Bayesian Adjustment for Unmeasured Environmental Confounding

Studies in environmental epidemiology are often concerned with understanding the health effects of environmental exposure in various forms. Because these studies are, by nature, observational, it is often difficult to make valid statistical conclusions. Additional complications arise from the presence of confounding variables, which relate to both the exposure and health effect, and hence complicate the relationship. Traditionally these confounders are controlled for by including them as explanatory variables in a statistical model. Bias-free conclusions become much more difficult, however, when some confounders are unmeasured or inadequately measured. A wide variety of environmental epidemiology studies have suffered from this problem, including, for instance, estimating the association between air pollution and mortality, between magnesium levels in drinking water and mortality from acute myocardial infarction, and between ethnicity, income and limiting long-term illness. The focus of Luke Bornn’s research is the development of a coherent, unified framework for modeling environmental risk exposure in the presence of unmeasured confounding. His model will account for spatial dependencies between adjacent geographical groups as well as other factors that are important for these studies, such as ecological bias and pure specification bias. His hypothesis is that by accounting for spatial dependence and unmeasured confounding under a comprehensive and unified framework, the risk estimates will more accurately estimate the true exposure risk and provide more appropriate estimates of the corresponding uncertainty. By developing a model through simulations, analytic results and application to real data sets, Mr. Bornn’s research will create a model that is both practical and useable for environmental epidemiology practitioners.