiang Mai,
Thailand: a time series study. Environmental Health, 11(36), doi:10.1186/1476-069X-11-36).
Guo Y, Jiang F, Peng L, Zhang J, Geng F, Xu J, Zhen C, Shen X, Tong S (2012). The
association between cold spells and pediatric outpatient visits for asthma in Shanghai, China.
PLoS ONE, (7): e42232.
Guo Y, Barnett AG, Pan X, Yu W, Tong S. (2011) The impact of temperature on mortality in
Tianjin, China: a case−crossover design with a distributed lag non-linear model.
Environmental Health Perspectives 119:1719–1725.
Guo Y, Barnett AG, Yu W, Pan X, Ye X, Huang C, Tong S. (2011) A large change in
temperature between neighbouring days increases the risk of mortality. PLoS ONE, 6(2),
e16511.
Guo Y, Barnett AG, Zhang Y, Tong S, Yu W, Pan X. (2010) The short-term effect of air
pollution on cardiovascular mortality in Tianjin, China: comparison of time series and case–
crossover analyses. Science of the Total Environment, 409(2), pp. 300–306.
Guo Y, Tong S, Zhang Y, Barnett AG, Jia Y, Pan X. (2010) The relationship between
particulate air pollution and emergency hospital visits for hypertension in Beijing, China.
Science of the Total Environment, 408(20), pp. 4446–4450.
ix
Guo Y, Tong S, Li S, Barnett AG, Yu W, Zhang Y, Pan X. (2010) Gaseous air pollution and
emergency hospital visits for hypertension in Beijing, China: a time-stratified case-crossover
study. Environmental Health, 9(1), pp. 57–63.
Guo Y, Barnett AG, Tong S. Spatiotemporal model or time series model for assessing city-
wide temperature effects on mortality? Environmental Research (in press), doi:
10.1016/j.envres.2012.09.001.
Guo Y, Barnett AG, Tong S. Associations between high temperatures and elderly mortality
differed by year, city and region in the United States. Scientific Reports (In revision).
Guo Y, Li S, Barnett AG, Jaakkola J, Tong S, Zhang Y, Gasparrini A, Pan X.The effects of
ambient temperature on cerebrovascular mortality: an epidemiologic study in four climatic
zones in China. American Journal of Epidemiology (In revision).
Kimlin M, Guo Y (2012). Assessing the impacts of lifetime sun exposure on skin damage
and skin aging using a non-invasive method. Science of the Total Environment 425: 35-41.
Zhang Y, Guo Y, Li G, Zhou J, Jin X, Wang W, Pan X (2012). The Spatial Characteristics
for Ambient Particulate Matter and Mortality in Urban Area of Beijing, China. Science of
The Total Environment, 7 (28);435-436C:14-20.
Tong S, Wang X, Guo Y (2012). Assessing the short-term effects of heatwaves on mortality
and morbidity in Brisbane, Australia: Comparison of case-crossover and time series analyses.
PLoS ONE, 7(5): e37500, doi:10.1371/journal.pone.0037500.
x
Banu S, Hu W, Guo Y, Zahirul Islam M, Tong S (2012). Space-time clusters of dengue fever
in Bangladesh. Tropical Medicine & International Health, 7 (19), doi: 10.1111/j.1365-
3156.2012.03038.x.
Xu Z, Eetzel R, Su H, Huang C, Guo Y, Tong S (2012). Impact of ambient temperature on
children's health: A systematic review, Environmental Research 2012, 8 (117):120-31.
Bi Y, Hu W, Liu H, Xiao Y, Guo Y, Chen S, Zhao L, Tong S. Can slide positivity rates
predict malaria transmission? Malaria Journal, 11(117), doi: 10.1186/1475-2875-11-117.
Madaniyzi L, Guo Y, Ye X, KIMDS, Zhang Y, Pan X (2012). The Effects of Metal
Components of Ambient Particulate Matter on Schoolchildren Lung Function in Inner
Mongolia of China, Journal of Occupational and Environmental Medicine (In press).
Yu W, Guo Y, Hu W, Mengersen K, Tong S (2011). The effect of various temperature
indicators on different mortality categories in a subtropical city of Brisbane, Australia.
Science of the Total Environment 409 (18): 3431-3437.
Yu W, Hu W, Mengersen K, Guo Y, Tong S (2011). Assessing the relationship between
global warming and mortality: Lag effects of temperature fluctuations by age and mortality
categories. Environmental Pollution 159 (2011): 1789-1793.
xi
Yu W, Mengersen K, Hu W, Guo Y, Tong S (2011). Time course of temperature effects on
cardiovascular mortality in Brisbane, Australia. Heart 97(13); 1089-93.
Huang C, Vaneckova P, Wang X, Guo Y, Shilu Tong. Constraints and barriers to public
health adaptation to climate change. American Journal of Preventive Medicine 402: 183–109.
Yu W, Mengersen K, Ye X, Guo Y, Pan X, Huang C, Wang X, Tong S. Daily average
temperature and mortality among the elderly: A meta-analysis and systematic review of
epidemiological literature. International Journal of Biometeorology: 1-13.
xii
CONFERENCE PRESENTATIONS
GuoY, Li S, Zhang Y, Pan X, Barnett A, Tong S. The effects of ambient temperature on
cerebrovascular deaths in five cities, China.
Oral presentation. The 2012 International Conference of International Society for
Environmental Epidemiology. Columbia, United States. 26–30, August 2011
Madaniyzi L, Guo Y, Ye X, KIMDS, Zhang Y, Pan X. The effects of metal components of
ambient particulate matter on schoolchildren lung function in Inner Mongolia of China.
Poster presentation. The 2012 International Conference of International Society for
Environmental Epidemiology. Columbia, United States. 26–30, August 2011
Zhang Y, Guo Y, Li G, Zhou J, Jin X, Wang W, Pan X. The Spatial Characteristics for
Ambient Particulate Matter and Mortality in Urban Area of Beijing, China
Poster presentation. The 2012 International Conference of International Society for
Environmental Epidemiology. Columbia, United States. 26–30, August 2011
Yu W, Megersen K, Ye X, Turner L, Hu W, Guo Y, Wang X, Tong S. Projecting Future
Transmission of Malaria under Climate Change Scenarios: Challenges and Opportunities
Poster presentation. The 2012 International Conference of International Society for
Environmental Epidemiology. Columbia, United States. 26–30, August 2011
Guo Y, Barnett AG, Tong S, Yu W, Pan X. The impacts of extreme cold and hot
temperatures on mortality in Tianjin, china: towards response for climate change.
xiii
Poster presentation. The 2011 International Conference of International Society for
Environmental Epidemiology. Barcelona, Spain. 13–16, September 2011
Guo Y, Barnett AG, Tong S. The effects of high temperatures on elderly mortality differed
by year, city and region in the United States.
Poster presentation. The 2011 International Conference of International Society for
Environmental Epidemiology. Barcelona, Spain. 13–16, September 2011
Zhang Y, Guo Y, Tao H, Wang L, Pan, X. The study on the relationship between personal
hygiene and intestinal infectious diseases of rural residents.
Poster presentation. The 2011 International Conference of International Society for
Environmental Epidemiology. Barcelona, Spain. 13–16, September 2011
Banu S, Tong S, Hu W, Hurst C, Guo Y, Islam MZ. Spatiotemporal clustering analysis of
dengue incidence in Bangladesh.
Oral presentation. The 2011 International Conference of International Society for
Environmental Epidemiology. Barcelona, Spain. 13–16, September 2011
Ye X, Tong S, Wolff R, Pan X, Guo Y, Vaneckova P. The effect of hot and cold
temperatures on emergency hospital admissions for respiratory and cardiovascular diseases in
Brisbane, Australia.
Oral presentation. The 2010 International Conference of International Society for
Environmental Epidemiology. Seoul, Korea August 28–September 1, 2010
xiv
Tong J, Su C, Guo Y, Wang J, Zhang M, Pan X. Study on the status and distribution of ultra-
fine particles during Beijing Olympics in 2008.
Oral presentation. The 2010 International Conference of International Society for
Environmental Epidemiology. Seoul, Korea August 28–September 1, 2010
xv
STATEMENT OF AUTHORSHIP
The work contained in this thesis has not been previously submitted for a degree or diploma
at any other higher education institute. To my best knowledge and belief, the thesis contains
no materials previously published or written by another person except where reference is
made.
Signature: ........................................................
xvi
ACKNOWLEDGEMENTS
I would like to thank the following people who have helped me make this thesis possible.
Without their help, I could not finish my PhD study.
Thanks to my supervisors, Professor Shilu Tong, Associate Professor Adrian Barnett, and
Professor Xiaochuan Pan for their experienced professional guidance. Professor Shilu Tong,
my principal supervisor, provided me with an opportunity to conduct my PhD study at QUT.
He gave me full freedom to realise my ideas for the PhD research and provided his best
support. He responded and revised my manuscripts and documents very promptly and
provided detailed feedback, even though he was very busy. Associate Professor Adrian
Barnett, my associate supervisor, spent a lot of time to teach me statistics. I learned greatly
from him, especially concerning the R language. I cannot forget how hard he helped me
check models at the regular meeting every week. I thank him for his flexibility and patience
for my PhD study. Professor Xiaochuan Pan, my external associated supervisor, was very
supportive and helped me greatly in data collection and data management.
I also want to thank Miss Xiaoyu Wang, Dr. Weiwei Yu and my other colleagues for their
help during my PhD study.
I would like to thank Queensland University of Technology for providing me with
scholarships to conduct my PhD study.
xvii
I want to thank School of Public Health and Institute of Health and Biomedical Innovation,
language advisors in learning and training department, High Performance Computer and
Research Support Unit, and IT help desk who have helped my research proceed smoothly.
I would like to thank my family and friends for their encouragement and care. They gave me
unconditional love and support.
xviii
LIST OF CONTENTS
ABSTRACT ............................................................................................................................... ii
KEY WORDS ........................................................................................................................... vi
PUBLICATIONS OR MANUSCRIPTS BY THE CANDIDATE ON MATTERS
RELEVENT TO THE THESIS ............................................................................................... vii
PUBLICATIONS OR MANUSCRIPTS BY THE CANDIDATE DURING PHD STUDY viii
CONFERENCE PRESENTATIONS ...................................................................................... xii
STATEMENT OF AUTHORSHIP ......................................................................................... xv
ACKNOWLEDGEMENTS .................................................................................................... xvi
LIST OF TABLES ................................................................................................................. xxii
LIST OF FIGURES .............................................................................................................. xxiv
LIST OF ABBREVIATION .............................................................................................. xxviii
CHAPTER 1: INTRODUCTION .............................................................................................. 1
1.1 BACKGROUND.............................................................................................................. 1
1.2 AIM AND OBJECTIVES ................................................................................................ 5
1.3 SIGNIFICANCE OF THE STUDY ................................................................................. 6
1.4 CONTENTS AND STRUCTURE OF THIS THESIS .................................................... 7
CHAPTER 2: THE EFFECTS OF AMBIENT TEMPERATURE ON MORTALITY: A
LITERATURE REVIEW .......................................................................................................... 8
2.1 CLIMATE CHANGE ...................................................................................................... 8
CLIMATE CHANGE, THE INDOOR ENVIRONMENT, AND HEALTH .......................... 10
2.2 THE RELATIONSHIP BETWEEN TEMPERATURE AND MORTALITY .............. 11
2.3 THE EFFECTS OF TEMPERATURE ON THE HUMAN BODY .............................. 15
2.4 MODELS FOR ASSESSING THE RELATIONSHIP BETWEEN TEMPERATURE
AND MORTALITY............................................................................................................. 20
xix
2.5 MODELS ASSESSING THE LAG EFFECTS OF TEMPERATURE IN MORTALITY
.............................................................................................................................................. 27
2.6 TEMPERATURE MEASURES AND MORTALITY .................................................. 31
2.7 INTERACTIVE EFFECTS BETWEEN TEMPERATURE AND AIR POLLUTION
ON MORTALITY ............................................................................................................... 33
2.8 GROUPS VULNERABLE TO TEMPERATURE EFFECTS ...................................... 34
2.9 SUMMARY ................................................................................................................... 37
2.10 REFERENCES ............................................................................................................. 38
CHAPTER 3: STUDY DESIGN AND METHODOLOGY ................................................... 57
3.1 STUDY POPULATION ................................................................................................ 57
3.3 DATA COLLECTION AND MANAGEMENT ........................................................... 61
3.4 DATA ANALYSIS ........................................................................................................ 64
3.4 RATIONALE FOR CHOOSING STUDY SITES OR TEMPERATURE MEASURES
.............................................................................................................................................. 67
3.5 REFERENCES ............................................................................................................... 69
CHAPTER 4: THE IMPACT OF TEMPERATURE ON MORTALITY IN TIANJIN,
CHINA: A CASE−CROSSOVER DESIGN WITH A DISTRIBUTED LAG NON-LINEAR
MODEL ................................................................................................................................... 71
4.2 INTRODUCTION.......................................................................................................... 74
4.3 MATERIALS AND METHODS ................................................................................... 76
4.4 RESULTS ...................................................................................................................... 80
4.5 DISCUSSION ................................................................................................................ 88
4.6 CONCLUSIONS ............................................................................................................ 94
4.7 REFERENCES ............................................................................................................... 95
4.8 SUPPLEMENTAL MATERIAL CHAPTER 4 ........................................................... 100
xx
CHAPTER 5: SPATIOTEMPORAL MODEL OR TIME SERIES MODEL FOR
ASSESSING CITY-WIDE TEMPERATURE EFFECTS ON MORTALITY? ................... 110
5.1 ABSTRACT ................................................................................................................. 111
5.2 INTRODUCTION........................................................................................................ 112
5.3 MATERIALS AND METHODS ................................................................................. 113
5.4 RESULTS .................................................................................................................... 120
5.5 DISCUSSION .............................................................................................................. 125
5.6 CONCLUSION ............................................................................................................ 130
5.7 REFERENCES ............................................................................................................. 131
5.8 SUPPLEMENTAL MATERIALS CHAPTER 5......................................................... 135
CHAPTER 6: A LARGE CHANGE IN TEMPERATURE BETWEEN NEIGHBOURING
DAYS INCREASES THE RISK OF MORTALITY ............................................................ 142
6.1 ABSTRACT ................................................................................................................. 143
6.2 INTRODUCTION........................................................................................................ 144
6.3 MATERIAL AND METHODS ................................................................................... 145
6.4 RESULTS .................................................................................................................... 148
6.5 DISCUSSION .............................................................................................................. 157
6.6 CONCLUSION ............................................................................................................ 161
6.7 REFERENCES ............................................................................................................. 162
6.8 SUPPLEMENTAL MATERIAL CHAPTER 6 ........................................................... 167
CHAPTER 7: ASSOCIATIONS BETWEEN HIGH TEMPERATURES AND ELDERLY
MORTALITY DIFFERED BY YEAR, CITY AND REGION IN THE UNITED STATES
................................................................................................................................................ 171
7.1 ABSTRACT ................................................................................................................. 172
7.2 INTRODUCTION........................................................................................................ 173
xxi
7.3 MATERIAL AND METHODS ................................................................................... 174
7.4 RESULTS .................................................................................................................... 178
7.5 DISCUSSION .............................................................................................................. 184
7.6 CONCLUSION ............................................................................................................ 188
7.7 REFERENCES ............................................................................................................. 189
7.8 SUPPLEMENTAL MATERIAL CHAPTER 7 ........................................................... 195
CHAPTER 8: GENERAL DISCUSSION ............................................................................. 198
8.1 METHODOLOGICAL DEVELOPMENT .................................................................. 198
8.2 IMPLICATION OF THE RESEARCH ....................................................................... 200
8.3 STRENGTHS OF THIS THESIS ................................................................................ 203
8.4 LIMITATIONS OF THIS THESIS ............................................................................. 205
8.5 RECOMMENDATIONS FOR FUTURE RESEARCH DIRECTIONS ..................... 206
8.6 CONCLUSIONS .......................................................................................................... 208
REFERENCES ...................................................................................................................... 210
xxii
LIST OF TABLES
TABLE 2.1: THE REPORTS FOR CLIMATE CHANGE AND HEALTH FROM LOCAL
GOVERNMENTS TO INTERNAL ORGANIZATIONS. ............................................. 10
TABLE 4.1: SUMMARY STATISTICS OF DAILY WEATHER CONDITIONS AND
MORTALITY IN TIANJIN, CHINA, 2005–2007 ................................................. 81
TABLE 4.2: SPEARMAN’S CORRELATION COEFFICIENTS BETWEEN WEATHER
CONDITIONS IN TIANJIN, CHINA, 2005–2007 ................................................. 81
TABLE 4.3: THE CUMULATIVE COLD AND HOT EFFECTS OF MEAN TEMPERATURE ON
MORTALITY CATEGORIES ALONG THE LAG DAYS, USING A “DOUBLE
THRESHOLD-NATURAL CUBIC SPLINE” DLNM WITH 4 DEGREES OF FREEDOM
NATURAL CUBIC SPLINE FOR LAG. .................................................................. 87
TABLE 5.1: SUMMARY STATISTICS FOR KRIGED TEMPERATURE, AVERAGED
TEMPERATURE, BRISBANE CENTRE’S TEMPERATURE, PM10, O3, RELATIVE
HUMIDITY, ELEVATION, AND MORTALITY IN BRISBANE CITY BETWEEN 2000
AND 2004 .................................................................................................... 124
TABLE 5.2: SPEARMAN CORRELATIONS BETWEEN KRIGED TEMPERATURE,
AVERAGED TEMPERATURE, BRISBANE CENTRE’S TEMPERATURE, PM10, O3,
AND RELATIVE HUMIDITY IN BRISBANE CITY BETWEEN 2000 AND 2004 ...... 126
TABLE 5.3: RELATIVE RISKS OF MORTALITY ASSOCIATED WITH HOT AND COLD
TEMPERATURES USING FOUR DIFFERENT MODELS ASSUMING A V-SHAPED
TEMPERATURE RISK WITH A THRESHOLD AT 28 °C ...................................... 127
xxiii
TABLE 6.1: SUMMARY STATISTICS FOR DAILY WEATHER CONDITIONS, AIR
POLLUTANTS, AND MORTALITY IN BRISBANE, AUSTRALIA AND LOS ANGELES,
UNITED STATES ........................................................................................... 149
TABLE 6.2: SPEARMAN’S CORRELATION BETWEEN DAILY WEATHER CONDITIONS
AND AIR POLLUTANTS IN BRISBANE, AUSTRALIA AND LOS ANGELES, UNITED
STATES ........................................................................................................ 150
TABLE 6.3: THE ASSOCIATIONS BETWEEN TEMPERATURE CHANGE AND MORTALITY
IN BRISBANE, AUSTRALIA AND LOS ANGELES, UNITED STATES.................. 151
TABLE 7.1: THE DISTRIBUTION OF YEARLY HIGH TEMPERATURE EFFECTS ON
ELDERLY MORTALITY BY REGION BETWEEN 1987 AND 2000 ....................... 179
TABLE 7.2: POOLED HIGH TEMPERATURE EFFECTS ON ELDERLY MORTALITY BY
REGION BETWEEN 1987 AND 2000 ............................................................... 180
xxiv
LIST OF FIGURES
FIGURE 2.1: COMMON SHAPES DESCRIBING THE RELATIONSHIP BETWEEN
TEMPERATURE AND MORTALITY. ................................................................... 12
FIGURE 2.2: FITTED DEATHS (SCALED TO BE A PERCENTAGE OF MEAN DAILY
DEATHS) IN SOFIA (LEFT) AND LONDON. PLOTTED AGAINST TEMPERATURE,
TWO DAY MEAN (TOP) AND TWO WEEK MEAN (PATTENDEN, NIKIFOROV, &
ARMSTRONG, 2003). ..................................................................................... 13
FIGURE 2.3: MODES OF HEAT TRANSFER BETWEEN HUMAN BODY AND
ENVIRONMENT. SOURCE: HTTP://WWW.THERMOANALYTICS.COM/HUMAN-
SIMULATION/THERMAL-MANIKIN. ................................................................. 16
FIGURE 2.4: HUMAN THERMOREGULATION WHEN PEOPLE EXPOSE TO HOT AND
COLD ENVIRONMENTAL TEMPERATURES. SOURCE:
HTTP://EXERCISEPHYSIOLOGIST.WORDPRESS.COM/2012/02/15/THE-HUMAN-
HOMOEOTHERMY/. ......................................................................................... 19
FIGURE 2.5: 3-D PLOT OF RR ALONG TEMPERATURE AND LAGS, USING DATA FROM
NMMAPS FOR CHICAGO DURING THE PERIOD 1987–2000. .......................... 28
FIGURE 2.6: PLOT OF OVERALL RR, USING DATA FROM THE NMMAPS FOR
CHICAGO DURING THE PERIOD 1987–2000. ................................................... 29
FIGURE 3.1: THE LOCATION OF TIANJIN, CHINA. .................................................. 58
FIGURE 3.2: THE LOCATION OF BRISBANE, AUSTRALIA. ...................................... 59
xxv
FIGURE 3.3: THE 83 LARGE CITIES AND 7 REGIONAL GROUPS IN UNITED STATES
FROM NMMAPS STUDY. .............................................................................. 60
FIGURE 4.1: RELATIVE RISKS OF MORTALITY TYPES BY MEAN TEMPERATURE (°C),
USING A NATURAL CUBIC SPLINE–NATURAL CUBIC SPLINE DLNM WITH 5 DF
NATURAL CUBIC SPLINE FOR TEMPERATURE AND 4 DF FOR LAG. (A)
NONACCIDENTAL, (B) CARDIOPULMONARY, (C) CARDIOVASCULAR, AND (D)
RESPIRATORY MORTALITY. ............................................................................ 83
FIGURE 4.2: THE ESTIMATED OVERALL EFFECTS OF MEAN TEMPERATURE (°C)
OVER 28 DAYS ON MORTALITY TYPES, USING A NATURAL CUBIC SPLINE–
NATURAL CUBIC SPLINE DLNM WITH 5 DF NATURAL CUBIC SPLINE FOR
TEMPERATURE AND 4 DF FOR LAG. (A) NONACCIDENTAL, (B)
CARDIOPULMONARY, (C) CARDIOVASCULAR, AND (D) RESPIRATORY
MORTALITY. THE BLACK LINES ARE THE MEAN RELATIVE RISKS, AND THE
BLUE REGIONS ARE 95% CIS. ........................................................................ 84
FIGURE 4.3: THE ESTIMATED EFFECTS OF A 1°C DECREASE IN MEAN TEMPERATURE
BELOW THE COLD THRESHOLD (LEFT) AND OF A 1°C INCREASE IN MEAN
TEMPERATURE ABOVE THE HOT THRESHOLD (RIGHT) ON MORTALITY TYPES
OVER 27 DAYS OF LAG, USING A DOUBLE THRESHOLD–NATURAL CUBIC SPLINE
DLNM WITH 4 DF NATURAL CUBIC SPLINE FOR LAG. (A) NONACCIDENTAL, (B)
CARDIOPULMONARY, (C) CARDIOVASCULAR, AND (D) RESPIRATORY
MORTALITY. THE BLACK LINES ARE MEAN RELATIVE RISKS, AND BLUE
xxvi
REGIONS ARE 95% CIS. THE COLD AND HOT THRESHOLDS WERE 0.8°C AND
24.9°C FOR NONACCIDENTAL MORTALITY (A), 0.1°C AND 25.3°C FOR
CARDIOPULMONARY MORTALITY (B), 0.6°C AND 25.1°C FOR
CARDIOVASCULAR MORTALITY (C), 0.7°C AND 24.8°C FOR RESPIRATORY
MORTALITY (D). ............................................................................................ 85
FIGURE 4.4: COMPARISON OF THE IMPACTS OF TEMPERATURE ON NONACCIDENTAL
MORTALITY IN DIFFERENT POPULATIONS ORDERED BY LATITUDE. ................. 89
FIGURE 5.1: THE 19 MONITORING SITES FOR TEMPERATURE IN OR AROUND
BRISBANE CITY, THE GREY REGIONS ARE STATISTICAL LOCAL AREAS OF
BRISBANE CITY, THE BLUE AREAS ARE WATER. ........................................... 116
FIGURE 5.2: MEAN DAILY MAXIMUM TEMPERATURES FOR THE 163 STATISTIC
LOCAL AREAS OF BRISBANE CITY BETWEEN JANUARY 2000 AND DECEMBER
2004. THE BLUE AREAS ARE WATER. ........................................................... 120
FIGURE 5.3: THE RELATIONSHIP BETWEEN TEMPERATURE AND MORTALITY IN
BRISBANE BETWEEN 2000 AND 2004, USING DIFFERENT MODELS WITH THREE
DEGREES OF FREEDOM FOR TEMPERATURE. ................................................. 121
FIGURE 5.4: THE RELATIONSHIP BETWEEN TEMPERATURE AND MORTALITY IN
BRISBANE BETWEEN 2000 AND 2004, USING DIFFERENT MODELS WITH FOUR
DEGREES OF FREEDOM FOR TEMPERATURE. ................................................. 122
FIGURE 6.1: THE ASSOCIATIONS BETWEEN TEMPERATURE CHANGE AND NON-
EXTERNAL MORTALITY, CARDIOVASCULAR MORTALITY, AND RESPIRATORY
xxvii
MORTALITY USING MODEL (6.1) IN BRISBANE, AUSTRALIA (LEFT SIDE) AND
LOS ANGELES, UNITED STATES (RIGHT SIDE). ............................................. 153
FIGURE 6.2: THE ASSOCIATIONS BETWEEN TEMPERATURE CHANGE AND NON-
EXTERNAL MORTALITY BY AGE GROUP USING MODEL (6.1) IN BRISBANE,
AUSTRALIA (LEFT SIDE) AND LOS ANGELES (RIGHT SIDE), UNITED STATES.
.................................................................................................................... 154
FIGURE 6.3: BIVARIATE RESPONSE SURFACES OF THE TEMPERATURE CHANGE AND
MEAN TEMPERATURE FOR NON-EXTERNAL MORTALITY, SUBGROUPS OF
MORTALITY USING MODEL (6.3) IN BRISBANE, AUSTRALIA. ........................ 155
FIGURE 6.4: BIVARIATE RESPONSE SURFACES OF THE TEMPERATURE CHANGE AND
MEAN TEMPERATURE FOR NON-EXTERNAL MORTALITY, SUBGROUPS OF
MORTALITY USING MODEL (6.3) IN LOS ANGELES, UNITED STATES. ........... 156
FIGURE 7.1: BOXPLOTS OF THE YEARLY HIGH TEMPERATURE EFFECTS ON ELDERLY
MORTALITY BY CITIES BETWEEN 1987 AND 2000. CITY ABBREVIATIONS ARE
EXPLAINED IN SUPPLEMENTAL MATERIAL CHAPTER 7, TABLE S7.1. .......... 181
FIGURE 7.2: MEAN HIGH TEMPERATURE EFFECTS ON ELDERLY MORTALITY BY
REGION BETWEEN 1987 AND 2000 USING A UNIVARIATE META-ANALYSIS. . 182
FIGURE 7.3: TREND IN THE EFFECTS OF HIGH TEMPERATURES ON THE ELDERLY
MORTALITY BY REGION BETWEEN 1987 AND 2000 USING A BAYESIAN
HIERARCHICAL MODEL. ............................................................................... 183
xxviii
LIST OF ABBREVIATION
ACF
Auto-correlation function
AIC
Akaike’s information criterion
CI
Confidence interval
CVM
Cardiovascular mortality
DF
Degree of freedom
DLNM
Distributed lag non-linear model
GAM
Generalized additive model
GAMM
Generalized additive mixed model
ICD
International Classification of Diseases
IPCC
Intergovernmental Panel on Climate Change
NEM
Non-external mortality
NMMAPS The National Morbidity Mortality Air Pollution Study
NO2
Nitrogen dioxide
O3
Ozone
PACF
Partial auto-correlation function
PM10
Particulate matter less than 10 μm in aerodynamic diameter
RM
Respiratory mortality
RR
Relative risk
SO2
Sulphur dioxide
UNEP
United Nations Environment Programme
WHO
World Health Organization
WMO
World Meteorological Organization
1
CHAPTER 1: INTRODUCTION
1.1 Background
Climate change is one of the most serious challenges for human health in the 21st century, as
it will directly or indirectly affect most populations (Costello et al., 2009). Future climate
change will increase the frequency, intensity and duration of heat waves (IPCC, 2007a).
Heat-related mortality has become a matter of increasing public health significance, as
climate change continues.
Studies have examined hot and cold temperatures in relation to total non-accidental deaths
and cause-specific deaths (Stafoggia, Forastiere, Agostini, Biggeri, Bisanti, Cadum, Caranci,
de' Donato, et al., 2006). Exposure to both low and high ambient temperatures increases the
risk of death and therefore the temperature-mortality relations appear J-, V- or U-shaped,
with thresholds corresponding to the lowest mortality. Temperature thresholds for an
increased mortality are generally higher in warmer climates (Patz, Campbell-Lendrum,
Holloway, & Foley, 2005; Yu, Mengersen, Wang, et al., 2011), as people adapt to their local
climates, through physiological, behavioural and cultural adaptation.
Many personal and environmental factors may modify the effects of temperature on human
health, including age, gender, chronic disease, economic disadvantage, demographic factors,
intensity of urban heat islands, housing characteristics, access to air conditioning and
availability of health care services (Kovats & Hajat, 2008). Populations in developing
countries are anticipated to be especially sensitive to impacts of climate change, as they have
limited adaptive capacity and more vulnerable people (Costello, et al., 2009).
2
Previous studies have identified that extreme temperatures have impacts on vulnerable people
(Basu, 2009b; Basu & Ostro, 2008a; Kovats & Hajat, 2008). The elderly and women are
particularly vulnerable to extreme temperatures (Hajat, Kovats, & Lachowycz, 2007b; P.
Vaneckova, Beggs, de Dear, & McCracken, 2008). People with particular diseases such as
cardiovascular, respiratory problems, diabetes, mental disorders are more sensitive to
temperature than healthy people (Basu, Dominici, & Samet, 2005; McMichael et al., 2008;
Stafoggia et al., 2006).
There are a few challenges in the assessment of temperature-mortality relationship. The
frequency, intensity and duration of weather extremes (e.g. heat waves, floods and cyclones)
are projected to increase as climate change continues (WHO/WMO/UNEP, 1996), and
unstable weather patterns (e.g. a significant drop/increase in temperature) are also more likely
to occur in the coming decades (Faergeman, 2008). However, less evidence is available on
the possible mortality effects due to temperature change between neighbouring days.
Epidemiological studies on heat-related mortality could be used by decision makers to
establish a warning system for high temperatures, by giving information on the heat threshold
and the expected increase in deaths above the threshold. Such studies are also useful for
estimating the potential health effects of climate change. However, most previous studies
only considered high temperature effects by averaging over the whole study period, and
ignored the variability in effects from year to year. Effects may vary from year to year
because of differences in the at-risk population (e.g., more elderly people), or because of
increased adaptation over time (Sheridan & Kalkstein; Stafoggia, Forastiere, Michelozzi, &
Perucci, 2009a).
3
Time series and case–crossover analyses are the most common methods used to estimate the
short-term effects of temperature (or air pollution) on health (Fung, Krewski, Chen, Burnett,
& Cakmak, 2003; J. Schwartz, 2004). Time series analysis allows for over dispersion
associated with the Poisson distribution and controls for long-term trend and seasonality
using nonparametric or parametric splines. The case−crossover design controls for seasonal
effects and secular trends by matching case and control days in relatively small time windows
(e.g., calendar month). This controls for season using a step-function rather than a smooth
spline function (Barnett & Dobson, 2010).
Most previous studies used the case–crossover design with relatively inflexible models to
investigate the effects of temperature on mortality, such as assuming a linear effect for
temperature in each season, with a single lag model, or moving average lag model (Basu,
Feng, & Ostro, 2008; Green et al., 2010). No study has examined non-linear and delayed
effects of temperature on mortality within a case–crossover design.
To estimate the impact of temperature on mortality, most studies used daily temperature data
from one monitoring site or daily mean values from a network of monitoring sites, which
may result in a measurement error for temperature exposure (Zhang et al., 2011). Studies
have shown that there is spatial variation in outdoor temperatures within cities and their
surroundings (Aniello, Morgan, Busbey, & Newland, 1995; Kestens et al., 2011; Lo,
Quattrochi, & Luvall, 1997). Urban areas usually have higher temperatures because of the
heat island effect (www.epa.gov/hiri/about/index.html). Smargiassi et al. (2009) found that
hotter areas within a city had a greater risk of heat-related deaths compared with cooler areas.
These results suggest that using temperature from one monitoring site or averaged values
4
from a network of sites may underestimate the risks of temperature on mortality. Geo-
statistical techniques have been used to model regional temperatures. Previously, these
models have been used to estimate the health effects of air pollution within cities (Lee &
Shaddick, 2010; Shaddick, Lee, Zidek, & Salway, 2008; Whitworth, Symanski, Lai, & Coker,
2011). However, few studies have used spatial methods to quantify the impact of temperature
on mortality (Kestens, et al., 2011; Smargiassi, et al., 2009).
5
1.2 Aim and objectives
Aim
This thesis aims to address methodological issues when examining the relationship between
temperature and mortality, and to assess the effects of temperature change between
neighbouring days on mortality, and to examine the variation in the effects of high
temperatures on mortality.
Specific objectives
1.2.1 To combine the case-crossover design and distributed lag non-linear model, I examined
the effects of temperature on cause-specific mortality in Tianjin, China.
1.2.2 To compare time series and spatiotemporal analyses, I examined the effects of
temperature on mortality in Brisbane, Australia.
1.2.3 To estimate the association between temperature change and mortality, I used time
series Poisson regression models for data from Brisbane, Australia and Los Angeles, United
States.
1.2.4 To examine the variation in the effects of high temperatures on elderly mortality
(age ≥ 75 years) by year, city, and region, I used data from 83 large US cities.
6
1.3 Significance of the study
This thesis will help gain a better understanding of some key methodological issues in the
assessment of the temperature effects on mortality. It will supply significant information for
future research on heat-related mortality concerning the model choice.
I also examined the effects of temperature change between neighbouring days on mortality,
which will provide additional information for the development of government policies and
public health strategies concerning temperature effects. In addition, I assessed the variation in
temperature effects on elderly mortality across years, which can be used to promote capacity
building for public health adaptation to cope with high temperature effects.
This thesis estimated the impacts of temperature on mortality in Tianjin. It provides
significant insights to assist policy makers in planning and communicating the health risks of
temperatures to the public in Tianjin. It may also promote considerations on capacity building
for adaptation in the face of extreme temperatures, by broadening and deepening the mindset
of stakeholders involved in the process of exploring possible futures.
Evidence-based assessment of temperature effects on mortality is a key challenge to both
society and government decision makers worldwide. In this thesis, I used most advanced
models to quantify both linear and non-linear, acute and delayed effects of temperature on
mortality. Additionally, I examined whether spatiotemporal model is better than time series
model in assessing temperature-mortality relationship. These findings may make a
contribution to this internationally important but challenging field.
7
1.4 Contents and structure of this thesis
I wrote this thesis using a publication style based on four manuscripts.
Chapter 2 provides a literature review for the temperature effects on mortality. This chapter
summarises previous research findings and current knowledge gaps when examining the
effects of temperature on mortality. Chapter 3 introduces the study design, materials and data
analysis.
The four manuscripts are presented in chapters 4–7. They are written in their conventional
publication style according to each particular journal. Chapter 4 combines a case–crossover
design with distributed lag non-linear model to examine the effects of temperature on cause-
specific mortality in Tianjin, China. Chapter 5 compares the time series and spatiotemporal
analyses for the effects of temperature on mortality in Brisbane, Australia. Chapter 6
examines the effects of temperature change between neighboring days on cause-specific
mortality in Brisbane, Australia, and Los Angeles, United States. Chapter 7 examines the
variation in high temperature effects on elderly mortality across years in 83 large cities in
United States. Chapter 8 discusses overall significance, limitations and probable implications.
The conclusions are made based on results of the four manuscripts. Some directions for
future research are also proposed.
8
CHAPTER 2: THE EFFECTS OF AMBIENT TEMPERATURE ON
MORTALITY: A LITERATURE REVIEW
2.1 Climate change
Global climate is rapidly changing within the last century, due to greenhouse gas emissions
largely driven by human activity. The Intergovernmental Panel on Climate Change has
concluded that warming of the climate system is unequivocal (IPCC, 2007b). From 1906 to
2005 the planet’s average temperature has increased by 0.74 °C, and the temperature has
increased by 0.55 °C from 1982 to 2007. Climate change is projected not only to increase the
global average temperature by between 1.1 °C and 6.4 °C by 2100, but also to increase the
frequency of extreme weather events (e.g., heat waves, cyclones and storms) (Medina-Ramón
& Schwartz, 2007).
Observational evidence from around the world shows that many systems are already being
affected by climate change, particularly temperature increases (IPCC, 2007a). Not only is
climate change an environmental issue, but it also affects human health directly and indirectly
through various pathways (e.g., extreme temperature, floods, droughts and infectious
diseases).
Climate change potentially affects every person, but there will likely be a greater impact on
vulnerable people (e.g., elderly, children, and people with chronic diseases) by climate
change (WHO, 2008). Vulnerability not only depends on the level of climate change, but also
population characteristics (e.g., age, gender and adaptation ability). The elderly and children
are more sensitive and have less adaptive capacity to deal with climate change (e.g., heat
9
waves, cold spells) (Adger, 2006; IPCC, 2007a). The health impacts will depend on the rate
and magnitude of changes in climate, and will be modified by social, economic, demographic
and infrastructure factors. All these factors can influence the sensitivity of populations to
climate change, and their adaptive capacity to manage the health effects of climate change
(Ebi, Kovats, & Menne, 2006a; Haines, Kovats, Campbell-Lendrum, & Corvalan, 2006;
Kovats & Hajat, 2008; Reid et al., 2009; Rey et al., 2009).
Policy makers from local and national governments to international organizations have been
increasing awareness of the influence of climate change on human health, and developed a
number of programs towards climate change mitigation and adaptation, and projected the
impacts of future climate change (Table 2.1).
There has been a growing interest in assessing how climate change influences health,
especially for relationships between ambient temperature and mortality and morbidity
(Barnett, Tong, & Clements, 2010; Michelozzi et al., 2006; Stafoggia, et al., 2009a). This
literature review assesses the effects of ambient temperature on mortality, focusing on the
methodological challenges and research opportunities in examining the relationship between
temperature and health.
10
Table 2.1: The reports for climate change and health from local governments to internal
organizations.
Organisation Year Title
Queensland,
Australia
2011
Climate change: Adaptation for Queensland
PwC Australia 2011
Protecting human health and safety during severe and
extreme heat events: A national framework
Marmot Review
Team, UK
2011
The health impacts of cold homes and fuel poverty
Committee on the
Effect of Climate
Change on Indoor
Air Quality and
Public Health, USA
2011
Climate change, the indoor environment, and health
WHO 2009 Protecting health from climate change: Connecting science,
policy and people
IPCC 2012 Managing the risks of extreme events and disasters to
advance climate change adaptation
11
2.2 The relationship between temperature and mortality
Evidence shows that temperature can directly or indirectly impact human health (Alderson,
1985; Baker-Blocker, 1982; Rogot & Blackwelder, 1970). Exposure to extreme temperatures
(heat waves or cold spells) is related to both mortality and morbidity (Luber & McGeehin,
2008). A number of epidemiological studies have reported both cold and high temperatures
were associated with non-accidental deaths (Baccini et al., 2008; Curriero et al., 2002a;
McMichael, et al., 2008; Stafoggia, et al., 2006), cause-specific deaths (Barnett, 2007; Pan,
Li, & Tsai, 1995; Rey et al., 2007), and other health outcomes such as emergency hospital
visits and hospital admissions (Hansen et al., 2008; Knowlton et al., 2009; Smith, Coyne,
Smith, & Mercier, 2003; Wang, Barnett, Hu, & Tong, 2009).
Extreme temperatures have significant impacts on health (Kovats & Hajat, 2008). For
example, over 700 excess people died during the 1995 Chicago heatwave (Semenza et al.,
1996a). Heatwaves in 2003 caused 15,000 excess deaths in France alone (Fouillet et al., 2007;
Tertre et al., 2006), and over 70,000 deaths across Europe countries (Conti et al., 2005;
Johnson et al., 2005). There were 274 excess cardiovascular deaths during the 1987 Czech
Republic cold spells (Kysely, Pokorna, Kyncl, & Kriz, 2009), and 370 excess deaths occurred
during the 2006 Moscow cold spells (Revich & Shaposhnikov, 2008a).
The relationship between temperature and mortality tend to be V-, U- or J-shaped in most
studies (Figures 2.1 and 2.2), with thresholds corresponding to the lowest mortality or
morbidity (Curriero et al., 2002; Kalkstein & Davis, 1989). The base of the V-, U- or J-shape
is the temperature (or temperature range) at which mortality rates are smallest, and from
12
which mortality levels will increase if the temperature increases or decreases (Kalkstein &
Davis, 1989).
Figure 2.1: Common shapes describing the relationship between temperature and mortality.
Cold and hot thresholds are generally higher in cities closer to the equator (Patz, et al., 2005),
as people have acclimatised to their local climates, through physiological, behavioural and
cultural adaptation. The optimum temperature varies according to the population, region and
climate type. For example, in the Netherlands during 1979–1997, the optimum temperature
13
was 16.5 °C for minimising non-accidental, cardiovascular and respiratory mortality, while
the optimum temperature in the younger age group was 15.5 °C for mortality due to
malignant neoplasm and 14.5 °C for non-accidental mortality (Huynen, 2001).
