Bayesian partitioning for mapping disease risk using a matched case-control approach to confounding.

Disease maps are useful for exploring geographical heterogeneity in health outcomes. Typically interest lies in unearthing atypical regions after adjusting for known confounders. This paper presents a Bayesian partitioning approach for analyses when individual-level matching has been used to control confounding. The model makes few assumptions about the surface form and, in particular, permits discontinuity. The specification is inherently parsimonious and posterior sampling permits direct assessment of surface uncertainty; additional unmatched covariates can also be incorporated. The method is used to investigate spatial variation in perinatal mortality in the North-West Thames region, England.