Department Seminar by Glen McGee

Please Note: This seminar will be given online.

Bayesian Multiple Index Models for Multi-Pollutant Mixtures

An important goal of environmental health research is to assess the risk posed by mixtures of pollutants. Two popular regression approaches for mixtures analyses are response surface methods and exposure index methods. Response surface methods, like Bayesian kernel machine regression (BKMR), estimate high dimensional surfaces and are thus highly flexible but difficult to interpret outside low-dimensional settings. In contrast, exposure index methods are highly interpretable---decomposing a linear model into an overall mixture effect and individual index weights---but permit only linear relationships and no interactions. We propose a Bayesian multiple index model framework that allows non-linear relationships of exposure indices as well as non-additive interactions between them---all while simplifying interpretation by reducing the dimensionality of the problem and estimating index weights. The proposed framework contains both BKMR as well as linear index approaches as special cases---thus unifying two analysis strategies previously considered distinct. In application to a motivating study of a mixture of 18 organic pollutants, we show how the proposed framework bridges the gap between flexible response surface methods and strict linear index methods---allowing one to select an analysis from a spectrum of models varying in flexibility and interpretability.