|
Abstract:
|
A key goal of environmental health research is to assess the risk posed by mixtures of pollutants. As studies of mixtures can be expensive to conduct, it behooves researchers to incorporate available knowledge about mixtures into their analyses. We propose several strategies for incorporating knowledge from toxicology in epidemiological analyses of mixtures. This builds on the recent Bayesian multiple index model, which combines the flexibility of response surface methods—allowing for non-linearity and interactions among groups of exposures—and the interpretability of linear index models—which decompose mixture effects into component contributions via interpretable weights. To incorporate prior knowledge, we: place constraints on weights, structure weights (eg impose partial effect rankings; smoothness over time; etc) via transformations, and adopt novel informative priors based on relative potency factors. When prior knowledge is correct, the proposed informative priors improve inferences, and when it is incorrect, they protect against misspecification suffered by naïve toxicological models. Different strategies may be combined for different indices to suit available information.
|
