Small area spatial modelling and mapping of health
Jun 27, 2015
The research described in this presentation involves small area estimation (SAE) statistical techniques applied to social and spatial health inequalities in developed countries to get around data gaps by performing spatial micro-simulation for synthetic populations. For health, the added value is the calculation of prevalence probabilities and projections based on health and census data. By Dianna Smith, QMUL
- 1. Small area spatial modelling &mapping of healthDianna Smith 11 March 2013
2. Research streams Social and spatial health inequalities inUS, UK, NZ Measuring demand/access to services:ArcGIS Modelling demand/access to services: spatialinteraction Filling in missing data: spatial microsimulationfor synthetic populations 3. Mathur, R., Noble, D., Smith, D., Greenhalgh, T., & Robson, J. (2012). Quantifying the risk of type 2 diabetes in EastLondon using the QDScore: a cross-sectional analysis. British Journal of General Practice, 62(603), e663-e670. The authors 4. Spatial interaction (gravity)m m m Identify the relativem S ij Oi Ai W j exp( d ij ) mattractiveness of adestination forsegments of thepopulation Calculate demand forservices Or location-allocationfor mobile services 5. Linking datasets and building estimates If there is a Health data Censusrepresentative(non-spatial) (spatial)population survey thatage agehas variables insex sexcommon with a ethnicitycensus, these variables ethnicitycan be used to calculatesocial gradesocial gradethe probabilities of each BMI (obesity)person in the surveydiabetesliving in each area. If the survey includesdata on obesity anddiabetes for eachperson, every time aperson is assigned toan area the prevalence 6. A good model The linking characteristics must be predictive of the health outcome for that population In Scotland, ethnicity is irrelevant for BMI estimation Helps if there is some scale where prevalence is known Aggregate the estimated data to check accuracy Cluster the areas with populations which are most similar in terms of predictive characteristics for the behaviour/outcome Global smoothing algorithm can create error in unusual populations 7. Area characteristics: clustersClustermembership 1:High % in DE, over 50 2: Low % DE, over 50 3: Low % DE, high % over 50 4: High % non-white, % DE, % under 50 5: High % DE, % under 500 3 6 12Kilometers 8. Estimated DiabetesLS29 LS21BD20 LS19 LS20BD21LS19BD16LS17 BD17LS16 LS18BD10 BD22 BD18 LS6 BD15LS5BD9BD2LS13LS4BD8 LS3 LS28 BD3 BD13BD1 LS12BD7 BD14BD5LS11 Diabetes SIRBD4 (Optimal model)BD675.5 - 92.8 BD11 LS2792.9 - 100.0 BD12100.1 - 150.0150.1 - 250.0BD190 1 246 8250.1 - 350.1Kilometers 9. Potential applications Building on incomplete datasets from currentrecords: linking data from multiple sources(building on Tatem et al 2012, Mappingpopulations at risk Estimating outcomes of interventions: scenariomodelling 10. Difficulties All very quantitative models Minimal applications (of the models) in LMICs, especially in terms of health outcomes 11. References Mathur, R., Noble, D., Smith, D., Greenhalgh, T., &Robson, J. (2012). Quantifying the risk of type 2diabetes in East London using the QDScore: a cross-sectional analysis. British Journal of GeneralPractice, 62(603), e663-e670. Smith, D. M., Pearce, J. R., & Harland, K. (2011). Cana deterministic spatial microsimulation model providereliable small-area estimates of health behaviours?An example of smoking prevalence in New Zealand.Health & Place, 17(2), 618-624. Tatem, A., Adamo, S., Bharti, N., Burgert, C., Castro,M., Dorelien, A., et al. (2012). Mapping populations atrisk: improving spatial demographic data for infectiousdisease modeling and metric derivation. PopulationHealth Metrics, 10(1), 8.