Figure 2.2: Fitted deaths (scaled to be a percentage of mean daily deaths) in Sofia (left) and
London. Plotted against temperature, two day mean (top) and two week mean (Pattenden,
Nikiforov, & Armstrong, 2003).
Several methods have been used to find cold and hot thresholds. One simple way is to plot
the association between temperature and mean mortality at each temperature. Visual
14
inspection can then be used to find the temperature threshold (Donaldson, Keatinge, &
Nayha, 2003). Many recent studies have used splines (such as natural cubic spline,
polynomial, B-spline, penalised spline) to examine the effects of temperature on mortality
and morbidity. Smoothed curves for the temperature-mortality relationship are estimated and
plotted using spline functions. Temperature points that correspond to the lowest mortality risk
were usually chosen as the temperature thresholds (El-Zein, Tewtel-Salem, & Nehme, 2004).
In recent years, advanced statistical methods have been developed to find temperature
thresholds. These models also considered the lag effects of temperature. A segmented method
was developed to find temperature thresholds (Michelozzi et al., 2006; Muggeo, 2003).
Another way is assuming the temperature-mortality relationship is linear in different seasons.
So the temperature thresholds were not tested, but instead the impact of temperature was
examined separately in Spring, Summer, Autumn, and Winter (Basu & Samet, 2002; Carson,
Hajat, Armstrong, & Wilkinson, 2006). Recently, studies used low and high percentiles of
temperature (e.g., 99th
against 90th
, 1st against 10
th) to examine cold and hot effects on
mortality (Anderson & Bell, 2009), as sometimes the non-linear function for temperature
might adequately capture the effect of temperature on mortality.
15
2.3 The effects of temperature on the human body
The human adjusts the core body temperature within a narrow range around 37 °C, and this
system is independent to the fluctuations in the ambient temperature (Sessler, 2009). The
human body constantly exchanges heat with its surrounding environments to keep the core
body temperature constant (Figure 2.3). Evaporation, radiation, conduction or convection are
the four main modes of heat transfer ("Metabolism, Energy Balance, and Temperature
Regulation," 2008). In a normal environment, about 30% of the total heat exchange of the
human body is by convection, and each person evaporates about 1 litre per day and dissipates
about one-quarter of the total daily loss of heat. Heat exchange through conduction and
radiation depends on the conductivity of objects and materials in contact with the skin, or
temperature difference between the skin and adjacent surfaces (Kroemer & Grandjean, 1997).
16
Figure 2.3: Modes of heat transfer between human body and environment. Source:
http://www.thermoanalytics.com/human-simulation/thermal-manikin.
17
Thermoregulation is a very complex process (Figure 2.4). When people are exposed to
extreme high temperatures above the body’s core temperature (i.e., 37 °C), blood may flow
increase to the skin’s vessels to increase heat loss which is simulated by homeostatic control.
When people are exposed to high temperatures for long periods the sweat glands may not
work well, and metabolic reaction may slow down (Jiang, Qu, Shang, & Zhang, 2004).
During continuous exposure to heat, the central nervous blood volume decreases as the
coetaneous vessels dilate. The stroke volume falls, while the heart rate increases to maintain
cardiac output. The effective circulatory volume also decreases as water is lost through
sweating (Parsons, 2003). The decrease in sweating promotes a further increase in core
temperature to beyond 38–39 °C where collapse of homeostatic control may occur.
Hyperthermia happens when the body temperature reaches about 40 °C (Axelrod & Diringer,
2006; http://en.wikipedia.org/wiki/Hyperthermia"), and heat stroke may occur when the
temperature is above 41 °C (Parsons, 2003).
Extreme high temperatures induced an acute event in people with previous myocardial
infarction or stroke (Muggeo & Hajat, 2009). The extra heat load can be fatal for people with
congestive heart failure (Näyhä, 2005). Exposure to high temperatures might cause
dehydration, salt depletion and increased surface blood circulation, which can lead to a
failure of thermoregulation (Bouchama & Knochel, 2002). High temperatures may also be
associated with elevated blood viscosity, cholesterol levels and sweating thresholds
(McGeehin & Mirabelli, 2001).
When people are exposed to cold temperatures, skin vessels contract and muscles shiver to
generate energy to maintain core body temperature (Figure 2.4). Shivering is an effective way
to increase the body’s heat production (Clark & Edholm, 1985). During prolonged cold
18
exposure, the body can activate another slower mechanism, by increasing the thyroid
hormone in the blood stream from the thyroid gland. The thyroid hormone reaches all the
cells of the body and increases their metabolic activity which increases heat production
(Wyndham, 1969). When the air temperature drops extensively in a short time the body finds
it difficult to cope. Hypothermia can occur when the air temperature drops the body
temperature below 35 °C (Parsons, 2003). The risk of death would be increased if the core
body temperature drops below 32 °C. When the core body temperature is less than 28 °C, the
life is threatened immediately if there is not any medical attention (Parsons, 2003).
Cold temperatures are also significantly associated with human health (Huynen, Martens,
Schram, Weijenberg, & Kunst, 2001; Kysely, et al., 2009). Cold temperatures increase the
rates of myocardial ischemia, myocardial infarction and sudden deaths (Hong et al., 2003;
Stewart, McIntyre, Capewell, & McMurray, 2002). Exposure to cold temperatures is
associated with an increase in blood pressure, blood cholesterol, heart rate, plasma fibrinogen,
platelet viscosity and peripheral vasoconstriction, (Ballester, Corella, Perez-Hoyos, Saez, &
Hervas, 1997a; Carder et al., 2005b). Skin cooling increases systematic vascular resistance,
heart rate and blood pressure.
19
Figure 2.4: Human thermoregulation when people expose to hot and cold environmental
temperatures. Source: http://exercisephysiologist.wordpress.com/2012/02/15/the-human-
homoeothermy/.
20
2.4 Models for assessing the relationship between temperature and mortality
A variety of models have been used to assess the impacts of temperature on mortality and
morbidity, such as descriptive models (Reid, et al., 2009), case-only models (Schwartz, 2005),
case-crossover models (Stafoggia, et al., 2006), time-series models (Hajat, Kovats, Atkinson,
& Haines, 2002) and spatial models (Vaneckova, Beggs, & Jacobson, 2010). In general, time-
series and case-crossover analyses are the most commonly used in a single or in multiple
locations over a time period from years to decades (Basu, et al., 2005). The main aim of these
analyses is to examine associations between health and the exposure (e.g. daily counts of
death and daily temperatures), after controlling for potential confounders such as temporal
trends and seasonality (Kinney, O'Neill, Bell, & Schwartz, 2008; Kovats & Hajat, 2008).
Season and long-term trends are considered as confounders in examining short-term effects
of temperature on deaths. Most previous studies have tried to control for both seasonality and
long-term trend (Rose, 1966; Anderson & Rochard, 1979). Some studies separated the data
into four seasons (spring, summer, autumn, and winter) to control for season. Time series
methods with a smooth function for calendar time are now commonly used to control for
season and long-term trend (Dominici, McDermott, Zeger, & Samet, 2002; El-Zein, et al.,
2004; Hales, Salmond, Town, Kjellstrom, & Woodward, 2000b; Kim & Jang, 2005; Revich
& Shaposhnikov, 2008). The other design is the case-crossover which controls for seasonal
effects and secular trends by matching case and control days in relatively small time windows
(e.g., calendar month). This controls for season using a step-function rather than the smooth
function used by time series (Barnett & Dobson, 2010).
2.4.1 Time series design
21
In the following sections I review the main statistical methods used to estimate the health
effects of temperature. Most studies in this area use daily data on deaths from a city or region
with a similar climate. These data sets are usually between 2 and 20 years long. The data are
time series of daily deaths and temperature, hence time series methods are usually applied.
The generalised linear time series model and generalised additive time series model are most
commonly used to examine the effects of temperature on mortality. Both models used
smoothing for calendar time to control for season and long-term trend.
The generalised linear model was first proposed by Nelder and Wedderburn in 1972 (Nelder
& Wedderburn, 1972). Generalised linear models can be used to fit regression models to non-
normal data with a minimum of extra complication compared with normal linear regression.
Generalised linear models are an extension of multiple linear models and are flexible enough
to include a wide range of common situations, including normal linear regression. This model
generalised the classic linear model to four distributions: normal, binomial (probit analysis,
etc.), Poison and Gamma, which can be transformed from an exponential family and link
functions to a linear basis (Nelder & Wedderburn, 1972). Zeger first used generalised linear
models to examine the effects of weather on human health in 1988 (Zeger, 1988).
The generalised additive model was developed by Hastie and Tibshirani to combine the
generalised linear model and additive model (Hastie & Tibshirani, 1990). The purpose was to
maximise the quality of prediction for various distributions, by using non-parametric
smoothing functions of the independent variables which are "connected" to the dependent
variable via a link function (Dominici, et al., 2002). Most smoothers attempt to mimic
category averaging through local averaging, that is, averaging the response-values of
22
observations having predictor values close to a target value. The averaging is done in
neighbourhoods around the target value (Hastie & Tibshirani, 1990).
Two main decisions need to be made when using a smoothing function: how to average the
response values in each neighborhood, and how big to make the neighborhoods. The former
is the question of which type of smoothing to use, and the latter concerns the degrees of
freedom expressed in terms of an adjustable smoothing parameter (Hastie, Tibshirani, &
Friedman, 2004). Methods for smoothing include nonparametric splines such as smoothing
splines, locally-weighted running-line smoothers (loess) and kernel splines; and parametric
splines such as natural cubic regression and B-splines (Hastie & Tibshirani, 1990).
Selecting an appropriate model is very important when examining the effects of temperature
on mortality, as the choice of model impacts on the prediction ability. Model choice can be
informed by model fit criteria such as residual testing, deviance statistics, Akaike’s
information criterion (AIC), or the partial auto-correlation function (PACF) using the
residuals to determine the degree of remaining autocorrelation (Gouveia, Hajat, &
Armstrong, 2003). Residual plots are also useful for finding outliers and non-random
variations.
The model selection sometimes depends on the study design. In general, simple models are
easy to interpret. The simple models are particularly attractive in multicity to compare
associations across cities. Complex models are sometimes required to better fit the data. The
flexible models are useful for single-city studies and can indicate to what extent there are
systematic effects of temperature beyond that captured in the simple model.
23
2.4.2 Case-crossover design
The case–crossover design has been used widely to examine the effects of temperature on
mortality and morbidity in the past decade (Basu, 2009b). The case–crossover design is a
special case of matched case-control study; each case in the case-crossover study is used as
their own control. Therefore, confounders related to individual characteristics that remain
relatively constant (e.g., age, sex and smoking) are controlled for by design.
For the time series data on deaths and temperature, the case–crossover design compares
temperatures on a case day when events occurred (e.g., deaths) with temperatures on nearby
control days to examine whether the events are associated with temperature. Because control
days are selected close to the case days, seasonality is controlled by design (which makes it
useful for studying the effects of short-term changes in temperature). There are many
different designs for choosing control days relative to a case day.
A unidirectional design selects fixed control day(s) per case day only before or after the case
day (with all controls either being selected before the case, or all controls selected after the
case). This design may not control for trends over time in temperature or health outcomes,
and so is subject to bias (Greenland, 1996).
Bidirectional designs include the full-stratum bidirectional (Navidi, 1998), symmetric
bidirectional (Bateson & Schwartz, 1999), and semi-symmetric case–crossover (Navidi &
Weinhandl, 2002). The full-stratum bidirectional case–crossover includes control days as all
days in the time series before and after the case days. This design controls for time trends in
24
exposure, but does not control for seasonal patterns in exposure or health outcomes (Bateson
& Schwartz, 1999).
The symmetric bidirectional design uses control days both before and after the case day. This
method can successfully control for seasonality in exposures and outcomes. However, there is
the potential for selection bias, because the case days at the beginning or end of the data
series have fewer control days for matching. Navidi and Weinhandl (2002) noted that the
symmetric case–crossover design might still be biased by time trends in the exposure.
The semi-symmetric design randomly selects a control day before or after the case day. This
design can also control for long-term trends and seasonality. However, because only one
control day is selected at a fixed interval, the estimates may still be biased (Levy et al., 2001).
Lumley and Levy (2000) illustrated how selection biases do not appear when cases may
occur at any time in the strata from which the controls are selected (time-stratified case–
crossover). Lumley and Levy also demonstrated that most of the other designs are biased
because the controls are not chosen independently of the case day. This bias is called the
‘overlap bias’ and occurs in case–crossover designs with non-disjointed strata (Lumley &
Levy, 2000).
The time-stratified case–crossover uses fixed and disjointed time strata (e.g., calendar month),
so the overlap bias is avoided. Janes et al. (Janes, Sheppard, & Lumley, 2005) demonstrated
that the overlap bias is not an issue for the time-stratified design. Time-stratified case–
crossover analyses are equivalent to time series analyses (Basu, et al., 2005; Fung, et al.,
2003).
25
Conditional logistic regression is usually used to estimate the model parameters for a case-
crossover design. The conditional logistic regression used in case–crossover analysis is a
special case of time series log-linear model (Lu, Symons, Geyh, & Zeger, 2008; Lu & Zeger,
2007). Hence log-linear models can also be used to estimate the parameters for a case–
crossover design. The advantage of log-linear model is that we can obtain the model residuals,
which can be examined to evaluate the adequacy of the model.
Most previous studies used the case–crossover design with relatively inflexible models to
investigate the effects of temperature on mortality, such as assuming a linear effect for
temperature in each season, with a single lag model, or a moving average lag model (Basu, et
al., 2008; Green, et al., 2010). Few studies have demonstrated how to fit non-linear and
delayed effects of temperature on mortality within a case–crossover design.
2.4.3 Spatiotemporal analysis
To estimate the impact of temperature on mortality, most studies used daily temperature data
from one monitoring site or daily mean values from a network of sites, which may result in a
measurement error for temperature exposure (Zhang, et al., 2011). Random measurement
error in temperature will bias the effect estimates towards the null (Hutcheon, Chiolero, &
Hanley, 2010). Studies have shown that there is spatial variation in ambient temperatures
within cities and their surroundings (Aniello, et al., 1995; Kestens, et al., 2011; Lo, et al.,
1997). Urban areas usually have higher temperatures because of the heat island effect
(www.epa.gov/hiri/about/index.html). Smargiassi et al. (2009) found that hotter areas within
a city had a greater risk of heat-related death compared with cooler areas (Smargiassi, et al.,
2009). These results suggest that using temperature from one monitoring site or averaged
26
values from a network of sites may underestimate the risks of temperature on mortality.
However, few studies have used spatial methods to quantify the impact of temperature on
mortality (Kestens, et al., 2011; Smargiassi, et al., 2009). If spatial exposures of temperature
are significantly more accurate than standard methods then they may improve our
understanding of the association between temperature and mortality.
Geo-statistical techniques have been used to model regional temperatures (Benavides,
Montes, Rubio, & Osoro, 2007; Zhang, et al., 2011). Recent studies have used spatial models
to examine climate variables like ambient temperature in the field of agriculture and forestry
science (Benavides, et al., 2007; Chuanyan, Zhongren, & Guodong, 2005). Different
techniques (inverse distance interpolation weighting, voronoi tessellation, regression analysis,
and geo-statistical methods) have been developed to predict regional temperature from station
data (Bhowmik & Cabral, 2011). These models have also been used to estimate the health
effects of air pollution within cities (Lee & Shaddick, 2010; Shaddick, et al., 2008;
Whitworth, et al., 2011). However, few studies have used spatiotemporal models to examine
the association between temperature and mortality.
27
2.5 Models assessing the lag effects of temperature in mortality
Mortality risk depends not only on exposure to the current day’s temperature, but also on
several previous days’ exposure (Anderson & Bell, 2009). The distributed lag model has been
applied to explore the delayed effect of temperature on mortality (Analitis et al., 2008;
Baccini, et al., 2008; Hajat, Armstrong, Gouveia, & Wilkinson, 2005). To overcome the
strong correlation between temperatures in the close days, constrained distributed lag
structures are used (Armstrong, 2006). The estimates are constrained by smoothing using
methods such as natural cubic splines, polynomials or stratified lags. Both unconstrained and
constrained distributed lag models assume a linear relationship between temperature below
(above) the cold (hot) threshold and mortality, so these models may not be sufficiently
flexible to capture the non-linear effects of temperature on mortality.
Recently, a distributed lag non-linear model (DLNM) was developed to simultaneously
estimate the non-linear and delayed effects of temperature on mortality (or morbidity)
(Armstrong, 2006; Gasparrini, Armstrong, & Kenward, 2010). DLNMs use a “cross-basis”
function that describes a two-dimensional temperature-response relationship along the
dimensions of temperature and lag. The choice of “cross-basis” functions for the temperature
and lag are independent, so spline or linear functions can be used for temperature, while the
polynomial functions can be used for the lag. The estimates can be plotted using a 3-
dimensional graph to show the relative risks along both temperature and lags (Figure 2.5).
We can estimate the relative risks for a certain temperature or lag, by extracting a “slice”
from the 3-dimensional graph. We can compute the overall effect by summing the log
relative risks of all lags (Figure 2.6).
28
Figure 2.5: 3-D plot of RR along temperature and lags, using data from NMMAPS for
Chicago during the period 1987–2000.
29
Figure 2.6: Plot of overall RR, using data from the NMMAPS for Chicago during the period
1987–2000.
30
One of the main advantages of DLNM is that it allows the model to contain detailed lag
effects of exposure on response, and provides the estimate of the overall effect that is
adjusted for harvesting (for example, a heat wave was followed by a decrease in mortality
during the subsequent days or weeks) (Gasparrini, et al., 2010). The DLNM can flexibly
show different temperature-mortality relationships for lags using smoothing functions. The
DLNM can adequately model the main effects of temperature (Armstrong, 2006).
There are also some issues in the selection of the DLNM, such as cross-basis type, maximum
lag, and degrees of freedom (knots and placement) for exposure and lag (Armstrong, 2006;
Gasparrini, et al., 2010). These values are generally pre-defined by the researcher based on
previous papers and biological plausibility. Because the DLNM is combined with a
regression model (e.g., Poisson regression), the residual deviance and autocorrelation plot,
maximum likelihood, Akaike’s information criteria or Bayesian information criteria can be
used to check the model, and the appropriateness of the selected lag and degrees of freedom.
The lag and degrees of freedom can be chosen according to the best model fit (e.g., AIC).
Previous studies recommend choosing a DLNM that is easy to interpret from an
epidemiological perspective (Armstrong, 2006; Gasparrini, et al., 2010). However, it is
sensible to conduct sensitivity analyses to assess the key conclusions for different lags and
degrees of freedom.
31
2.6 Temperature measures and mortality
Previous studies examining the relationship between temperature and mortality have used
maximum, mean, or minimum temperatures as an indicator of exposure, while controlling for
relative humidity (Anderson & Bell, 2009). Other studies have used apparent temperature, the
Humidex and temporal synoptic index as exposure measures (Zanobetti & Schwartz, 2008).
A large study of mortality in the US found that the different measures of temperature had a
similar ability to predict the impacts of temperature on mortality (Barnett, Tong, & Clements,
2010). The authors suggested that the best measure of temperature should be that with the
least amount of missing data and best spatial coverage of the study area. Other US studies
considered the best predictor of mortality in heat waves and found a great variability in the
best temperature model between cities (Basu, et al., 2008; Bobb, Dominici, & Peng, 2011;
Hajat & Kosatky, 2010).
Mean temperature is the average of maximum and minimum temperature or 24-hour
monitoring averaged mean temperature. Many studies have showed that mean temperature is
a better predictor of mortality and morbidity than minimum and maximum temperatures
(Gouveia, et al., 2003). This might be because mean temperature covers the whole day and
night, so it can represent the whole day’s exposure compared with minimum or maximum
temperature which only occur for a relatively short time. However, some studies suggest that
minimum temperature should be better at predicting heat-related health effects, while
maximum temperature should be used to examine the cold effects (Kinney, et al., 2008).
Studies examining heat wave effects used both minimum and maximum temperature at the
same model, as the small differences between minimum and maximum temperatures are
related to mortality (Filleul et al., 2006). A study found that the maximum temperature was
32
more closely correlated with mortality than minimum temperature in extreme cold events in
Madrid (DÃaz et al., 2004).
The diurnal temperature difference is the difference between a day’s maximum and minimum
temperatures, and so represents the variation in temperature for each day. Several studies
have shown that diurnal temperature had negative effects on mortality (Kan et al., 2007;
Revich & Shaposhnikov, 2008). Kan et al. (2007) found that a large diurnal temperature
change significantly increased the risk of non-accidental mortality. The estimated effect of
diurnal temperature was not changed when adjusted for ambient temperature days (Haidong
Kan, et al., 2007).
Recently, some studies have used other temperature measures that combined air temperature
and other weather metrics together (Kalkstein & Valimont, 1986; Stafoggia, Forastiere,
Michelozzi, & Perucci, 2009b), for example, apparent temperature and Steadman index.
Some studies have used sensitivity analysis to selected a better temperature measure for a
specific climate (Anderson & Bell, 2009). And some studies used multiple temperature
measures at the same time (Medina-Ramón & Schwartz, 2007; Michelozzi et al., 2000).
Overall there is no one temperature measure that works best in all climates. This is probably
because of the strong correlation between daily estimates such as mean and maximum
temperature (Barnett, Tong, et al., 2010).
33
2.7 Interactive effects between temperature and air pollution on mortality
There is a consistent association between air pollution and increased mortality across the
world (Bell, McDermott, Zeger, Samet, & Dominici, 2004; Dominici, McDermott, Daniels,
Zeger, & Samet, 2005; Hales, Salmond, Town, Kjellstrom, & Woodward, 2000a; Hong et al.,
2002; Keatinge & Donaldson, 2001; Leah & Scott, 2005; Neas, Schwartz, & Dockery, 1999;
Rossi et al., 1999; Saldiva & Pope, 1995; Samoli et al., 2001; Schwartz, 2000; Schwartz et
al., 2001; Tsai, Chen, Hsieh, Chang, & Yang, 2006; Yang et al., 2004; Zanobetti et al., 2003).
Air pollution includes particular matter with aerodynamic diameters less than 10 μg/m3
(PM10), sulphur dioxide (SO2), nitrogen dioxide (NO2) and ozone (O3). Mortality includes
mortality from all-causes, respiratory and cardiovascular diseases.
The interactive effects between air pollution and temperature on mortality have been known
for many years since the disaster of fog in London in 1952. Researchers found that the low
temperature together with air pollution killed thousands of people at that time (Wilson, 2003).
Some air pollutants, like ozone for example, are dependent on the temperature (Sartor,
Snacken, Demuth, & Walckiers, 1995). There are interactive effects between O3, PM10 and
temperature (Ren, Williams, Morawska, Mengersen, & Tong, 2008b; Ren, Williams, &
Tong, 2006; J. Samet, Zeger, Kelsall, Xu, & Kalkstein, 1998) on mortality, however, little
literature has adjusted for interactive effects when examine the relationship between
temperature and mortality.
34
2.8 Groups vulnerable to temperature effects
Previous studies have identified particular health outcomes and vulnerable subgroups to heat
(Basu, 2009b; Basu & Ostro, 2008a; Kovats & Hajat, 2008). People with cardiovascular,
respiratory problems, diabetes, chronic mental disorders or other pre-existing medical
conditions are at greater risk from heat exposure (Basu, et al., 2005; McMichael, et al., 2008;
Stafoggia, et al., 2006). Studies have also shown that the effects of thermal stress were
highest in the elderly following heat waves (Hajat, et al., 2007b; Vaneckova, et al., 2008).
Other factors can also affect the risks of heat-related mortality, such as: socio-economic
status, income, education, social isolation, intensity of urban heat islands, housing
characteristics, access to air conditioning, and availability of health care services (Ebi, 2007;
Ebi & Schmier, 2005; Kovats & Hajat, 2008; Luber & McGeehin, 2008; Vandentorren et al.,
2006; WHO, 2009).
2.8.1 Age
The elderly are susceptible to the impact of ambient temperature (Bull & Morton, 1975).
Almost all the literature has confirmed this result regardless of time periods, regions and
methods used (Guy, 1858; Kovats & Hajat, 2008), although some articles showed a slightly
weaker effect in the oldest age group compare with all ages (Hales, et al., 2000b; Vaneckova,
Beggs, Dear, & McCracken, 2007).
There has been heterogeneity of the vulnerability to temperature in the elderly. People older
than 85 years old were more affected by temperature than other elderly groups (Hajat,
Kovats, & Lachowycz, 2007a; Stafoggia et al., 2008). A US study found that elderly females
35
were more sensitive to heat exposure while elderly males were more vulnerable in cold
climates (Macey & Schneider, 1993). Compared to white elderly, non-white older people had
an increased risk of temperature-related mortality (Martinez et al., 1989).
The elderly are more sensitive than younger people to the climate both in terms of physiology
and behaviour (Macey & Schneider, 1993). Their body temperature lowers with age and their
comfort zone becomes narrower (Watts, 1972). The thermal regulation system weakens with
age, for example, skin sensory perception may diminish and thermal homeostasis may decline
(Collins, 1987). The blood vessel muscles lack vigor with increasing age (Collins, 1987).
This means the elderly cannot easily get relief from heat or cold stress. A higher prevalence
of cardiovascular and respiratory morbidity and mortality in older people is another reason
why the elderly are vulnerable to temperature (Wyndham, 1978).
2.8.2 Gender
There have been no uniform results in difference in temperature-risk by gender. Some studies
found that females were sensitive to heat while males were more sensitive to cold (Hajat, et
al., 2007a; Ishigami et al., 2008; Koutsavlis & Kosatsky, 2003; Macey & Schneider, 1993;
Mannino & Washburn, 1989; Stafoggia, Forastiere, Agostini, Biggeri, Bisanti, Cadum,
Caranci, Lisio, et al., 2006; Vaneckova, Beggs, Dear, & McCracken, 2008). Some
researchers found that men were more at a risk of coronary mortality than women in cold
climates (Gorjanc, Flanders, VanDerslice, Hersh, & Malilay, 1999; Gyllerup, Lanke,
Lindholm, & Schersten, 1991), while others found there is no difference between the
sensitivity of men and women to the cold or heat (Basu & Ostro, 2008b; Kan, et al., 2007;
O'Neill, Zanobetti, & Schwartz, 2003; Stafoggia, Forastiere, et al., 2008). One reason that
36
women may be found to be more sensitive to temperature is because of their generally older
age. So observed differences due to gender may be caused by confounding with age.
2.8.3 Socio-economic status
Economic disadvantage was proposed as an explanation of temperature-related mortality
(Healy, 2003), but its real role remains to be determined. Low education attainment
(Curriero, et al., 2002; Ramón, Zanobetti, Cavanagh, & Schwartz, 2006; O'Neill, Zanobetti,
& Schwartz, 2003), low income (Stafoggia, Forastiere, et al., 2008), living in poverty
(Curriero, et al., 2002) and black ethnicity (Basu & Ostro, 2008b; O'Neill, et al., 2003)
significantly increased susceptibility to the effects of temperature. But other researchers
found no evidence that there was a relation between socioeconomic deprivation and mortality
(Gemmell, McLoone, Boddy, Dickinson, & Watt, 2000; Hajat, et al., 2007a; Healy, 2003;
Lawlor, Maxwell, & Wheeler, 2002; Shah & Peacock, 1999; Stafoggia, Forastiere, et al.,
2008).
2.8.4 Physiologic acclimatisation
There are several ways to protect people from death in the uncomfortable temperatures. The
physical way is to have the appropriate skin colour and body shape for the local climate
(Voelker, 1995). Clothing, shelter and fire are the natural and effective human adaptations to
the cold (Donaldson, Tchernjavskii, Ermakov, Bucher, & Keatinge, 1998; Wyndham, 1969).
The thermoregulation reaction can become adapted when someone has lived in the same
place for a long time, as the core body temperature can change according to the environment
(Wyndham, 1969).
37
2.9 Summary
As the global average temperature continues to increase over the coming decades, there has
been increasing interest in assessing the relationship between temperature and mortality.
Generally, the temperature-mortality relationships were U-, V-, and J-shaped, with thresholds
corresponding to lowest mortality. The elderly are more vulnerable to temperature than the
young. Temperature-mortality relationships vary across different cities. Therefore, we need to
consider the climate adaptation to examine city-specific temperature effects on mortality.
Statistical models for examining temperature-related morality are well established. However,
it remains unknown whether time series and spatiotemporal models produce similar effect
estimates. No study has used case-crossover design to examine non-linear temperature effects
on mortality. No study has examined the effects of temperature change between neighbouring
days on mortality. No study has considered the variation in temperature effects on mortality
by years. There is a need to address these issues in further research.
38
2.10 References
Adger, W. N. (2006). Vulnerability. Global Environmental Change, 16(3), 268-281.
Alderson, M. R. (1985). Season and mortality. Health Trends, 17, 87-96.
Analitis, A., Katsouyanni, K., Biggeri, A., Baccini, M., Forsberg, B., Bisanti, L., et al. (2008).
Effects of cold weather on mortality: results from 15 European cities within the
PHEWE project. American Journal of Epidemiology, 168(12), 1397-1408.
Anderson, B. G., & Bell, M. L. (2009). Weather-related mortality: how heat, cold, and heat
waves affect mortality in the United States. Epidemiology, 20(2), 205-213.
Aniello, C., Morgan, K., Busbey, A., & Newland, L. (1995). Mapping micro-urban heat
islands using Landsat TM and a GIS. Computers & Geosciences, 21(8), 965-967.
Armstrong, B. (2006). Models for the relationship between ambient temperature and daily
mortality. Epidemiology, 17(6), 624-631.
Axelrod, Y. K., & Diringer, M. N. (2006). Temperature management in acute neurologic
disorders. Critical Care Clinics, 22(4), 767-785.
Baccini, M., Biggeri, A., Accetta, G., Kosatsky, T., Katsouyanni, K., Analitis, A., et al.
(2008). Heat effects on mortality in 15 European cities. Epidemiology, 19(5), 711-719.
Baker-Blocker, A. (1982). Winter weather and cardiovascular mortality in Minneapolis-St.
Paul. American Journal Of Public Health, 72(3), 261-265.
Ballester, F., Corella, D., Perez-Hoyos, S., Saez, M., & Hervas, A. (1997). Mortality as a
function of temperature, a study in Valencia, Spain, 1991-1993. International Journal
of Epidemiology, 26(3), 551-561.
Barnett, A. G. (2007). Temperature and cardiovascular deaths in the US elderly: changes over
time. Epidemiology, 18(3), 369-372.
39
Barnett, A. G., & Dobson, A. J. (2010). Analysing Seasonal Health Data. Berlin, Heidelberg:
Springer.
Barnett, A. G., Tong, S., & Clements, A. (2010). What measure of temperature is the best
predictor of mortality? Environmental research, 110(6), 604-611.
Basu, R. (2009). High ambient temperature and mortality: a review of epidemiologic studies
from 2001 to 2008. Environmental Health, 8, 40.
Basu, R., Dominici, F., & Samet, J. M. (2005). Temperature and mortality among the elderly
in the United States: a comparison of epidemiologic methods. Epidemiology, 16(1),
58-66.
Basu, R., Feng, W. Y., & Ostro, B. D. (2008). Characterizing temperature and mortality in
nine California counties. Epidemiology, 19(1), 138-145.
Basu, R., & Ostro, B. D. (2008). A multicounty analysis identifying the populations
vulnerable to mortality associated with high ambient temperature in California.
American Journal of Epidemiology, 168, 632-637.
Basu, R., & Samet, J. M. (2002). Relation between elevated ambient temperature and
mortality: a review of the epidemiologic evidence. Epidemiologic Reviews, 24, 190–
202.
Bateson, T. F., & Schwartz, J. (1999). Control for seasonal variation and time trend in case-
crossover studies of acute effects of environmental exposures. Epidemiology, 10(5),
539-544.
Bell, M. L., McDermott, A., Zeger, S. L., Samet, J. M., & Dominici, F. (2004). Ozone and
Short-term Mortality in 95 US Urban Communities, 1987-2000. Journal of the
American Medical Association, 292(19), 2372-2378.
40
Benavides, R., Montes, F., Rubio, A., & Osoro, K. (2007). Geostatistical modelling of air
temperature in a mountainous region of Northern Spain. Agricultural and Forest
Meteorology, 146(3-4), 173-188.
Bhowmik, A., & Cabral, P. (2011). Statistical Evaluation of Spatial Interpolation Methods for
Small-Sampled Region: A Case Study of Temperature Change Phenomenon in
Bangladesh. Computational Science and Its Applications-ICCSA 2011, 44-59.
Bobb, J. F., Dominici, F., & Peng, R. D. (2011). A Bayesian model averaging approach for
estimating the relative risk of mortality associated with heat waves in 105 U.S. cities.
Biometrics, 67(4), 1605-1616.
Bouchama, A., & Knochel, J. P. (2002). Heat stroke. New England Journal of Medicine,
346(25), 1978-1988.
Bull, G. M., & Morton, J. (1975). Relationships of temperature with death rates from all
causes and from certain respiratory and arteriosclerotic diseases in different age
groups. Age Ageing, 4(4), 232-246.
Carder, M., McNamee, R., Beverland, I., Elton, R., Cohen, G. R., Boyd, J., et al. (2005). The
lagged effect of cold temperature and wind chill on cardiorespiratory mortality in
Scotland. Occupational Environmental Medicine, 62(10), 702-710.
Carson, C., Hajat, S., Armstrong, B., & Wilkinson, P. (2006). Declining vulnerability to
temperature-related mortality in London over the 20th century. American Journal of
Epidemiology 164(1), 77-84.
Chuanyan, Z., Zhongren, N., & Guodong, C. (2005). Methods for modelling of temporal and
spatial distribution of air temperature at landscape scale in the southern Qilian
mountains, China. Ecological modelling, 189(1-2), 209-220.
Clark, R. P., & Edholm, O. G. (1985). Man and his thermal environment: Edward Arnold
London.
41
Collins, K. J. (1987). Effects of cold on old people. British Journal Of Hospital Medicine,
38(6), 506.
Conti, S., Meli, P., Minelli, G., Solimini, R., Toccaceli, V., Vichi, M., et al. (2005).
Epidemiologic study of mortality during the Summer 2003 heat wave in Italy.
Environmental Research, 98(3), 390-399.
Curriero, F. C., Heiner, K. S., Samet, J. M., Zeger, S. L., Strug, L., & Patz, J. A. (2002).
Temperature and mortality in 11 cities of the eastern United States. American Journal
of Epidemiology 155(1), 80-87.
DÃaz, J., GarcÃa, R., López, C. S., Linares, C., TobÃas, A., & Prieto, L. (2004). Mortality
impact of extreme winter temperatures. International Journal of Biometeorology,
49(3), 179-183.
Dessai, S. (2003). Heat stress and mortality in Lisbon Part II. An assessment of the potential
impacts of climate change. International Journal of Biometeorology, 48(1), 37-44.
Dominici, F., McDermott, A., Daniels, M., Zeger, S. L., & Samet, J. M. (2005). Revised
Analyses of the National Morbidity, Mortality, and Air Pollution Study: Mortality
Among Residents Of 90 Cities. Journal of Toxicology & Environmental Health: Part
A, 68(13/14), 1071-1092.
Dominici, F., McDermott, A., Zeger, S. L., & Samet, J. M. (2002). On the Use of Generalized
Additive Models in Time-Series Studies of Air Pollution and Health. American
Journal of Epidemiology, 156(3), 193-203.
Donaldson, G. C., Keatinge, W. R., & Nayha, S. (2003). Changes in summer temperature and
heat-related mortality since 1971 in North Carolina, South Finland, and Southeast
England. Environmental Research, 91(1), 1-7.
42
Donaldson, G. C., Tchernjavskii, V. E., Ermakov, S. P., Bucher, K., & Keatinge, W. R.
(1998). Winter mortality and cold stress in Yekaterinburg, Russia: interview survey.
British Medical Journal, 316(7130): 514–518.
Ebi, K. L. (2007). Towards an early warning system for heat events. Journal of Risk Research,
10(5), 729-744.
Ebi, K. L., Kovats, R. S., & Menne, B. (2006). An approach for assessing human health
vulnerability and public health interventions to adapt to climate change.
Environmental Health Perspectives, 114(12), 1930-1934.
Ebi, K. L., & Schmier, J. K. (2005). A stitch in time: improving public health early warning
systems for extreme weather events. Epidemiological Reviews, 27, 115-121.
El-Zein, A., Tewtel-Salem, M., & Nehme, G. (2004). A time-series analysis of mortality and
air temperature in Greater Beirut. Science of The Total Environment, 330, 71-80.
Filleul, L., Cassadou, S., Medina, S., Fabres, P., Lefranc, A., Eilstein, D., et al. (2006). The
relation between temperature, ozone, and mortality in nine French cities during the
heat wave of 2003. Environmental Health Perspectives, 114(9), 1344-1347.
Fouillet, A., Rey, G., Jougla, E., Frayssinet, P., Bessemoulin, P., & Hemon, D. (2007). A
predictive model relating daily fluctuations in summer temperatures and mortality
rates. BMC Public Health, 7, 114.
Fung, K. Y., Krewski, D., Chen, Y., Burnett, R., & Cakmak, S. (2003). Comparison of time
series and case-crossover analyses of air pollution and hospital admission data.
International Journal of Epidemiology, 32(6), 1064-1070.
Gasparrini, A., Armstrong, B., & Kenward, M. G. (2010). Distributed lag non-linear models.
Statistics in Medicine, 29(21), 2224-2234.
43
Gemmell, I., McLoone, P., Boddy, F. A., Dickinson, G. J., & Watt, G. C. M. (2000).
Seasonal variation in mortality in Scotland. International Journal of Epidemiology,
29(2), 274-279.
Gorjanc, M. L., Flanders, W. D., VanDerslice, J., Hersh, J., & Malilay, J. (1999). Effects of
temperature and snowfall on mortality in Pennsylvania. American Journal of
Epidemiology, 149(12), 1152-1160.
Gouveia, N., Hajat, S., & Armstrong, B. (2003). Socioeconomic differentials in the
temperature-mortality relationship in Sao Paulo, Brazil. International Journal of
Epidemiology, 32(3), 390-397.
Green, R. S., Basu, R., Malig, B., Broadwin, R., Kim, J. J., & Ostro, B. (2010). The effect of
temperature on hospital admissions in nine California counties. International Journal
of Public Health, 55(2), 113-121.
Greenland, S. (1996). Confounding and exposure trends in case-crossover and case-time-
control designs. Epidemiology, 7(3), 231-239.
Guest, C. S., Wilson, K., Woodward, A. J., Hennessy, K., Kalkstein, L. S., Skinner, C., et al.
(1999). Climate and mortality in Australia: retrospective study, 1979-1990, and
predicted impacts in five major cities in 2030. Climate Research, 13, 1-15.
Guy, W. A. (1858). On the annual fluctuations in the number of deaths from various diseases,
compared with like fluctuations in crime, and in other events within andbeyond the
control of the humanwill. Journal of the Statistical Society of London, 21, 52-
86.
Gyllerup, S., Lanke, J., Lindholm, L. H., & Schersten, B. (1991). High coronary mortality in
cold regions of Sweden. Journal of internal medicine, 230(6), 479-485.
Haines, A., Kovats, R., Campbell-Lendrum, D., & Corvalan, C. (2006). Climate change and
human health: impacts, vulnerability, and mitigation. Lancet, 367(9528), 2101-2109.
44
Hajat, S., Armstrong, B. G., Gouveia, N., & Wilkinson, P. (2005). Mortality displacement of
heat-related deaths: a comparison of Delhi, Sao Paulo, and London. Epidemiology,
16(5), 613-620.
Hajat, S., & Kosatky, T. (2010). Heat-related mortality: a review and exploration of
heterogeneity. Journal of Epidemiology and Community Health, 64(9), 753-760.
Hajat, S., Kovats, R. S., Atkinson, R. W., & Haines, A. (2002). Impact of hot temperatures on
death in London: a time series approach. Journal of Epidemiology and Community
Health, 56(5): 367-372.
Hajat, S., Kovats, R. S., & Lachowycz, K. (2007). Heat-related and cold-related deaths in
England and Wales: who is at risk? Occupational and Environmental Mededicine.,
64(2), 93-100.
Hales, S., Salmond, C., Town, G. I., Kjellstrom, T., & Woodward, A. (2000a). Daily
mortality in relation to weather and air pollution in Christchurch, New Zealand.
Australia and New Zealand journal of Public Health, 24(1):89-91.
Hansen, A. L., Bi, P., Ryan, P., Nitschke, M., Pisaniello, D., & Tucker, G. (2008). The effect
of heat waves on hospital admissions for renal disease in a temperate city of Australia.
International Journal of Epidemiology, 37(6), 1359-1365.
Hastie, T., Tibshirani, R., & Friedman, J. (2004). The Elements of Statistical Learning: Data
Mining, Inference, and Prediction. BeiJing: Publishing House of Electronics Industry.
Hastie, T. J., & Tibshirani, R. J. (1990). Generalized additive models: Chapman & Hall/CRC.
Healy, J. D. (2003). Excess winter mortality in Europe: a cross country analysis identifying
key risk factors. Journal of Epidemiology and Community Health, 57(10):784-9.
Hong, Y.-C., Lee, J.-T., Kirn, H., Ha, E.-H., Schwartz, J., & Christiani, D. C. (2002). Effects
of air pollutants on acute stroke mortality. Environmental Health Perspect, 110(2),
187-191.
45
Hong, Y. C., Rha, J. H., Lee, J. T., Ha, E. H., Kwon, H. J., & Kim, H. (2003). Ischemic
stroke associated with decrease in temperature. Epidemiology, 14(4), 473-478.
