Some Epidemiologic Applications of Kriging

Environmental epidemiology is the study of the association between exposure to environmental agents and the occurrence of disease. Exposure data typically are expensive and difficult to collect, and epidemiologists try to make efficient use of all available data. Nonetheless, quantification of exposure is often the weakest link in the investigation of the chain of causation, and exposure assessment is the area most subject to improvement. In this study, I try to improve the use of exposure data by exploring methods which may increase the statistical power of the epidemiologic assessment. Since many environmental exposures are coherent in space (e.g., a plume of polluted air downwind of a factory, a path of polluted water downstream from a discharge point), the application of geostatistics offers a wide range of opportunities for improving exposure models and providing more meaningful epidemiologic assessment. In this paper, I discuss two applications of kriging to epidemiology from a recent investigation of a cancer excess on Cape Cod, Massachusetts. I note at the outset that this study is mainly exploratory and, given the nature of the environmental monitoring data used, substantive results should be interpreted cautiously and viewed as suggestive at best.