Spatial Epidemiology

The spatial distribution of cases of disease often captures the imagination of health researchers and the general public, based primarily on the notion that the observed pattern of incidence or prevalence can provide insight in to the underlying mechanisms driving disease incidence, its progression, and the design and implementation of effective public health responses. Historical examples include John Snow’s famous maps of cholera incidence in London neighborhoods and early maps of yellow fever incidence in relation to features of cities and docks (Walter 2000, Koch 2005). More recent examples include reports of clusters of cancer cases near hazardous waste sites. However, quantifying such hypotheses through statistical inference is a difficult task due to oftentimes subtle signals within multiple layers of noisy, nonindependent, observational data from multiple agencies collected for multiple purposes, typically other than the spatial epidemiology question at hand. As a result, most applications do not have the luxury of a research design optimized for the questions of interest, nor an experimental setting within which to conduct inference controlling for potential confounding factors. For these reasons, spatial epidemiologic studies encounter many complications in addition to those in traditional studies, including some unique to the geographic setting. The field of spatial statistics involves the statistical analysis of observations with associated locations in space. Often these observations do not follow a Gaussian distribution and are not independent, two mainstays in the development of statistical methods. In addition, asymptotic results take on a different flavor depending on whether we consider