Hutcheon, J. A., Chiolero, A., & Hanley, J. A. (2010). Random measurement error and
regression dilution bias. British Medical Journal, 340, c2289.
Huynen, M. M., Martens, P., Schram, D., Weijenberg, M. P., & Kunst, A. E. (2001). The
impact of heat waves and cold spells on mortality rates in the Dutch population.
Environmental Health Perspectives, 109(5), 463-470.
Huynen, M. M. (2001). The impact of heat waves and cold spells on mortality rates in the
Dutch population. Environmental Health Perspectives, 109(5), 463.
IPCC. (2007a). Climate change 2007: impacts, adaptation and vulnerability. Contribution of
Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on
Climate Change. Cambridge, UK: Cambridge University Press.
IPCC. (2007b). Climate change 2007: synthesis report. Contribution of Working Groups I, II
and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate
Change. Geneva, Switzerland: IPCC.
Ishigami, A., Hajat, S., Kovats, R. S., Bisanti, L., Rognoni, M., Russo, A., et al. (2008). An
ecological time-series study of heat-related mortality in three European cities.
Environmental Health, 7, 5-11.
Janes, H., Sheppard, L., & Lumley, T. (2005). Overlap bias in the case-crossover design, with
application to air pollution exposures. Statitics in Medicine, 24(2), 285-300.
Jiang, G., Qu, T., Shang, Z., & Zhang, X. (2004). A circuit simulating method for heat
transfer mechanism in human body. Proceedings of the 26th Annual International
Conference of the IEEE EMBS (September 1-5), 5274-5276.
46
Johnson, H., Kovats, R. S., McGregor, G., Stedman, J., Gibbs, M., Walton, H., et al. (2005).
The impact of the 2003 heat wave on mortality and hospital admissions in England.
Health Statistics Quaterly, 25, 6-11.
Kalkstein, L., & Greene, J. (1997). An evaluation of climate/mortality relationships in large
US cities and the possible impacts of a climate change. Environmental Health
Perspectives, 105(1), 84-93.
Kalkstein, L. L., & Valimont, K. M. (1986). An evaluation of summer discomfort in the
United-States using a relative climatological index. Journal of the American Medical
Association, 67: 842-848.
Kalkstein, L. S., & Davis, R. E. (1989). Weather and human mortality: an evaluation of
demographic and interregional responses in the United States. Annals of
the Association of American Geographers, 79(1), 44-64.
Kan, H., London, S. J., Chen, H., Song, G., Chen, G., Jiang, L., et al. (2007). Diurnal
temperature range and daily mortality in Shanghai, China. Environmental Research,
103(3), 424-431.
Keatinge, W. R., & Donaldson, G. C. (2001). Mortality Related to Cold and Air Pollution in
London After Allowance for Effects of Associated Weather Patterns. Environmental
Research, 86(3):209-16.
Kestens, Y., Brand, A., Fournier, M., Goudreau, S., Kosatsky, T., Maloley, M., et al. (2011).
Modelling the variation of land surface temperature as determinant of risk of heat-
related health events. International Journal of Health Geographics, 10, 7.
Kim, S., & Jang, J. Y. (2005). Temperature and summer mortality association in Seoul.
Epidemiology, 16(5), S112-S112
47
Kinney, P. L., O'Neill, M. S., Bell, M. L., & Schwartz, J. (2008). Approaches for estimating
effects of climate change on heat-related deaths: challenges and opportunities.
Environmental Science & Policy, 11(1), 87-96.
Knowlton, K., Rotkin-Ellman, M., King, G., Margolis, H. G., Smith, D., Solomon, G., et al.
(2009). The 2006 California heat wave: impacts on hospitalizations and emergency
department visits. Environmental Health Perspectives, 117(1), 61-67.
Koutsavlis, A. T., & Kosatsky, T. (2003). Environmental-Temperature Injury in a Canadian
Metropolis. Journal of Environmental Health, 66(5), 40-45.
Kovats, R. S., & Hajat, S. (2008). Heat stress and public health: a critical review. Annual
Review of Public Health, 29, 41-55.
Kroemer, K. H. E., & Grandjean, E. (1997). Fitting the task to the human. A textbook of
occupational ergonomics (Fifth edition ed. Vol. Chapter 20). London:
Taylor&Francis Ltd.,.
Kysely, J., Pokorna, L., Kyncl, J., & Kriz, B. (2009). Excess cardiovascular mortality
associated with cold spells in the Czech Republic. BMC Public Health, 9, 19.
Lawlor, D. A., Maxwell, R., & Wheeler, B. W. (2002). Rurality, deprivation, and excess
winter mortality: An ecological study. Journal of Epidemiology and Community
Health, 56(5), 373-374.
Le Tertre, A., Lefranc, A., Eilstein, D., Declercq, C., Medina, S., Blanchard, M., et al. (2006).
Impact of the 2003 heatwave on all-cause mortality in 9 French cities. Epidemiology,
17(1), 75-79.
Leah, J. W., & Scott, L. Z. (2005). Are the Acute Effects of Particulate Matter on Mortality in
the National Morbidity, Mortality, and Air Pollution Study the Result of Inadequate
Control for Weather and Season? A Sensitivity Analysis using Flexible Distributed
Lag Models. American Journal of Epidemiology, 162(1), 80.
48
Lee, D., & Shaddick, G. (2010). Spatial modeling of air pollution in studies of its short-term
health effects. Biometrics, 66(4), 1238-1246.
Levy, D., Sheppard, L., Checkoway, H., Kaufman, J., Lumley, T., Koenig, J., et al. (2001). A
case-crossover analysis of particulate matter air pollution and out-of-hospital primary
cardiac arrest. Epidemiology, 12(2), 193-199.
Lo, C. P., Quattrochi, D. A., & Luvall, J. C. (1997). Application of high-resolution thermal
infrared remote sensing and GIS to assess the urban heat island effect. International
Journal of Remote Sensing, 18(2), 287-304.
Lu, Y., Symons, J. M., Geyh, A. S., & Zeger, S. L. (2008). An approach to checking case-
crossover analyses based on equivalence with time-series methods. Epidemiology,
19(2), 169-175.
Lu, Y., & Zeger, S. L. (2007). On the equivalence of case-crossover and time series methods
in environmental epidemiology. Biostatistics, 8(2), 337-344.
Luber, G., & McGeehin, M. (2008). Climate change and extreme heat events. American
Journal of Prevetion Medicine, 35(5), 429-435.
Lumley, T., & Levy, D. (2000). Bias in the case-crossover design: implications for studies of
air pollution. Environmetrics, 11, 689-704.
Macey, S. M., & Schneider, D. F. (1993). Deaths from excessive heat and excessive cold
among the elderly. The Gerontologist, 33(4), 497.
Mannino, J. A., & Washburn, R. A. (1989). Environmental temperature and mortality from
acute myocardial infarction. International Journal of Biometeorology, 33(1), 32-35.
Martinez, B. F., Annest, J. L., Kilbourne, E. M., Kirk, M. L., Lui, K. J., & Smith, S. M.
(1989). Geographic distribution of heat-related deaths among elderly persons. Use of
county-level dot maps for injury surveillance and epidemiologic research. Journal of
the American Medical Association, 262(16), 2246-2250.
49
McGeehin, M., & Mirabelli, M. (2001). The potential impacts of climate variability and
change on temperature-related morbidity and mortality in the United States.
Environmental Health Perspectives, 109(Suppl 2), 185-189.
McMichael, A. J., Wilkinson, P., Kovats, R. S., Pattenden, S., Hajat, S., Armstrong, B., et al.
(2008). International study of temperature, heat and urban mortality: the
'ISOTHURM' project. International Journal of Epidemiology, 37(5), 1121-1131.
Medina--Ramón, M., Zanobetti, A., Cavanagh, D. P., & Schwartz, J. (2006). Extreme
temperatures and mortality: Assessing effect modification by personal characteristics
and specific cause of death in a multi-city case-only analysis. Environmental Health
Perspectives, 114(9), 1331-1336.
Medina-Ramón, M., & Schwartz, J. (2007). Temperature, temperature extremes, and
mortality: a study of acclimatisation and effect modification in 50 US cities.
Occupational and Environmental Medicine, 64(12), 827-833.
Metabolism, Energy Balance, and Temperature Regulation. (2008, April 10, 2008).
Retrieved 11 May,2010.
Michelozzi, P., De Sario, M., Accetta, G., de'Donato, F., Kirchmayer, U., D'Ovidio, M., et al.
(2006). Temperature and summer mortality: geographical and temporal variations in
four Italian cities. Journal of Epidemiology and Community Health, 60(5), 417-423.
Michelozzi, P., Fano, V., Forastiere, F., Barca, A., Kalkstein, L. S., & Perucci, C. A. (2000).
Weather conditions and elderly mortality in Rome during summer. Bulletin of World
Meteorology Organization, 49, 348-355.
Muggeo, V. M., & Hajat, S. (2009). Modelling the non-linear multiple-lag effects of ambient
temperature on mortality in Santiago and Palermo: a constrained segmented
distributed lag approach. British Medical Journal, 66(9), 584-591.
50
Muggeo, V. M. R. (2003). Estimating regression models with unknown break-points.
Statistics in Medicine, 22(19), 3055-3071.
Navidi, W. (1998). Bidirectional case-crossover designs for exposures with time trends.
Biometrics, 54(2), 596-605.
Navidi, W., & Weinhandl, E. (2002). Risk set sampling for case-crossover designs.
Epidemiology, 13(1), 100-105.
Näyhä, S. (2005). Environmental temperature and mortality. International Journal of
Circumpolar Health, 64(5), 451-458.
Neas, L. M., Schwartz, J., & Dockery, D. (1999). A Case-Crossover Analysis of Air Pollution
and Mortality in Philadelphia. Environmental Health Perspectives, 107(8), 629-31.
Nelder, J. A., & Wedderburn, R. W. M. (1972). Generalized Linear Models. Journal of the
Royal Society Interface, 135(3), 370-384.
O'Neill, M. S., Zanobetti, A., & Schwartz, J. (2003). Modifiers of the temperature and
mortality association in seven US cities. American Journal of Epidemiology, 157(12),
1074-1082.
Pan, W., Li, L., & Tsai, M. (1995). Temperature extremes and mortality from coronary heart
disease and cerebral infarction in elderly Chinese. Lancet, 345(8946), 353-355.
Parsons, K. C. (2003). Huamn thermal environment. The effect of hot, moderate and cold
environments n human health,comfort and performance (Second edition ed.). London
and New York: Taylor&Francis.
Patz, J. A., Campbell-Lendrum, D., Holloway, T., & Foley, J. A. (2005). Impact of regional
climate change on human health. Nature, 438(7066), 310-317.
Reid, C., O’Neill, M., Gronlund, C., Brines, S., Brown, D., Diez-Roux, A., et al. (2009).
Mapping community determinants of heat vulnerability. Environmental Health
Perspectives, 117(11), 1730-1736.
51
Ren, C., Williams, G. M., Morawska, L., Mengersen, K., & Tong, S. (2008). Ozone modifies
associations between temperature and cardiovascular mortality: analysis of the
NMMAPS data. Occupational and Environmental Medicine, 65(4), 255-260.
Ren, C., Williams, G. M., & Tong, S. (2006). Does particulate matter modify the association
between temperature and cardiorespiratory diseases? Environmental Health
Perspectives, 114(11), 1690-1696.
Revich, B., & Shaposhnikov, D. (2008). Excess mortality during heat waves and cold spells
in Moscow, Russia. Occupational and Environmental Medicine, 65(10), 691-696.
Revich , B., & Shaposhnikov, D. (2008). Temperature-induced excess mortality in Moscow,
Russia International Journal of Biometeorology.
Rey, G., Fouillet, A., Bessemoulin, P., Frayssinet, P., Dufour, A., Jougla, E., et al. (2009).
Heat exposure and socio-economic vulnerability as synergistic factors in heatwave-
related mortality. European Journal of Epidemiology, 24(9), 495-502.
Rey, G., Jougla, E., Fouillet, A., Pavillon, G., Bessemoulin, P., Frayssinet, P., et al. (2007).
The impact of major heat waves on all-cause and cause-specific mortality in France
from 1971 to 2003. International archives of occupational and environmental health,
80(7), 615-626.
Rogot, E., & Blackwelder, W. C. (1970). Associations of cardiovascular mortality with
weather in Memphis, Tennessee. Public Health Reports, 85(1), 25.
Rose, G. (1966). Cold weather adn ischaemic heart disease. Brit. J. Prev.soc.Med, 20, 97-100.
Rossi, G., Vigotti, M. A., Zanobetti, A., Repetto, F., Gianelle, V., & Schwartz, J. (1999). Air
Pollution and Cause-Specific Mortality in Milan, Italy, 1980-1989. Archives of
Environmental Health, 54(3), 158-164.
52
Saldiva, P. H. N., & Pope Iii, C. A. (1995). Air Pollution and Mortality in Elderly People: A
Time-Series Study in Sao Paulo, Brazil. Archives of Environmental Health, 50(2),
159-163.
Samet, J., Zeger, S., Kelsall, J., Xu, J., & Kalkstein, L. (1998). Does Weather Confound or
Modify the Association of Particulate Air Pollution with Mortality?: An Analysis of
the Philadelphia Data, 1973-1980. Environmental Research, 77(1), 9-19.
Samoli, E., Schwartz, J., Wojtyniak, B., Touloumi, G., Spix, C., Balducci, F., et al. (2001).
Investigation Regional Differences in Short-Term Effects of Air Pollution on Daily
Mortality in the APHEA Project: A Sensitivity Analysis for Controlling Long-Term
Trends and Seasonality. Environmental Health Perspectives, 109(4), 349-353.
Sartor, F., Snacken, R., Demuth, C., & Walckiers, D. (1995). Temperature, Ambient Ozone
Levels, and Mortality during Summer, 1994, in Belgium. Environmental Research,
70(2), 105-113.
Schwartz, J. (2000). Assessing Confounding, Effect Modification, and Thresholds in the
Association between Ambient Particles and Daily Deaths. Environmental Health
Perspectives, 108(6), 563-568.
Schwartz, J. (2005). Who is sensitive to extremes of temperature?: A case-only analysis.
Epidemiology, 16(1), 67.
Schwartz, J., Ballester, F., Saez, M., Prez-Hoyos, S., Bellido, J., Cambra, K., et al. (2001).
The Concentration--Response Relation between Air Pollution and Daily Deaths.
Environmental Health Perspectives, 109(10), 1001-1006.
Semenza, J. C., Rubin, C. H., Falter, K. H., Selanikio, J. D., Flanders, W. D., Howe, H. L., et
al. (1996). Heat-related deaths during the July 1995 heat wave in Chicago. The New
England Journal Of Medicine, 335(2), 84-90.
53
Sessler, D. (2009). Thermoregulatory defense mechanisms. Critical care medicine, 37(7),
S203-210.
Shaddick, G., Lee, D., Zidek, J. V., & Salway, R. (2008). Estimating exposure response
functions using ambient pollution concentrations. The Annals of Applied Statistics,
2(4), 1249-1270.
Shah, S., & Peacock, J. (1999). Deprivation and excess winter mortality. Journal of
Epidemiology and Community Health, 53(8), 499-502.
Smargiassi, A., Goldberg, M. S., Plante, C., Fournier, M., Baudouin, Y., & Kosatsky, T.
(2009). Variation of daily warm season mortality as a function of micro-urban heat
islands. Journal of Epidemiology and Community Health, 63(8), 659-664.
Smith, W. R., Coyne, P., Smith, V. S., & Mercier, B. (2003). Temperature changes,
temperature extremes, and their relationship to emergency department visits and
hospitalizations for sickle cell crisis. Pain Management Nursing, 4(3), 106-111.
Stafoggia, M., Forastiere, F., Agostini, D., Biggeri, A., Bisanti, L., Cadum, E., et al. (2006).
Vulnerability to heat-related mortality: a multicity, population-based, case-crossover
analysis. Epidemiology, 17(3), 315-323.
Stafoggia, M., Forastiere, F., Agostini, D., Caranci, N., de'Donato, F., Demaria, M., et al.
(2008). Factors affecting in-hospital heat-related mortality: a multi-city case-
crossover analysis. Journal of Epidemiology and Community Health, 62(3), 209-215.
Stafoggia, M., Forastiere, F., Michelozzi, P., & Perucci, C. A. (2009). Summer temperature-
related mortality: effect modification by previous winter mortality. Epidemiology
20(4), 575-583.
Stewart, S., McIntyre, K., Capewell, S., & McMurray, J. J. V. (2002). Heart failure in a cold
climate* 1:: Seasonal variation in heart failure-related morbidity and mortality.
Journal of the American College of Cardiology, 39(5), 760-766.
54
Takahashi, K., Honda, Y., & Emori, S. (2007). Assessing mortality risk from heat stress due
to global warming. Journal of Risk Research, 10(3), 339-354.
Terence W. Anderson, B., B CH, & Rochard, C. (1979). Cold snap, snowfall and sudden
death from ischemic heart disease. Canadian Medical Association Journal, 121,
1580-1583.
Tsai, S. S., Chen, C. C., Hsieh, H. J., Chang, C. C., & Yang, C. Y. (2006). Air Pollution and
Postneonatal Mortality in a Tropical City: Kaohsiung, Taiwan. Inhalation Toxicology,
18(3), 185-9.
Vandentorren, S., Bretin, P., Zeghnoun, A., Mandereau-Bruno, L., Croisier, A., Cochet, C., et
al. (2006). August 2003 heat wave in France: risk factors for death of elderly people
living at home. The European Journal of Public Health, 16(6), 583-591.
Vaneckova, P., Beggs, P. J., de Dear, R. J., & McCracken, K. W. (2007). The Effect of
Temperature on Morbidity and Mortality in Sydney, Australia, Between 1993-2005.
Epidemiology, 18(5)(Suppl), S198.
Vaneckova, P., Beggs, P. J., de Dear, R. J., & McCracken, K. W. J. (2008). Effect of
temperature on mortality during the six warmer months in Sydney, Australia, between
1993 and 2004. Environmental Research, 108(3), 361-369.
Vaneckova, P., Beggs, P. J., & Jacobson, C. R. (2010). Spatial analysis of heat-related
mortality among the elderly between 1993 and 2004 in Sydney, Australia. Social
Science Medicine, 70(2), 293-304.
Voelker, R. (1995). Probe of heat wave deaths under way. Journal of the American Medical
Association, 274(8), 595-596.
Wang, X. Y., Barnett, A. G., Hu, W., & Tong, S. (2009). Temperature variation and
emergency hospital admissions for stroke in Brisbane, Australia, 1996-2005.
International Journal of Biometeorology, 53(6), 535-541.
55
Watts, A. J. (1972). Hypothermia in the aged: A study of the role of cold-sensitivity.
Environmental Research, 5(1), 119-126.
Whitworth, K. W., Symanski, E., Lai, D., & Coker, A. L. (2011). Kriged and modeled
ambient air levels of benzene in an urban environment: an exposure assessment study.
Environmental Health, 10(1), 21.
WHO. (2008). Protecting health from climate change: World Health Day 2008. Geneva:
World Health Organization.
WHO. (2009). Improving public health responses to extreme weather heat-waves EuroHEAT.
Copenhagen Ø, Denmark: WHO Regional Office for Europe.
Wilson, J. (2003). Weather and health division. The Journal of Meterology- International
Weather and health special issue, 28(283).
Wyndham, C. H. (1969). Adaptation to heat and cold. Environmental Research, 2(5-6), 442-
469.
Wyndham CH, F. S. (1978). Climate and disease. South African Medical Journal, 53(26),
1051-1061.
Yang, C. Y., Chang, C. C., Chuang, H. Y., Tsai, S. S., Wu, T. N., & Ho, C. K. (2004).
Relationship between air pollution and daily mortality in a subtropical city: Taipei,
Taiwan. Environment International, 30(4):519-23.
Zanobetti, A., & Schwartz, J. (2008). Temperature and mortality in nine US cities.
Epidemiology, 19(4), 563-570.
Zanobetti, A., Schwartz, J., Samoli, E., Gryparis, A., Touloumi, G., Peacock, J., et al. (2003).
The Temporal Pattern of Respiratory and Heart Disease Mortality in Response to Air
Pollution. Environmental Health Perspectives, 111(9), 1188-1193.
Zeger, S. L. (1988). A regression model for time series of counts. Biometrika, 75(4), 621-629.
56
Zhang, K., Oswald, E. M., Brown, D. G., Brines, S. J., Gronlund, C. J., White-Newsome, J.
L., et al. (2011). Geostatistical exploration of spatial variation of summertime
temperatures in the Detroit metropolitan region. Environmental Research, 111(8),
1046-1053.
57
CHAPTER 3: STUDY DESIGN AND METHODOLOGY
This chapter introduces the study populations, data collection, data management and analysis.
The detailed and specific materials and methodologies are described in each corresponding
chapter.
3.1 Study population
This thesis was conducted in three settings: Tianjin, China; Brisbane, Australia; and 83 cities
in the United States.
Tianjin is a city in northeastern China (Figure 3.1), and is adjacent to Beijing and Hebei
Province, along the coast of Bohai Gulf (39° 07' North, 117° 12' East). Tianjin has four
distinct seasons, with cold, windy, dry winters influenced by the vast Siberian anticyclone,
and hot, humid summers due to the monsoon. It is the fifth largest Chinese city in terms of
urban land area. The population in the urban area was 4.2 million in 2005.
Brisbane is the capital city of the state of Queensland in Australia, and is located on the east
coast of the country (27° 30' south, 153° 00' east) (Figure 3.2). It has a humid subtropical
climate, with the average temperature of 25 °C in summer (Dec–Feb).
The third data set was obtained from the publicly available National Morbidity and Mortality
Air Pollution Study (NMMAPS) study (Samet, Dominici, Zeger, Schwartz, & Dockery, 2000;
Samet et al., 2000) (Figure 3.3). This study included daily climatic conditions, air pollution
levels, and mortality in 108 cities in the United States We excluded data for small cities
58
(population under 200,000) and cities with more than 0.5% missing data for air or dew point
temperature, which left 83 cities. We stratified the cities into seven regions (Industrial
Midwest, North East, North West, South East, South West, Southern California, and Upper
Midwest).
Figure 3.1: The location of Tianjin, China.
61
3.3 Data collection and management
3.3.1 Tianjin data for chapter 4
Mortality data was obtained from the China Information System for Death Register and
Report of Chinese Centre for Disease Control and Prevention from January 1, 2005 to
December 31, 2007. The mortality data were from six urban districts of Tianjin (Heping,
Hedong, Hexi, Nankai, Hebei and Hongqiao). Non-accidental mortality was classified
according to the International Classification of Diseases, 10th revision (ICD-10: A00–R99)
(World Health Organization, 2007). Cardiopulmonary (ICD-10:I00–I99 and ICD-10:J00–
J99), cardiovascular mortality (ICD-10:I00–I99) and respiratory mortality (ICD-10:J00–J99)
were examined separately.
Daily meteorological data on maximum, mean and minimum temperature, and relative
humidity, were obtained from the China Meteorological Data Sharing Service System
(http://cdc.cma.gov.cn). Daily air pollution data on particulate matter less than 10 μm in
aerodynamic diameter (PM10), sulphur dioxide (SO2) and nitrogen dioxide (NO2) were
obtained from the Tianjin Environmental Monitoring Centre.
3.3.2 Brisbane data for chapters 5 and 6
For chapter 5, daily morality data were obtained for non-accidental causes between January
2000 and December 2004 at the level of statistical local areas in Brisbane city from the Office
of Economic and Statistical Research of the Queensland Treasury. There are 163 statistical
local areas in Brisbane city. The causes of non-accidental mortality were coded according to
62
the International Classification of Diseases, ninth version tenth version (ICD-10) (ICD-10:
A00–R99). All deaths were residents of Brisbane city. Population data were obtained from
the Australia Bureau of Statistics for each statistical local area. The census of population and
housing for each statistical local area is conducted once every five years. I used the 1996
census for the year 2000, and the 2001 census for years 2001−2004.
Daily data on maximum temperature were obtained from the Australian Bureau of
Meteorology at 19 sites in or around Brisbane city. Daily data on relative humidity were
obtained from one monitoring site (Brisbane airport). Daily data on air pollution were
obtained from the Queensland Environmental Protection Agency. The daily PM10
concentrations were averaged from 13 monitoring sites. The daily O3 concentrations were
averaged from 10 monitoring sites.
For chapter 6, the Brisbane data on daily deaths of non-external causes between Jan, 1996
and Dec, 2004 were gathered from the Office of Economic and Statistical Research of the
Queensland Treasury. The causes of non-external mortality were coded according to the
International Classification of Diseases, ninth version (ICD-9) (ICD-9: 001–799) before
December 1996 and tenth version (ICD-10) (ICD-10: A00–R99) between December 1996
and December 2004. Cardiovascular mortality (CVM, ICD-9:390–459, ICD-10:I00–I79) and
respiratory mortality (RM: ICD-9: 460–519, ICD-10:J00–J99) were extracted from the
mortality database. Influenza deaths (ICD-9: 487.0–487.8 or ICD-10: J10–J11) were
excluded from respiratory mortality. All deaths were for residents of Brisbane city. Non-
external mortality were stratified by gender and age (3 groups: 0–64, 65–74, and ≥ 75 years).
63
Values of temperature change were calculated using the current day’s mean temperature
minus the previous day’s mean temperature. Temperature change between the neighbouring
days is a measure of temperature stability, with large positive and negative values indicating
an unstable temperature. The air pollutants including daily mean ozone (O3) and PM10 were
monitored at a central site in Brisbane. We collected these data from the Queensland
Environmental Protection Agency.
3.3.3 NMMAPS data for chapters 6 and 7
The NMMAPS (The National Morbidity Mortality Air Pollution Study) included daily
climatic conditions, air pollution levels, and mortality in 108 cities in the United States from
1987 to 2000. Data on maximum and minimum temperatures came from the National
Climatic Data Center, and daily mortality counts came from the National Center for Health
Statistics. Daily non-external deaths consisted of death counts among residents, excluding
injuries and external causes. More information is available from the NMMAPS web site
(http://www.ihapss.jhsph.edu).
For chapter 6, Los Angeles’ data were obtained from NMMAPS database. Mean temperature,
relative humidity, O3, NEM, CVM, RM, and NEM in age groups (0–64, 65–74, and ≥ 75
years) were used here. Mortality counts were not split by gender in the NMMAPS, so the
impact of temperature change on mortality by gender could not be analysed in Los Angeles.
For chapter 7, I limited analyses to elderly morality (age ≥ 75 years) in the warm season
(1 May–30 September) as I was interested in the effects of heat on a susceptible population.
64
Mean temperature (i.e., average of maximum and minimum temperatures) was used as the
main exposure variable.
3.4 Data analysis
3.4.1 Case−crossover design and distributed lag non-linear model
Chapter 4 introduces how to combine the time-stratified case–crossover and a distributed lag
non-linear model. These models were demonstrated using a motivating example of the
temperature-mortality relationship in Tianjin, China. A Poisson regression model that allows
for over-dispersion was used to combine the case–crossover design with a distributed lag
non-linear model, as the case–crossover using conditional logistic regression is a special case
of time series analysis (Lu & Zeger, 2007).
The distributed lag non-linear model was used to get the predicted effects and standard errors
for combinations of temperature and lags. Graphs, summaries, and statistical inference were
obtained from the estimates and standard errors of distributed lag non-linear model
(Armstrong, 2006).
3.4.2 Comparing time series and spatiotemporal analyses
Chapter 5 demonstrates the comparison of time series and spatiotemporal analyses, using the
effects of temperature on mortality in Brisbane as an example.
65
Ordinary kriging was used to interpolate the daily values of temperature to the centroids of
163 locations across Brisbane city from the values of 19 monitoring sites.
A generalized additive mixed model (GAMM) with Poisson regression was used to examine
the association between the spatial temperature and mortality (Augustin et al., 2009), to
perform the spatiotemporal model,
To examine association between non-spatial temperature exposure and mortality, I used
temperature data from single monitoring site (Brisbane centre) using a generalized additive
model (GAM).
To examine the effects of non-spatial temperature exposure using multiple monitoring sites
on morality, I used daily temperature averaged from 3 sites (Brisbane centre, Brisbane airport,
Archerfield Airport)
Squared Pearson scaled residuals were used to compare the fit of the spatiotemporal and time
series models.
3.4.3 The effect of temperature change on mortality
Chapter 6 describes the effect of temperature change between neighbouring days on mortality.
Values of temperature change were calculated using the current day’s mean temperature
minus the previous day’s mean temperature. Temperature change between the neighbouring
66
days is a measure of temperature stability, with large positive and negative values indicating
unstable temperatures.
Firstly, a Poisson generalized additive model (GAM) was used to examine the effects of
short-term changes in temperature between neighbouring days on mortality. Regression
spline was used for temperature change.
Secondly, as an alternative model temperature change was categorised into 3 groups: a drop
of more than 3 °C; a rise of more than 3 °C; a change in either direction of less than 3 °C.
3.4.4 The variation in the effects of temperature on mortality
Chapter 7 describes the variation in the effects of temperature on elderly mortality in 83 US
cities.
A city-specific Poisson regression model was used to examine each year’s high temperature
effect on elderly mortality. In the city-specific model, the heat effects were divided into a
“main effect” and an “added effect” according to a previous study (Gasparrini & Armstrong,
2011b). These estimates (main effect and added effect separately) were then combined using
a univariate meta-analysis to create yearly estimates for each region and for the entire US.
The pooled yearly national and regional main effects and added effects were plotted from
1987 to 2000.
A Bayesian hierarchical model was used to estimate an overall high temperature effect (main
effect and added effect separately) for each region and the nation by combining the yearly
67
estimated effects of high temperatures within each city and incorporating the estimates’
variance (Everson & Morris, 2000). Each city was given a random intercept to model its
mean heat effect, and a random linear effect of time to model linear trends over time. The
model fit was assessed using the Deviance Information Criteria (Spiegelhalter, Best, Carlin,
& Van Der Linde, 2002).
3.4 Rationale for choosing study sites or temperature measures
In chapter 4, we used Tianjin data as attractive example to illustrate how to combine case–
crossover and a distributed lag non-linear model, because no study has been conducted in this
city. We also examined which temperature measure is the best to predict mortality. On one
hand, we showed the process for combining case–crossover design and distributed lag non-
linear models. On the other hand, we provided useful information on the association between
temperature and mortality in Tianjin city.
In chapter 5, we used Brisbane data to compare spatiotemporal and time series models, as we
collected high quality data on spatial temperature and mortality in this city. We used
maximum temperature as exposure, because previous studies shown that mean, minimum and
maximum temperature had similar ability to predictive mortality. Also, the maximum
temperature can be freely downloaded from the website of Australia Bureau of Meteorology
(http://www.bom.gov.au/climate/data/).
In chapter 6, we used Brisbane and Los Angles data to examine the effects of temperature
change on mortality, because the two cities have different subtropical weather. We used two
cities in different countries with different subtropical climates to confirm the findings.
68
In chapter 7, we used 83 cities’ data from The National Morbidity Mortality Air Pollution
Study to examine the variability in the temperature effects on mortality across years, cities,
and regions, because this dataset is free and provides high quality data on temperature and
mortality. Results can be compared within cities and regions. Evidence from large dataset is
should be reliable.
69
3.5 References
Armstrong, B. (2006). Models for the relationship between ambient temperature and daily
mortality. Epidemiology, 17(6), 624-631.
Augustin, N. H., Musio, M., von Wilpert, K., Kublin, E., Wood, S. N., & Schumacher, M.
(2009). Modeling spatiotemporal forest health monitoring data. Journal of the
American Statistical Association, 104(487), 899-911.
Everson, P. J., & Morris, C. N. (2000). Inference for multivariate normal hierarchical models.
Journal of the Royal Statistical Society: Series B (Statistical Methodology), 62(2),
399-412.
Gasparrini, A., & Armstrong, B. (2011). The impact of heat waves on mortality.
Epidemiology, 22(1), 68-73.
Lu, Y., Symons, J. M., Geyh, A. S., & Zeger, S. L. (2008). An approach to checking case-
crossover analyses based on equivalence with time-series methods. Epidemiology,
19(2), 169-175.
Lu, Y., & Zeger, S. L. (2007). On the equivalence of case-crossover and time series methods
in environmental epidemiology. Biostatistics, 8(2), 337-344.
Peng, R. D., & Dominici, F. (2008). Statistical Methods for Environmental Epidemiology in
R: A Case Study in Air Pollution and Health: New York, NY: Springer Publishing
Company.
Rothpearl, A. (1989). The jackknife technique in statistical analysis. Chest, 95(4), 940.
Samet, J. M., Dominici, F., Zeger, S. L., Schwartz, J., & Dockery, D. W. (2000). The
National Morbidity, Mortality, and Air Pollution Study. Part I: Methods and
methodologic issues. Res Rep Health Eff Inst(94 Pt 1), 5-14; discussion 75-84.
70
Samet, J. M., Zeger, S. L., Dominici, F., Curriero, F., Coursac, I., Dockery, D. W., et al.
(2000). The National Morbidity, Mortality, and Air Pollution Study. Part II:
Morbidity and mortality from air pollution in the United States. Res Rep Health Eff
Inst, 94(Pt 2), 5-70; discussion 71-79.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian
measures of model complexity and fit. Journal of the Royal Statistical Society: Series
B (Statistical Methodology), 64(4), 583-639.
World Health Organization. (2007). International Statistical Classification of Diseases and
Related Health Problems, 10th Revision, Version for 2007.
Http://apps.who.int/classifications/apps/icd/icd10online/.
71
CHAPTER 4: THE IMPACT OF TEMPERATURE ON MORTALITY IN TIANJIN,
CHINA: A CASE−CROSSOVER DESIGN WITH A DISTRIBUTED LAG NON-
LINEAR MODEL
Citation:
Guo Y, Barnett AG, Pan X, Yu W, Tong S. (2011) The impact of temperature on mortality in
Tianjin, China: a case−crossover design with a distributed lag non-linear model.
Environmental Health Perspectives 119:1719-1725.
Statement of the joint authorship:
Yuming Guo (Candidate): Conceived and conducted the study design, performed data
analysis and wrote the manuscript;
Adrian Barnett: Performed data analysis, reviewed, edited, and revised the manuscript;
Xiaochuan Pan: Reviewed, edited, and revised the manuscript;
Weiwei Yu: Reviewed, edited, and revised the manuscript;
Shilu Tong: Reviewed, edited, and revised the manuscript.
72
4.1 Abstract
Background: There has been increasing interest in assessing the impacts of temperature on
mortality. However, few studies have used a case–crossover design to examine non-linear
and distributed lag effects of temperature on mortality. Additionally, little evidence is
available on the temperature-mortality relationship in China, or what temperature measure is
the best predictor of mortality.
Objectives: To use a distributed lag non-linear model (DLNM) as a part of case–crossover
design. To examine the non-linear and distributed lag effects of temperature on mortality in
Tianjin, China. To explore which temperature measure is the best predictor of mortality;
Methods: The DLNM was applied to a case−crossover design to assess the non-linear and
delayed effects of temperatures (maximum, mean and minimum) on deaths (non-accidental,
cardiopulmonary, cardiovascular and respiratory).
Results: A U-shaped relationship was consistently found between temperature and mortality.
Cold effects (significantly increased mortality associated with low temperatures) were
delayed by 3 days, and persisted for 10 days. Hot effects (significantly increased mortality
associated with high temperatures) were acute and lasted for three days, and were followed
by mortality displacement for non-accidental, cardiopulmonary, and cardiovascular deaths.
Mean temperature was a better predictor of mortality (based on model fit) than maximum or
minimum temperature.
Conclusions: In Tianjin, extreme cold and hot temperatures increased the risk of mortality.
Results suggest that the effects of cold last longer than the effects of heat. It is possible to
combine the case−crossover design with DLNMs. This allows the case−crossover design to
flexibly estimate the non-linear and delayed effects of temperature (or air pollution) whilst
controlling for season.
73
Key words: Cardiovascular mortality; Case−crossover; Distributed lag non-linear model; Mortality;
Respiratory mortality; Temperature
74
4.2 Introduction
Heat-related mortality has become a matter of increasing public health significance,
especially in the light of climate change. Studies have examined hot and cold temperatures in
relation to total non-accidental deaths and cause-specific deaths (Stafoggia, Forastiere,
Agostini, Biggeri, Bisanti, Cadum, Caranci, Donato, et al., 2006). The city- or region-specific
temperature-mortality relationship is often V-, U- or J-shaped, with increases in mortality at
temperatures below (above) the cold (hot) threshold (Hajat & Kosatky, 2010). The
temperature-mortality relationship varies greatly by geographic, climate and population
characteristics (The Eurowinter Group, 1997). Social, economic, demographic and
infrastructure factors can influence the sensitivity of populations to temperature (Ebi, Kovats,
& Menne, 2006b). In China, only a few studies on temperature-mortality relationship have
been conducted in Shanghai (Kan, Jia, & Chen, 2003), Hong Kong (Chan, Goggins, Kim, &
Griffiths, 2010) and Beijing (Liu et al., 2011). No research has been undertaken in Tianjin,
one of the largest cities in northeastern China.
A previous study found that no temperature measure (maximum, mean or minimum
temperature) was consistently better at predicting mortality in the US. The best temperature
measure differed by age group, season and region (Barnett, Tong, et al., 2010). It is unknown
which temperature measure is the best predictor of mortality in Tianjin.
Mortality risk depends not only on exposure to the current day’s temperature, but also on
several previous days’ exposure (Anderson & Bell, 2009). The distributed lag model has been
applied to explore the delayed effect of temperature on mortality (Analitis, et al., 2008;
Baccini, et al., 2008; Hajat, et al., 2005). To overcome the strong correlation between daily
75
temperatures over short time periods, constrained distributed lag structures are used in time
series regressions (Armstrong, 2006). The estimates are constrained by smoothing using
methods such as natural cubic splines, polynomials, or stratified lag. Both unconstrained and
constrained distributed lag models assume a linear relationship between temperature below
(above) the cold (hot) threshold and mortality, so these models may not be sufficiently
flexible to capture the effects of temperature on mortality.
Recently, a distributed lag non-linear model (DLNM) was developed to simultaneously
estimate the non-linear and delayed effects of temperature (or air pollution) on mortality (or
morbidity) (Armstrong, 2006; Gasparrini, et al., 2010). DLNMs use a “cross-basis” function
that describes a two-dimensional temperature-response relationship along the dimensions of
temperature and lag. The choice of “cross-basis” functions for the temperature and lag are
independent, so the spline or linear functions can be used for temperature, while the
polynomial functions can be used for the lag. The estimates can be plotted using a 3-
dimensional graph to show the relative risks along both temperature and lags. We can predict
the relative risks for a certain temperature or lag, by extracting a “slice” from the 3-
dimensional graph. We can compute the overall effect by summing the log relative risks of
each lag. Separate smoothing functions are applied to time in order to control for season and
secular trends.
The case−crossover design controls for seasonal effects and secular trends by matching case
and control days in relatively small time windows (e.g., calendar month). This controls for
season using a step-function rather than a smooth spline function (Barnett & Dobson, 2010).
Most previous studies used the case–crossover design with relatively inflexible models to
investigate the effects of temperature on mortality, such as assuming a linear effect for
76
temperature in each season, with a single lag model, or moving average lag model (Basu, et
al., 2008; Green, et al., 2010). Few studies have demonstrated how to fit non-linear and
delayed effects of temperature on mortality within a case–crossover design.
We used DLNMs combined with the case–crossover design, making it possible to fit more
sophisticated estimates of the effects of temperature (or air pollution) using a case–crossover
design. We demonstrated these models here using a motivating example of the temperature-
mortality relationship in Tianjin, China, and also investigated which temperature measure had
the best predictive ability for mortality.
4.3 Materials and methods
4.3.1 Data collection
Tianjin is a city in northeastern China, and is adjacent to Beijing and Hebei Province, along
the coast of Bohai Gulf (39° 07' North, 117° 12' East). Tianjin has four distinct seasons, with
cold, windy, dry winters influenced by the vast Siberian anticyclone, and hot, humid
summers due to the monsoon. It is the fifth largest Chinese city in terms of urban land area.
The population in the urban area was 4.2 million in 2005.
Mortality data was obtained from the China Information System for Death Register and
Report of Chinese Centre for Disease Control and Prevention from January 1, 2005 to
December 31, 2007. The mortality data were from six urban districts of Tianjin (Heping,
Hedong, Hexi, Nankai, Hebei and Hongqiao). Non-accidental mortality was classified
according to the International Classification of Diseases, 10th revision (ICD-10: A00–R99)
(World Health Organization, 2007). Cardiopulmonary (ICD-10:I00–I99 and ICD-10:J00–
77
J99), cardiovascular mortality (ICD-10:I00–I99) and respiratory mortality (ICD-10:J00–J99)
were examined separately.
Daily meteorological data on maximum, mean and minimum temperature, and relative
humidity, were obtained from the China Meteorological Data Sharing Service System
(http://cdc.cma.gov.cn). Daily air pollution data on particulate matter less than 10 μm in
aerodynamic diameter (PM10), sulphur dioxide (SO2) and nitrogen dioxide (NO2) were
obtained from the Tianjin Environmental Monitoring Centre.
