Marek - Spatial analyses of health data: From points to models
-
Upload
swenney -
Category
Data & Analytics
-
view
37 -
download
0
description
Transcript of Marek - Spatial analyses of health data: From points to models
www.geoinformatics.upol.cz
Spatial analyses of health Spatial analyses of health data: From points to data: From points to
models models
Lukáš MAREK
www.geoinformatics.upol.cz
Why Geographical Information Systems?
• Advanced methods for spatial analyses
• Spatial statistics
• Exploration of spatial pattern
• Visualization and presentation for non-
geographers (doctors, specialist)
www.geoinformatics.upol.cz
Spatial Analyses of Health Data• Disease mapping
– Visual description of spatial variability of the disease incidence
– Maps of incidence risk, identification of areas with high risk
• Geographic correlation studies– Analysis of associations among the incidence and
environmental factors
• Analyses of spatial pattern– Exploration of spatial and spatio-temporal patterns in data
– Disease clusters, randomness, …
www.geoinformatics.upol.cz
Objectives
• Disease occurence as the event– Space, time, attributes– Spatial (point) pattern
• Spatio-temporal evaluation of Campylobacter infection in the Czech Republic
• Association with environmental / social factors
www.geoinformatics.upol.cz
Tools
www.geoinformatics.upol.cz
Campylobacteriosis
• Campylobacter spp. (C. jejuni)
• Most frequent gastroenteritis in developed countries
• Often foodborne
• Symptoms are similar to salmonella
• Poultry, fresh milk products, sewage, wild animals …
www.geoinformatics.upol.cz
Campylobacteriosis
• Children are by far the most vulnerable group
• Seasonality• Underreported
– 225 cases / 100 000 population
– Havelaar et al. (2013)• 11.3 times more cases in
the Czech Republic• 45.7 times more case in
EU
www.geoinformatics.upol.cz
Data
• EPIDAT database– Mandatory records about infectious diseases and
patients, manually fulfilled– Age, Sex, Date, Profession, Place of residence, infection,
isolation, …– 2008 - 2012– ≈ 100 000 records– (weakly) Anonymized
• Aggregation– Regular square network, municipal districts
• Statistical data– National census 2011– EUROSTAT population grid
www.geoinformatics.upol.cz
Data Privacy
• Health and medical data = private, confidential and sensitive data
• Keeping all available records but prevent their re-identification
• Usefulness of the local scale analysis X privacy protection
• Unlikely to explore the relations on the individual level (and not necessary)
• Availability, accessibility and restrictions
www.geoinformatics.upol.cz
Time
www.geoinformatics.upol.cz
Spatial visualization
www.geoinformatics.upol.cz
Spatio-temporal visualization
www.geoinformatics.upol.cz
Spatio-temporal kriging
• Continuous incidence surface in populated places of the Czech Republic
• Exploration of the phenomenon‘s autocorrelation simultaneously in space and time
• Interpolation of logarithm of standardized incidence into 4 km2 grid
• Ordinary global spatio-temporal kriging
www.geoinformatics.upol.cz
Spatio-temporal kriging
• Exponential model / Metric model
• nugget = 0.15, partial sill = 1.94, range = 14150.46 m and space-time anisotropy = 544.58
www.geoinformatics.upol.cz
Spatio-temporal kriging
• Vysledky• Bud sada obrazku nebo potom
screen video
www.geoinformatics.upol.cz
Spatio-temporal clustering• Visual overview of the space-time pattern
in Google Earth
• Spatial Scan Statistics– Identification of clusters of high and low rate areas
together in the continuous geographical areas and time
• Age/sex stratified cases, age/sex stratified population, coordinates of centroids
• Weekly aggregated data
www.geoinformatics.upol.cz
Spatio-temporal clustering• Retrospective analysis based on Poisson
probability model
• Clusters of maximum size of 3% of population, max. 50% of time or entire time period
• Non-parametric temporal adjustment
• Monte Carlo simulation, p-value < 0.05
• Comparison based on relative risk
www.geoinformatics.upol.cz
Spatial clustering
www.geoinformatics.upol.cz
Spatio-temporal clustering
www.geoinformatics.upol.cz
Spatio-temporal clustering
Cluster Type Time period Region Count Observed Expected Relative risk Population
1* H 2008/01/01 – 2012/12/31 Ostrava 31 5975 2861 2.16 292,978
2 H 2008/01/01 – 2012/12/31 North Wallachia - Lachia 70 5414 2788 2.00 277,236
3 H 2008/01/01 – 2012/12/31 Silesia 16 4773 2534 1.93 256,657
4 H 2008/01/01 – 2012/12/31 Prague - centre 1 1006 245 4.13 29,948
5 H 2008/05/13 – 2010/11/01 South Moravia 167 2274 1432 1.60 292,885
6 H 2008/01/01 – 2012/12/31 Dražíč 1 72 2 41.92 214
7 H 2008/01/01 – 2012/12/31 Brno - centre 19 3951 2590 1.55 271,742
8 H 2008/01/01 – 2012/12/31 Opava 37 1714 877 1.97 87,203
9 H 2008/01/01 – 2012/12/31 Haná 66 3828 2526 1.54 256,721
10 H 2009/04/14 – 2011/09/05 South Wallachia 90 1596 932 1.72 196,522
11 H 2010/01/12 – 2010/02/22 Budweis 60 194 36 5.41 157,425
12 H 2008/01/01 – 2012/12/31 Benešov 15 640 313 2.05 31,115
13 H 2010/04/06 – 2010/10/04 Brno - surroundings 224 568 286 1.99 284,346
14 H 2011/05/03 – 2011/11/14 Pilsen 22 394 201 1.96 197,263
www.geoinformatics.upol.cz
Model
• Models help to understand reality– Although they do not have to fully describe its variation
• Models help to clarify significance of most likely associated factors
• Models help to describe the future using the data from the past
• Models help to identify vulnerable areas
www.geoinformatics.upol.czArsenault, J. (2010): Épidémiologie spatiale de la campylobactériose au Québec.
www.geoinformatics.upol.cz
www.geoinformatics.upol.cz
Conclusions
• Space – time aggregation and visualization for visual analytics
• Continuous incidence surface describing spatial and temporal progress of the disease
• Scan statistics identifying high and low rate clusters as well as their temporal support
• Visualization in Google Earth
www.geoinformatics.upol.cz
Problems / Challenges
• Geocoding• Aggregation• Age / Population standardization• Neighbourhood estimation• Modifiable areal unit problem• Probability distribution of the disease occurrence• Underlying processes• Under / Overestimation of results leading to
misinterpretation