Multilevel and Clustering Analysis of Health Outcomes in Small Areas

This paper considers models of the variable incidence of health outcomes in geographical areas and of variable regression effects of socio-economic variables on such outcomes. It adopts a Bayesian approach to variation in relative risk and regression effects, and assesses different prior specifications of risk (e.g. a latent class structure versus a spatially correlated structure). Implications are considered for smoothing and mapping rare health outcomes. The analysis is for electoral wards in London, with the health-deprivation link forming the focus for regression effects. Implications for inferences about risk factors and for health-need ratings (before and after smoothing) are also considered.