4.3.2 Data analysis
The time-stratified case–crossover using a fixed and disjointed window (e.g., calendar month)
avoids the “overlap bias” (Janes, et al., 2005). The case–crossover using conditional logistic
regression is a special case of time series analysis (Lu & Zeger, 2007). This equivalence
provides computational convenience and permits model checking for the case–crossover
design using standard log-linear model diagnostics (Lu, et al., 2008). We used a Poisson
regression model that allows for over-dispersion to combine the case–crossover design with a
DLNM:
Yt ~ Poisson(μt)
Log (μt) = α + βTt,l + S(RHt, 3) + S(PM10t, 3) + S(SO2t, 3) + S(NO2t, 3)
+ λStratat + ηDOWt + υHolidayt + δInfluenzat
= α + βT t,l + COVs , [4.1]
where t is the day of the observation; Yt is the observed daily death counts on day t; α is the
intercept; Tt,l is a matrix obtained by applying the DLNM to temperature, β is vector of
coefficients for Tt,l, and l is the lag days. S(.) is a natural cubic spline. Three degrees of
78
freedom were used to smooth relative humidity, PM10, NO2, and SO2 according to previous
studies (Anderson & Bell, 2009; Stafoggia, Schwartz, Forastiere, & Perucci, 2008). Stratat is
a categorical variable of the year and calendar month used to control for season and trends,
and λ is vector of coefficients. DOWt is day of the week on day t, and η is vector of
coefficients. Holidayt is a binary variable that is “1” if day t was a holiday. Influenzat is a
binary variable that is “1” if there were any influenza deaths on day t.
Based on the vector of estimated coefficients β in model [4.1], the DLNM was used to get the
predicted effects and standard errors for combinations of temperature and lags. Graphs,
summaries, and statistical inference can be obtained from the DLNM estimates and standard
errors (Armstrong, 2006).
We used a “natural cubic spline-natural cubic spline” DLNM that modelled both the non-
linear temperature effect and the lagged effect using a natural cubic spline. We placed spline
knots at equal spaces in the temperature range to allow enough flexibility in the two ends of
temperature distribution. We placed spline knots at equal intervals in the log scale of lags to
allow more flexible lag effects at shorter delays. To completely capture the overall
temperature effect and adjust for any potential harvesting (heat-related excesses of mortality
were followed by deficits), we used lags up to 27 days according to a previous study
(Armstrong, 2006). The median value of temperature was defined as the baseline temperature
(“centering value”) for calculating the relative risks. To choose the degree of freedom (knots)
for temperature and lag, we used Akaike information criterion (AIC) for quasi-Poisson
models (Gasparrini, et al., 2010; Peng, Dominici, & Louis, 2006). We found that 5 degrees of
freedom for temperature and 4 degrees of freedom for lag produced the best model fitting.
We plotted the relative risks against temperature and lags to show the entire relationship
79
between temperature and mortality. We also plotted the overall effect of temperature on
morality summed over lag days.
Our initial analysis found that the temperature-mortality relationships were U-shaped, with
potential cold and hot thresholds. Thus we also used a “double threshold-natural cubic spline”
DLNM that assumes the effect of cold temperature is linear below the cold threshold while
the effect of high temperature is linear above the hot threshold, and models the lag effects
using a natural cubic spline with 4 degrees of freedom. Formula [4.1] was altered by
modifying the βTi,l term into two linear threshold terms:
Log (μt) = α + βcTCt,l + βHTHt,l + COVs , [4.2]
where TCt,l (THt,l) is a matrix obtained by applying the “double threshold-natural cubic
spline” DLNM to temperatures below the cold threshold and above the hot threshold.
Temperature thresholds used in the model [4.2] were determined by testing multiple
thresholds. For example, for mean temperature, our initial analysis indicated that the potential
cold threshold was within −5 to 5 °C, and the potential hot threshold was within 19 to 29 °C.
Hence we examined combinations of cold thresholds from −5.0 to 5.0 °C (in 0.1 °C gaps) and
hot thresholds from 19.0 to 29.0 °C (in 0.1 °C gaps) to identify the combination that
minimised the residual deviance. We then estimated the relative risks of mortality for a 1 °C
decrease in temperature below the cold threshold and a 1 °C increase above the hot threshold.
The temperature-mortality relationships for combinations of temperature measures
(maximum, mean, and minimum temperatures) and mortality categories (Non-accidental,
cardiopulmonary, cardiovascular, and respiratory deaths) were each examined using the
80
above steps. The AIC was used to choose the temperature measure that best predicted
mortality.
Sensitivity analyses were performed by changing the window length in the case–crossover
from calendar month to 30, 28 and 21 days to control for season, and varying the maximum
lags to 20 and 30 days for the DLNM.
All statistical tests were two-sided and values of P<0.05 were considered statistically
significant. Spearman’s correlation coefficients were used to summarize the similarities in
daily weather conditions. The R software (version 2.12.1, R Development Core Team 2009)
was used to fit all models, with the “dlnm” package to create the DLNM (Gasparrini &
Armstrong, 2011a).
A detailed explanation of how to combine the case–crossover with DLNM is provided in the
supplemental material (see Supplemental Material Chapter 4, R code).
4.4 Results
The average daily maximum temperature was 19 °C, mean temperature 13 °C, minimum
temperature 8 °C, and relative humidity 60%. On average there were 56 daily non-accidental
deaths, 34 cardiopulmonary deaths, 30 cardiovascular deaths, and 4 respiratory deaths (Table
4.1). The three temperature measures were strongly correlated (Table 4.2).
81
Table 4.1: Summary statistics of daily weather conditions and mortality in Tianjin, China,
2005–2007
Variables Minimum 25% Median 75% Maximum Mean SD
Maximum temperature (°C) –6 8 21 30 40 19 12
Mean temperature (°C) –11 3 14 24 31 13 11
Minimum temperature (°C) –14 –2 10 19 29 8 11
Humidity (%) 13 46 61 74 97 60 19
Non-accidental death 26 46 55 66 106 56 14
Cardiopulmonary death 13 27 33 40 77 34 9
Cardiovascular death 9 24 29 35 67 30 8
Respiratory death 0 3 4 6 15 4 2
Influenza death 0 0 0 0 2 0 0.1
SD = standard deviation
Table 4.2: Spearman’s correlation coefficients between weather conditions in Tianjin, China,
2005–2007
Temperature measures Mean
temperature
Minimum
temperature
Humidity
Maximum temperature 0.98** 0.94** 0.16*
Mean temperature 0.98** 0.24*
Minimum temperature 0.32*
*P<0.05
**P<0.01
82
Mean temperature generally gave the lowest AIC values (i.e., had the best predictive ability
for mortality) in Tianjin (see Supplemental Material Chapter 4, Table S4.1). The “double
threshold-natural cubic spline” DLNM generally fit the data better than the “natural cubic
spline-natural cubic spline” DLNM (see Supplemental Material Chapter 4, Table S4.1).
Therefore we report results for associations with mean temperature only.
The 3-dimensional plots show the entire surface between mean temperature and mortality
categories at all lag days (Figure 4.1). The estimated effects of temperature were non-linear
for all mortality types, with higher relative risks at hot and cold temperatures. For example,
extreme hot temperature (30 °C) was positively associated with non-accidental mortality on
current day, whilst extreme cold temperature (–6 °C) significantly increased non-accidental
mortality after 3-days lag. Neither hot effects (i.e., significant increases in mortality
associated with hot temperatures) nor cold effects (i.e., significant increases in mortality
associated with cold temperatures) were apparent after a 20-day lag, with relative risks close
to one across the entire range of temperatures (see Supplemental Material Chapter 4, Figure
S4.1).
83
Figure 4.1: Relative risks of mortality types by mean temperature (°C), using a natural cubic
spline–natural cubic spline DLNM with 5 df natural cubic spline for temperature and 4 df for
lag. (A) Nonaccidental, (B) cardiopulmonary, (C) cardiovascular, and (D) respiratory
mortality.
84
Figure 4.2: The estimated overall effects of mean temperature (°C) over 28 days on mortality
types, using a natural cubic spline–natural cubic spline DLNM with 5 df natural cubic spline
for temperature and 4 df for lag. (A) Nonaccidental, (B) cardiopulmonary, (C) cardiovascular,
and (D) respiratory mortality. The black lines are the mean relative risks, and the blue regions
are 95% CIs.
85
Figure 4.3: The estimated effects of a 1°C decrease in mean temperature below the cold
threshold (left) and of a 1°C increase in mean temperature above the hot threshold (right) on
mortality types over 27 days of lag, using a double threshold–natural cubic spline DLNM
with 4 df natural cubic spline for lag. (A) Nonaccidental, (B) cardiopulmonary, (C)
cardiovascular, and (D) respiratory mortality. The black lines are mean relative risks, and
blue regions are 95% CIs. The cold and hot thresholds were 0.8°C and 24.9°C for
nonaccidental mortality (A), 0.1°C and 25.3°C for cardiopulmonary mortality (B), 0.6°C and
25.1°C for cardiovascular mortality (C), 0.7°C and 24.8°C for respiratory mortality (D).
86
Figure 4.2 shows the estimated effect of mean temperature over 28 days on mortality. There
were U-shaped relationships between mean temperature and all mortality types, with large
“comfortable” temperature ranges where the relative risks of mortality were close to one. The
cold and hot thresholds (i.e., the temperatures below and above which estimates were
constrained to be linear by the model, which do not necessarily coincide with temperatures
associated with increased mortality by model [4.1]) were 0.8 °C and 24.9 °C for non-
accidental mortality, 0.1°C and 25.3 °C for cardiopulmonary mortality, 0.6 °C and 25.1 °C
for cardiovascular mortality, 0.7 °C and 24.8 °C for respiratory mortality.
Significant cold effects appeared after after a 3-day lag, while significant hot effects occurred
within 0 to 3 days (Figure 4.3). Associations between cold and mortality lasted longer than
associations with heat. Heat-related excesses of non-accidental, cardiopulmonary, and
cardiovascular mortality were followed by deficits in mortality, consistent with some
mortality displacement caused by hot temperatures.
We calculated the overall effects of mean temperature on non-accidental, cardiopulmonary,
cardiovascular and respiratory mortality along the lags (Table 4.3). For cold effects over lag
0–18 days, a 1 °C decrease in mean temperature below the cold thresholds was associated
with a 2.99% (95% confidence interval (CI): 0.85–5.17%) increase in non-accidental deaths,
5.49% (95% CI: 2.29–8.79%) increase in cardiopulmonary deaths, 4.05% (95% CI: 1.14–
7.06%) increase in cardiovascular deaths, and 9.25% (95% CI: 1.70–17.37%) increase in
respiratory deaths. For hot effects over lag 0–2 days, a 1 °C increase in mean temperature
above the hot thresholds was associated with a 2.03% (95% CI: 0.70–3.38%) increase in non-
accidental deaths, 3.04% (95% CI: 1.24–4.87%) increase in cardiopulmonary deaths, 2.80%
(95% CI: 0.95–4.68%) in cardivascular deaths, and 3.36% (95% CI: –0.77 to 7.67%) increase
87
in respiratory deaths. In general, cold effects of lag 0–27 days were greater than hot effects of
lag 0-27 days except for respiratory mortality.
Table 4.3: The cumulative cold and hot effects of mean temperature on mortality categories
along the lag days, using a “double threshold-natural cubic spline” DLNM with 4 degrees of
freedom natural cubic spline for lag.
Effects
Lag
(days)
% increase in mortality (95% CI)
Non-accidental Cardiopulmonary Cardiovascular Respiratory
Cold effect a 0–2 –0.27 (–1.25, 0.72) –0.19 (–1.49, 1.12) –0.14 (–1.43, 1.17) –1.65 (–4.75, 1.55)
0–18 2.99 (0.85, 5.17)* 5.49 (2.29, 8.79)* 4.05 (1.14, 7.06)* 9.25 (1.70, 17.37)*
0–27 2.13 (–0.44, 4.78) 4.16 (0.27, 8.21)* 2.66 (–0.86, 6.30) 7.99 (–1.08, 17.9)
Hot effect b 0–2 2.03 (0.70, 3.38)* 3.04 (1.24, 4.87)* 2.80 (0.95, 4.68)* 3.36 (–0.77, 7.67)
0–18 –0.78 (–4.20, 2.77) 2.32 (–2.59, 7.49) 0.86 (–4.02, 5.98) 8.60 (–2.78, 21.31)
0–27 0.31 (–3.48, 4.24) 3.83 (–1.75, 9.72) 2.47 (–2.99, 8.24) 8.79 (–3.62, 22.80)
*P<0.05
a The percent increase in mortality for a 1 °C of temperature decrease below the cold thresholds
(0.8 °C for non-accidental, 0.1 °C for cardiopulmonary 0.6 °C for cardiovascular, and 0.7 °C
respiratory mortality).
b The percent increase in mortality for a 1 °C of temperature increase above the hot thresholds
(24.9 °C for non-accidental, 25.3 °C for cardiopulmonary 25.1 °C for cardiovascular, and 24.8 °C for
respiratory mortality).
88
Sensitivity analysis
We changed the window length of calendar month in the case–crossover to 30, 28, and 21
days, which gave similar results (data not shown). In addition, we changed the maximum lag
to 20 and 30 days, which gave similar results (data not shown). Consequently, we believe that
the models used in this study adequately captured the main effects of temperature on
mortality.
4.5 Discussion
4.5.1 Temperature-mortality relationship
The temperature-mortality relationship in Tianjin was U-shaped, with a large range of
temperatures that were not associated with excess mortality. Significant associations between
cold temperatures and mortality (cold effects) appeared after 3 days and lasted longer than the
associations between high temperatures and mortality (hot effects), which were acute and of
short duration. There was evidence of some mortality displacement due to effects of high
temperatures on non-accidental, cardiopulmonary, and cardiovascular deaths.
Many studies have examined the temperature-mortality relationship worldwide, but few are
from China (Hajat & Kosatky, 2010). We compared our results with studies that examined
both cold and hot effects using mean temperature for non-accidental mortality (Curriero et al.,
2002b; El-Zein, Tewtel-Salem, & Nehme, 2004; Revich & Shaposhnikov, 2008b; Rocklov &
Forsberg, 2008; Yu, Mengersen, Hu, et al., 2011) (Figure 4.4). Results show that estimated
temperature effects varied by region and population. Compared with populations living at
similar latitudes, our results suggest a stronger cold effect and smaller hot effect. The reason
89
might be that short lags were used in other studies, while we examined overall cold and hot
effects of lag 0–27 days. Studies using short lags may have underestimated the cold effect, as
in our results the estimated cold effect was delayed by three days and lasted for 10 days.
Studies using short lags may overestimate the hot effect, as in our results there was evidence
of some mortality displacement which can only be captured by using longer lags (Anderson
& Bell, 2009). Compared with other median or lower income populations (e.g., Bangkok,
Mexico City, São Paulo, Delhi, Santiago, and Cape Town), Tianjin had lower cold and hot
effects. The reason might be that people in Tianjin used protection measures in summer and
winter (e.g., air conditioning and heating system) (McMichael, et al., 2008).
Figure 4.4: Comparison of the impacts of temperature on nonaccidental mortality in different
populations ordered by latitude.
90
We can compare our results with those from similar cities in China. Kan et al. (2003) found a
V-shaped relationship between lag 0–2 days’ temperature and non-accidental mortality in
Shanghai, with an optimum temperature of 26.7 °C. A 1 °C decrease (increase) in
temperature below (above) 26.7 °C increased non-accidental mortality by 1.21% (0.73%).
Liu et al. (2011) found both cold and hot temperatures were associated with increased
cardiopulmonary mortality in Beijing, which has a climate that is similar to Tianjin’s. They
also found an acute and short-term hot effect followed by some mortality displacement for
cardiovascular mortality, consistent with our results.
An interesting finding is that the range of temperatures that are not associated with increased
mortality is quite large in Tianjin, but extreme temperatures still had adverse effects on
mortality. The exchange of heat between the body and surrounding temperature is regulated
constantly by physiological control. Extreme high temperatures may cause a failure of
thermoregulation, which may be impaired by dehydration, salt depletion and increased
surface blood circulation (Bouchama & Knochel, 2002). Elevated blood viscosity, cholesterol
levels and sweating thresholds may also be the cause of heat-related mortality (McGeehin &
Mirabelli, 2001). Cold temperatures increase the heart rate, peripheral vasoconstriction, blood
pressure, blood cholesterol levels, plasma fibrinogen concentrations, and platelet viscosity
(Ballester, et al., 1997a; Carder, et al., 2005b). In Tianjin urban city, eighty-three percent of
houses had central heating in winter (Tianjin Statistic Bureau, 2005) and ninety percent of
homes had air conditioners (Tianjin Statistic Bureau, 2004). However, although the majority
of the urban population were potentially protected from the weather, there were still some
increased risks during extreme cold and hot days.
91
We investigated lag effects over 28 days on mortality for both hot and cold days. In general,
cold effects lasted about 10 days after the extreme cold days. Previous studies also reported
similarly delayed cold effects on mortality (Anderson & Bell, 2009; Goodman, Dockery, &
Clancy, 2004). The findings indicate that using short lags cannot completely capture the cold
effect, and so longer lags are required to examine the cold impact.
The hot effects were more acute and short-term. Studies have shown that hot temperatures
induce an acute event in people with pre-existing diseases (e.g., a previous myocardial
infarction or stroke) and in those who may find it difficult to deal with heat (e.g., the elderly)
(Muggeo & Hajat, 2009). In people with congestive heart failure, the extra heat load may lead
to fatal consequences (Näyhä, 2005). The hot effect also led mortality displacement for non-
accidental, cardiopulmonary, and cardiovascular deaths, which is in agreement with studies
conducted in Europe (Hajat, et al., 2005; Pattenden, Nikiforov, & Armstrong, 2003) and US
(Braga, Zanobetti, & Schwartz, 2001). Therefore, using short lags cannot adequately assess
the hot effects, as the harvesting effects were ignored.
Studies of heat-related mortality have examined maximum, mean, or minimum temperatures,
controlling for relative humidity (Anderson & Bell, 2009). Other studies have used apparent
temperature, the humidex and temporal synoptic index (Zanobetti & Schwartz, 2008). A large
study of mortality in the US found that the different measures of temperature had a similar
ability to predict the impacts of temperature on mortality (Barnett, Tong, et al., 2010). We
found that maximum, mean, and minimum temperatures had similar predictive ability,
probably because of their strong correlation. Overall, mean temperature performed best
according to the AIC.
92
4.5.2 Case−crossover design and DLNM
Many models have been used to assess the impacts of temperature and air pollution on
mortality and morbidity, such as descriptive (Reid, et al., 2009), case-only (Schwartz, 2005),
case–crossover (Stafoggia, Forastiere, Agostini, Biggeri, Bisanti, Cadum, Caranci, Donato, et
al., 2006), time-series (Hajat, et al., 2002) and spatial analysis (Vaneckova, et al., 2010).
Generally, time-series and case–crossover designs are the most commonly used in a single or
in multiple locations over a time period. The main aim of both analyses is to examine
associations between health and temperature, after controlling for potential confounding
factors such as secular trends and seasonal cycles (Basu, et al., 2005). Using the case–
crossover design each subject is their own control, and so any confounding by fixed
characteristics is removed. Another advantage of the case-crossover is that it controls for
long-term and seasonal trends by design through short-interval strata (e.g. calendar month).
We compared the case–crossover design and a time series design using a natural cubic spline
with 7 degrees of freedom for time per year. The case–crossover design performed better than
time series analysis for this particular data based on AIC and residuals. However, we cannot
conclude the case–crossover is better than time series for other data. We suggest checking the
model fit and residuals when using case–crossover or time series designs. In this study, we
illustrated how to combine the DLNM with a case−crossover design. This allows
sophisticated non-linear and delayed temperatures to be fitted using the case−crossover
design.
One of the main advantages of DLNM is that it allows the model to contain detailed lag
effects of exposure on response, and provides the estimate of the overall effect that is
93
adjusted for harvesting (Gasparrini, et al., 2010). The DLNM can flexibly show different
temperature-mortality relationships for lags using different smoothing functions. The DLNM
can adequately model the main effects of temperature (Armstrong, 2006).
There are also some issues in the selection of the DLNM, such as cross-basis type, maximum
lag day, and degrees of freedom (knots and placement) for exposure and lag (Armstrong,
2006; Gasparrini, et al., 2010). Because the DLNM is combined with a regression model (e.g.,
Poisson regression), the residual deviance and autocorrelation plot, maximum likelihood,
Akaike’s information criteria or Bayesian information criteria can be used to check the model.
The options for the DLNM can be chosen according to the best model fit. Previous studies
recommend choosing a DLNM that is easy to interpret from an epidemiological perspective
(Armstrong, 2006; Gasparrini, et al., 2010). However, it is necessary to conduct sensitivity
analyses to assess the key conclusions on model choice. In this study, we used AIC to select
the degrees of freedom, and used residual deviance to choose both cold and hot thresholds,
but used a priori arguments to choose cross-basis type and maximum lag day.
4.5.3 Strengths and limitations
We examined both cold and hot lag effects on four types of mortality, and explored which
temperature measure was the best predictor of mortality. Our findings can be used to promote
capacity building for local response for extreme temperatures.
A limitation is that the data are only from one city, so it is difficult to generalise our results to
other cities or to rural areas. We used the data on temperature and air pollution from fixed
sites rather than individual exposure, so there may be some inevitable measurement error.
94
The influence of ozone was not controlled for, because data on ozone were unavailable. In
previous research, hot effects were slightly reduced when ozone was controlled for, but cold
effects were not changed (Anderson & Bell, 2009). Some studies found a potential interaction
between temperature and ozone (Ren, Williams, Morawska, Mengersen, & Tong, 2008a).
Further study needs to be conducted for this issue.
4.6 Conclusions
DLNM can be applied in a case−crossover design, so that the case−crossover can be used to
examine sophisticated non-linear and delayed effects of exposure (e.g., temperature or air
pollution). Even though there was a relatively large temperature range that was not associated
with excess mortality, extreme cold and hot temperatures were associated with an increased
risk of mortality in Tianjin, China. Cold temperatures had longer lasting effects on mortality,
while hot temperatures had acute and short-term effects.
95
4.7 References
Analitis A, Katsouyanni K, Biggeri A, Baccini M, Forsberg B, Bisanti L, et al. (2008).
Effects of cold weather on mortality: results from 15 European cities within the
PHEWE project. American Journal of Epidemiology, 168(12):1397-1408.
Anderson BG, Bell ML. (2009). Weather-related mortality: how heat, cold, and heat waves
affect mortality in the United States. Epidemiology, 20(2):205-213.
Armstrong B. (2006). Models for the relationship between ambient temperature and daily
mortality. Epidemiology, 17(6):624-631.
Baccini M, Biggeri A, Accetta G, Kosatsky T, Katsouyanni K, Analitis A, et al. (2008). Heat
effects on mortality in 15 European cities. Epidemiology, 19(5):711-719.
Ballester F, Corella D, Perez-Hoyos S, Saez M, Hervas A. (1997). Mortality as a function of
temperature, a study in Valencia, Spain, 1991-1993. International Journal of
Epidemiology, 26(3):551-561.
Barnett AG, Dobson AJ. (2010). Analysing Seasonal Health Data. Berlin, Heidelberg:
Springer.
Barnett AG, Tong S, Clements ACA. (2010). What measure of temperature is the best
predictor of mortality? Environmental Research, 110(6):604-611.
Basu R, Dominici F, Samet JM. (2005). Temperature and mortality among the elderly in the
United States: a comparison of epidemiologic methods. Epidemiology, 16(1):58-66.
Basu R, Feng WY, Ostro BD. (2008). Characterizing temperature and mortality in nine
California counties. Epidemiology, 19(1):138-145.
Bouchama A, Knochel JP. (2002). Heat stroke. N Engl J Med 346(25):1978-1988.
Braga AL, Zanobetti A, Schwartz J. (2001). The time course of weather-related deaths.
Epidemiology, 12(6):662-667.
96
Carder M, McNamee R, Beverland I, Elton R, Cohen GR, Boyd J, et al. (2005). The lagged
effect of cold temperature and wind chill on cardiorespiratory mortality in Scotland.
Occupational and Environmental Medicine, 62(10):702-710.
Chan EY, Goggins WB, Kim JJ, Griffiths SM. (2010). A study of intracity variation of
temperature-related mortality and socioeconomic status among the Chinese
population in Hong Kong. Journal of Epidemiology and Community Health,
doi:10.1136/jech.2008.085167.
Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz JA. (2002). Temperature and
mortality in 11 cities of the eastern United States. American Journal of
Epidemiology, 155(1):80-87.
Ebi KL, Kovats RS, Menne B. (2006). An approach for assessing human health vulnerability
and public health interventions to adapt to climate change. Environmental Health
Perspectives, 114(12):1930-1934.
El-Zein A, Tewtel-Salem M, Nehme G. (2004). A time-series analysis of mortality and air
temperature in Greater Beirut. Science of The Total Environment, 330(1-3):71-80.
Gasparrini A, Armstrong B. (2011). Distributed lag non-linear models in R: the package
dlnm.
Gasparrini A, Armstrong B, Kenward MG. (2010). Distributed lag non-linear models.
Statistics in Medicine, 29(21):2224-2234.
Goodman PG, Dockery DW, Clancy L. (2004). Cause-Specific Mortality and the Extended
Effects of Particulate Pollution and Temperature Exposure. Environmental Health
Perspectives, 112(2):179-185.
Green RS, Basu R, Malig B, Broadwin R, Kim JJ, Ostro B. (2010). The effect of temperature
on hospital admissions in nine California counties. International Journal of Public
Health, 55(2):113-121.
97
The Eurowinter Group. (1997). Cold exposure and winter mortality from ischaemic heart
disease, cerebrovascular disease, respiratory disease, and all causes in warm and
cold regions of Europe. The Eurowinter Group. Lancet, 349(9062):1341-1346.
Hajat S, Kosatky T. (2010). Heat-related mortality: a review and exploration of
heterogeneity. Journal of Epidemiology and Community Health, 64(9):753-760.
Hajat S, Kovats RS, Atkinson RW, Haines A. (2002). Impact of hot temperatures on death in
London: a time series approach. Journal of Epidemiology and Community Health,
367-372.
Hajat S, Armstrong BG, Gouveia N, Wilkinson P. (2005). Mortality displacement of heat-
related deaths: a comparison of Delhi, Sao Paulo, and London. Epidemiology,
16(5):613-620.
Janes H, Sheppard L, Lumley T. (2005). Overlap bias in the case-crossover design, with
application to air pollution exposures. Statistics in Medicine, 24(2):285-300.
Kan H.D, Jia J, Chen B.H. (2003). Temperature and daily mortality in Shanghai: a time-series
study. Biomedical and Environmental Science, 16(2):133-139.
Liu L, Breitner S, Pan X, Franck U, Leitte A, Wiedensohler A, et al. (2011). Associations
between Air Temperature and Cardio-Respiratory Mortality in the Urban Area of
Beijing, China: A Time-Series Analysis. Environmental Health, 10(1):51.
Lu Y, Zeger S.L. (2007). On the equivalence of case-crossover and time series methods in
environmental epidemiology. Biostatistics, 8(2):337-344.
Lu Y, Symons J.M, Geyh A.S, Zeger S.L. (2008). An approach to checking case-crossover
analyses based on equivalence with time-series methods. Epidemiology, 19(2):169-
175.
98
McGeehin M, Mirabelli M. (2001). The potential impacts of climate variability and change
on temperature-related morbidity and mortality in the United States. Environmental
Health Perspectives, 109(Suppl 2):185-189.
McMichael A.J, Wilkinson P, Kovats R.S, Pattenden S, Hajat S, Armstrong B, et al. (2008).
International study of temperature, heat and urban mortality: the 'ISOTHURM'
project. International Journal of Epidemiology, 37(5):1121-1131.
Muggeo VM, Hajat S. (2009). Modelling the non-linear multiple-lag effects of ambient
temperature on mortality in Santiago and Palermo: a constrained segmented
distributed lag approach. British Medical Journal, 66(9):584-591.
Näyhä S. (2005). Environmental temperature and mortality. International Journal of
Circumpolar Health, 64(5):451-458.
Pattenden S, Nikiforov B, Armstrong B.G. (2003). Mortality and temperature in Sofia and
London. Journal of Epidemiology and Community Health, 57(8):628-633.
Peng R.D, Dominici F, Louis T.A. (2006). Model choice in time series studies of air pollution
and mortality. Journal of the Royal Statistical Society: Series A, 169(2):179-203.
Reid C, O’Neill M, Gronlund C, Brines S, Brown D, Diez-Roux A, et al. (2009). Mapping
community determinants of heat vulnerability. Environmental Health Perspectives,
117(11):1730-1736.
Ren C, Williams G.M, Morawska L, Mengersen K, Tong S. (2008). Ozone modifies
associations between temperature and cardiovascular mortality: analysis of the
NMMAPS data. Occupational and Environmenal Medicine, 65(4):255-260.
Revich B, Shaposhnikov D. (2008). Temperature-induced excess mortality in Moscow,
Russia. International Journal of Biometeorology, 52(5):367-374.
99
Rocklov J, Forsberg B. (2008). The effect of temperature on mortality in Stockholm 1998--
2003: a study of lag structures and heatwave effects. Scandinavian Journal of Public
Health, 36(5):516-523.
Schwartz J. (2005). Who is sensitive to extremes of temperature?: A case-only analysis.
Epidemiology, 16(1):67.
Stafoggia M, Schwartz J, Forastiere F, Perucci C.A. (2008). Does temperature modify the
association between air pollution and mortality? A multicity case-crossover analysis
in Italy. American Journal of Epidemiology, 167(12):1476-1485.
Stafoggia M, Forastiere F, Agostini D, Biggeri A, Bisanti L, Cadum E, et al. (2006).
Vulnerability to heat-related mortality: a multicity, population-based, case-crossover
analysis. Epidemiology, 17(3):315-323.
Tianjin Statistic Bureau. 2004. Tianjin Statistical Bulletin. http://www.stats-
tj.gov.cn/Article/tjgb/stjgb/200612/5371.html.
Tianjin Statistic Bureau. 2005. Tianjin Statistical Bulletin. http://www.stats-
tj.gov.cn/Article/tjgb/stjgb/200612/5375.html.
Vaneckova P, Beggs P.J, Jacobson C.R. (2010). Spatial analysis of heat-related mortality
among the elderly between 1993 and 2004 in Sydney, Australia. Social Science
Medicine, 70(2):293-304.
World Health Organization. (2007). International Statistical Classification of Diseases and
Related Health Problems, 10th Revision, Version for 2007.
Yu W, Mengersen K, Hu W, Guo Y, Pan X, Tong S. (2011). Assessing the relationship
between global warming and mortality: Lag effects of temperature fluctuations by
age and mortality categories. Environmental Pollution, 159(7):1789-1793.
Zanobetti A, Schwartz J. (2008). Temperature and mortality in nine US cities. Epidemiology,
19(4):563-570.
100
4.8 Supplemental Material Chapter 4
Supplemental Material Chapter 4, Table S4.1: Akaike information criteria (AIC) values for
the relationship between temperature measures and mortality categories by DLNM type
DLNM type Temperature measure
AIC
Non-accidental Cardiopulmonary Cardiovascular Respiratory
Natural cubic
spline-natural
cubic spline a
Maximum temperature 7494 6860 6679 4562
Mean temperature 7472 6841 6658 4570
Minimum temperature 7472 6840 6660 4580
Double threshold-
natural cubic
spline b
Maximum temperature 7488 6849 6662 4568
Mean temperature 7473 6833 6653 4558
Minimum temperature 7481 6845 6666 4556
a Using “natural cubic spline-natural cubic spline” DLNM with smoothing of 5 degrees of
freedom for temperature and 4 degrees of freedom for lag;
b Using “double threshold-natural cubic spline” DLNM with smoothing of 4 degrees of
freedom for lag; the cold and hot thresholds are shown in Supplemental Material, Table 2.
101
Supplemental Material Chapter 4, Table S4.2: Cold and hot thresholds (°C) used by the
“double threshold-natural cubic spline”
Threshold type Temperature measure
Mortality type
Non-accidental Cardiopulmonary Cardiovascular Respiratory
Cold threshold
(°C)
Maximum temperature 4.8 4.3 4.5 4.7
Mean temperature 0.8 0.1 0.6 0.7
Minimum temperature –3.9 –4.1 –3.8 –3.5
Hot threshold
(°C)
Maximum temperature 31.2 31.3 31.1 31.5
Mean temperature 24.9 25.3 25.1 24.8
Minimum temperature 20.1 22.0 21.9 21.6
102
Supplemental Material Chapter 4, Figure S4.1: Relative risks by lag at specific mean
temperatures (left) and relative risks by mean temperature at specific lags (right) for non-
accidental mortality, using a “natural cubic spline-natural cubic spline” DLNM with 5
degrees of freedom for temperature and 4 degrees of freedom for lag. The reference
temperature is 14 °C.
103
Supplemental Material Chapter 4, R code
As our data from Tianjin is not publicly available, we used data from Jersey city as an
example. The data were from the National Morbidity, Mortality, and Air Pollution Study
(NMMAPS) (J. M. Samet, Dominici, et al., 2000; J. M. Samet, Zeger, et al., 2000).
Load packages and prepare dataset:
>library(dlnm); library (NMMAPSlite)
>initDB()
>cities <- listCities()
# Jersey City: jers (city number 43)
>data <- readCity(cities[43], collapseAge = TRUE)
>data <- data[,c("city","date","death","inf","tmpd","rhum","so2mean","pm10trend")]
>data$temp <- (data$tmpd-32)*5/9 # Transfer temperature to Celsius
>data$time<-1:length(data[,1]) # Create time
>data$dow<-as.numeric(format(data$date,"%w")) # Create day of the week
>data$year<-as.numeric(format(data$date,"%Y")) # Create year
>data$month<-as.numeric(format(data$date,"%m")) # Create month
>data$strata<-data$year*100+data$month # Case-Control strata
Create Cross-basis matrix using “natural cubic spline-natural cubic spline” DLNM
with 5 df for temperature and 4 df for lag
>range <- range(data$temp,na.rm=T)
>nknots<-4 # Number of knots for temperature
>nlagknots<-2 # Number of knots for lag
104
>ktemp <- range[1] + (range[2]-range[1])/(nknots+1)*1:nknots # Knots for temperature
>klag<-exp((log(27))/(nlagknots+2)*1:nlagknots) # Knots for lag
>basis.temp <- crossbasis(data$temp, vartype="ns", varknots=ktemp,
cenvalue=median(data$temp,na.rm=T), lagtype="ns", lagknots=klag,maxlag=27)
Combine the case-crossover design with DLNM
>model.month <- glm(death ~ basis.temp + ns(rhum,df=3) + ns(pm10trend,df=3) +
ns(so2mean,df=3) + as.factor(I(inf>0)) + as.factor(strata)+as.factor(dow),
family=quasipoisson(), data)
Derive the predicted effects and standard errors for temperature and lags using DLNM
>pred.month <- crosspred(basis.temp, model.month, at=-16:32)
Plot 3D and overall effect graphics
> plot (pred.month,"3d",zlab="Relative Risk", r=90, d=0.3, col="red", xlab="Temperature",
main="3D graphic for Jersey City", expand=0.6,lwd=0.5)
>plot(pred.month,"overall", xlab="Temperature (°C)", ylab=" Relative Risk ",
main="Overall effect of temperature on mortality\n between 1987-2000 for Jersey City")
Determine the cold and hot thresholds (in °C) using “double threshold-natural cubic
spline” DLNM
Based on the above 3D plot and overall effect plot, there are two potential thresholds for
temperature. The cold threshold is somewhere between 0 to 8 °C, and hot threshold is
somewhere between 19 to 26 °C. We used the following models to determine which
combination of cold and hot thresholds gave the lowest residual deviance.
105
>cold.thr<-0:8 # In 1°C increments (In our study, we used 0.1°C increments)
>hot.thr<-19:26 # In 1°C increments (In our study, we used 0.1°C increments)
>deviance.matrix<-matrix(data = NA, nrow = length(cold.thr), ncol = length(hot.thr), byrow
= FALSE, dimnames = list(paste("cold.thr", cold.thr,sep="."),
paste("hot.thr", hot.thr,sep=".")))
>for (i in 1:length(cold.thr)){
for (j in 1:length(hot.thr)){
basis.try <- crossbasis(data$temp, vartype="dthr",varknots=c(cold.thr[i],hot.thr[j]),
lagtype="ns", lagknots=klag, maxlag=27)
model <- glm(death ~ basis.try + ns(rhum,df=3) + ns(pm10trend,df=3) + ns(so2mean,df=3)
+ as.factor(I(inf>0)) + as.factor(strata)+as.factor(dow), family=quasipoisson(), data)
deviance.matrix[i,j]<-model$deviance
}
}
>row.col <- arrayInd(which.min(deviance.matrix), dim(deviance.matrix))
>rowname<-rownames(deviance.matrix)[row.col[,1]]
>colname<-colnames(deviance.matrix)[row.col[,2]]
>rowname;colname # Get the cold and hot thresholds
[1] "cold.thr.4" # The best cold threshold is 4°C
[1] "hot.thr.22" # The best hot threshold is 22 °C
Examine the cold (hot) effects below (above) the cold (hot)threshold using “Double
threshold-natural cubic spline” DLNM
106
The cold threshold 4 °C and hot threshold 22 °C are used for a “Double threshold-natural
cubic spline” DLNM.
>basis.cold.hot<- crossbasis(data$temp, vartype="dthr",varknots=c(4,22),
lagtype="ns", lagknots=klag, maxlag=27)
>model.cold.hot <- glm(death ~ basis.cold.hot + ns(rhum,df=3) + ns(pm10trend,df=3) +
ns(so2mean,df=3) + as.factor(I(inf>0)) + as.factor(strata)+as.factor(dow),
family=quasipoisson(), data)
>cold.hot.pred <- crosspred(basis.cold.hot,model.cold.hot,at=-16:32)
> plot(cold.hot.pred,"3d",zlab="Relative Risk", r=90,d=0.3,col="red",xlab="Temperature",
main="\n3D graphic for Jersey City\nfor double threshold",expand=0.6,lwd=0.5) # 3D plot
>par(mfrow=c(2,1))
>plot(cold.hot.pred,"slices",var=c(3),main="Cold effect", xlab="", ylab=" Relative Risk ",
ylim=range(0.99,1.01))
>plot(cold.hot.pred,"slices",var=c(23),main="Hot effect",xlab="Lag (day)",
ylab=" Relative Risk", ylim=range(0.99,1.01))
Sensitivity analysis using 20 days as the maximum lag
> nlagknots<-2 # Number of knots for lag
> klag.20<-exp(log(20)/(nlagknots+2)*1:nlagknots) # Knots for lag
> basis.temp.20 <- crossbasis(data$temp, vartype="ns", varknots=ktemp,
cenvalue=median(data$temp,na.rm=T), lagtype="ns",lagknots=klag.20,maxlag=20)
> model.month.20 <- glm(death ~ basis.temp.20 + ns(rhum,df=3) + ns(pm10trend,df=3) +
ns(so2mean,df=3) + as.factor(I(inf>0)) +as.factor(strata)+as.factor(dow),
family=quasipoisson(), data)
107
> pred.month.20 <- crosspred(basis.temp.20, model.month.20, at=-16:32)
> plot(pred.month.20,"overall", xlab="Temperature (°C)", ylab="Relative risk",
main="Overall effect of temperature on mortality\n between 1987-2000 for Jersey
City using maximum lag of 20 days")
Sensitivity analysis using 30 days as strata
>strata30<-floor((data$time-min(data$time))/30) # Create strata as 30 days
>model.strata30<- glm(death ~ basis.temp + ns(rhum,df=3) + ns(pm10trend,df=3) +
ns(so2mean,df=3) + as.factor(I(inf>0)) +as.factor(strata30)+as.factor(dow),
family=quasipoisson(), data)
>pred.strata30<- crosspred(basis.temp, model.strata30, at=-16:32, cumul=T)
>plot(pred.strata30,"overall", xlab="Temperature (°C)", ylab=" Relative risk ",
main="Overall effect of temperature on mortality\n between 1987-2000 for Jersey
City using 30 days as strata")
Comparison of time series and case–crossover design
# ignore humidity & pollution to remove influence of missing values
# case-crossover using calendar month as strata
>model.month <- glm(death ~ basis.temp + as.factor(I(inf>0))
+as.factor(strata)+as.factor(dow), family=quasipoisson(), data)
# time series with 7 degrees of freedom for time per year
>model.ts <- glm(death ~ basis.temp + as.factor(I(inf>0)) +ns(time,98)+as.factor(dow),
family=quasipoisson(), data)
108
Plot the residual distribution
>par(mfrow=c(2,1))
> hist(resid(model.month),main="Residual distribution for case-crossover design\nusing
calendar month as strata", xlim=range(-4,5),ylim=range(0,1100),xlab="Residuals",col="red",
font.lab=2,las=1)
>hist(resid(model.ts),main="Residual distribution for time series design\nusing 7 df for time
per year", xlim=range(-4,5),ylim=range(0,195),xlab="Residuals",col="red",font.lab=2,las=1)
>par(mfrow=c(1,1))
Calculate AIC value for case-crossover
>AIC.cc<- -2*sum( dpois( model.month$y, model.month$fitted.values, log=TRUE))+
2*summary(model.month)$df[3]*summary(model.month)$dispersion
AIC.cc="26364.29"
Calculate AIC value for time series
>AIC.ts <- -2*sum( dpois( model.ts $y, model.ts $fitted.values, log=TRUE))+
2*summary(model.ts )$df[3]*summary(model.ts )$dispersion
AIC.ts =" 26297.70"
For Jersey City, a time series design performs better than case-crossover as judged by the
AIC. However, both designs give similar residuals. (For Tianjin, a case–crossover performed
better than a time series according to both the AIC and residuals)
109
References
Samet JM, Dominici F, Zeger SL, Schwartz J, Dockery DW. 2000a. The National Morbidity,
Mortality, and Air Pollution Study. Part I: Methods and methodologic issues. Res Rep
Health Eff Inst(94 Pt 1): 5-14; discussion 75-84.
Samet JM, Zeger SL, Dominici F, Curriero F, Coursac I, Dockery DW, et al. 2000b. The
National Morbidity, Mortality, and Air Pollution Study. Part II: Morbidity and
mortality from air pollution in the United States. Res Rep Health Eff Inst 94(Pt 2): 5-70;
discussion 71-79.
110
CHAPTER 5: SPATIOTEMPORAL MODEL OR TIME SERIES MODEL FOR
ASSESSING CITY-WIDE TEMPERATURE EFFECTS ON MORTALITY?
Citation:
Guo Y, Barnett AG, Tong S. (2011) Spatiotemporal model or time series model for assessing
city-wide temperature effects on mortality? Environmental Research (in press), doi:
10.1016/j.envres.2012.09.001.
Statement of the joint authorship:
Yuming Guo (Candidate): Conceived and conducted the study design, performed data
analysis and wrote the manuscript;
Adrian Barnett: Performed data analysis, reviewed, edited, and revised the manuscript;
Shilu Tong: Reviewed, edited, and revised the manuscript.
111
5.1 Abstract
Most studies examining the temperature-mortality association in a city used temperatures
from one site or the average from a network of sites. This may cause measurement error as
temperature varies across a city due to effects such as urban heat islands. We examined
whether spatiotemporal models using spatially resolved temperatures produced different
associations between temperature and mortality compared with time series models that used
non-spatial temperatures. We obtained daily mortality data in 163 areas across Brisbane city,
Australia from 2000 to 2004. We used ordinary kriging to interpolate spatial temperature
variation across the city based on 19 monitoring sites. We used a spatiotemporal model to
examine the impact of spatially resolved temperatures on mortality. Also, we used a time
series model to examine non-spatial temperatures using a single site and the average
temperature from three sites. We used squared Pearson scaled residuals to compare model fit.
We found that kriged temperatures were consistent with observed temperatures.
Spatiotemporal models using kriged temperature data yielded slightly better model fit than
time series models using a single site or the average of three sites’ data. Despite this better fit,
spatiotemporal and time series models produced similar associations between temperature
and mortality. In conclusion, time series models using non-spatial temperatures were equally
good at estimating the city-wide association between temperature and mortality as
spatiotemporal models.
Key words: Mortality; Spatiotemporal model; Temperature effect; Time series model
112
5.2 Introduction
It is widely recognised that the Earth is becoming warmer in response to an accumulation of
greenhouse gas emissions (IPCC, 2007c). Climate change will have wide ranging impacts on
health, including increased heat-related mortality which has become a matter of increasing
public health concern. Studies have examined the association between temperature and
mortality, and both high temperatures and cold temperatures increase the risks of mortality
(Baccini, et al., 2008; Curriero, et al., 2002a; McMichael, et al., 2008; Stafoggia, et al., 2006).
To estimate the impact of temperature on mortality, most studies used daily temperature data
from one monitoring site or daily mean values from a network of sites, which may result in a
measurement error for temperature exposure (Zhang, et al., 2011). Random measurement
error in temperature will bias the effect estimates towards the null (Hutcheon, et al., 2010).
Studies have shown that there is spatial variation in outdoor temperatures within cities and
their surroundings (Aniello, et al., 1995; Kestens, et al., 2011; Lo, et al., 1997). Urban areas
usually have higher temperatures because of the heat island effect
(www.epa.gov/hiri/about/index.html). Smargiassi et al. (2009) found that hotter areas within
a city had a greater risk of heat-related death compared with cooler areas (Smargiassi, et al.,
2009). These results suggest that using temperature from one monitoring site or averaged
values from a network of sites may underestimate the risks of temperature on mortality.
Geo-statistical techniques have been used to model regional temperatures (Benavides, et al.,
2007; Zhang, et al., 2011). Recent studies have used spatial models to examine climate
variables like ambient temperature in the field of agriculture and forestry science (Benavides,
et al., 2007; Chuanyan, et al., 2005). Different techniques (Inverse distance interpolation
113
weighting, Voronoi tessellation, regression analysis, and geo-statistical methods) have been
developed to predict regional temperature from station data (Bhowmik & Cabral, 2011).
Previously, these models have also been used to estimate the health effects of air pollution
within cities (Lee & Shaddick, 2010; Shaddick, et al., 2008; Whitworth, et al., 2011).
Few studies have used spatial methods to quantify the impact of temperature on mortality
(Kestens, et al., 2011; Smargiassi, et al., 2009). If spatial exposures of temperature are
significantly more accurate than standard methods then they may improve our understanding
of the association between temperature and mortality. In this study, we examined the
temperature effect on mortality in Brisbane using spatial temperatures, single site’s
temperatures and averaged temperatures, and compared the risk estimate using a
spatiotemporal model with those using time series models.
5.3 Materials and methods
5.3.1Data collection
Brisbane is the capital city of the state of Queensland in Australia, and is on the east coast of
the country (27° 30' south, 153° 00' east) (Figure 5.1). It has a humid subtropical climate.
We obtained daily morality data on non-accidental causes between January 2000 and
December 2004 at the level of statistical local areas in Brisbane city from the Office of
Economic and Statistical Research of the Queensland Treasury. There are 163 statistical local
areas in Brisbane city (Figure 5.1). The causes of non-accidental mortality were coded
according to the International Classification of Diseases, ninth version tenth version (ICD-10)
114
(ICD-10: A00–R99). All deaths were residents of Brisbane city. We obtained population data
from the Australia Bureau of Statistics for each statistical local area. The census of
population and housing for each statistical local area is conducted once every five years. We
used the 1996 census for the year 2000, and the 2001 census for years 2001−2004.
We obtained daily data on maximum temperature from the Australian Bureau of Meteorology
at 19 sites in or around Brisbane city (Figure 5.1). We used maximum temperature in this
study, because a previous study found that all temperature measures (mean, minimum,
maximum, apparent) have a similar ability to predict mortality (A.G. Barnett, S. Tong, et al.,
2010). We obtained daily data on relative humidity from one monitoring site (Brisbane
airport). We obtained daily data on air pollution from the Queensland Environmental
Protection Agency. The daily PM10 concentrations were averaged from 13 monitoring sites.
The daily O3 concentrations were averaged from 10 monitoring sites.
5.3.2Data analysis
5.3.2.1Modelling spatio-temporal temperatures
Ordinary kriging is a geo-statistical technique that interpolates the value of a random field
with observed spatial information (e.g., elevation, longitude, latitude, and land use) into
nearby unobserved locations (Bhowmik & Cabral, 2011; Lefohn, Knudsen, & McEvoy,
1988). Ordinary kriging assumes that the spatial variation is statistically homogeneous
throughout the study area (Isaaks & Srivastava, 1989; Wartenberg, Uchrin, & Coogan, 1991).
We used ordinary kriging to interpolate the daily values of temperature to the centroids of
163 locations across Brisbane city from the values of 19 monitoring sites.
115
To verify the reliability of the kriged temperatures we compared the kriged and observed
temperatures at the 19 monitoring sites using Spearman correlations and the absolute
difference. We also compared ordinary kriging with universal kriging, inverse distance
weighting models, land-use regression models, and spatiotemporal models.
116
Figure 5.1: The 19 monitoring sites for temperature in or around Brisbane city, the grey
regions are statistical local areas of Brisbane city, the blue areas are water.
The monitoring sites are Amberley Amo (AMA), Cape Moreton Lighthouse (CML),
University of Queensland Gatton (UQG), Point Lookout (PL), Archerfield Airport (ARA),
Redlands Hrs (RH), Beerburrum Forest Station (BFS), Gatton Qdpi Research Station
(GQRS), Hinze Dam (HD), Jimna Forestry (JF), Coolangatta (CA), Gold Coast Seaway
(GCS), Brisbane Airport (BA), Baroon Pocket Dam (BPD), Logan City Water Treatment
Plant (LCWTP), Sunshine Coast Airport (SCA), Brisbane Centre (BC), Toowoomba Airport
(TA), Murwillumbah (MWB).
117
5.3.2.2Modelling the relationships between temperatures and mortality
To perform the spatiotemporal model, a generalized additive mixed model (GAMM) with
Poisson regression was used to examine the association between the spatial temperature and
mortality (Augustin, et al., 2009). To examine the non-linear temperature-mortality
relationship (Guo, Barnett, Pan, Yu, & Tong, 2011a), we used a natural cubic spline for
temperature (Gasparrini, et al., 2010). Hot effects are acute and short-term, whereas cold
effects occur late and last longer. Hence we used moving average lag 0–3 days and lag 0–10
days respectively to examine the temperature-mortality relationships. We used GAMMs to
model the random effects of locations using a random intercept, in order to model those areas
with higher death rates (Ruppert, Wand, & Carroll, 2003).
We assumed the daily number of deaths had an over-dispersed Poisson distribution. A natural
cubic spline for time was used to control for long-term trends and seasonal patterns in
mortality (Daniels, Dominici, Samet, & Zeger, 2000). We controlled for day of the week
(DOW), relative humidity, PM10 and O3. The population in each location was modelled using
an offset. The model was:
Log (E (Yi, t)) = α + S(Tempi,t,l, 3) + S(RHt, 3) + βPM10 t + λO3t + S(t, 7×Years)
+ µDOWt + log(popi, t) + πZi,
= α + S(Tempi,t,l, 3) + COVs + πZi , [5.1]
where i is the location (Statistical local area=1, …, 163); t is the day. Yi, t is the daily death
counts in location i on day t; α is the intercept; S(.) is a natural cubic spline; Tempi,t,l is the
maximum temperature in location i on day t, and l is the lag in days. Three degrees of
freedom (df) spline was used for temperature; RHt is relative humidity on day t, and three df
118
spline was used for relative humidity; PM10 t is averaged PM10 from 13 monitoring sites on
day t; O3 t is averaged O3 from 10 monitoring sites on day t; 7 df spline per year for time was
used to control for season and long-term trend; DOWt is the categorical variable day of the
week on day t; popi, t is the population in location i on day t; and Zi is a random intercept for
locations.
To examine a non-spatial temperature exposure using the time series model, we used
temperature data from single monitoring site (Brisbane centre) using a generalized additive
model (GAM) as follows:
Log (E (Yi, t)) = α + S(Tempt,l, 3) + S(RHt, 3) + βPM10t + λO3t + S(t, 7×Years)
+ µDOWt + log(popi, t),
= α + S(Tempt,l, df) + COVs , [5.2]
Tempt,l is the maximum temperature at Brisbane centre on day t, and l is the lag in days.
To examine a non-spatial temperature exposure using multiple monitoring sites, we used
daily temperature averaged from 3 sites (Brisbane centre, Brisbane airport, Archerfield
Airport; Figure 5.1). The model is as equation [5.2] but Tempt,l is the maximum temperature
averaged over 3 sites.
Our initial results showed that the threshold for the temperature-mortality relationship was
28 °C. We used the above models to assess the linear hot (cold) temperature-mortality
relationship above (below) 28 °C. We examined the relative risk of mortality associated with
a 1 °C increase (decrease) above (below) the 28 °C threshold.
119
Squared Pearson scaled residuals were used to compare the fit of the spatiotemporal and time
series models. Relative risks (RRs) of mortality and confidence intervals (CIs) were
calculated. All statistical tests were two-sided and values of P<0.05 were considered
statistically significant.
Sensitivity analyses
Sensitivity analyses were used to check our main findings. We varied the degrees of freedom
for temperature (2– 5 degrees of freedom) and time (4– 9 degrees of freedom per year). We
used natural cubic splines with 2 to 4 degrees of freedom for relative humidity, PM10 and O3,
respectively.
On the suggestion of a reviewer, we used an area and time stratified case–crossover model to
verify the results from the GAMM. Control days were matched to case days from the same
calendar month and day of the week. Temperatures on the case and control days were
compared within the statistical local areas.
Software
The R software (version 2.15.0, R Development Core Team 2009) was used to fit all models.
The “mgcv” package was used to fit the GAMM and GAM. The “maptools” package was
used to create the maps. A High Performance Computer was used to run the GAMMs, as this
particular model used a large internal memory (220 GigaBytes) and long running time (5
hours).
120
Figure 5.2: Mean daily maximum temperatures for the 163 statistic local areas of Brisbane
city between January 2000 and December 2004. The blue areas are water.
5.4 Results
There were only small differences in temperatures among the 19 monitoring sites
(supplemental material chapter 5, Table S5.1), and the correlations in daily temperatures
between sites were strong (supplemental material chapter 5, Table S5.2).
121
Figure 5.3: The relationship between temperature and mortality in Brisbane between 2000
and 2004, using different models with three degrees of freedom for temperature.
RR = relative risk. BC is Brisbane centre.
122
Figure 5.4: The relationship between temperature and mortality in Brisbane between 2000
and 2004, using different models with four degrees of freedom for temperature.
RR = relative risk. BC is Brisbane centre.
123
Ordinary kriging gave better spatial predictions of temperature than universal kriging, inverse
distance weighting models, land-use regression models or spatiotemporal models according
to differences between predicted and observed temperatures (supplemental material chapter 5,
Table S5.3). Ordinary kriging using latitude, longitude and elevation (transferred into log
scale) as explanatory variables performed slightly better than including greening rate,
distance to water, and population density. The predicted temperatures from ordinary kriging
were strongly correlated with observed temperatures (generally r>0.95, P<0.01)
(supplemental material chapter 5, Table S5.4). The differences between predicted and kriged
temperatures were not statistically significant (supplemental material chapter 5, Table S5.4).
Predicted daily temperatures were similar to observed temperatures (supplemental material
chapter 5, Figure S5.1).
Generally, locations close to the bay had lower temperatures (Figure 5.2). The kriged
temperature had a wider range (10.9 to 44 °C) than single site’s temperature (12 to 41.7 °C)
and averaged temperature (12.4 to 41.2 °C) (Table 5.1). The kriged spatial temperature was
strongly correlated with single site’s temperature and averaged temperature (generally r>0.97,
P<0.01, Table 5.2).
Figure 5.3 shows the relationship between temperature and mortality using three models with
3 df for lag 0–3 and lag 0–10. Three models produced similar temperature-mortality
relationships for lag 0–3 and lag 0–10, respectively. Both lag 0–3 and lag 0–10 temperatures
had no cold effect. However, spatiotemporal models (GAMM) using spatial temperatures
gave the best model fit according to the Pearson scaled residuals (supplemental material
chapter 5, Table S5.5).
124
Table 5.1: Summary statistics for kriged temperature, averaged temperature, Brisbane
centre’s temperature, PM10, O3, relative humidity, elevation, and mortality in Brisbane city
between 2000 and 2004a
Variables Distribution
Min 25% Median 75% max Mean SD
Kriged temperature (°C) 10.9 23.3 26.1 28.7 44.0 26.1 3.7
Averaged temperature (°C) 12.4 23.1 26.0 28.7 41.2 26.0 3.7
BC’s temperature (°C) 12.0 24.0 27.0 29.0 41.7 26.6 3.8
Relative humidity (%) 23.8 66.8 73.1 79.0 96.3 72.0 1.1
PM10 (µg/ml) 5.0 13.7 16.6 20.7 149.6 18.1 8.1
O3 (ppt) 4.8 12.0 14.9 18.4 36.8 15.4 4.5
Elevation (m) 4 15 26 44 177 33 25
Deaths (N) 1 13 15 18 42 15 4
a SD: Standard deviation; BC: Brisbane centre.
Table 5.3 shows the cold (hot) effects below (above) 28 °C on mortality using the three
models for lag 0–3 and lag 0–10. All models produced similar relative risks for hot and cold
effects for lag 0–3 and lag 0–10, respectively. The relative risks due to hot temperatures using
a shorter lag of 0–3 days were greater than those using a longer lag of 0–10 days.
We conducted sensitivity analyses to check our findings. We changed the degrees of freedom
for temperature (2, 4 and 5 df), spatiotemporal models still gave similar temperature-mortality
relationships as time series models (Figure 5.4). When we changed the degrees of freedom
for time (4, 5 and 6 degrees of freedom per year), the estimates of temperature effects were
similar to those using 7 degrees of freedom. Even we used distributed lag non-linear models
125
for temperature and a spline for relative humidity, PM10 and O3, spatiotemporal models still
produced similar temperature-mortality relationships as time series models. Spatial
temperatures modelled by universal kriging, inverse distance weighting models, land-use
regression models gave a similar temperature-mortality relationship as ordinary kriging.
We used an area and time stratified case-crossover to verify the results of the GAMM, and
the results were similar (supplemental material chapter 5, Figure S5.2).
5.5 Discussion
We used ordinary kriging to predict daily spatial temperatures in 163 locations across
Brisbane city. The predicted temperatures were consistent with observed temperatures at 19
monitoring sites. The kriged spatial temperatures across Brisbane city had a wider range than
single or average temperature, but were strongly correlated with single site’s and averaged
temperatures. We compared the association between temperature and mortality using
spatiotemporal with spatial temperature and time series models using non-spatial temperature.
The spatiotemporal models had the best model fit, but gave similar relative risks as the non-
spatial models.
126
Table 5.2: Spearman correlations between kriged temperature, averaged temperature,
Brisbane centre’s temperature, PM10, O3, and relative humidity in Brisbane city between
2000 and 2004a
Averaged temperature BC’ temperature Relative humidity PM10 O3
Kriged temperature 0.97** 0.97** 0.06 0.21* 0.20*
Averaged temperature 0.99** 0.03 0.21* 0.18*
BC’ temperature 0.08 0.19* 0.20*
Relative humidity −0.23* −0.29*
PM10 0.28*
a BC: Brisbane centre;
*P<0.05;
**P<0.01;
Meteorological data are routinely monitored in or around urban areas throughout the world.
These data are often used in epidemiological studies to examine the impacts of temperature
on health outcomes (mortality or morbidity) using time series and case–crossover analyses
(Basu & Ostro, 2008a; Guo et al., 2011). Temperature records from a single site or averaged
from a network of monitor sites are required on each day. This means that these sites are
assumed to be representative of the study region. However, temperature can vary over space
due to the differences in altitude, and environment (e.g., living near a park or ocean). Studies
have demonstrated hotter areas within a city due to the urban heat island effect
(www.epa.gov/hiri/about/index.html). To account for the spatial variation of temperature,
geo-statistical models have been used to predict temperatures using observed temperatures
from a network of monitors in or around the study region (Benavides, et al., 2007; Chuanyan,
et al., 2005; Ustrnul & Czekierda, 2005).
127
Table 5.3: Relative risks of mortality associated with hot and cold temperatures using four
different models assuming a V-shaped temperature risk with a threshold at 28 °C a
Models Relative risk
Hot effect b Cold effect
c
Lag 0−3 Spatiotemporal model (kriged
temperature)
1.044 (1.029, 1.058)* 1.004 (0.993, 1.015)
Time series model (BC’s
temperature)
1.040 (1.026,1.054)* 1.004 (0.992, 1.015)
Times series model (averaged
temperature)
1.045 (1.030, 1.060)* 1.002 (0.990, 1.014)
Lag 0−10 Spatiotemporal model (kriged
temperature)
1.027 (1.006, 1.047)* 1.004 (0.988, 1.020)
Time series model (BC’s
temperature)
1.023 (1.002, 1.044)* 1.003 (0.986, 1.021)
Times series model (averaged
temperature)
1.028 (1.006, 1.051)* 1.002 (0.984, 1.020)
a BC: Brisbane centre;
b Relative risk associated with a 1 °C increase in temperature above 28 °C.
c Relative risk associated with a 1 °C decrease in temperature below 28 °C.
*P<0.05
We used ordinary kriging to predict temperature in each statistical local area of Brisbane city,
as it gave a better model fit than universal kriging, inverse distance weighting models, land-
use regression models, and spatiotemporal models. These results are supported by a previous
study in the Detroit, USA (Zhang, et al., 2011). However, in Brisbane, the ordinary kriging
using latitude, longitude and elevation (transferred into log scale) as explanatory variables
performed slightly better than including greening rate, distance to water, and population
128
density. The reason might be that green place and distance to water have small variability in
Brisbane.
The time series analysis using non-spatial single site’s temperature or averaged temperature
may not produce an accurate temperature effect across a city, especially given the potentially
variation in temperature at different places. For example, if a monitor is close to city centre
(which is generally warmer), then temperature from this monitoring site may overestimate the
population’s temperature exposure. Our results show there was spatial variation in ambient
temperature (Figure 5.2). Time series analyses using single site’s temperature, and averaged
temperature gave similar hot and cold effects. The lack of difference could be due to the fact
that daily temperatures across the city had a very strong correlation (supplemental material
chapter 5, Table S5.2), and so one site’s temperature and averaged temperature can be used to
substitute other places’ temperature.
Considering the random effect of location, spatiotemporal models using spatial temperature
gave similar hot and cold effects as time series models using single site’s temperature, and
averaged temperature. Lee and Shaddick (2010) compared a Bayesian spatiotemporal model
with the time series analysis for the effect of air pollution on mortality. They investigated the
impact of spatial variation, monitor placement, and measurement error in the pollution data,
and found that, for pollutants such as O3 that do not vary greatly by monitor sites,
spatiotemporal analysis for spatial-time data and time series analysis for averaged data
produced similar risk estimates. In this case, they suggested that time series analysis using the
non-spatial averaged pollution data is the easiest method, and should be used in future studies.
In our study, we also found spatiotemporal model using spatial temperature produced similar
temperature effects on mortality as time series models using averaged temperature and single
129
site’s temperature. That means time series analysis using non-spatial estimates of exposure
might be equally good at estimating the temperature effect on mortality as spatiotemporal
models.
The spatiotemporal and time series models gave similar estimates for the city-wide
temperature-mortality association in Brisbane. However, this does not mean that the
association is the same in all areas. There may be area-level heterogeneity according to local
characteristics such as building density (Smargiassi et al., 2009), socioeconomic status (Yu et
al., 2010), or the use of air condition. Also, there might be Berkson error for time series
models using one site’s temperature or averaged temperature from a network of sites.
Berkson error is a random error in measurement, which reduces the power of a study, but risk
estimates are not attenuated. Further studies are needed to compare temperature effects across
different areas, for example, examining urban heat islands.
Strengths and limitations
We examined whether time series models using single site’s temperature or averaged
temperature from a network of monitoring sites are appropriate to assess the temperature
effect on human health. Also, we compared spatiotemporal model using spatial temperature
with time series models that used single site’s temperature and averaged temperature. We
found that all the models had similar ability to predict the temperature effect on mortality,
even though the spatiotemporal model performed better than others in terms of model fit. We
conclude that previous studies using single site’s temperatures or averaged temperatures were
appropriate for examining temperature effects.
130
There are some limitations to this study. Although kriged spatial temperatures can be used to
assess the temperature-mortality relationship, they are still not the observed temperatures, and
hence there will still be some measurement error. The findings of this study might not be
generalisable to other cities, particularly places with high variability in green place, distance
to water, and topography.
5.6 Conclusion
Spatiotemporal analysis using spatial temperature produced similar temperature-mortality
relationships as time series analyses using single site temperature and averaged temperature.
Although the spatiotemporal model using spatial temperatures had better model fit, we still
recommend using time series analysis with averaged temperatures or single site temperatures
in future studies, as it is easy to fit and does not require a high performance computer as our
spatial model did.
131
5.7 References
Aniello C, Morgan K, Busbey A, Newland L. Mapping micro-urban heat islands using
Landsat TM and a GIS. Computers & Geosciences 1995, 21: 965-967.
Augustin NH, Musio M, von Wilpert K, Kublin E, Wood SN, Schumacher M. Modeling
spatiotemporal forest health monitoring data. Journal of the American Statistical
Association 2009, 104: 899-911.
Baccini M, Biggeri A, Accetta G, Kosatsky T, Katsouyanni K, Analitis A, et al. Heat effects
on mortality in 15 European cities. Epidemiology 2008, 19: 711-9.
Barnett AG, Tong S, Clements A. What measure of temperature is the best predictor of
mortality? Environmental research 2010, 110: 604-611.
Basu R, Ostro BD. A multicounty analysis identifying the populations vulnerable to mortality
associated with high ambient temperature in California. American Journal of
Epidemiology 2008, 168: 632-7.
Benavides R, Montes F, Rubio A, Osoro K. Geostatistical modelling of air temperature in a
mountainous region of Northern Spain. Agricultural and Forest Meteorology 2007,
146: 173-188.
Bhowmik A, Cabral P. Statistical Evaluation of Spatial Interpolation Methods for Small-
Sampled Region: A Case Study of Temperature Change Phenomenon in Bangladesh.
Computational Science and Its Applications-ICCSA 2011 2011: 44-59.
Chuanyan Z, Zhongren N, Guodong C. Methods for modelling of temporal and spatial
distribution of air temperature at landscape scale in the southern Qilian mountains,
China. Ecological modelling 2005, 189: 209-220.
132
Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz JA. Temperature and mortality
in 11 cities of the eastern United States. American Journal of Epidemiology 2002, 155:
80.
Daniels MJ, Dominici F, Samet JM, Zeger SL. Estimating particulate matter-mortality dose-
response curves and threshold levels: an analysis of daily time-series for the 20 largest
US cities. American Journal of Epidemiology 2000, 152: 397.
Gasparrini A, Armstrong B, Kenward MG. Distributed lag non-linear models. Statistics in
Medicine 2010, 29: 2224-34.
Guo Y, Barnett AG, Pan X, Yu W, Tong S. The Impact of Temperature on Mortality in
Tianjin, China: A Case-crossover Design with A Distributed Lag Non-linear Model.
Environmental Health Perspectives 2011a, 119:1719-1725.
Guo Y, Barnett AG, Yu W, Pan X, Ye X, Huang C, et al. A Large Change in Temperature
between Neighbouring Days Increases the Risk of Mortality. PLoS One 2011b, 6:
e16511.
Hutcheon JA, Chiolero A, Hanley JA. Random measurement error and regression dilution
bias. British Medical Journal 2010, 340: c2289.
IPCC. Climate change 2007: the physical science basis. Contribution of Working Group I to
the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.
Cambridge, UK: Cambridge University Press, 2007.
Isaaks EH, Srivastava RM. Applied geostatistics. Vol 2: Oxford University Press New York,
1989.
Kestens Y, Brand A, Fournier M, Goudreau S, Kosatsky T, Maloley M, et al. Modelling the
variation of land surface temperature as determinant of risk of heat-related health
events. International Journal of Health Geographics 2011, 10: 7.
133
Lee D, Shaddick G. Spatial modeling of air pollution in studies of its short-term health effects.
Biometrics 2010, 66: 1238-46.
Lefohn AS, Knudsen HP, McEvoy LR, Jr. The use of kriging to estimate monthly ozone
exposure parameters for the Southeastern United States. Environmenal Pollution 1988,
53: 27-42.
Lo CP, Quattrochi DA, Luvall JC. Application of high-resolution thermal infrared remote
sensing and GIS to assess the urban heat island effect. International Journal of
Remote Sensing 1997, 18: 287-304.
McMichael AJ, Wilkinson P, Kovats RS, Pattenden S, Hajat S, Armstrong B, et al.
International study of temperature, heat and urban mortality: the 'ISOTHURM' project.
International Journal of Epidemiology 2008, 37: 1121-31.
Ruppert D, Wand M, Carroll R. Semiparametric regression: Cambridge University Press,
2003.
Shaddick G, Lee D, Zidek JV, Salway R. Estimating exposure response functions using
ambient pollution concentrations. The Annals of Applied Statistics 2008, 2: 1249-1270.
Smargiassi A, Goldberg MS, Plante C, Fournier M, Baudouin Y, Kosatsky T. Variation of
daily warm season mortality as a function of micro-urban heat islands. Journal of
Epidemiology and Community Health 2009, 63: 659-64.
Stafoggia M, Forastiere F, Agostini D, Biggeri A, Bisanti L, Cadum E, et al. Vulnerability to
heat-related mortality: a multicity, population-based, case-crossover analysis.
Epidemiology 2006, 17: 315-323.
Ustrnul Z, Czekierda D. Application of GIS for the development of climatological air
temperature maps: an example from Poland. Meteorological applications 2005, 12:
43-50.
134
Wartenberg D, Uchrin C, Coogan P. Estimating exposure using kriging: a simulation study.
Environmental Health Perspectives 1991, 94: 75-82.
Whitworth KW, Symanski E, Lai D, Coker AL. Kriged and modeled ambient air levels of
benzene in an urban environment: an exposure assessment study. Environmental
Health 2011, 10: 21.
Zhang K, Oswald EM, Brown DG, Brines SJ, Gronlund CJ, White-Newsome JL, et al.
Geostatistical exploration of spatial variation of summertime temperatures in the
Detroit metropolitan region. Environmental Research 2011, 111: 1046-53.
135
5.8 Supplemental Materials Chapter 5
Supplemental material chapter 5, Table S5.1: Summary statistics for daily observed
maximum temperatures (°C) at 19 monitoring stations in or around Brisbane, Australia
between 2000 and 2004
Stations Distribution Mean SD
Min 25% Median 75% max
AMA 12.9 24.1 27.6 30.5 42.3 27.55 4.45
CML 15.2 21.4 24.0 26.6 35.4 23.96 3.31
UQG 14.7 23.9 27.8 31.2 42.6 27.78 4.91
PL 11.9 22.6 25.1 27.6 36.5 25.03 3.32
ARA 13.5 23.5 26.7 29.3 41.8 26.59 4.0
RH 11.9 23.0 25.7 28.0 37.0 25.58 3.54
BFS 12.0 23.9 26.8 29.0 42.0 26.57 4.06
GQRS 13.8 23.2 27.1 30.7 41.9 27.16 4.88
HD 13.2 22.4 25.4 28.2 40.9 25.64 4.12
JF 8.0 20.0 24.0 27.0 39.0 23.80 4.78
CA 15.7 22.2 24.6 26.9 40.0 24.58 3.21
GCS 16.2 23.1 25.7 28.0 40.5 25.72 3.57
BA 11.8 23.0 25.7 28.0 40.2 25.55 3.43
BPD 11.9 22.2 25.1 27.9 40.2 25.26 4.03
LCWTP 12.3 23.3 26.1 28.6 41.0 26.06 3.74
SCA 15.7 22.6 25.5 27.6 38.7 25.18 3.42
BC 12.0 24.0 27.0 29.0 41.7 26.58 3.78
TA 9.6 19.0 23.3 26.8 37.7 23.14 5.12
MWB 11.8 23.0 25.9 28.8 42.9 26.09 4.18
136
Supplemental material chapter 5, Table S5.2: Spearman correlations between daily maximum temperatures at 19 monitoring stations in or
around Brisbane city between 2000 and 2004a
Stations AMA CML UQG PL ARA RH BFS GQRS HD JF CA GCS BA BPD LCWTP SCA BC TA
CML 0.82
UQG 0.97 0.82
PL 0.81 0.94 0.79
ARA 0.97 0.87 0.94 0.85
RH 0.82 0.85 0.81 0.84 0.85
BFS 0.95 0.84 0.92 0.82 0.96 0.83
GQRS 0.98 0.84 0.98 0.81 0.95 0.82 0.93
HD 0.87 0.77 0.85 0.74 0.87 0.86 0.86 0.86
JF 0.94 0.84 0.93 0.80 0.93 0.82 0.94 0.94 0.85
CA 0.85 0.89 0.82 0.89 0.90 0.84 0.86 0.84 0.80 0.83
GCS 0.88 0.87 0.85 0.86 0.91 0.82 0.89 0.87 0.83 0.86 0.90
BA 0.91 0.90 0.87 0.90 0.95 0.88 0.93 0.89 0.83 0.88 0.93 0.90
BPD 0.90 0.82 0.88 0.79 0.91 0.86 0.92 0.89 0.87 0.91 0.82 0.84 0.88
LCWTP 0.95 0.86 0.92 0.85 0.97 0.86 0.95 0.93 0.88 0.92 0.90 0.92 0.95 0.90
SCA 0.88 0.89 0.85 0.87 0.91 0.85 0.92 0.87 0.82 0.88 0.90 0.87 0.94 0.88 0.91
BC 0.95 0.87 0.92 0.87 0.98 0.86 0.95 0.93 0.86 0.92 0.91 0.91 0.97 0.90 0.97 0.92
TA 0.96 0.83 0.97 0.80 0.93 0.82 0.90 0.98 0.85 0.94 0.82 0.85 0.86 0.88 0.91 0.85 0.91
MWB 0.94 0.82 0.91 0.81 0.94 0.81 0.93 0.92 0.88 0.90 0.87 0.91 0.88 0.88 0.94 0.86 0.92 0.90 a All the correlations were statistically significant (P<0.01); see figure 5.1 for site acronyms.
137
Supplemental material chapter 5, Table S5.3: Differences between predicted and observed
temperatures using five spatial models of temperature at 19 monitoring sites during 2000 to
2004.
Site Mean difference (standard deviation) (°C)
Ordinary
kriging
Universal
kriging
Inverse distance
weighting
Land use
model
Spatiotemporal
AMA −0.02
(0.26)
−0.41
(0.36)
−0.35 (0.31) −0.26 (0.29) −0.13 (0.32)
CML 0.44
(1.01)
0.50
(1.10)
0.52 (1.06) −0.48 (1.05) 0.46 (1.03)
UQG −0.20
(0.69)
−0.32
(0.072)
−0.26 (0.71) −0.18 (0.62) −0.30 (0.77)
PL −0.48
(0.84)
−0.53
(0.086)
0.52 (0.82) −0.46 (0.82) 0.58 (0.95)
ARA −0.08
(0.55)
−0.10
(0.60)
0.12 (0.57) −0.20 (0.63) −0.15 (0.63)
RH 0.27
(1.35)
0.31
(1.37)
0.36 (1.41) 0.42 (1.42) −0.36 (1.38)
BFS −0.24
(0.33)
−0.22
(0.35)
−0.26 (0.37) −0.29 (0.42) −0.33 (0.42)
GQRS 0.42
(0.67)
−0.46
(0.70)
0.38 (0.62) 0.50 (0.71) −0.42 (0.75)
HD −0.27
(0.78)
−0.38
(0.88)
0.32 (0.76) −0.41 (0.92) −0.29 (0.79)
JF −0.29
(0.85)
−0.53
(0.96)
−0.43 (0.87) −0.36 (0.89) −0.39 (0.94)
CA 0.50
(0.87)
−0.55
(0.89)
−0.52 (0.89) 0.61 (0.93) −0.48 (0.67)
GCS −0.37
(0.64)
−0.32
(0.61)
−0.43 (0.67) −0.38 (0.96) 0.47 (0.75)
BA 0.37
(0.63)
0.46
(0.67)
0.41 (0.62) 0.39 (0.98) 0.42 (0.69)
BPD −0.86
(1.13)
−0.92
(1.17)
0.88 (1.12) −0.81 (1.09) −0.79 (1.15)
LCWTP −0.04
(0.86)
−0.12
(0.95)
−0.08 (0.87) 0.12 (0.98) 0.20 (0.91)
SCA −0.01
(0.18)
−0.09
(0.23)
−0.01 (0.18) 0.08 (0.22) −0.06 (0.25)
BC −0.22
(0.39)
−0.31
(0.46)
−0.45 (0.48) −0.36 (0.56) −0.37 (0.49)
TA 0.71
(0.74)
0.73
(0.77)
0.86 (0.91) 0.82 (0.69) −0.79 (0.81)
MWB −0.24
(0.40)
−0.56
(0.53)
−0.33 (0.45) −0.35 (0.47) −0.52 (0.48)
138
Supplemental material chapter 5, Table S5.4: Spearman correlations and differences between
kriged and observed temperatures at 19 monitoring sites during 2000 to 2004 a
Sites Correlation Difference (°C)
Mean SD
AMA 1.00 −0.02 0.26
CML 0.95 0.44 1.01
UQG 0.99 −0.20 0.69
PL 0.97 −0.48 0.84
ARA 0.99 −0.08 0.55
RH 0.95 0.27 1.35
BFS 1.00 −0.24 0.33
GQRS 0.99 0.42 0.67
HD 0.98 −0.27 0.78
JF 0.99 −0.29 0.85
CA 0.97 0.50 0.87
GCS 0.99 −0.37 0.64
BA 0.98 0.37 0.63
BPD 0.96 −0.86 1.13
LCWTP 0.97 −0.04 0.86
SCA 1.00 −0.01 0.18
BC 0.99 −0.22 0.39
TA 0.99 0.71 0.74
MWB 1.00 −0.24 0.40
a All the correlations were statistically significant (P<0.01); see figure 5.1 for site acronyms;
SD: Standard deviation.
139
Supplemental material chapter 5, Table S5.5: Descriptive statistics for Pearson scaled
residuals using spatial and non-spatial models
Models Summary statistic χ2 a
Min 25% Median Mean 75% Max
Lag 0−3 GAMM (kriged temperature) −0.8 −0.3 −0.3 0.0 −0.1 23 290929
GAM (BC’s temperature) −2.0 −0.3 −0.2 0.1 −0.2 20 398215
GAM (averaged temperature) −2.0 −0.3 −0.2 0.1 −0.2 20 398713
Lag 0−10 GAMM (kriged temperature) −0.7 −0.3 −0.3 0.0 −0.1 22 289802
GAM (BC’s temperature) −2.0 −0.3 −0.2 0.1 −0.2 20 396230
GAM (averaged temperature) −2.0 −0.3 −0.2 0.1 −0.2 20 397538
a Squared Pearson scaled residuals.
140
Supplemental material chapter 5, Figure S5.1: Predicted daily temperatures using kriging
(blue circles) and observed (red lines) temperatures at 10 monitoring stations in or around
Brisbane city during 2000 and 2004.
141
Supplemental material chapter 5, Figure S5.2: The association between temperature and
mortality in Brisbane between 2000 and 2004, using an area and time-stratified case-
crossover model with three degrees of freedom for temperature and a maximum lag of 3 and
10 days.
142
CHAPTER 6: A LARGE CHANGE IN TEMPERATURE BETWEEN
NEIGHBOURING DAYS INCREASES THE RISK OF MORTALITY
Citation:
Guo Y, Barnett AG, Yu W, Pan X, Ye X, Huang C, Tong S. (2011) A large change in
temperature between neighbouring days increases the risk of mortality. PLoS ONE, 6(2),
e16511.
Statement of the joint authorship:
Yuming Guo (Candidate): Conceived and conducted the study design, performed data
analysis and wrote the manuscript;
Adrian Barnett: Performed data analysis, reviewed, edited, and revised the manuscript;
Weiwei Yu: Reviewed, edited, and revised the manuscript;
Xiaochuan Pan: Reviewed, edited, and revised the manuscript;
Xiaofang Ye: Reviewed, edited, and revised the manuscript;
Cunrui Huang: Reviewed, edited, and revised the manuscript;
Shilu Tong: Reviewed, edited, and revised the manuscript.
143
6.1 Abstract
Background: Previous studies have found high temperatures increase the risk of mortality in
summer. However, little is known about whether a sharp decrease or increase in temperature
between neighbouring days has any effect on mortality.
Method: Poisson regression models were used to estimate the association between
temperature change and mortality in summer in Brisbane, Australia during 1996–2004 and
Los Angeles, United States during 1987–2000. The temperature change was calculated as the
current day’s mean temperature minus the previous day’s mean.
Results: In Brisbane, a drop of more than 3 °C in temperature between days was associated
with relative risks (RRs) of 1.157 (95% confidence interval (CI): 1.024, 1.307) for total non-
external mortality (NEM), 1.186 (95%CI: 1.002, 1.405) for NEM in females, and 1.442
(95%CI: 1.099, 1.892) for people aged 65–74 years. An increase of more than 3 °C was
associated with RRs of 1.353 (95%CI: 1.033, 1.772) for cardiovascular mortality and 1.667
(95%CI: 1.146, 2.425) for people aged < 65 years. In Los Angeles, only a drop of more than
3 °C was significantly associated with RRs of 1.133 (95%CI: 1.053, 1.219) for total NEM,
1.252 (95%CI: 1.131, 1.386) for cardiovascular mortality, and 1.254 (95%CI: 1.135, 1.385)
for people aged ≥75 years. In both cities, there were joint effects of temperature change and
mean temperature on NEM.
Conclusion: A significant change in temperature of more than 3 °C, whether positive or
negative, has an adverse impact on mortality even after controlling for the current
temperature.
Key words: Temperature change; Summer; Climate change; Mortality; Cardiovascular;
Respiratory;
144
6.2 Introduction
Climate change is unequivocal, with a general increase in both mean temperature and
temperature variability over the last half a century. These changes are primarily due to
emissions of greenhouse gases caused by human activity (McMichael, 1993; WHO, 2008).
The frequency, intensity and duration of weather extremes (e.g. heat waves, floods and
cyclones) are projected to increase as climate change continues (WHO/WMO/UNEP, 1996),
and unstable weather patterns (e.g. a significant drop/increase in temperature) are also more
likely to occur in the coming decades (Faergeman, 2008). As well as being an enormous
environmental issue, climate change affects human health via extreme weather events and
associated socio-ecological changes (Intergovernmental Panel on Climate Change (IPCC),
2001; WHO, 2008).
Much recent research has assessed the relationship between temperature and human health.
Morbidity and mortality are known to be seasonal, with excess morbidity and mortality
during cold winters and hot summers (Kalkstein & Greene, 1997; Kilbourne, 1999). The
effects of temperature on mortality and morbidity have been examined in various climates,
and J-, V-, or U-shaped associations have been observed (Baccini, et al., 2008; Braga,
Zanobetti, & Schwartz, 2002; Curriero, et al., 2002b; Huynen, Martens, Schram, Weijenberg,
& Kunst, 2001; Keatinge, Donaldson, Cordioli, et al., 2000; Nakaji et al., 2004). However,
less evidence is available on the possible effects on mortality due to temperature change
between neighbouring days.
In this study we hypothesized that if the temperature changed sharply between neighbouring
days, it would result in adverse impacts on human health. Because the effects of temperature
145
are strongly dependent on season, we only analysed the relationship between temperature
change and mortality in summer. Poisson regression models were used to examine the effects
on mortality due to short-term changes in temperature between neighbouring days in
Brisbane, Australia and Los Angeles, United States.
6.3 Material and methods
6.3.1 Data collection
Brisbane is the capital city of the state of Queensland in Australia, and is on the east coast of
the country (27° 30' south, 153° 00' east). It has a humid subtropical climate, with the average
temperature of 25 °C in summer (Dec–Feb). Los Angeles is the largest city in the state of
California and the Western United States (34° 03' north, 118° 15' west). Los Angeles has a
dry-summer subtropical climate, with an average temperature of 21 °C in summer (Jun–Aug).
We chose these two cities, because they have a sub-tropical climate pattern and we aimed to
explore whether the temperature change between neighbouring days has health effects in both
the Northern and Southern Hemisphere.
We gathered the Brisbane data on daily deaths of non-external causes in summers between
Jan, 1996 and Dec, 2004 from the Office of Economic and Statistical Research of the
Queensland Treasury. The causes of non-external mortality (NEM) were coded according to
the International Classification of Diseases, ninth version (ICD-9) (ICD-9: 001–799) before
December 1996 and tenth version (ICD-10) (ICD-10: A00–R99) between December 1996
and December 2004. Cardiovascular mortality (CVM; ICD-9:390–459, ICD-10:I00–I79) and
respiratory mortality (RM; ICD-9: 460–519, ICD-10:J00–J99) were extracted from the
146
mortality database. Influenza deaths (ICD-9: 487.0–487.8 or ICD-10: J10–J11) were
excluded from respiratory mortality. All deaths were for residents of Brisbane city. We
stratified NEM by gender and age (3 groups: 0–64, 65–74, and ≥75 years).
We gathered daily meteorological data including mean temperature and mean relative
humidity (RH) from the Australian Bureau of Meteorology. Values of temperature change
were calculated using the current day’s mean temperature minus the previous day’s mean
temperature. Temperature change between the neighbouring days is a measure of temperature
stability, with large positive and negative values indicating an unstable temperature. The air
pollutants including daily mean ozone (O3) and particulate matter less than 10 μm in
aerodynamic diameter (PM10) were monitored at a central site in Brisbane. We collected
these data from the Queensland Environmental Protection Agency.
Los Angeles’ data were obtained from the National Morbidity and Mortality Air Pollution
Study (NMMAPS) which is publicly available and covers the years 1987 to 2000. Mean
temperature, relative humidity, O3, NEM, CVM, RM, and NEM in age groups (0–64, 65–74,
and ≥75 years) were used here. We excluded PM10, because there was a large number of
missing values, reflecting requirements for regulatory monitoring. Mortality counts were not
split by gender in the NMMAPS, so the impact of temperature change on mortality by gender
could not be analysed in Los Angeles.
6.3.2 Data analysis
Poisson generalized additive models (GAMs) were used to examine the effects of short-term
changes in temperature between neighbouring days on mortality. We used GAMs because the
147
associations between temperature change and mortality are non-linear, and the daily number
of deaths has an over-dispersed Poisson distribution. We adjusted for day of the week (DOW)
using a categorical variable. Regression splines for calendar time and year were used to
control for long-term trends and seasonal patterns (Ren, et al., 2006). We controlled for
relative humidity, PM10, and O3 using regression splines. To assess the effects of temperature
change on mortality, we used the following model:
[ ( | )] α ( ) ( ) ( ) ( )
( ) ( ) ( )
α ( ) ( ) [6.1]
where i is the day of the observation; j is the lag days; E(Yi|X) are the estimates of daily death
counts on day i; α is the intercept; s ( , £) is a regression spline with £ degrees of freedom.
TCi-j is temperature change, MEANTi-j is daily mean temperature, RHi-j is relative humidity,
PM10i-j, is particulate matter, and O3i-j is ozone; timei is days of calendar time on day i; yeari is
the year on day i; DOWi is the day of the week on day i, and κ is vector of coefficients for
DOW; COVs represents all other covariates in the model.
As an alternative model to compare the effects of large changes in temperature on mortality
with moderate changes, temperature change was categorised into 3 groups: a drop of more
than 3 °C; a rise of more than 3 °C; a change in either direction of less than 3 °C. We used
3 °C as cut-off value because values less than 3 °C produced non-significant effect estimates.
Model (6.1) was altered by modifying the single terms of TCi-j into a categorical variable as
follows:
[ ( | )] α ( ) [6.2]
where TCi-j is a categorical variable, λ is vector of coefficients for categories of temperature
change.
148
The joint effects of temperature change and mean temperature were estimated using GAMs.
We plotted these estimates to assess whether there was an interaction between temperature
change and mean temperature on mortality. Model (6.1) was modified by changing the single
terms of TCi-j and MEANTi-j into a bivariate term as follows:
[ ( | )] α ( ) [6.3]
where s (TCi-j, MEANTi-j,6) is the joint effect of temperature change and mean temperature on
mortality, which we modelled using a regression spline with 6 degrees of freedom.
The Akaike information criterion (AIC) was used to measure goodness of fit. Residuals were
examined to evaluate the adequacy of the models. Sensitivity analyses were performed
through changing degrees of freedom for time and removing PM10 from the Brisbane data.
Relative risks (RRs) and confidence intervals (CIs) were calculated. All statistical tests were
two-sided. Values of P<0.05 were considered statistically significant. Spearman correlation
coefficients were used to summarize the correlations between daily weather conditions and
air pollutants in each city. The R software (version 2.10.1, R Development Core Team 2009)
was used to fit all models.
6.4 Results
Table 6.1 summarises the daily weather conditions, air pollutants, and mortality in summers
in Brisbane from 1996 to 2004 and Los Angeles from 1987 to 2000. The temperature change
ranged from –6.5 °C to 5.0 °C in Brisbane, and from –5.3 °C to 5.8 °C in Los Angeles. The
mean temperature was higher in Brisbane (24.4 °C) than in Los Angeles (21.3 °C). There
149
were, on average, 16 daily deaths from non-external causes in Brisbane, and 138 in Los
Angeles.
Table 6.1: Summary statistics for daily weather conditions, air pollutants, and mortality in
Brisbane, Australia and Los Angeles, United States
City Variable Frequency distribution Mean SD Sum
Min 25% Median 75% Max
Brisbane TC (°C) -6.5 -0.6 0.1 0.8 5.0 0.01 1.2 —
MEANT (°C) 18.8 23.2 24.4 25.6 31.9 24.4 1.8 —
RH (%) 38.9 68.0 73.4 79.1 97.5 73.5 8.3 —
O3 (ppb) 0.0 8.0 10.8 14.0 45. 11.4 5.2 —
PM10 (μg/m3) 3.9 12.6 15.5 19.1 84.5 16.9 7.4 —
NEM 1 13 15 18 43 16 4.4 12,364
CVM 0 5 6 8 31 6 2.9 5,076
RM 0 0 1 2 6 1 1.1 916
Age <65 years 0 2 3 4 12 3 1.7 2,372
Age 65–74 years 0 1 2 4 11 3 1.8 2,133
Age ≥75 years 1 8 10 10 12 37 3.5 7,859
Male 1 5 8 9 20 8 2.9 6,093
Female 1 6 8 9 30 8 3.0 6,271
Los Angeles TC (°C) -5.3 -0.56 0 0.56 5.8 0 1.0 —
MEANT (°C) 14.3 20.0 21.1 22.4 29.4 21.3 2.1 —
RH (%) 34.2 71.9 76.3 79.6 89.6 75.2 6.8 —
O3 (ppb) -18.2 4.9 10.0 15.2 44.9 10.2 7.9 —
NEM 95 12.9 138 147 217 138 13.4 177,384
CVM 34 57 63 70 106 64 9.4 81,913
RM 3 9 11 14 22 12 3.6 14,917
Age <65 years 17 34 39 44 70 39 7.4 50,163
Age 65–74 years 14 25 28 32 50 29 5.8 37,021
Age ≥75 years 44 64 70 76 109 70 9.4 90,200
150
Table 6.2 shows the Spearman’s correlations between daily weather conditions and air
pollutants. Temperature change was positively correlated with mean temperature in both
cities. In Brisbane, there were no statistically significant correlations between temperature
change and humidity, while temperature change was negatively correlated with humidity but
positively with O3.
Table 6.2: Spearman’s correlation between daily weather conditions and air pollutants in
Brisbane, Australia and Los Angeles, United States
Brisbane Los Angeles
MEANT TC RH O3 MEANT TC RH
TC 0.30** 0.20**
RH 0.21** 0.05 -0.10** -0.28**
O3 0.19** 0.03 –0.15** 0.04 0.17** 0.21**
PM10 0.20** 0.07 –0.24** 0.40** —— —— ——
**P<0.01
In both cities, there was little effect of temperature change on mortality, when the
temperature change ranged from –3 °C to 3 °C (Figure 6.1). Therefore, we divided
temperature change into three categories: less than –3 °C, –3 °C to 3 °C, and more than 3 °C.
Figure 6.2 shows the association between temperature change and NEM by age group. In
Brisbane, people aged <65 years appeared to be vulnerable to a sharp increase in temperature,
while those aged 65–74 years were sensitive to a sudden drop in temperature. In Los Angeles,
151
both people aged 65–74 and ≥75 years were vulnerable to a sudden temperature drop, while
no significant effects of temperature change were found for those aged <65 years.
Table 6.3: The associations between temperature change and mortality in Brisbane, Australia
and Los Angeles, United States
RR (95% CI)
1°C increase in TC (°C)a TC < –3 °C
b TC > 3 °C
b
Brisbane NEM 0.993 (0.977, 1.008) 1.157 (1.024, 1.307)** 1.198 (0.997, 1.438)
CVM 0.986 (0.962, 1.011) 1.115 (0.923, 1.347) 1.353 (1.033, 1.772)**
RM 0.997 (0.941, 1.057) 1.202 (0.774, 1.867) 1.608 (0.925, 2.794)
Age<65 years 1.021 (0.985, 1.059) 1.135 (0.859, 1.501) 1.667 (1.146, 2.425)**
Age 65–74 years 0.971 (0.935, 1.009) 1.442 (1.099, 1.892)** 1.016 (0.631, 1.634)
Age ≥75 years 0.990 (0.971, 1.010) 1.088 (0.930, 1.273) 1.118 (0.885, 1.413)
Male 1.001 (0.978, 1.024) 1.131 (0.949, 1.348) 1.225 (0.941, 1.596)
Female 0.985 (0.963, 1.007) 1.186 (1.002, 1.405)* 1.174 (0.910, 1.513)
Los Angeles NEM 0.994 (0.989, 1.000) 1.133 (1.053, 1.219)** 1.039 (0.971, 1.112)
CVM 0.988 (0.979, 0.997)** 1.252 (1.131, 1.386)** 1.031 (0.933, 1.140)
RM 1.008 (0.988, 1.029) 1.006 (0.767, 1.321) 1.002 (0.792, 1.266)
Age<65 years 1.001 (0.990, 1.013) 0.957 (0.825, 1.108) 1.014 (0.893, 1.153)
Age 65–74 years 1.004 (0.991, 1.018) 1.092 (0.929, 1.284) 1.106 (0.955, 1.280)
Age ≥75 years 0.986 (0.978, 0.995)** 1.254 (1.135, 1.385)** 1.026 (0.933, 1.129)
*P<0.05; **P<0.01;
aTC as a continuous variable, using model (6.1);
bTC as a categorical variable, using model (6.2);
The delayed effect of temperature change on mortality was examined using model (6.1). The
change in temperature between current day and previous day (lag 0) had the highest impact
on current day’s mortality (results not shown). Therefore, we only show the associations
152
between temperature change and mortality on the current day (Table 6.3). As a continuous
variable, temperature change only had statistically significant effects on CVM and NEM in
those aged ≥75 years in Los Angeles but not on other groups. However, further analyses
using model (6.2) show that a temperature drop of more than 3 °C had statistically significant
adverse impacts on total NEM, NEM among those aged 65–74 years, and women in Brisbane;
on total NEM, CVM, and NEM among those aged ≥75 years in Los Angeles. A temperature
increased of more than 3 °C was significantly associated with CVM and NEM among those
aged < 65 years in Brisbane.
Figure 6.3 and Figure 6.4 illustrate the joint effects of temperature change and mean
temperature on NEM and subgroups of NEM using model (6.3). The adverse effects of mean
temperature on mortality occurred when the mean temperature was under 26 °C in Brisbane
and under 24 °C in Los Angeles. In contrast, when we used model (6.1), mean temperature
had no adverse effect on mortality in the temperature range under 26 °C in Brisbane and
24 °C in Los Angeles (Supplemental material chapter 6, Figure S6.1 and Figure S6.2). The J-
shaped relationships between mean temperature and mortality in model (6.1) become
approximately U-shaped relationships when the joint effect of temperature change and mean
temperature was modelled (except for RM and NEM among those aged ≥75 years in
Brisbane). These results suggest that there were joint effects of temperature change and mean
temperature on mortality.
In order to perform sensitivity analyses, we changed degrees of freedom for time and
removed PM10 from Brisbane data. The results showed that there were no substantial change
in effect estimates. Also, the residual analyses showed that models were a good fit to the data.
153
Figure 6.1: The associations between temperature change and non-external mortality,
cardiovascular mortality, and respiratory mortality using model (6.1) in Brisbane, Australia
(left side) and Los Angeles, United States (right side).
154
Figure 6.2: The associations between temperature change and non-external mortality by age
group using model (6.1) in Brisbane, Australia (left side) and Los Angeles (right side),
United States.
155
Figure 6.3: Bivariate response surfaces of the temperature change and mean temperature for
non-external mortality, subgroups of mortality using model (6.3) in Brisbane, Australia.
156
Figure 6.4: Bivariate response surfaces of the temperature change and mean temperature for
non-external mortality, subgroups of mortality using model (6.3) in Los Angeles, United
States.
157
6.5 Discussion
This study examined the effect of temperature change on mortality, and explored the joint
effects of temperature change and mean temperature on mortality in Brisbane, Australia, and
Los Angeles, United States. In Brisbane, a relatively large decrease in temperature between
neighbouring days increased the risk of total NEM, and NEM among those aged 65–74 years
and in women overall. A sharp increase in temperature was significantly associated with
increased CVM and NEM among those aged < 65 years. A significant drop in temperature
increased the risks of total NEM, CVM, and NEM among those aged ≥75 in Los Angeles.
Also, joint effects of temperature change and mean temperature on mortality were found in
both locations.
These increased risks of death during periods of temperature fluctuations highlight the
importance of not only considering hot absolute temperatures in relation to human health, but
also sudden changes in temperature, particularly for a relatively large temperature changes
(more than 3 C).
We assessed whether temperature change had an adverse impact on mortality in different
subtropical climates and in different locations. Both Brisbane and Los Angeles have a
subtropical summer climate, but Brisbane is humid while Los Angeles is dry. The non-linear
pattern in the increased risk of mortality for a change in temperature was similar in the two
cities, although there were some differences. These differences might be caused by
population characteristics (e.g. racial composition), geographic location, and living
conditions including air conditioning and family income, as well as access to health care
(Stafoggia, Forastiere, et al., 2008).
158
Some previous studies have examined the effects of sudden changes in temperature on
cardiovascular disease, and found similar results. For example, Kyobutungi et al. (2005)
investigated the relationship between ischemic stroke occurrence and the temperature change
in 24 hours, but without controlling for season. The results showed that sudden temperature
changes of more than 5 °C, regardless of whether the change was negative or positive, were
associated with an increased risk of acute ischemic stroke. Schneider et al. (2008) carried out
a longitudinal study to examine the impact of weather parameters on cardiovascular patients.
Results showed that a rise or fall in air temperature was associated with an increase in heart
rate. Ebi et al. (2004) found that a 3 °C increase in minimum temperature or decrease in
maximum temperature caused a significant increase in hospital admissions for cardiovascular
diseases and stroke in three Californian regions, with a stronger association for the oldest age
group. However, Plavcova et al. (2010) only found a significant increase in mortality for
large increases in temperature.
A large change in temperature might impact on mortality, whether it is positive or negative,
because the automatic thermoregulation system cannot adapt to sudden temperature change,
particularly for people with certain medical conditions. Sudden changes in temperature have
been associated with risk factors for human health, such as increases in blood cholesterol
levels, blood pressure, plasma fibrinogen concentrations, peripheral vasoconstriction, heart
rate, platelet viscosity, and reducing the immune system’s resistance (Ballester, Corella,
Perez-Hoyos, Saez, & Hervas, 1997b; Carder et al., 2005a; Schneider, et al., 2008).
In this study, we found females were more sensitive to a drop in temperature than males in
Brisbane. Previous studies have found that gender can modify the association between
159
temperature and health (Goodman, et al., 2004; Ishigami, et al., 2008; Stafoggia, Forastiere,
Agostini, Biggeri, Bisanti, Cadum, Caranci, De Lisio, et al., 2006; Vaneckova, et al., 2008).
There is evidence that women are more vulnerable to heat-related mortality than men
(Goodman, et al., 2004; Ishigami, et al., 2008; Stafoggia, Forastiere, Agostini, Biggeri,
Bisanti, Cadum, Caranci, Lisio, et al., 2006; Vaneckova, et al., 2008). Studies have also
found that women have higher risks for ischemic, arrhythmic and blood pressure effects
associated with the weather (Diaz et al., 2002; Douglas, Dunnigan, Allan, & Rawles, 1995).
However, Basu (2009a) pointed out that the differences of the effect of temperature on
women and men was dependent on location and population. For example, the impact of hot
temperature on mortality was higher for men in São Paulo, but higher for women in Mexico
City (Bell et al., 2008).
The association between temperature change and mortality varied by age group, and the
effect of age differed between Brisbane and Los Angeles. This may be due to different life
styles, living conditions, family income, as well as access to health care. Many studies have
shown that age is a modifier of the association between temperature and health (Goodman, et
al., 2004; Stafoggia, Forastiere, Agostini, Biggeri, Bisanti, Cadum, Caranci, Lisio, et al., 2006;
Vaneckova, et al., 2008). Keatinge et al. (2000) found that people aged 65–74 years were the
most vulnerable subgroup to cold in seven European countries. Hajat et al. (2007b) found that
the elderly were the most vulnerable group to temperature in England and Wales, both for
cold and hot weather.
Our findings suggest that people with cardiovascular diseases are more vulnerable to short-
term changes in temperature than those with respiratory diseases (Figure 6.3 and Figure 6.4).
Many studies have shown that temperature is associated with physiological changes in the
160
circulatory system, including blood pressure, heart rate, blood cholesterol levels, plasma
fibrinogen concentrations, peripheral vasoconstriction, and platelet viscosity (Ballester, et al.,
1997b; Carder, et al., 2005a; Schneider, et al., 2008). These factors are directly associated
with cardiovascular function. Respiratory mortality is generally attributed to the immune
system’s resistance to respiratory infections caused by exposure to cold or hot temperatures
(Curriero, et al., 2002b). Therefore, people with some pre-existing cardiovascular disease
might be more sensitive than those with pre-existing respiratory disease to short-term changes
in temperature.
We controlled for mean temperature as many studies have illustrated a consistent relationship
between temperature and human health (Baccini, et al., 2008; Braga, et al., 2002; Curriero, et
al., 2002b; Huynen, et al., 2001; Keatinge, Donaldson, Cordioli, et al., 2000; Nakaji, et al.,
2004). Saez et al. (1995) found a 1 °C increase in temperature was associated with 1.7%,
4.2%, and 13.2% increase in NEM, CVM, and RM respectively. Schwartz (2005) found that
people with cardiovascular diseases, chronic obstructive pulmonary disease, or diabetes
appeared more vulnerable to the effects of hot weather. Mean temperature was also
associated with mortality in the present study (Supplemental material chapter 6, Table S6.1).
The mechanisms of heat-related deaths may result from failure in the thermoregulation which
may be impaired by dehydration, salt depletion and increased surface blood circulation
during hot period (Basu, 2009a; Bouchama & Knochel, 2002). Elevated blood viscosity,
cholesterol levels and sweating thresholds might also trigger heat-related mortality (Basu,
2009a; McGeehin & Mirabelli, 2001). The reduced sweat gland output and skin blood flow,
reduction in cardiac output and less redistribution of blood flow from renal and splanchnic
circulations will impair thermoregulation. We only examined the effect of mean temperature,
not maximum, minimum, and apparent temperature. A recent study has shown how these
161
different measures of temperature gave similar results for predicting mortality, so we would
anticipate similar results if we used alternative temperature measures (Barnett, Tong, &
Clements, 2010).
This study has some limitations. The findings of this study may not be generalisable to other
locations, particularly places with different climates. We used the data on temperature and air
pollution from fixed sites rather than individual exposure. Therefore, there might be the
potential for exposure measurement bias.
6.6 Conclusion
In conclusion, we found there were adverse effects due to relatively large changes in
temperature on NEM, particularly for females and people aged 65–74 years in summer in
Brisbane, as well as on NEM, CVM, and NEM among people aged ≥ 75 years in Los Angeles.
A significant increase in temperature was also associated with CVM and NEM in those < 65
years in Brisbane. In addition, there were joint effects of temperature change and mean
temperature on NEM and most subgroups in both cities. These findings suggest that people
should not only pay attention to the increases in absolute temperature in summer, but also to
temperature changes of 3 C of more. The findings might provide an important impetus for
evaluating population vulnerability, and improving the climate change adaption strategies.
162
6.7 References
Baccini, M., Biggeri, A., Accetta, G., Kosatsky, T., Katsouyanni, K., Analitis, A., et al.
(2008). Heat effects on mortality in 15 European cities. Epidemiology, 19(5), 711-719.
Ballester, F., Corella, D., Perez-Hoyos, S., Saez, M., & Hervas, A. (1997). Mortality as a
function of temperature. A study in Valencia, Spain, 1991-1993. International
Journal of Epidemiology, 26(3), 551-561.
Barnett, A. G., Tong, S., & Clements, A. C. (2010). What measure of temperature is the best
predictor of mortality? Environmental Research, 110(6), 604-611.
Basu, R. (2009). High ambient temperature and mortality: a review of epidemiologic studies
from 2001 to 2008. Environmental Health, 8, 40-52.
Bell, M. L., O'Neill, M. S., Ranjit, N., Borja-Aburto, V. H., Cifuentes, L. A., & Gouveia, N.
C. (2008). Vulnerability to heat-related mortality in Latin America: a case-crossover
study in Sao Paulo, Brazil, Santiago, Chile and Mexico City, Mexico. International
Journal of Epidemiology, 37(4), 796-804.
Bouchama, A., & Knochel, J. P. (2002). Heat stroke. New England Journal of Medicine,
346(25), 1978-1988.
Braga, A. L., Zanobetti, A., & Schwartz, J. (2002). The effect of weather on respiratory and
cardiovascular deaths in 12 U.S. cities. Environmental Health Perspectives, 110(9),
859-863.
Carder, M., McNamee, R., Beverland, I., Elton, R., Cohen, G. R., Boyd, J., et al. (2005). The
lagged effect of cold temperature and wind chill on cardiorespiratory mortality in
Scotland. Occupational and Environmental Medicine, 62(10), 702-710.
163
Curriero, F. C., Heiner, K. S., Samet, J. M., Zeger, S. L., Strug, L., & Patz, J. A. (2002).
Temperature and mortality in 11 cities of the eastern United States. American Journal
of Epidemiology, 155(1), 80-87.
Diaz, J., Garcia, R., Velazquez de Castro, F., Hernandez, E., Lopez, C., & Otero, A. (2002).
Effects of extremely hot days on people older than 65 years in Seville (Spain) from
1986 to 1997. International Journal of Biometeorology, 46(3), 145-149.
Douglas, A. S., Dunnigan, M. G., Allan, T. M., & Rawles, J. M. (1995). Seasonal variation in
coronary heart disease in Scotland. Journal of Epidemiology and Community Health,
49(6), 575-582.
Ebi, K. L., Exuzides, K. A., Lau, E., Kelsh, M., & Barnston, A. (2004). Weather changes
associated with hospitalizations for cardiovascular diseases and stroke in California,
1983-1998. International Journal of Biometeorology, 49(1), 48-58.
Faergeman, O. (2008). Climate change, disease and public health. Ugeskrift for Læger,
170(35), 2667-2668.
Goodman, P. G., Dockery, D. W., & Clancy, L. (2004). Cause-Specific Mortality and the
Extended Effects of Particulate Pollution and Temperature Exposure. Environmental
Health Perspectives, 112(2), 179-185.
Hajat, S., Kovats, R. S., & Lachowycz, K. (2007). Heat-related and cold-related deaths in
England and Wales: who is at risk? Occupational and Environmental Medicine, 64(2),
93-100.
Huynen Mm Fau - Martens, P., Martens P Fau - Schram, D., Schram D Fau - Weijenberg, M.
P., Weijenberg Mp Fau - Kunst, A. E., & Kunst, A. E. (2001). The impact of heat
waves and cold spells on mortality rates in the Dutch population. Environmental
Health Perspectives, 109(5), 463-470.
164
Intergovernmental Panel on Climate Change(IPCC). (2001). Climate Change 2001: Third
Assessment Report(Volume 1). Cambridge: UK Cambridge University Press.
Ishigami, A., Hajat, S., Kovats, R. S., Bisanti, L., Rognoni, M., Russo, A., et al. (2008). An
ecological time-series study of heat-related mortality in three European cities.
Environmental Health, 7, 5-11.
Kalkstein, L. S., & Greene, J. S. (1997). An evaluation of climate/mortality relationships in
large U.S. cities and the possible impacts of a climate change. Environmental Health
Perspectives, 105(1), 84-93.
Keatinge, W. R., Donaldson, G. C., Bucher, K., Jendritzky, G., Cordioli, E., Martinelli, M., et
al. (2000). Winter mortality in relation to climate. International Journal of
Circumpolar Health, 59(3-4), 154-159.
Keatinge, W. R., Donaldson, G. C., Cordioli, E., Martinelli, M., Kunst, A. E., Mackenbach, J.
P., et al. (2000). Heat related mortality in warm and cold regions of Europe:
observational study. British Medical Journal, 321(7262), 670-673.
Kilbourne, E. M. (1999). The spectrum of illness during heat waves. American Journal of
Preventive Medicine, 16(4), 359-360.
Kyobutungi, C., Grau, A., Stieglbauer, G., & Becher, H. (2005). Absolute temperature,
temperature changes and stroke risk: a case-crossover study. European Journal of
Epidemiology, 20(8), 693-698.
McGeehin, M. A., & Mirabelli, M. (2001). The potential impacts of climate variability and
change on temperature-related morbidity and mortality in the United States.
Environmental Health Perspectives, 109(Suppl 2), 185-189.
McMichael, A. J. (1993). Global environmental change and human population health: a
conceptual and scientific challenge for epidemiology. International Journal of
Epidemiology, 22(1), 1-8.
165
Nakaji, S., Parodi, S., Fontana, V., Umeda, T., Suzuki, K., Sakamoto, J., et al. (2004).
Seasonal changes in mortality rates from main causes of death in Japan (1970--1999).
European Journal of Epidemiology, 19(10), 905-913.
Plavcova, E., & Kysely, J. (2010). Relationships between sudden weather changes in summer
and mortality in the Czech Republic, 1986-2005. International Journal of
Biometeorology, 54(5):539-51.
Ren, C., Williams, G. M., & Tong, S. (2006). Does particulate matter modify the association
between temperature and cardiorespiratory diseases? Environmental Health
Perspectives, 114(11), 1690-1696.
Saez, M., Sunyer, J., Castellsague, J., Murillo, C., & Anto, J. M. (1995). Relationship
between weather temperature and mortality: a time series analysis approach in
Barcelona. Interational Journal of Epidemiology, 24(3), 576-582.
Schneider, A., Schuh, A., Maetzel, F. K., Ruckerl, R., Breitner, S., & Peters, A. (2008).
Weather-induced ischemia and arrhythmia in patients undergoing cardiac
rehabilitation: another difference between men and women. International Journal of
Biometeorology, 52(6), 535-547.
Schwartz, J. (2005). Who is sensitive to extremes of temperature?: A case-only analysis.
Epidemiology, 16(1), 67-72.
Stafoggia, M., Forastiere, F., Agostini, D., Biggeri, A., Bisanti, L., Cadum, E., et al. (2006).
Vulnerability to heat-related mortality: A multicity, population-based, case-crossover
analysis. Epidemiology, 17(3), 315-323.
Stafoggia, M., Forastiere, F., Agostini, D., Caranci, N., de'Donato, F., Demaria, M., et al.
(2008). Factors affecting in-hospital heat-related mortality: a multi-city case-
crossover analysis. Journal of Epidemiology and Community Health, 62(3), 209-215.
166
Vaneckova, P., Beggs, P. J., de Dear, R. J., & McCracken, K. W. J. (2008). Effect of
temperature on mortality during the six warmer months in Sydney, Australia, between
1993 and 2004. Environmental Research, 108(3), 361-369.
WHO/WMO/UNEP. (1996). Climate and health: The potential impacts of climate change.
Geneva, Swizterland.
World Health Organization. (2008). Protecting Health from Climate Change - World Health
Day 2008.
167
6.8 Supplemental Material Chapter 6
Supplemental material chapter 6, Table S6.1: The associations between a 1 °C increase in
mean temperature and mortality in Brisbane, Australia and Los Angeles, United States
RR (95% CI)
Brisbane Los Angeles
NEM 1.029 (1.017, 1.041)** 1.004 (1.001, 1.007 )**
CVM 1.026 (1.008, 1.044)** 1.004 (1.000, 1.008)
RM 1.066 (1.023, 1.110)** 1.006 (0.996, 1.016)
Age <65 years 1.019 (0.993, 1.045) 1.001 (0.996, 1.007)
Age 65–74 years 1.031 (1.003, 1.059)* 1.001 (0.995, 1.007)
Age ≥75 years 1.034 (1.019, 1.049)** 1.008 (1.003,1.012)**
Male 1.020 (1.004, 1.036)* ———————
Female 1.036 (1.019, 1.053)** ———————
**P<0.01; *P<0.05
168
Supplemental material chapter 6, Figure S6.1: The associations between the mean
temperature and non-external mortality, cardiovascular mortality, and respiratory mortality
using model (6.1) in Brisbane, Australia (left side) and Los Angeles, United States (right
side).
169
Supplemental material chapter 6, Figure S6.2: The associations between the mean
temperature and age groups of non-external mortality using model (6.1) in Brisbane,
Australia (left side) and Los Angeles, United States (right side).
171
CHAPTER 7: ASSOCIATIONS BETWEEN HIGH TEMPERATURES AND
ELDERLY MORTALITY DIFFERED BY YEAR, CITY AND REGION IN THE
UNITED STATES
Citation:
Guo Y, Barnett AG, Tong S. (2011) Associations between high temperatures and elderly
mortality differed by year, city and region in the United States. Scientific Reports (In
revision).
Statement of the joint authorship:
Yuming Guo (Candidate): Conceived and conducted the study design, performed data
analysis and wrote the manuscript;
Adrian Barnett: Performed data analysis, reviewed, edited, and revised the manuscript;
Shilu Tong: Reviewed, edited, and revised the manuscript.
172
7.1 Abstract
Studies examining the impacts of high temperatures on mortality are useful for establishing
warning systems to prevent heat effects. However, most previous studies have rarely
considered the variability in effects across different years. In this study, We aimed to examine
the variation in the effects of high temperatures on elderly mortality (age ≥ 75 years) by year,
city and region for 83 large US cities between 1987 and 2000. We defined high temperature
days as two or more consecutive days with temperatures above the 90th
percentile for each
city during each warm season (May 1 to September 30). We used a Poisson regression model
and decomposed the mortality risk for high temperatures into: a “main effect” due to high
temperatures using a distributed lag non-linear function, and an “added effect” due to
consecutive high temperature days. We pooled yearly effects across regions and overall
effects at both regional and national levels. We found that the high temperature effects (both
main and added effects) on elderly mortality varied greatly by year, city and region. The
results also show that the years with higher heat-related mortality were often followed by
those with relatively lower mortality. In conclusion, it is important to take the variability in
high temperature effect into account in the development of heat-warning systems.
Key words: High temperature effect; elderly mortality; climate change;
173
7.2 Introduction
There is increasing public health concern for heat-related mortality worldwide, as the climate
is rapidly changing (Gosling, Lowe, McGregor, Pelling, & Malamud, 2009; Kan, Chen, &
Tong, 2011). The Intergovernmental Panel on Climate Change has concluded that, without
increased investments in countermeasures, heat waves will cause increased adverse health
impacts, including heat-related mortality (IPCC, 2007a). Urban areas may be particularly
vulnerable to heat waves because of high concentrations of susceptible population groups and
the urban heat island effect (Kinney, et al., 2008; Luber & McGeehin, 2008).
Many studies have shown that high temperatures are related to non-accidental deaths (Baccini,
et al., 2008; Curriero, et al., 2002b; McMichael, et al., 2008; Stafoggia, et al., 2006), cause-
specific deaths (Barnett, 2007; Rey, et al., 2007), and morbidity such as emergency hospital
visits and hospital admissions (Hansen, et al., 2008; Knowlton, et al., 2009; Smith, et al.,
2003). Other studies have specifically examined the impacts of heat waves on mortality
(Kaiser et al., 2007; Vandentorren et al., 2004; Weisskopf et al., 2002). Gasparrini and
Armstrong (2011) decomposed the effects of heat waves into two parts: a “main effect”
related to the independent effects of high temperatures, and an “added effect” due to heat
waves. Results showed that the “main effect” contributed the most to the excess risk of
mortality, while a smaller added effect appeared during heat wave days. Joacim et al. (2010)
found an added effect of heat wave days due to two or more days above the 98th
percentile of
temperature. Anderson and Bell (2011) examined the modification of heat wave
characteristics (e.g., intensity, duration, and timing in summer) on the effects of heat waves
on mortality. They found higher mortality risks for heat waves that were more intense or
longer, or occurred in early summer
174
Epidemiological studies on heat-related mortality could be used by decision makers to
establish a warning system for high temperatures, by giving information on the heat threshold
and the expected increase in deaths above the threshold. Such studies are also useful for
estimating the potential health effects of climate change. However, most previous studies
only considered high temperature effects in the whole study period, and ignored the potential
variability of effects from year to year. Effects may vary from year to year because of
differences in the at-risk population (e.g., more elderly people), or because of increased
adaptation over time (Sheridan & Kalkstein; Stafoggia, et al., 2009a).
Temperature-related deaths are more pronounced in the elderly (Anderson & Bell, 2009), as
they are more sensitive to temperature (Macey & Schneider, 1993). The thermal regulation
system weakens with age; skin sensory perception may diminish and thermal homeostasis
may decline (Collins, 1987). This means that the elderly are not as well equipped to get relief
from heat or cold stress. A higher incidence of pre-existing cardiovascular and respiratory
disease in the elderly might be another reason why they are vulnerable to temperature
extremes (Wyndham CH, 1978). Studies have shown that the effects of thermal stress were
highest in the elderly following heat waves (Hajat, et al., 2007b; P. Vaneckova, et al., 2008).
Therefore, in this study, we examined the variation in high temperature effects on elderly
mortality by year, city and region in the United States.
7.3 Material and methods
7.3.1 Data collection
175
We used the data from the publicly available National Morbidity and Mortality Air Pollution
Study (NMMAPS) study (Samet, Dominici, et al., 2000; Samet, Zeger, et al., 2000). This
study included daily climatic conditions, air pollution levels, and mortality in 108 cities in the
United States from 1987 to 2000. Data on maximum and minimum temperatures came from
the National Climatic Data Center, and daily mortality counts came from the National Center
for Health Statistics. Daily non-external deaths consisted of death counts among residents,
excluding injuries and external causes. More information is available from the NMMAPS
web site (http://www.ihapss.jhsph.edu).
We excluded data for small cities (population under 200,000) and cities with more than 0.5%
missing data for air or dew point temperature, which left 83 cities. We stratified the cities into
seven regions (Industrial Midwest, North East, North West, South East, South West,
Southern California, and Upper Midwest) (See Supplemental Material chapter 7, Table S7.1).
We limited analyses to elderly morality (age ≥ 75 years) in the warm season (1 May–30
September) as we were interested in the effects of heat on a susceptible population. Mean
temperature (i.e., average of maximum and minimum temperatures) was used as the main
exposure variable.
7.3.2 Data analysis
A city-specific Poisson regression model was used to examine each year’s high temperature
effect on elderly mortality. In the city-specific model, the heat effects were divided into a
“main effect” and an “added effect” according to a previous study (Gasparrini & Armstrong,
2011b). These estimates (main effect and added effect separately) were then combined using
a univariate meta-analysis to create yearly estimates for each region and for the entire US.
176
The estimates were then combined using a Bayesian meta-analysis to create overall estimates
for each region and the entire US.
To examine the city-specific “main effect” and “added effect” in each warm season, we used
the following Poisson regression model:
Yt ~ Poisson(μt), t =1,...,5114,
Log (μt) = α + βTt,l + υHTt + S(timet, 3) + λDOW + εt, [7.1]
where Yt is the observed daily death count on day t; α is the intercept; Tt,l is a vector obtained
by applying the distributed lag non-linear model (DLNM) to mean temperature, β is vector of
coefficients for Tt,l, and l is the lag days; S(timet, 3) is natural cubic spline of time, with 3
degrees of freedom (df) per warm season (1 May–30 September) used to control for trends
and seasonal patterns in mortality; HTt is a categorical variable for high temperature days on
day t, HT=0 if day t was a non-high temperature day, HT=1 if day t was a high temperature
day (defined below); DOW is a categorical variable for day of the week, and λ is vector of
coefficients. ε are the residuals.
We modelled the main effect of high temperatures on day t using the term βTt,l which is fitted
using a DLNM. A DLNM is a two-dimensional spline which models the main effect of heat
along both dimensions of temperature and lags. DLNMs overcome the fact that temperatures
within a couple of days are strongly correlated, and constrain the effect of temperature using
a spline (Armstrong, 2006). We used a cubic spline with 3 df to model the U-shaped
relationship between temperature and mortality. A cubic spline with 5 df was used to model
the lagged (delayed) effect of temperature on mortality up to 10 days (Gasparrini &
Armstrong, 2011b). The DLNM provides an estimate of the overall effect along lag days
(Gasparrini, et al., 2010). We modelled the main effect for each city’s warm season from the
177
term βTt,l. We estimated the percent change of mortality between the median temperature
among high temperature days against the 75th percentile of temperature.
The added effect of high temperatures was modelled by υHTt. We defined high temperatures
as two or more than two consecutive days with temperature above the 90th
percentile for each
city during each warm season (May 1 to September 30).
To examine the variation in the high temperature effects on mortality from year to year, we
pooled the yearly national and regional high temperature effects (main effect and added effect
separately) across cities using a univariate meta-analysis with a random effect for each city.
We plotted the pooled yearly national and regional main effects and added effects from 1987
to 2000.
We used a Bayesian hierarchical model to estimate an overall high temperature effect (main
effect and added effect separately) for each region and the nation by combining the yearly
estimated effects of high temperatures within each city and incorporating the estimates’
variance (Everson & Morris, 2000). Each city was given a random intercept to model its
mean heat effect, and a random linear effect of time to model linear trends over time. A
sensitivity analysis was conducted to examine random non-linear effects of time. The model
fit was assessed using the Deviance Information Criteria (Spiegelhalter, et al., 2002).
Bayesian hierarchical models have been increasingly used to combine estimated effects
across communities in air pollution and temperature studies (Anderson & Bell, 2011; Barnett,
2007).
7.3.2 Sensitivity Analyses
178
Sensitivity analyses were carried out on the parameters for the city-specific model to test the
robustness of the results to our assumptions concerning the temperature-mortality relationship.
We modified the df of smoothing for time (4 to 8), and varied the df for the splines for
temperature (4 to 8) and lag (3 to 8) in the DLNM.
To test the sensitivity of our results to the definition of high temperatures, three alternative
high temperature definitions were used: 95th
and 97th
percentile for at least 2 consecutive days.
The Akaike information criterion was used to measure goodness of fit for city-specific
models. Residuals were examined to evaluate the adequacy of the city-specific models. The R
software (version 2.11.0, R Development Core Team 2009) was used to fit city specific
models. The “dlnm” package of R software was used to create DLNM. The “metafor”
package of R software was used to fit univariate meta-analyses. WinBUGS software (version
1.4) was used to fit the Bayesian hierarchical model.
7.4 Results
Figure 7.1 shows the variability in the estimated percent increase in elderly mortality due to
the high temperature effects (including both main and added effects) between 1987 and 2000.
Each city’s yearly high temperature effects on elderly mortality varied greatly, and there were
large differences between cities (supplemental Material chapter 7, Figures S7.1 and S7.2).
For example, the main effects varied greatest in Lexington (the percentage change in elderly
mortality due to main effect ranged from −80% to 150%), while there was a smallest
179
variation in San Diego (the percentage change in elderly mortality due to added effect range
from −57% to 29%).
Table 7.1: The distribution of yearly high temperature effects on elderly mortality by region
between 1987 and 2000
Region Main effect (percent change %) Added effect (percent change %)
Min 25% Median 75% Max Min 25% Median 75% Max
Industrial Midwest –8.1 0.5 6.5 8.9 16.4 –6.9 –5.1 –2.2 –0.4 7.8
North East –16.4 –1.4 5.5 14.2 19.1 –9.8 –5.8 0.0 7.0 20.7
North West –8.5 –1.1 6.6 15.3 32.3 –16.1 –3.5 1.7 3.6 14.2
South East –10.9 –6.4 0.5 7.6 12.9 –11.3 –3.4 0.4 7.0 8.7
South West –12.9 –5.7 0.0 11.7 32.4 –16.6 –8.2 –2.6 5.5 7.9
Southern California –6.9 4.8 7.0 13.1 15.9 –9.8 –4.2 –2.0 0.6 11.1
Upper Midwest –30.2 –1.5 1.4 9.7 38.3 –14.7 –0.7 5.8 15.1 22.8
National –0.3 2.6 4.1 7.7 12.2 –5.3 –1.4 0.1 0.6 2.2
We found a similar variation in high temperature effects on elderly morality at the regional
level (Table 7.1). Figure 7.2 shows the trend of the estimated percent increase in elderly
mortality due to the high temperature effects by year and region. There was a great variation
from year to year in the mean effects of high temperatures. In every region the main and
added effects of high temperatures were negative in at least one year (meaning that heat was
associated with a decreased risk of death). Very high increases in mortality were often
followed by much lower increases, and vice versa.
According to the Deviance Information Criteria, the Bayesian hierarchical model fit was not
improved by using a random nonlinear term for trends over time, so the combined estimates
180
were based on modelling a linear trend over time. Figure 7.3 shows the linear trends over
time of high temperature effects by geographical region using the Bayesian hierarchical
model. The main effect declined over time in all of the regions.
Table 7.2: Pooled high temperature effects on elderly mortality by region between 1987 and
2000
Region % increase (95% CI)
Main effect Added effect
Industrial Midwest 6.8 (2.1, 11.5) –1.6 (–4.6, 1.4)
North East 3.3 (–1.9, 8.5) 3.7 (0.3, 7.2)
North West 9.0 (1.9, 16.0) 0.7 (–4.0, 5.4)
South East 3.6 (–0.6, 7.9) 0.3 (–3.0, 3.4)
South West 1.1 (–5.1, 7.3) –1.9 (–6.8, 2.9)
Southern California 8.5 (3.5, 13.6) –1.2 (–5.2, 2.9)
Upper Midwest 3.8 (–7.3, 15.0) 6.4 (–1.0, 13.7)
National 5.1 (2.9, 7.3) 0.4 (–1.1, 1.9)
CI = confidence interval
Table 7.2 shows the overall high temperature effects pooled from each city across different
geographical regions and the whole United States. The high temperature effects differed by
geographical regions. The highest main effects were in the North West with a 9.0% (95%
confidence interval (CI): 1.9%, 16.0%) increase in elderly mortality, while the highest added
effect occurred in the Upper Midwest with an average 6.4% (95% CI: –1.0%, 13.7%)
increase in elderly morality. For the whole USA, the main effect of high temperature was
greater than the added effect.
181
Figure 7.1: Boxplots of the yearly high temperature effects on elderly mortality by cities
between 1987 and 2000. City abbreviations are explained in Supplemental Material Chapter 7,
Table S7.1.
182
Figure 7.2: Mean high temperature effects on elderly mortality by region between 1987 and
2000 using a univariate meta-analysis.
183
Figure 7.3: Trend in the effects of high temperatures on the elderly mortality by region between
1987 and 2000 using a Bayesian hierarchical model.
184
7.5 Discussion
7.5.1 Variation in high temperature effects
This study characterized the excess risk for elderly mortality due to high temperature effects
on a yearly basis. We found that the both main and added effects varied by year, city and
geographic region (Figure 7.2).
Our results have potentially important consequences for heat warning systems. The great
variation of high temperature effects shown here means that heat warning systems, which
warn of approaching dangerous temperatures (Hajat, O'Connor, & Kosatsky, 2010; Hajat et
al., 2010; O'Neill et al., 2009), are likely to suffer false positives as in some years an
anticipated spike in deaths will not occur (whereas in other years the spike could be much
greater than expected). These false alarms have financial implications, as much of the
preventive actions planned under the current system (such as freeing-up hospital beds) are
wasted. Unnecessarily raising warnings may also undermine confidence in the system if the
warnings are perceived to be too frequent and without any real need (Kovats & Kristie, 2006).
There is a regular pattern of heat effects which often had “bad” years followed by relatively
“good” years (Figure 7.2). The great variations of heat effects found here might be caused by
the interaction between temperature-related mortality in summer and the previous summer or
winter (Ha, Kim, & Hajat, 2011; Rocklov, Forsberg, & Meister, 2009; Stafoggia, et al.,
2009a). A summer with a relatively high level of mortality leaves less people susceptible to
heat-related mortality in the following winter or summer. Studies have also shown that high
winter mortality leads to lower mortality in the following summer (Ha, et al., 2011; Rocklov,
185
et al., 2009; Stafoggia, et al., 2009a), as well as high summer mortality following a lower
winter mortality (Valleron & Boumendil, 2004).
We found that the regional patterns of high temperature-mortality relationship (both main and
added effects) were different. The heterogeneity in high temperature effects between cities or
geographic regions might be due to differences in: physical acclimatization to high
temperatures (e.g., air conditioning use, housing structure, clothing type); exposure to high
temperatures; city-level responses to extreme high temperatures; sociodemographic
characteristics, or meteorological factors within cities or regions that might influence or
modify temperature–mortality relationships (Anderson & Bell, 2011).
7.5.2 Overall high temperature effects and trends over time
The main high temperature effect was greater than added effect. In general, we observed that
the main effect was higher in the Northwest and South California, while the added effect was
higher in Northeast and Upper Midwest. These findings are consistent with other studies of
heat waves or high temperatures (Anderson & Bell, 2009; Braga, et al., 2001). For example,
Anderson and Bell (2011) found that heat wave had smaller effects on mortality in the South
than elsewhere.
There was a decline in the main effect of high temperature-related death from 1987 to 2000 in
all regions (Figure 7.3). This is consistent with previous US studies (Barnett, 2007; Davis,
Knappenberger, Michaels, & Novicoff, 2003). An increased use of air conditioning may be
an underlying reason for this reduction in heat-related mortality (Barnett, 2007; Davis, et al.,
2003; O’Neill, 2003). US studies have shown that the effect of hot temperatures on mortality
186
was related to the level of air conditioning use (Braga, et al., 2001; Curriero, et al., 2002b;
Nunes, Paixao, Dias, Nogueira, & Marinho Falcao, 2011), and that the use of air conditioning
steadily increased in all areas of the United States during 1980 and 2000 (Barnett, 2007).
During the 1995 Chicago heat waves, moving from unventilated, indoor locations to places
with air conditioning decreased the mortality risk (Chan, Stacey, Smith, Ebi, & Wilson, 2001;
Semenza et al., 1996b).
We defined high temperature days as 2 or more consecutive days with temperature above the
90th
percentile during each warm season (May 1 to September 30), as there were only 153
days in each warm season. If we used a more strict definition (e.g., 2 or more consecutive
days with temperature above the 95th
percentile), there were often no high temperature days
in some warm seasons in some cities. However, the findings of our study are consistent with
previous study (Gasparrini & Armstrong, 2011b), which showed that the main effect of high
temperatures is greater than the added heat wave effect.
We decomposed the heat-mortality association into main and added effects, because main
effect itself cannot completely capture the effects of temperature on mortality. We did not
consider the impact of high temperature characteristics on mortality. Previous studies have
shown that individual days of extreme heat increased the risk of mortality (Baccini, et al.,
2008; Hajat et al., 2006), and that the added high temperature effect on mortality was greater
for consecutive high temperature days compared with non-consecutive high temperature days
(Anderson & Bell, 2011; Gasparrini & Armstrong, 2011b). Additionally, Anderson and Bell
(2011) illustrated that the added effect on mortality were associated with a heat wave’s
intensity, duration, and timing in the season (Anderson & Bell, 2011).
187
In this study, air pollution was not controlled for, as a previous study showed that heat effects
on mortality in the United States were robust to air pollution (Anderson & Bell, 2009). Other
related studies also did not control for air pollution when assessing the impact of temperature
on mortality (Anderson & Bell, 2011; Gasparrini & Armstrong, 2011b; Ha, et al., 2011).
7.5.3 Sensitivity analyses
Sensitivity analyses were carried out by modifying the df of smoothing for time, and varying
the df of the splines for temperature and lag in the DLNM. The results remained broadly
similar (results not shown). The model fit appeared better in some regions as judged by
akaike information criterion, but worse in other regions.
We examined the main and added effects using different high temperature definitions (95th
and 97th
percentile with 2 consecutive days of duration). The results showed that both the
main and added effects of high temperature increased if the definition of 97th
percentile with
2 days of duration was used compared with the definition of 95th
percentile with 2 days of
duration (results not shown).
7.5.4 Strengthens and limitations
This study has several strengths. The high temperature effect on elderly mortality is highly
variable from year to year. This finding is important from a public health perspective,
because this issue needs to be taken into account when establishing a warning system for high
temperatures and projecting the impact of future climate change. We used a flexible DLNM
188
to capture the main effect of high temperature on elderly mortality and consider lagged
effects. We used a large database of 85 cities in the United States.
This study also has some limitations. We only used 14 years of data. A longer period may be
better to model the exact fluctuation in high temperature-related mortality over years. We did
not have individual exposure data, and so assumed a common daily exposure to temperature
for people in the same city. Additionally, the influence of air pollution and socioeconomic
factors was not controlled for. We did not consider the modification of high temperature
intensity, duration, and timing in the season, as a previous study has addressed this question
(Anderson & Bell, 2011).
7.6 Conclusion
Both the main and added effects of high temperatures on elderly mortality varied greatly from
year to year, making it difficult to predict the health effects of heat in any one year. There
was a gradual reduction in high temperature-related elderly mortality (particularly the main
effect) over the study period. Cities using heat warning systems need to be aware of both the
long-term changes in heat-related mortality and the year-to-year variations.
189
7.7 References
Anderson, B. G., & Bell, M. L. (2009). Weather-related mortality: how heat, cold, and heat
waves affect mortality in the United States. Epidemiology, 20(2), 205-213.
Anderson, G. B., & Bell, M. L. (2011). Heat Waves in the United States: Mortality Risk
during Heat Waves and Effect Modification by Heat Wave Characteristics in 43 U.S.
Communities. Environmental Health Perspectives, 119(2), 210-218.
Armstrong, B. (2006). Models for the relationship between ambient temperature and daily
mortality. Epidemiology, 17(6), 624-631.
Baccini, M., Biggeri, A., Accetta, G., Kosatsky, T., Katsouyanni, K., Analitis, A., et al.
(2008). Heat effects on mortality in 15 European cities. Epidemiology, 19(5), 711-719.
Barnett, A. G. (2007). Temperature and cardiovascular deaths in the US elderly: changes over
time. Epidemiology, 18(3), 369-372.
Braga, A. L., Zanobetti, A., & Schwartz, J. (2001). The time course of weather-related deaths.
Epidemiology, 12(6), 662-667.
Chan, N. Y., Stacey, M. T., Smith, A. E., Ebi, K. L., & Wilson, T. F. (2001). An empirical
mechanistic framework for heat-related illness. Climate Research, 16(2), 133-143.
Collins, K. J. (1987). Effects of cold on old people. British Journal Of Hospital Medicine,
38(6), 506.
Curriero, F. C., Heiner, K. S., Samet, J. M., Zeger, S. L., Strug, L., & Patz, J. A. (2002).
Temperature and mortality in 11 cities of the eastern United States. American Journal
of Epidemiology, 155(1), 80-87.
Davis, R. E., Knappenberger, P. C., Michaels, P. J., & Novicoff, W. M. (2003). Changing
heat-related mortality in the United States. Environmental Health Perspectives,
111(14), 1712-1718.
190
Everson, P. J., & Morris, C. N. (2000). Inference for multivariate normal hierarchical models.
Journal of the Royal Statistical Society: Series B (Statistical Methodology), 62(2),
399-412.
Gasparrini, A., & Armstrong, B. (2011). The impact of heat waves on mortality.
Epidemiology, 22(1), 68-73.
Gasparrini, A., Armstrong, B., & Kenward, M. G. (2010). Distributed lag non-linear models.
Statistics in Medicine, 29(21), 2224-2234.
Gosling, S. N., Lowe, J. A., McGregor, G. R., Pelling, M., & Malamud, B. D. (2009).
Associations between elevated atmospheric temperature and human mortality: a
critical review of the literature. Climatic Change, 92(3), 299-341.
Ha, J., Kim, H., & Hajat, S. (2011). Effect of Previous-Winter Mortality on the Association
between Summer Temperature and Mortality in South Korea. Environmental Health
Perspectives, 119:542-546.
Hajat, S., Armstrong, B., Baccini, M., Biggeri, A., Bisanti, L., Russo, A., et al. (2006).
Impact of high temperatures on mortality: is there an added heat wave effect?
Epidemiology, 17(6), 632-638.
Hajat, S., Kovats, R. S., & Lachowycz, K. (2007). Heat-related and cold-related deaths in
England and Wales: who is at risk? Occupational and Environmental Medicine, 64(2),
93-100.
Hajat, S., O'Connor, M., & Kosatsky, T. (2010). Health effects of hot weather: from
awareness of risk factors to effective health protection. Lancet, 375(9717), 856-863.
Hajat, S., Sheridan, S. C., Allen, M. J., Pascal, M., Laaidi, K., Yagouti, A., et al. (2010).
Heat-health warning systems: a comparison of the predictive capacity of different
approaches to identifying dangerously hot days. American Journal of Public Health,
100(6), 1137-1144.
191
Hansen, A. L., Bi, P., Ryan, P., Nitschke, M., Pisaniello, D., & Tucker, G. (2008). The effect
of heat waves on hospital admissions for renal disease in a temperate city of Australia.
Internationa Journal of Epidemiology, 37(6), 1359-1365.
IPCC. (2007). Climate change 2007: impacts, adaptation and vulnerability. Contribution of
Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on
Climate Change. Cambridge, UK: Cambridge University Press.
Kaiser, R., Le Tertre, A., Schwartz, J., Gotway, C. A., Daley, W. R., & Rubin, C. H. (2007).
The effect of the 1995 heat wave in Chicago on all-cause and cause-specific mortality.
American Journal of Public Health, 97, 158-162.
Kan, H., Chen, R., & Tong, S. (2011). Ambient air pollution, climate change, and population
health in China. Environment International, .
Kinney, P. L., O'Neill, M. S., Bell, M. L., & Schwartz, J. (2008). Approaches for estimating
effects of climate change on heat-related deaths: challenges and opportunities.
Environmental Science & Policy, 11(1), 87-96.
Knowlton, K., Rotkin-Ellman, M., King, G., Margolis, H. G., Smith, D., Solomon, G., et al.
(2009). The 2006 California heat wave: impacts on hospitalizations and emergency
department visits. Environmental Health Perspectives, 117(1), 61-67.
Kovats, R. S., & Kristie, L. E. (2006). Heatwaves and public health in Europe. The European
Journal of Public Health, 16(6), 592.
Luber, G., & McGeehin, M. (2008). Climate change and extreme heat events. American
Journal of Prevetive Medicine, 35(5), 429-435.
Macey, S. M., & Schneider, D. F. (1993). Deaths from excessive heat and excessive cold
among the elderly. The Gerontologist, 33(4), 497.
192
McMichael, A. J., Wilkinson, P., Kovats, R. S., Pattenden, S., Hajat, S., Armstrong, B., et al.
(2008). International study of temperature, heat and urban mortality: the
'ISOTHURM' project. International Journal of Epidemiology, 37(5), 1121-1131.
Nunes, B., Paixao, E., Dias, C. M., Nogueira, P., & Marinho Falcao, J. (2011). Air
conditioning and intrahospital mortality during the 2003 heatwave in Portugal:
evidence of a protective effect. Occupational and Environmental Medicine, 68(3),
218-223.
O'Neill, M. S., Carter, R., Kish, J. K., Gronlund, C. J., White-Newsome, J. L., Manarolla, X.,
et al. (2009). Preventing heat-related morbidity and mortality: new approaches in a
changing climate. Maturitas, 64(2), 98-103.
O’Neill, M. (2003). Air conditioning and heat-related health effects. Applied Environmental
Science and Public Health, 1, 9-12.
Rey, G., Jougla, E., Fouillet, A., Pavillon, G., Bessemoulin, P., Frayssinet, P., et al. (2007).
The impact of major heat waves on all-cause and cause-specific mortality in France
from 1971 to 2003. International archives of occupational and environmental health,
80(7), 615-626.
Rocklov, J., Forsberg, B., & Meister, K. (2009). Winter mortality modifies the heat-mortality
association the following summer. European Respiratory Journal, 33(2), 245-251.
Samet, J. M., Dominici, F., Zeger, S. L., Schwartz, J., & Dockery, D. W. (2000). The
National Morbidity, Mortality, and Air Pollution Study. Part I: Methods and
methodologic issues. Research report (Health Effects Institute) 94, 5-14; discussion
75-84.
Samet, J. M., Zeger, S. L., Dominici, F., Curriero, F., Coursac, I., Dockery, D. W., et al.
(2000). The National Morbidity, Mortality, and Air Pollution Study. Part II:
193
Morbidity and mortality from air pollution in the United States. Research report
(Health Effects Institute), 94(Pt 2), 5-70; discussion 71-79.
Semenza, J. C., Rubin, C. H., Falter, K. H., Selanikio, J. D., Flanders, W. D., Howe, H. L., et
al. (1996). Heat-related deaths during the July 1995 heat wave in Chicago. New
England Journal of Medcine, 335(2), 84-90.
Sheridan, S. C., & Kalkstein, A. J. Seasonal variability in heat-related mortality across the
United States. Natural Hazards, 1-15.
Smith, W. R., Coyne, P., Smith, V. S., & Mercier, B. (2003). Temperature changes,
temperature extremes, and their relationship to emergency department visits and
hospitalizations for sickle cell crisis. Pain Management Nursing, 4(3), 106-111.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian
measures of model complexity and fit. Journal of the Royal Statistical Society: Series
B (Statistical Methodology), 64(4), 583-639.
Stafoggia, M., Forastiere, F., Agostini, D., Biggeri, A., Bisanti, L., Cadum, E., et al. (2006).
Vulnerability to heat-related mortality: a multicity, population-based, case-crossover
analysis. Epidemiology, 17(3), 315-323.
Stafoggia, M., Forastiere, F., Michelozzi, P., & Perucci, C. A. (2009). Summer temperature-
related mortality: effect modification by previous winter mortality. Epidemiology,
20(4), 575-583.
Valleron, A. J., & Boumendil, A. (2004). Epidemiology and heat waves: analysis of the 2003
episode in France. Comptes Rendus Biologies, 327(12), 1125-1141.
Vandentorren, S., Suzan, F., Medina, S., Pascal, M., Maulpoix, A., Cohen, J. C., et al. (2004).
Mortality in 13 French cities during the August 2003 heat wave. American Journal of
Public Health, 94(9), 1518-1520.
194
Vaneckova, P., Beggs, P. J., de Dear, R. J., & McCracken, K. W. (2008). Effect of
temperature on mortality during the six warmer months in Sydney, Australia, between
1993 and 2004. Environmental Research, 108(3), 361-369.
Weisskopf, M. G., Anderson, H. A., Foldy, S., Hanrahan, L. P., Blair, K., Torok, T. J., et al.
(2002). Heat wave morbidity and mortality, Milwaukee, Wis, 1999 vs 1995: an
improved response? American Journal of Public Health, 92(5), 830-833.
Wyndham CH, F. S. (1978). Climate and disease. South African Medocal Journal, 53(26),
1051-1061.
195
7.8 Supplemental Material Chapter 7
Supplemental Material chapter 7, Table S7.1: List of the 83 US cities by region.
Region City Abbreviation State Region City Abbreviation State
Industrial Midwest Akron akr OH South East Atlanta atla GA
Buffalo buff NY Baton Rouge batr LA
Chicago chic IL Birmingham birm AL
Cincinnati cinc OH Cayce cayc SC
Cleveland clev OH Charlotte char NC
Columbus clmo OH Dallas/Fort Worth dlft TX
Dayton dayt OH Greensboro grnb NC
Detroit det MI Houston hous TX
Fort Wayne ftwa IN Huntsville hunt AL
Grand Rapids gdrp MI Jackson jcks MS
Indianapolis indi IN Jacksonville jckv FL
Lexington lex KY Knoxville knox TN
Louisville loui KY Memphis memp TN
Madison madi WI Miami miam FL
Milwaukee milw WI Mobile mobi AL
Pittsburgh pitt PA Nashville nash TN
St. Louis stlo MO New Orleans no LA
Toledo tole OH Orlando orla FL
Raleigh ral NC
North East Baltimore balt MD Shreveport shr LA
Boston bost MA St. Petersburg stpe FL
Jersey City jers NJ Tampa tamp FL
Norfolk nor VA Tulsa tuls OK
Newark nwk NJ
New York ny NY South West Albuquerque albu NM
Philadelphia phil PA Austin aust TX
Providence prov RI Corpus Christi corp TX
Rochester roch NY El Paso elpa TX
Syracuse syra NY Las Vegas lasv NV
Washington dc DC Lubbock lubb TX
Oklahoma City okla OK
North West Denver denv CO Phoenix phoe AZ
Oakland oakl CA San Antonio sana TX
Portland port OR Tucson tucs AZ
Sacramento sacr CA
Salt Lake City salt UT Southern California Fresno fres
San Francisco sanf CA Los Angeles la CA
San Jose sanj CA Riverside rive CA
Seattle seat WA San Bernardino sanb CA
Spokane spok WA San Diego sand CA
Tacoma taco WA Santa Ana/Anaheim staa CA
Upper Midwest Des Moines desm IA
Kansas City kan MO
Lincoln linc NE
Minneapolis/St. Paul minn MN
Wichita wich KS
196
Supplemental Material chapter 7, Figure S7.1: Main effects of high temperature on elderly
mortality by city between 1987 and 2000. The white colour represents high risk of mortality
while the red colour represents low risk of mortality. City abbreviations are explained in
Supplemental Material Chapter 7, Table S7.1.
197
Supplemental Material chapter 7, Figure S7.2: Added effects of high temperature on elderly
mortality by city between 1987 and 2000. The white colour represents high risk of mortality
while the red colour represents low risk of mortality. City abbreviations are explained in
Supplemental Material Chapter 7, Table S7.1.
198
CHAPTER 8: GENERAL DISCUSSION
Each of the previous results chapter (Chapters 4–7) has its own discussion section in which
major findings were discussed in relation to the literature, interpretation of the results, and
public health implications. This chapter discusses the methodological development,
implications of the overall research, and the strengths and limitations of the thesis. It also
includes recommendations for future research directions.
8.1 Methodological development
Although a number of studies have examined the impact of temperature on human health,
such as mortality and morbidity, a few key methodological issues remain to be resolved.
The case–crossover design has its advantages. For example, case–crossover studies use each
subject as their own control, so any fixed confounding is removed and exposure
misclassification can be minimised; in the mean time, long-term and seasonal trends were
controlled by using short-interval strata. However, most previous case–crossover studies used
linear parametric models to examine the effects of temperature on mortality and used a single
lag model, or moving average lag model (Basu, et al., 2008; Green, et al., 2010). These
models are not adequate for exploring the complex relationship between temperature and
mortality, which is generally non-linear. Additionally, exposure to ambient temperature
usually has lagged effects which might also be non-linear. This thesis combined a distributed
lag non-linear model with a case–crossover design, making it possible to use such a design to
fit more sophisticated estimates of the effects of temperature. The distributed lag non-linear
model is developed on the basis of a ‘‘cross-basis’’ function, which allows simultaneously
199
estimating the non-linear effects of temperature at each lag and also the non-linear effects
across lags (Gasparrini , Armstrong , & Kenward, 2010).
Another methodological issue is to examine whether there is a significant difference in
assessment of temperature effects using a spatiotemporal approach or a non-spatial approach.
There is variation in ambient temperature within most cities. Some time series studies
consider this spatial difference, but the standard approach is to average the observed data to
create a non-spatial daily exposure. Some time series studies only used one monitoring site’s
data, which similarly does not consider the spatial variance in temperature across the study
region. There have been some concerns that simple models using averaged temperature or a
single site’s temperature might not accurately estimate the effects of temperature on mortality
across the region, which may bias the effect estimates. In this thesis, I used an ordinary
kriging approach to model the spatial temperature across Brisbane city. Then, I compared
simple time series and spatiotemporal models, and found that the two models had a similar
ability to predict the effects of temperature on mortality. Therefore, I conclude that time
series studies using single site’s temperature or averaged temperature are generally
appropriate for examining temperature effects. This finding will be useful for future studies
aiming to estimate the effects of temperature on mortality, especially as time series models
are relatively easier to implement than spatiotemporal models. However, the lack of a
difference between the simple and spatial models here may be because of a lack of spatial
detail in both the temperature monitoring and mortality data. Other cities may still show a
spatial pattern in the effects of temperature if they use more detailed spatial data, or have a
strong spatial heterogeneity in housing characteristics (e.g., particular suburbs with low rates
of air conditioning), and city heat island.
200
8.2 Implication of the research
8.2.1 Heat warning system
This thesis has some implications for the prevention and control of heat-related mortality.
Currently, health authorities issue public warnings of heat-related mortality risk based on the
association between temperature and mortality using relatively long time series (e.g., 10 years
data or 20 years data). This thesis found that the associations between temperature and
mortality were different by year within the same city. Previous studies similarly found that
the 2006 heat waves in western Europe had much less impact on mortality than the 2003 heat
waves (Empereur-Bissonnet, Salines, Berat, Caillere, & Josseran, 2006). In Chicago, there
were fewer heat-related deaths in 1999 heatwaves, compared with 1995 heatwaves. The
impacts of heat waves on mortality varied either across cities or over time in the same city
(Delaroziere & Sanmarco, 2004; Kovats & Kristie, 2006). This variation in heat-related
mortality might be attributed to the successful implementation of prevention measures, such
as the opening of cooling centres, and increase of air conditioning use over recent years.
Another explanation is a significant increase in public health messages and general level of
awareness of heat wave impacts (Palecki, Changnon, & Kunkel, 2001). This means that cities
need to constantly update their heat warnings (for example, reanalysing data at the end of
every summer) as heat-related health risks change over time.
Identification and prevention of the heat-related mortality is a very important public health
challenge, and general warnings should be made at the beginning of every summer.
Intervention policies should best geared towards the local climate and population
characteristics by the local health agencies and social services (Kovats & Hajat, 2008).
201
Efficient warnings of heat-related mortality risk should be generated after taking into account
the variation in temperature effects on mortality over time. Ignoring this variation could lead
to false positives, as in some years an anticipated spike in deaths will not occur (whereas in
other years the spike could be much greater than expected) (Hajat, O'Connor, et al., 2010;
Hajat, Sheridan, et al., 2010; O'Neill, et al., 2009).
This thesis showed that the effects of high temperature on mortality were highly variable
from year to year. This finding suggests that future estimates of climate change may not have
correctly accounted for the uncertainty in the association between heatwaves and mortality,
and so their confidence intervals may be too narrow. Policy makers for public health need to
consider this issue when they establish a warning system for high temperatures and projecting
the impact of future climate change.
In this thesis, I found there is an apparent pattern of heat effects which often had “high risk”
years followed by relatively “low risk” years. This variation in heat effects may be related to
the differences in the at-risk population (e.g., more elderly people), or because of increased
adaptation over time (Sheridan & Kalkstein; Stafoggia, et al., 2009). To conform if this is a
regular pattern we would need a longer time series (possibly 50 years or more). However, this
finding is potentially important for heat warning systems. If policy makers consider this
variation in temperature effects, the heat warning systems might better predict heat-related
mortality.
8.2.2 Health effects of unstable weather
202
Unstable weather is anticipated to increase, as climate change is accelerating. Understanding
whether unstable weather is associated with increased or decreased mortality can lead to
improved understanding of climate change effects on human health. Thus, this thesis
examined mortality risks when people exposed to an unstable temperatures between
neighbouring days. I found that unstable temperatures (relatively large increases or decreases
in temperature between neighbouring days) were associated with an increase in mortality.
That means unstable temperature is hazardous to human health. Public health policy makers
should consider the impacts of unstable temperatures on population health when they make
policy choices. For example, weather forecast announcers could remind residents to pay
attention to an unstable temperature before a sharp increase or decrease in temperature occurs.
People, especially the elderly, young children or those with chronic diseases, who are most
vulnerable to temperature should stay in a controlled temperature environment or take other
measures to maintain an appropriate temperature range in order to protect them from the
harm of unstable temperatures.
203
8.3 Strengths of this thesis
This thesis has six major strengths:
I detailed how to apply a distributed lag non-linear model in the case−crossover design. This
allows the case−crossover design to flexibly estimate the non-linear and delayed effects of
temperature (or air pollution) on human health.
I compared time series and spatiotemporal models in assessing the association between
temperature and mortality. It provides important information for the future research on
temperature and mortality. For example, in general, time series models using single site’s
temperature or averaged temperature can be used to examine the temperature effects on
mortality.
I examined whether temperature change between neighbouring days increases the risk of
mortality in summer. We used two cities in different countries with different subtropical
climates to confirm the findings. The findings suggest not only absolute temperature affects
human health, but also temperature changes (both significant temperature drops and
temperature increases) had significant impacts on human health. This study provides an
important impetus for evaluating population vulnerability, and improving public health
adaptation strategies for reducing the burden of temperature.
I found that the high temperature effect on mortality is highly variable from year to year. This
finding is important for policy makers to develop public health strategies, as this issue needs
204
to be taken into account when establishing a weather/health warning system and projecting
the health impact of future climate change.
I assessed the relationship between temperature and mortality in Tianjin, China using
advanced statistical models. The findings can be used to promote capacity building for local
response to extreme temperatures.
Finally, this thesis used comprehensive data including weather, air pollution, and mortality
from many cities (Tianjin, Brisbane, and 83 large US cities), and including whole population
in these cities as study samples, so the statistical power is high.
205
8.4 Limitations of this thesis
This thesis also has some limitations:
Like other ecological studies, this thesis lacked information on individual data, such as
personal behaviours, indoor and outdoor activities, housing characteristics, and individual
medical history. The environmental data were also limited, as I used the data on ambient
temperature and air pollution from fixed sites rather than individual exposure, so there may
be some inevitable measurement error.
This thesis only used a single city’s data to compare time series and spatiotemporal models.
The finding might be difficult to be generalised to other cities that have great spatial variation
in temperature or mortality. But I provided detailed information on how to choose such
models for other studies.
This thesis did not control some potential confounders, such as fine particulate matter and
Ozone, because these data were not available in Tianjin, and had missing values in US cities.
206
8.5 Recommendations for future research directions
As global climate continues to change, assessing the health effects of temperature is a very
broad and complex topic. Based on the findings from this thesis, I draw a number of
recommendations for future directions.
Many personal and environmental factors can modify the effects of temperature on human
health. Populations acclimatize to their local climates, so what is considered an extreme hot
day in some cities may be regarded as a normal summer day in others (Kovats & Hajat, 2008).
Therefore, it is necessary to examine city-specific temperature-mortality association
worldwide, paying particular attention to differences in local climate patterns and socio-
economic conditions.
Populations in developing countries are anticipated to be especially sensitive to impacts of
climate change, as they have limited adaptive capacity and more vulnerable people (Costello,
et al., 2009). To date there has been little research on temperature effects in developing
countries (McMichael, et al., 2008), including China (Guo, Barnett, Pan, Yu, & Tong, 2011b).
This thesis found both cold and high temperatures are related to mortality increase in Tianjin,
China. Future studies are needed to examine the associations between temperature and
mortality in other Chinese cities and other developing countries.
Studies on the associations between ambient temperature and mortality are mainly from
urban cities, either in developed counties or developing countries (Basu & Samet, 2002;
McMichael, et al., 2008). Few studies have examined the relationship between temperature
and mortality in rural areas, especially in developing countries (Hashizume et al., 2009). In
207
fact, 57% of the population are living in rural area in developing countries (United Nations.,
2006). Populations living in rural areas of developing countries may be vulnerable to
temperature effects, because of the level of socioeconomic development and access to
medical care. I therefore suggest examining the effects of ambient temperature on human
health in rural areas, especially for developing countries. This will be more challenging than
working with data from cities because rural areas have a much wider spatial range, and are
unlikely to have the network of temperature monitoring stations that exist in most cities.
Given the large amount of extra computing time needed to fit spatiotemporal models, I
recommend that future studies use the simple time series model to examine the effects of
temperature on mortality. This may not be applicable to data with more spatially detailed
estimates of temperature and/or health. Therefore, comparison between time series and
spatiotemporal models still need to be confirmed in other cities, especially for those with
large variations in socio-economic status.
208
8.6 Conclusions
There has been increasing concern for the health impacts of exposure to ambient temperature,
because climate change is continuing. Case−crossover and time series models are most
commonly used to examine the effects of temperatures on mortality. Previously, few studies
have used case−crossover to examine the delayed (distributed lag) and non-linear relationship
between temperature and mortality. This thesis made it possible to combine a case–crossover
design and distributed lag non-linear model, which may have significant benefits for future
research to assess temperature related health effects.
Most studies used daily temperature data from one monitoring site or daily mean values from
a network of monitoring sites to estimate the impact of temperature on mortality, which may
result in a measurement error for temperature exposure (Zhang, et al., 2011). In this thesis,
results show that that time series and spatiotemporal models produced similar effect estimates
of temperature exposure. The findings may also be important, because the time series models
are relatively easy to fit and take less time than spatial models.
Climate change may not only increase the frequency, intensity and duration of weather
extremes (e.g. heat waves, floods and cyclones) (WHO/WMO/UNEP, 1996), but also
increase unstable weather patterns (e.g. a significant drop/increase in temperature) in the
coming decades (Faergeman, 2008). However, less evidence is available on the possible
mortality effects due to temperature change between neighbouring days. In this thesis, results
show that a significant change in temperature between the neighbouring days might increase
the risk of mortality in Brisbane, Australia and Los Angeles, USA, which suggests that
209
temperature change is a significant risk factor for mortality. People sensitive to unstable
weather should protect themselves from the temperature change.
Most previous studies only considered high temperature effects by averaging over the whole
study period, and ignored the variability in effects from year to year. In this thesis, results
show that high temperature effects on mortality varied greatly by year, city and region in US
cities, which demonstrates that it is important to take the variability in high temperature
effects into account in the development of weather/health warning systems. Policy makers for
public health need to consider this issue when they develop a warning system for high
temperatures and projecting the impact of future climate change.
210
REFERENCES
Adger, W. N. (2006). Vulnerability. Global Environmental Change, 16(3), 268-281.
Alderson, M. R. (1985). Season and mortality. Health Trends, 17, 87-96.
Analitis, A., Katsouyanni, K., Biggeri, A., Baccini, M., Forsberg, B., Bisanti, L., et al. (2008).
Effects of cold weather on mortality: results from 15 European cities within the
PHEWE project. American Journal of Epidemiology, 168(12), 1397-1408.
Anderson, B. G., & Bell, M. L. (2009). Weather-related mortality: how heat, cold, and heat
waves affect mortality in the United States. Epidemiology, 20(2), 205-213.
Anderson, G. B., & Bell, M. L. (2011). Heat Waves in the United States: Mortality Risk
during Heat Waves and Effect Modification by Heat Wave Characteristics in 43 U.S.
Communities. Environmental Health Perspectives, 119(2), 210-218.
Aniello, C., Morgan, K., Busbey, A., & Newland, L. (1995). Mapping micro-urban heat
islands using Landsat TM and a GIS. Computers & Geosciences, 21(8), 965-967.
Armstrong, B. (2006). Models for the relationship between ambient temperature and daily
mortality. Epidemiology, 17(6), 624-631.
Augustin, N. H., Musio, M., von Wilpert, K., Kublin, E., Wood, S. N., & Schumacher, M.
(2009). Modeling spatiotemporal forest health monitoring data. Journal of the
American Statistical Association, 104(487), 899-911.
Axelrod, Y. K., & Diringer, M. N. (2006). Temperature management in acute neurologic
disorders. Critical Care Clinics, 22(4), 767-785.
Baccini, M., Biggeri, A., Accetta, G., Kosatsky, T., Katsouyanni, K., Analitis, A., et al.
(2008). Heat effects on mortality in 15 European cities. Epidemiology, 19(5), 711-719.
Baker-Blocker, A. (1982). Winter weather and cardiovascular mortality in Minneapolis-St.
Paul. American Journal Of Public Health, 72(3), 261-265.
211
Ballester, F., Corella, D., Perez-Hoyos, S., Saez, M., & Hervas, A. (1997b). Mortality as a
function of temperature. A study in Valencia, Spain, 1991-1993. International
Journal of Epidemiology, 26(3), 551-561.
Barnett, A. G. (2007). Temperature and cardiovascular deaths in the US elderly: changes over
time. Epidemiology, 18(3), 369-372.
Barnett, A. G., & Dobson, A. J. (2010). Analysing Seasonal Health Data. Berlin, Heidelberg:
Springer.
Barnett, A. G., Tong, S., & Clements, A. (2010). What measure of temperature is the best
predictor of mortality? Environmental research, 110(6), 604-611.
Basu, R. (2009a). High ambient temperature and mortality: a review of epidemiologic studies
from 2001 to 2008. Environmental Health, 8, 40-52.
Basu, R., Dominici, F., & Samet, J. M. (2005). Temperature and mortality among the elderly
in the United States: a comparison of epidemiologic methods. Epidemiology, 16(1),
58-66.
Basu, R., Feng, W. Y., & Ostro, B. D. (2008). Characterizing temperature and mortality in
nine California counties. Epidemiology, 19(1), 138-145.
Basu, R., & Ostro, B. D. (2008). A multicounty analysis identifying the populations
vulnerable to mortality associated with high ambient temperature in California.
American Journal of Epidemiology, 168, 632-637.
Basu, R., & Samet, J. M. (2002). Relation between elevated ambient temperature and
mortality: a review of the epidemiologic evidence. Epidemiologic Reviews, 24, 190–
202.
Bateson, T. F., & Schwartz, J. (1999). Control for seasonal variation and time trend in case-
crossover studies of acute effects of environmental exposures. Epidemiology, 10(5),
539-544.
212
Bell, M. L., McDermott, A., Zeger, S. L., Samet, J. M., & Dominici, F. (2004). Ozone and
Short-term Mortality in 95 US Urban Communities, 1987-2000. Journal of the
American Medical Association, 292(19), 2372-2378.
Bell, M. L., O'Neill, M. S., Ranjit, N., Borja-Aburto, V. H., Cifuentes, L. A., & Gouveia, N.
C. (2008). Vulnerability to heat-related mortality in Latin America: a case-crossover
study in Sao Paulo, Brazil, Santiago, Chile and Mexico City, Mexico. International
Journal of Epidemiology, 37(4), 796-804.
Benavides, R., Montes, F., Rubio, A., & Osoro, K. (2007). Geostatistical modelling of air
temperature in a mountainous region of Northern Spain. Agricultural and Forest
Meteorology, 146(3-4), 173-188.
Bhowmik, A., & Cabral, P. (2011). Statistical Evaluation of Spatial Interpolation Methods for
Small-Sampled Region: A Case Study of Temperature Change Phenomenon in
Bangladesh. Computational Science and Its Applications-ICCSA 2011, 44-59.
Bobb, J. F., Dominici, F., & Peng, R. D. (2011). A Bayesian model averaging approach for
estimating the relative risk of mortality associated with heat waves in 105 U.S. cities.
Biometrics, 67(4), 1605-1616.
Bouchama, A., & Knochel, J. P. (2002). Heat stroke. New England Journal of Medicine,
346(25), 1978-1988.
Braga, A. L., Zanobetti, A., & Schwartz, J. (2001). The time course of weather-related deaths.
Epidemiology, 12(6), 662-667.
Braga, A. L., Zanobetti, A., & Schwartz, J. (2002). The effect of weather on respiratory and
cardiovascular deaths in 12 U.S. cities. Environmental Health Perspectives, 110(9),
859-863.
213
Bull, G. M., & Morton, J. (1975). Relationships of temperature with death rates from all
causes and from certain respiratory and arteriosclerotic diseases in different age
groups. Age Ageing, 4(4), 232-246.
Carder, M., McNamee, R., Beverland, I., Elton, R., Cohen, G. R., Boyd, J., et al. (2005a).
The lagged effect of cold temperature and wind chill on cardiorespiratory mortality in
Scotland. Occupational and Environmental Medicine, 62(10), 702-710.
Carder, M., McNamee, R., Beverland, I., Elton, R., Cohen, G. R., Boyd, J., et al. (2005b).
The lagged effect of cold temperature and wind chill on cardiorespiratory mortality in
Scotland. Occupational and Environmental Medicine, 62(10), 702-710.
Carson, C., Hajat, S., Armstrong, B., & Wilkinson, P. (2006). Declining vulnerability to
temperature-related mortality in London over the 20th century. American Journal of
Epidemiology 164(1), 77-84.
Chan, E. Y., Goggins, W. B., Kim, J. J., & Griffiths, S. M. (2010). A study of intracity
variation of temperature-related mortality and socioeconomic status among the
Chinese population in Hong Kong. Journal of Epidemiology and Community Health.
Chan, N. Y., Stacey, M. T., Smith, A. E., Ebi, K. L., & Wilson, T. F. (2001). An empirical
mechanistic framework for heat-related illness. Climate Research, 16(2), 133-143.
Chuanyan, Z., Zhongren, N., & Guodong, C. (2005). Methods for modelling of temporal and
spatial distribution of air temperature at landscape scale in the southern Qilian
mountains, China. Ecological modelling, 189(1-2), 209-220.
Clark, R. P., & Edholm, O. G. (1985). Man and his thermal environment: Edward Arnold
London.
Collins, K. J. (1987). Effects of cold on old people. British Journal Of Hospital Medicine,
38(6), 506.
214
Conti, S., Meli, P., Minelli, G., Solimini, R., Toccaceli, V., Vichi, M., et al. (2005).
Epidemiologic study of mortality during the Summer 2003 heat wave in Italy.
Environmental Research, 98(3), 390-399.
Costello, A., Abbas, M., Allen, A., Ball, S., Bell, S., Bellamy, R., et al. (2009). Managing the
health effects of climate change. Lancet, 373(9676), 1693-1733.
Curriero, F. C., Heiner, K. S., Samet, J. M., Zeger, S. L., Strug, L., & Patz, J. A. (2002).
Temperature and mortality in 11 cities of the eastern United States. American Journal
of Epidemiology 155(1), 80-87.
DÃaz, J., GarcÃa, R., López, C. s., Linares, C., TobÃas, A., & Prieto, L. (2004). Mortality
impact of extreme winter temperatures. International Journal of Biometeorology,
49(3), 179-183.
Daniels, M. J., Dominici, F., Samet, J. M., & Zeger, S. L. (2000). Estimating particulate
matter-mortality dose-response curves and threshold levels: an analysis of daily time-
series for the 20 largest US cities. American journal of epidemiology, 152(5), 397.
Davis, R. E., Knappenberger, P. C., Michaels, P. J., & Novicoff, W. M. (2003). Changing
heat-related mortality in the United States. Environmental Health Perspectives,
111(14), 1712-1718.
Delaroziere, J. C., & Sanmarco, J. L. (2004). Excess mortality in people over 65 years old
during summer heat waves in Marseille. Comparison before and after a preventive
campaign. Presse Medicine, 33(1), 13-16.
Diaz, J., Garcia, R., Velazquez de Castro, F., Hernandez, E., Lopez, C., & Otero, A. (2002).
Effects of extremely hot days on people older than 65 years in Seville (Spain) from
1986 to 1997. International Journal of Biometeorology, 46(3), 145-149.
Dominici, F., McDermott, A., Daniels, M., Zeger, S. L., & Samet, J. M. (2005). Revised
Analyses of the National Morbidity, Mortality, and Air Pollution Study: Mortality
215
Among Residents Of 90 Cities. Journal of Toxicology & Environmental Health: Part
A, 68(13/14), 1071-1092.
Dominici, F., McDermott, A., Zeger, S. L., & Samet, J. M. (2002). On the Use of Generalized
Additive Models in Time-Series Studies of Air Pollution and Health. American
Journal of Epidemiology, 156(3), 193-203.
Donaldson, G. C., Keatinge, W. R., & Nayha, S. (2003). Changes in summer temperature and
heat-related mortality since 1971 in North Carolina, South Finland, and Southeast
England. Environmental Research, 91(1), 1-7.
Donaldson, G. C., Tchernjavskii, V. E., Ermakov, S. P., Bucher, K., & Keatinge, W. R.
(1998). Winter mortality and cold stress in Yekaterinburg, Russia: interview survey.
British Medical Journal, 316(7130): 514–518.
Douglas, A. S., Dunnigan, M. G., Allan, T. M., & Rawles, J. M. (1995). Seasonal variation in
coronary heart disease in Scotland. Journal of Epidemiology and Community Health,
49(6), 575-582.
Ebi, K. L. (2007). Towards an early warning system for heat events. [Article]. Journal of Risk
Research, 10(5), 729-744.
Ebi, K. L., Exuzides, K. A., Lau, E., Kelsh, M., & Barnston, A. (2004). Weather changes
associated with hospitalizations for cardiovascular diseases and stroke in California,
1983-1998. International Journal of Biometeorology, 49(1), 48-58.
Ebi, K. L., Kovats, R. S., & Menne, B. (2006). An approach for assessing human health
vulnerability and public health interventions to adapt to climate change.
Environmental Health Perspectives, 114(12), 1930-1934.
Ebi, K. L., & Schmier, J. K. (2005). A stitch in time: improving public health early warning
systems for extreme weather events. Epidemiological Review, 27, 115-121.
216
El-Zein, A., Tewtel-Salem, M., & Nehme, G. (2004). A time-series analysis of mortality and
air temperature in Greater Beirut. Science of The Total Environment, 330(1-3), 71-80.
Empereur-Bissonnet, P., Salines, G., Berat, B., Caillere, N., & Josseran, L. (2006). Heatwave
in France, July 2006: 112 excess deaths so far attributed to the heat. European
communicable disease bulletin, 11(8), E060803. 060803.
Everson, P. J., & Morris, C. N. (2000). Inference for multivariate normal hierarchical models.
Journal of the Royal Statistical Society: Series B (Statistical Methodology), 62(2),
399-412.
Faergeman, O. (2008). Climate change, disease and public health. Ugeskr Laeger, 170(35),
2667-2668.
Filleul, L., Cassadou, S., Medina, S., Fabres, P., Lefranc, A., Eilstein, D., et al. (2006). The
relation between temperature, ozone, and mortality in nine French cities during the
heat wave of 2003. Environmental Health Perspectives, 114(9), 1344-1347.
Fouillet, A., Rey, G., Jougla, E., Frayssinet, P., Bessemoulin, P., & Hemon, D. (2007). A
predictive model relating daily fluctuations in summer temperatures and mortality
rates. BMC Public Health, 7, 114.
Fung, K. Y., Krewski, D., Chen, Y., Burnett, R., & Cakmak, S. (2003). Comparison of time
series and case-crossover analyses of air pollution and hospital admission data.
International Journal of Epidemiology, 32(6), 1064-1070.
Gasparrini, A., & Armstrong, B. (2011a). Distributed lag non-linear models in R: the package
dlnm. Retrieved from http://cran.r-
project.org/web/packages/dlnm/vignettes/dlnmOverview.pdf
Gasparrini, A., & Armstrong, B. (2011b). The impact of heat waves on mortality.
Epidemiology, 22(1), 68-73.
217
Gasparrini , A., Armstrong , B., & Kenward, M. (2010). Distributed lag non-linear models.
Statistics in Medicine, 29(21), 2224-2234.
Gemmell, I., McLoone, P., Boddy, F. A., Dickinson, G. J., & Watt, G. C. M. (2000).
Seasonal variation in mortality in Scotland. International Journal of Epidemiology,
29(2), 274-279.
Goodman, P. G., Dockery, D. W., & Clancy, L. (2004). Cause-Specific Mortality and the
Extended Effects of Particulate Pollution and Temperature Exposure. Environmental
Health Perspectives, 112(2), 179-185.
Gorjanc, M. L., Flanders, W. D., VanDerslice, J., Hersh, J., & Malilay, J. (1999). Effects of
temperature and snowfall on mortality in Pennsylvania. American Journal of
Epidemiology, 149(12), 1152-1160.
Gosling, S. N., Lowe, J. A., McGregor, G. R., Pelling, M., & Malamud, B. D. (2009).
Associations between elevated atmospheric temperature and human mortality: a
critical review of the literature. Climatic Change, 92(3), 299-341.
Gouveia, N., Hajat, S., & Armstrong, B. (2003). Socioeconomic differentials in the
temperature-mortality relationship in Sao Paulo, Brazil. International Journal of
Epidemiology, 32(3), 390-397.
Green, R. S., Basu, R., Malig, B., Broadwin, R., Kim, J. J., & Ostro, B. (2010). The effect of
temperature on hospital admissions in nine California counties. International Journal
of Public Health, 55(2), 113-121.
Greenland, S. (1996). Confounding and exposure trends in case-crossover and case-time-
control designs. Epidemiology, 7(3), 231-239.
Group, E. (1997). Cold exposure and winter mortality from ischaemic heart disease,
cerebrovascular disease, respiratory disease, and all causes in warm and cold regions
of Europe. The Eurowinter Group. Lancet, 349(9062), 1341-1346.
218
Guo, Y., Barnett, A. G., Pan, X., Yu, W., & Tong, S. (2011). The impact of temperature on
mortality in tianjin, china: a case-crossover design with a distributed lag nonlinear
model. Environmental Health Perspectives, 119(12), 1719-1725.
Guo, Y., Barnett, A. G., Yu, W., Pan, X., Ye, X., Huang, C., et al. (2011). A Large Change in
Temperature between Neighbouring Days Increases the Risk of Mortality. PLoS One,
6(2), e16511.
Guy, W. A. (1858). On the annual fluctuations in the number of deaths from various diseases,
compared with like fluctuations in crime, and in other events within and
beyond the control of the humanwill. Journal of the Statistical Society of London, 21,
52-86.
Gyllerup, S., Lanke, J., Lindholm, L. H., & Schersten, B. (1991). High coronary mortality in
cold regions of Sweden. Journal of internal medicine, 230(6), 479-485.
Ha, J., Kim, H., & Hajat, S. (2011). Effect of Previous-Winter Mortality on the Association
between Summer Temperature and Mortality in South Korea. Environmental Health
Perspectives, 119:542-546.
Haines, A., Kovats, R., Campbell-Lendrum, D., & Corvalan, C. (2006). Climate change and
human health: impacts, vulnerability, and mitigation. Lancet, 367(9528), 2101-2109.
Hajat, S., Armstrong, B., Baccini, M., Biggeri, A., Bisanti, L., Russo, A., et al. (2006).
Impact of high temperatures on mortality: is there an added heat wave effect?
Epidemiology, 17(6), 632-638.
Hajat, S., Armstrong, B. G., Gouveia, N., & Wilkinson, P. (2005). Mortality displacement of
heat-related deaths: a comparison of Delhi, Sao Paulo, and London. Epidemiology,
16(5), 613-620.
Hajat, S., & Kosatky, T. (2010). Heat-related mortality: a review and exploration of
heterogeneity. Journal of Epidemiology and Community Health, 64(9), 753-760.
219
Hajat, S., Kovats, R. S., Atkinson, R. W., & Haines, A. (2002). Impact of hot temperatures on
death in London: a time series approach. Journal of Epidemiology and Community
Health, 56(5): 367-372.
Hajat, S., Kovats, R. S., & Lachowycz, K. (2007). Heat-related and cold-related deaths in
England and Wales: who is at risk? Occupational and Environmental Mededicine.,
64(2), 93-100.
Hajat, S., O'Connor, M., & Kosatsky, T. (2010). Health effects of hot weather: from
awareness of risk factors to effective health protection. Lancet, 375(9717), 856-863.
Hajat, S., Sheridan, S. C., Allen, M. J., Pascal, M., Laaidi, K., Yagouti, A., et al. (2010).
Heat-health warning systems: a comparison of the predictive capacity of different
approaches to identifying dangerously hot days. American Journal of Public Health,
100(6), 1137-1144.
Hales, S., Salmond, C., Town, G. I., Kjellstrom, T., & Woodward, A. (2000a). Daily
mortality in relation to weather and air pollution in Christchurch, Australia and New
Zealand journal of Public Health, 24 (1), 89-91.
Hansen, A. L., Bi, P., Ryan, P., Nitschke, M., Pisaniello, D., & Tucker, G. (2008). The effect
of heat waves on hospital admissions for renal disease in a temperate city of Australia.
International Journal of Epidemiology, 37(6), 1359-1365.
Hashizume, M., Wagatsuma, Y., Hayashi, T., Saha, S. K., Streatfield, K., & Yunus, M.
(2009). The effect of temperature on mortality in rural Bangladesh--a population-
based time-series study. International Journal of Epidemiology, 38(6), 1689-1697.
Hastie, T., Tibshirani, R., & Friedman, J. (2004). The Elements of Statistical Learning: Data
Mining, Inference, and Prediction. BeiJing: Publishing House of Electronics Industry.
Hastie, T. J., & Tibshirani, R. J. (1990). Generalized Additive Models. In Chanpman&Hall
(Ed.), AT&T Bell Laboratories.
220
Hastie, T. J., & Tibshirani, R. J. (1990). Generalized additive models: Chapman & Hall/CRC.
Healy, J. D. (2003). Excess winter mortality in Europe: a cross country analysis identifying
key risk factors. Journal of Epidemiology and Community Health, 57(10):784-9.
Hong, Y.-C., Lee, J.-T., Kirn, H., Ha, E.-H., Schwartz, J., & Christiani, D. C. (2002). Effects
of air pollutants on acute stroke mortality. Environmental Health Perspectives, 110(2),
187-191.
Hong, Y. C., Rha, J. H., Lee, J. T., Ha, E. H., Kwon, H. J., & Kim, H. (2003). Ischemic
stroke associated with decrease in temperature. Epidemiology, 14(4), 473-478.
Hutcheon, J. A., Chiolero, A., & Hanley, J. A. (2010). Random measurement error and
regression dilution bias. British Medical Journal, 340, c2289.
Huynen, M. M., Martens, P., Schram, D., Weijenberg, M. P., & Kunst, A. E. (2001). The
impact of heat waves and cold spells on mortality rates in the Dutch population.
Environmental Health Perspectives, 109(5), 463-470.
Intergovernmental Panel on Climate Change(IPCC). (2001). Climate Change 2001: Third
Assessment Report(Volume 1). Cambridge: UK Cambridge University Press.
IPCC. (2007a). Climate change 2007: impacts, adaptation and vulnerability. Contribution of
Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on
Climate Change. Cambridge, UK: Cambridge University Press.
IPCC. (2007b). Climate change 2007: synthesis report. Contribution of Working Groups I, II
and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate
Change. Geneva, Switzerland: IPCC.
IPCC. (2007c). Climate change 2007: the physical science basis. Contribution of Working
Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate
Change. Cambridge, UK: Cambridge University Press.
221
Isaaks, E. H., & Srivastava, R. M. (1989). Applied geostatistics (Vol. 2): Oxford University
Press New York.
Ishigami, A., Hajat, S., Kovats, R. S., Bisanti, L., Rognoni, M., Russo, A., et al. (2008). An
ecological time-series study of heat-related mortality in three European cities.
Environmental Health, 7, 5-11.
Janes, H., Sheppard, L., & Lumley, T. (2005). Overlap bias in the case-crossover design, with
application to air pollution exposures. Statistics in Medicine, 24(2), 285-300.
Jiang, G., Qu, T., Shang, Z., & Zhang, X. (2004). A circuit simulating method for heat
transfer mechanism in human body. Proceedings of the 26th Annual International
Conference of the IEEE EMBS(September 1-5), 5274-5276.
Johnson, H., Kovats, R. S., McGregor, G., Stedman, J., Gibbs, M., Walton, H., et al. (2005).
The impact of the 2003 heat wave on mortality and hospital admissions in England.
Health Statistics Quaterly, 25, 6-11.
Kaiser, R., Le Tertre, A., Schwartz, J., Gotway, C. A., Daley, W. R., & Rubin, C. H. (2007).
The effect of the 1995 heat wave in Chicago on all-cause and cause-specific mortality.
American Journal of Public Health, 97, S158-162.
Kalkstein, L. L., & Valimont, K. M. (1986). An evaluation of summer discomfort in the
United-States using a relative climatological index. Journal of the American Medical
Association, 67 842 -848.
Kalkstein, L. S., & Davis, R. E. (1989). Weather and human mortality: an evaluation of
demographic and interregional responses in the United States. Annals of
the Association of American Geographers, 79(1), 44-64.
Kalkstein, L. S., & Greene, J. S. (1997). An evaluation of climate/mortality relationships in
large U.S. cities and the possible impacts of a climate change. Environmental Health
Perspectives, 105(1), 84-93.
222
Kan, H., Chen, R., & Tong, S. (2011). Ambient air pollution, climate change, and population
health in China. Environment International.
Kan, H., London, S. J., Chen, H., Song, G., Chen, G., Jiang, L., et al. (2007). Diurnal
temperature range and daily mortality in Shanghai, China. Environmental Research,
103(3), 424-431.
Kan H.D., Jia J, Chen B.H. 2003. Temperature and daily mortality in Shanghai: a time-series
study. Biomedical and Environmental Science, 16(2):133-139.
Keatinge, W. R., & Donaldson, G. C. (2001). Mortality Related to Cold and Air Pollution in
London After Allowance for Effects of Associated Weather Patterns. Environmental
Research, 86(3):209-16.
Keatinge, W. R., Donaldson, G. C., Bucher, K., Jendritzky, G., Cordioli, E., Martinelli, M., et
al. (2000). Winter mortality in relation to climate. International Journal of
Circumpolar Health, 59(3-4), 154-159.
Keatinge, W. R., Donaldson, G. C., Cordioli, E., Martinelli, M., Kunst, A. E., Mackenbach, J.
P., et al. (2000). Heat related mortality in warm and cold regions of Europe:
observational study. British Medical Journal, 321(7262), 670-673.
Kestens, Y., Brand, A., Fournier, M., Goudreau, S., Kosatsky, T., Maloley, M., et al. (2011).
Modelling the variation of land surface temperature as determinant of risk of heat-
related health events. International Journal of Health Geographics, 10, 7.
Kilbourne, E. M. (1999). The spectrum of illness during heat waves. American Journal of
Preventive Medicine, 16(4), 359-360.
Kim, S., & Jang, J. Y. (2005). Temperature and summer mortality association in Seoul.
Epidemiology, 16(5), S112-S112
223
Kinney, P. L., O'Neill, M. S., Bell, M. L., & Schwartz, J. (2008). Approaches for estimating
effects of climate change on heat-related deaths: challenges and opportunities.
Environmental Science & Policy, 11(1), 87-96.
Knowlton, K., Rotkin-Ellman, M., King, G., Margolis, H. G., Smith, D., Solomon, G., et al.
(2009). The 2006 California heat wave: impacts on hospitalizations and emergency
department visits. Environmental Health Perspectives, 117(1), 61-67.
Koutsavlis, A. T., & Kosatsky, T. (2003). Environmental-Temperature Injury in a Canadian
Metropolis. Journal of Environmental Health, 66(5), 40-45.
Kovats, R. S., & Hajat, S. (2008). Heat stress and public health: A critical review. Annual
Review of Public Health, 29, 41-55.
Kovats, R. S., & Kristie, L. E. (2006). Heatwaves and public health in Europe. European
Journal of Public Health, 16(6), 592-599.
Kroemer, K. H. E., & Grandjean, E. (1997). Fitting the task to the human. A textbook of
occupational ergonomics (Fifth edition ed. Vol. Chapter 20). London:
Taylor&Francis Ltd.
Kyobutungi, C., Grau, A., Stieglbauer, G., & Becher, H. (2005). Absolute temperature,
temperature changes and stroke risk: a case-crossover study. European Journal of
Epidemiology, 20(8), 693-698.
Kysely, J., Pokorna, L., Kyncl, J., & Kriz, B. (2009). Excess cardiovascular mortality
associated with cold spells in the Czech Republic. BMC Public Health, 9, 19.
Lawlor, D. A., Maxwell, R., & Wheeler, B. W. (2002). Rurality, deprivation, and excess
winter mortality: An ecological study. Journal of Epidemiology and Community
Health, 56(5), 373-374.
224
Le Tertre, A., Lefranc, A., Eilstein, D., Declercq, C., Medina, S., Blanchard, M., et al. (2006).
Impact of the 2003 heatwave on all-cause mortality in 9 French cities. Epidemiology,
17(1), 75-79.
Leah, J. W., & Scott, L. Z. (2005). Are the Acute Effects of Particulate Matter on Mortality in
the National Morbidity, Mortality, and Air Pollution Study the Result of Inadequate
Control for Weather and Season? A Sensitivity Analysis using Flexible Distributed
Lag Models. American Journal of Epidemiology, 162(1), 80.
Lee, D., & Shaddick, G. (2010). Spatial modeling of air pollution in studies of its short-term
health effects. Biometrics, 66(4), 1238-1246.
Lefohn, A. S., Knudsen, H. P., & McEvoy, L. R., Jr. (1988). The use of kriging to estimate
monthly ozone exposure parameters for the Southeastern United States.
Environmental Pollution, 53(1-4), 27-42.
Levy, D., Sheppard, L., Checkoway, H., Kaufman, J., Lumley, T., Koenig, J., et al. (2001). A
case-crossover analysis of particulate matter air pollution and out-of-hospital primary
cardiac arrest. Epidemiology, 12(2), 193-199.
Liu, L., Breitner, S., Pan, X., Franck, U., Leitte, A., Wiedensohler, A., et al. (2011).
Associations between Air Temperature and Cardio-Respiratory Mortality in the Urban
Area of Beijing, China: A Time-Series Analysis. Environmental Health, 10(1), 51.
Lo, C. P., Quattrochi, D. A., & Luvall, J. C. (1997). Application of high-resolution thermal
infrared remote sensing and GIS to assess the urban heat island effect. International
Journal of Remote Sensing, 18(2), 287-304.
Lu, Y., Symons, J. M., Geyh, A. S., & Zeger, S. L. (2008). An approach to checking case-
crossover analyses based on equivalence with time-series methods. Epidemiology,
19(2), 169-175.
225
Lu, Y., & Zeger, S. L. (2007). On the equivalence of case-crossover and time series methods
in environmental epidemiology. Biostatistics, 8(2), 337-344.
Luber, G., & McGeehin, M. (2008). Climate change and extreme heat events. American
Journal of Preventive Medicine, 35(5), 429-435.
Lumley, T., & Levy, D. (2000). Bias in the case-crossover design: implications for studies of
air pollution. Environmetrics, 11, 689-704.
Macey, S. M., & Schneider, D. F. (1993). Deaths from excessive heat and excessive cold
among the elderly. The Gerontologist, 33(4), 497.
Mannino, J. A., & Washburn, R. A. (1989). Environmental temperature and mortality from
acute myocardial infarction. International Journal of Biometeorology, 33(1), 32-35.
Martinez, B. F., Annest, J. L., Kilbourne, E. M., Kirk, M. L., Lui, K. J., & Smith, S. M.
(1989). Geographic distribution of heat-related deaths among elderly persons. Use of
county-level dot maps for injury surveillance and epidemiologic research. Journal of
the American Medical Association, 262(16), 2246-2250.
McGeehin, M. A., & Mirabelli, M. (2001). The potential impacts of climate variability and
change on temperature-related morbidity and mortality in the United States.
Environmental Health Perspectives, 109, 185-189.
McMichael, A. J. (1993). Global environmental change and human population health: a
conceptual and scientific challenge for epidemiology. International Journal of
Epidemiology, 22(1), 1-8.
McMichael, A. J., Wilkinson, P., Kovats, R. S., Pattenden, S., Hajat, S., Armstrong, B., et al.
(2008). International study of temperature, heat and urban mortality: the
'ISOTHURM' project. International Journal of Epidemiology, 37(5), 1121-1131.
Medina--Ramón, M., Zanobetti, A., Cavanagh, D. P., & Schwartz, J. (2006). Extreme
temperatures and mortality: Assessing effect modification by personal characteristics
226
and specific cause of death in a multi-city case-only analysis. Environmental Health
Perspectives, 114(9), 1331-1336.
Medina-Ramón, M., & Schwartz, J. (2007). Temperature, temperature extremes, and
mortality: a study of acclimatisation and effect modification in 50 US cities.
Occupational and Environmental Medicine, 64(12), 827-833.
Metabolism, Energy Balance, and Temperature Regulation. (2008, April 10, 2008).
Retrieved 11 May,2010
Michelozzi, P., De Sario, M., Accetta, G., de'Donato, F., Kirchmayer, U., D'Ovidio, M., et al.
(2006). Temperature and summer mortality: geographical and temporal variations in
four Italian cities. Journal of Epidemiology and Community Health, 60(5), 417-423.
Michelozzi, P., Fano, V., Forastiere, F., Barca, A., Kalkstein, L. S., & Perucci, C. A. (2000).
Weather conditions and elderly mortality in Rome during summer. Bulletin of World
Meteorology Organization, 49, 348-355.
Muggeo, V. M., & Hajat, S. (2009). Modelling the non-linear multiple-lag effects of ambient
temperature on mortality in Santiago and Palermo: a constrained segmented
distributed lag approach. British Medical Journal, 66(9), 584-591.
Muggeo, V. M. R. (2003). Estimating regression models with unknown break-points.
Statistics in Medicine, 22(19), 3055-3071.
Nakaji, S., Parodi, S., Fontana, V., Umeda, T., Suzuki, K., Sakamoto, J., et al. (2004).
Seasonal changes in mortality rates from main causes of death in Japan (1970--1999).
European Journal of Epidemiology, 19(10), 905-913.
Navidi, W. (1998). Bidirectional case-crossover designs for exposures with time trends.
Biometrics, 54(2), 596-605.
Navidi, W., & Weinhandl, E. (2002). Risk set sampling for case-crossover designs.
Epidemiology, 13(1), 100-105.
227
Näyhä, S. (2005). Environmental temperature and mortality. International Journal of
Circumpolar Health, 64(5), 451-458.
Neas, L. M., Schwartz, J., & Dockery, D. (1999). A Case-Crossover Analysis of Air Pollution
and Mortality in Philadelphia. Environmental Health Perspectives, 107(8), 629-31.
Nelder, J. A., & Wedderburn, R. W. M. (1972). Generalized Linear Models. J R Soc, 135(3),
370-384.
Nunes, B., Paixao, E., Dias, C. M., Nogueira, P., & Marinho Falcao, J. (2011). Air
conditioning and intrahospital mortality during the 2003 heatwave in Portugal:
evidence of a protective effect. Occupational and Environmental Medicine, 68(3),
218-223.
O'Neill, M. S., Carter, R., Kish, J. K., Gronlund, C. J., White-Newsome, J. L., Manarolla, X.,
et al. (2009). Preventing heat-related morbidity and mortality: new approaches in a
changing climate. Maturitas, 64(2), 98-103.
O'Neill, M. S., Zanobetti, A., & Schwartz, J. (2003). Modifiers of the temperature and
mortality association in seven US cities. American Journal of Epidemiology, 157(12),
1074-1082.
O’Neill, M. (2003). Air conditioning and heat-related health effects. Applied Environmental
Science and Public Health, 1, 9-12.
Palecki, M. A., Changnon, S. A., & Kunkel, K. E. (2001). The nature and impacts of the July
1999 heat wave in the midwestern United States: learning from the lessons of 1995.
Bulletin of the American Meteorological Society, 82(7), 1353-1367.
Pan, W., Li, L., & Tsai, M. (1995). Temperature extremes and mortality from coronary heart
disease and cerebral infarction in elderly Chinese. Lancet, 345(8946), 353-355.
228
Parsons, K. C. (2003). Huamn thermal environment. The effect of hot, moderate and cold
environments n human health,comfort and performance (Second edition ed.). London
and New York: Taylor&Francis.
Pattenden, S., Nikiforov, B., & Armstrong, B. G. (2003). Mortality and temperature in Sofia
and London. Journal of Epidemiology and Community Health, 57(8), 628-633.
Patz, J. A., Campbell-Lendrum, D., Holloway, T., & Foley, J. A. (2005). Impact of regional
climate change on human health. Nature, 438(7066), 310-317.
Peng, R. D., Dominici, F., & Louis, T. A. (2006). Model choice in time series studies of air
pollution and mortality. Journal of the Royal Statistical Society: Series A (Statistics in
Society), 169(2), 179-203.
Physiological bases of human performance during work and exercise. (2008). Taylor & H.
Groeller (Eds.), (pp. 321-487). Philadephia: Churchill Livingstone Elsevier limited.
Plavcova, E., & Kysely, J. (2010). Relationships between sudden weather changes in summer
and mortality in the Czech Republic, 1986-2005. International Journal of
Biometeorology, 54(5):539-51.
Reid, C., O’Neill, M., Gronlund, C., Brines, S., Brown, D., Diez-Roux, A., et al. (2009).
Mapping community determinants of heat vulnerability. Environmental Health
Perspectives, 117(11), 1730-1736.
Ren, C., Williams, G. M., Morawska, L., Mengersen, K., & Tong, S. (2008). Ozone modifies
associations between temperature and cardiovascular mortality: analysis of the
NMMAPS data. [Article]. Occupational and Environmental Medicine, 65(4), 255-260.
Ren, C., Williams, G. M., & Tong, S. (2006). Does particulate matter modify the association
between temperature and cardiorespiratory diseases? Environmental Health
Perspectives, 114(11), 1690-1696.
229
Revich, B., & Shaposhnikov, D. (2008a). Excess mortality during heat waves and cold spells
in Moscow, Russia. Occupational and Environmental Medicine, 65(10), 691-696.
Revich , B., & Shaposhnikov, D. (2008). Temperature-induced excess mortality in Moscow,
International Journal of Biometeorology, 52(5):367-374.
Rey, G., Fouillet, A., Bessemoulin, P., Frayssinet, P., Dufour, A., Jougla, E., et al. (2009).
Heat exposure and socio-economic vulnerability as synergistic factors in heatwave-
related mortality. European Journal of Epidemiology, 24(9), 495-502.
Rey, G., Jougla, E., Fouillet, A., Pavillon, G., Bessemoulin, P., Frayssinet, P., et al. (2007).
The impact of major heat waves on all-cause and cause-specific mortality in France
from 1971 to 2003. International archives of occupational and environmental health,
80(7), 615-626.
Rocklov J, Forsberg B. 2008. The effect of temperature on mortality in Stockholm 1998--
2003: a study of lag structures and heatwave effects. Scandinavian Journal of Public
Health, 36(5):516-523.
Rocklov, J., Forsberg, B., & Meister, K. (2009). Winter mortality modifies the heat-mortality
association the following summer. European Respiratory Journal, 33(2), 245-251.
Rogot, E., & Blackwelder, W. C. (1970). Associations of cardiovascular mortality with
weather in Memphis, Tennessee. Public Health Reports, 85(1), 25.
Rose, G. (1966). Cold weather adn ischaemic heart disease. Brit. J. Prev.soc.Med, 20, 97-100.
Rossi, G., Vigotti, M. A., Zanobetti, A., Repetto, F., Gianelle, V., & Schwartz, J. (1999). Air
Pollution and Cause-Specific Mortality in Milan, Italy, 1980-1989. Archives of
Environmental Health, 54(3), 158-164.
Ruppert, D., Wand, M., & Carroll, R. (2003). Semiparametric regression: Cambridge
University Press.
230
Saez, M., Sunyer, J., Castellsague, J., Murillo, C., & Anto, J. M. (1995). Relationship
between weather temperature and mortality: a time series analysis approach in
Barcelona. International Journal of Epidemiology, 24(3), 576-582.
Saldiva, P. H. N., & Pope Iii, C. A. (1995). Air Pollution and Mortality in Elderly People: A
Time-Series Study in Sao Paulo, Brazil. Archives of Environmental Health, 50(2),
159-163.
Samet, J., Zeger, S., Kelsall, J., Xu, J., & Kalkstein, L. (1998). Does Weather Confound or
Modify the Association of Particulate Air Pollution with Mortality?: An Analysis of
the Philadelphia Data, 1973-1980. Environmental Research, 77(1), 9-19.
Samet, J. M., Dominici, F., Zeger, S. L., Schwartz, J., & Dockery, D. W. (2000). The
National Morbidity, Mortality, and Air Pollution Study. Part I: Methods and
methodologic issues. Research report (Health Effects Institute) 94, 5-14; discussion
75-84.
Samet, J. M., Zeger, S. L., Dominici, F., Curriero, F., Coursac, I., Dockery, D. W., et al.
(2000). The National Morbidity, Mortality, and Air Pollution Study. Part II:
Morbidity and mortality from air pollution in the United States. Research report
(Health Effects Institute), 94(Pt 2), 5-70; discussion 71-79.
Samoli, E., Schwartz, J., Wojtyniak, B., Touloumi, G., Spix, C., Balducci, F., et al. (2001).
Investigation Regional Differences in Short-Term Effects of Air Pollution on Daily
Mortality in the APHEA Project: A Sensitivity Analysis for Controlling Long-Term
Trends and Seasonality. Environmental Health Perspectives, 109(4), 349-353.
Sartor, F., Snacken, R., Demuth, C., & Walckiers, D. (1995). Temperature, Ambient Ozone
Levels, and Mortality during Summer, 1994, in Belgium. Environmental Research,
70(2), 105-113.
231
Schneider, A., Schuh, A., Maetzel, F. K., Ruckerl, R., Breitner, S., & Peters, A. (2008).
Weather-induced ischemia and arrhythmia in patients undergoing cardiac
rehabilitation: another difference between men and women. International Journal of
Biometeorology, 52(6), 535-547.
Schwartz, J. (2000). Assessing Confounding, Effect Modification, and Thresholds in the
Association between Ambient Particles and Daily Deaths. Environmental Health
Perspectives, 108(6), 563-568.
Schwartz, J. (2004). The effects of particulate air pollution on daily deaths: a multi-city case
crossover analysis. Occupational and Environmental Medicine, 61(12), 956-961.
Schwartz, J. (2005). Who is sensitive to extremes of temperature?: A case-only analysis.
Epidemiology, 16(1), 67-72.
Schwartz, J., Ballester, F., Saez, M., P茅 rez-Hoyos, S., Bellido, J., Cambra, K., et al. (2001).
The Concentration--Response Relation between Air Pollution and Daily Deaths.
Environmental Health Perspectives, 109(10), 1001-1006.
Semenza, J. C., Rubin, C. H., Falter, K. H., Selanikio, J. D., Flanders, W. D., Howe, H. L., et
al. (1996). Heat-related deaths during the July 1995 heat wave in Chicago. New
England Journal Of Medicine, 335(2), 84-90.
Sessler, D. (2009). Thermoregulatory defense mechanisms. Critical care medicine, 37(7),
S203-210.
Shaddick, G., Lee, D., Zidek, J. V., & Salway, R. (2008). Estimating exposure response
functions using ambient pollution concentrations. The Annals of Applied Statistics,
2(4), 1249-1270.
Shah, S., & Peacock, J. (1999). Deprivation and excess winter mortality. Journal of
Epidemiology and Community Health, 53(8), 499-502.
232
Sheridan, S. C., & Kalkstein, A. J. Seasonal variability in heat-related mortality across the
United States. Natural Hazards, 1-15.
Smargiassi, A., Goldberg, M. S., Plante, C., Fournier, M., Baudouin, Y., & Kosatsky, T.
(2009). Variation of daily warm season mortality as a function of micro-urban heat
islands. Journal of Epidemiology and Community Health, 63(8), 659-664.
Smith, W. R., Coyne, P., Smith, V. S., & Mercier, B. (2003). Temperature changes,
temperature extremes, and their relationship to emergency department visits and
hospitalizations for sickle cell crisis. Pain Management Nursing, 4(3), 106-111.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian
measures of model complexity and fit. Journal of the Royal Statistical Society: Series
B (Statistical Methodology), 64(4), 583-639.
Stafoggia, M., Forastiere, F., Agostini, D., Biggeri, A., Bisanti, L., Cadum, E., et al. (2006).
Vulnerability to heat-related mortality: a multicity, population-based, case-crossover
analysis. Epidemiology, 17(3), 315-323.
Stafoggia, M., Forastiere, F., Agostini, D., Biggeri, A., Bisanti, L., Cadum, E., et al. (2006).
Vulnerability to heat-related mortality: A multicity, population-based, case-crossover
analysis. Epidemiology, 17(3), 315-323.
Stafoggia, M., Forastiere, F., Michelozzi, P., & Perucci, C. A. (2009). Summer temperature-
related mortality: effect modification by previous winter mortality. Epidemiology,
20(4), 575-583.
Stafoggia, M., Schwartz, J., Forastiere, F., & Perucci, C. A. (2008). Does temperature modify
the association between air pollution and mortality? A multicity case-crossover
analysis in Italy. American Journal of Epidemiology, 167(12), 1476-1485.
233
Stewart, S., McIntyre, K., Capewell, S., & McMurray, J. J. V. (2002). Heart failure in a cold
climate 1: Seasonal variation in heart failure-related morbidity and mortality. Journal
of the American College of Cardiology, 39(5), 760-766.
Terence W. Anderson, B., B CH, & Rochard, C. (1979). Cold snap, snowfall and sudden
death from ischemic heart disease. Canadian Medical Association Journal, 121,
1580-1583.
Tianjin Statistic Bureau (Producer). (2004) Tianjin Statistical Bulletin. retrieved from
http://www.stats-tj.gov.cn/Article/tjgb/stjgb/200612/5370.html
Tianjin Statistic Bureau. (2005). Tianjin Statistical Bulletin.
Tsai, S. S., Chen, C. C., Hsieh, H. J., Chang, C. C., & Yang, C. Y. (2006). Air Pollution and
Postneonatal Mortality in a Tropical City: Kaohsiung, Taiwan. Inhalation Toxicology,
18(3), 185-189.
United Nations. (2006). World urbanization prospects: the 2005 revision: United Nations.
Ustrnul, Z., & Czekierda, D. (2005). Application of GIS for the development of
climatological air temperature maps: an example from Poland. Meteorological
applications, 12(1), 43-50.
Valleron, A. J., & Boumendil, A. (2004). Epidemiology and heat waves: analysis of the 2003
episode in France. Comptes Rendus Biologies, 327(12), 1125-1141.
Vandentorren, S., Bretin, P., Zeghnoun, A., Mandereau-Bruno, L., Croisier, A., Cochet, C., et
al. (2006). August 2003 heat wave in France: risk factors for death of elderly people
living at home. The European Journal of Public Health, 16(6), 583-591.
Vandentorren, S., Suzan, F., Medina, S., Pascal, M., Maulpoix, A., Cohen, J. C., et al. (2004).
Mortality in 13 French cities during the August 2003 heat wave. American Journal of
Public Health, 94(9), 1518-1520.
234
Vaneckova, P., Beggs, P. J., de Dear, R. J., & McCracken, K. W. (2007). The Effect of
Temperature on Morbidity and Mortality in Sydney, Australia, Between 1993-2005.
Epidemiology, 18(5), S198.
Vaneckova, P., Beggs, P. J., de Dear, R. J., & McCracken, K. W. (2008). Effect of
temperature on mortality during the six warmer months in Sydney, Australia, between
1993 and 2004. Environmental Research, 108(3), 361-369.
Vaneckova, P., Beggs, P. J., & Jacobson, C. R. (2010). Spatial analysis of heat-related
mortality among the elderly between 1993 and 2004 in Sydney, Australia. Social
Science Medicine, 70(2), 293-304.
Voelker, R. (1995). Probe of heat wave deaths under way. Journal of the American Medical
Association, 274(8), 595-596.
Wang, X. Y., Barnett, A. G., Hu, W., & Tong, S. (2009). Temperature variation and
emergency hospital admissions for stroke in Brisbane, Australia, 1996-2005.
International Journal of Biometeorology, 53(6), 535-541.
Wartenberg, D., Uchrin, C., & Coogan, P. (1991). Estimating exposure using kriging: a
simulation study. Environmental Health Perspectives, 94, 75-82.
Watts, A. J. (1972). Hypothermia in the aged: A study of the role of cold-sensitivity.
Environmental Research, 5(1), 119-126.
Weisskopf, M. G., Anderson, H. A., Foldy, S., Hanrahan, L. P., Blair, K., Torok, T. J., et al.
(2002). Heat wave morbidity and mortality, Milwaukee, Wis, 1999 vs 1995: an
improved response? American Journal of Public Health, 92(5), 830-833.
Whitworth, K. W., Symanski, E., Lai, D., & Coker, A. L. (2011). Kriged and modeled
ambient air levels of benzene in an urban environment: an exposure assessment study.
Environmental Health, 10(1), 21.
235
WHO. (2008). Protecting health from climate change: World Health Day 2008. Geneva:
World Health Organization.
WHO. (2009). Improving public health responses to extreme weather heat-waves EuroHEAT.
Copenhagen Ø, Denmark: WHO Regional Office for Europe.
WHO/WMO/UNEP. (1996). Climate and health: The potential impacts of climate change.
Geneva, Swizterland.
Wilson, J. (2003). Weather and health division. The Journal of Meterology- International
Weather and health special issue, 28(283).
WHO. (2007). International Statistical Classification of Diseases and Related Health
Problems, 10th Revision, Version for 2007.
http://apps.who.int/classifications/apps/icd/icd10online/
WHO. (2008). Protecting Health from Climate Change - World Health Day 2008.
Wyndham, C. H. (1969). Adaptation to heat and cold. Environmental Research, 2, 442-469.
Wyndham CH, F. S. (1978). Climate and disease. South African Medical Journal, 53(26),
1051-1061.
Yang, C. Y., Chang, C. C., Chuang, H. Y., Tsai, S. S., Wu, T. N., & Ho, C. K. (2004).
Relationship between air pollution and daily mortality in a subtropical city: Taipei,
Taiwan. Environment International, 30(4):519-23.
Yu, W., Mengersen, K., Hu, W., Guo, Y., Pan, X., & Tong, S. (2011). Assessing the
relationship between global warming and mortality: Lag effects of temperature
fluctuations by age and mortality categories. Environmental Pollution, 159(7), 1789-
1793.
Yu, W., Mengersen, K., Wang, X., Ye, X., Guo, Y., Pan, X., et al. (2011). Daily average
temperature and mortality among the elderly: a meta-analysis and systematic review
of epidemiological evidence. International Journal of Biometeorology, 1-13.
236
Zanobetti, A., & Schwartz, J. (2008). Temperature and mortality in nine US cities.
Epidemiology, 19(4), 563-570.
Zanobetti, A., Schwartz, J., Samoli, E., Gryparis, A., Touloumi, G., Peacock, J., et al. (2003).
The Temporal Pattern of Respiratory and Heart Disease Mortality in Response to Air
Pollution. Environmental Health Perspectives, 111(9), 1188-1193.
Zeger, S. L. (1988). A regression model for time series of counts. Biometrika, 75(4), 621-629.
Zhang, K., Oswald, E. M., Brown, D. G., Brines, S. J., Gronlund, C. J., White-Newsome, J.
L., et al. (2011). Geostatistical exploration of spatial variation of summertime
temperatures in the Detroit metropolitan region. Environmental Research, 111(8),
1046-1053.