Statistical Methods for Estimating the Effects of Multi-Pollutant Exposures in Children's Health Research

Abstract

We develop statistical strategies to explore how time-varying exposures to heavy metal mixtures affects cognition and cognitive trajectories in children. In chapter 1, we develop a Bayesian model, called Lagged Kernel Machine Regression (LKMR), to identify time windows of susceptibility to exposures of metal mixtures. In chapter 2, we develop a Mean Field Variational Bayesian (MFVB) inference procedure for LKMR. We demonstrate large computational gains under MFVB as opposed to Markov chain Monte Carlo (MCMC) inference for LKMR, which allows for the analysis of large datasets while maintaining accuracy. In chapter 3, we present a Bayesian hierarchical model, called Bayesian Varying Coefficient Kernel Machine Regression, to investigate the impact of exposure to heavy metal mixtures on cognitive growth trajectories in children. Simulation studies demonstrate the effectiveness of these methods, and the methods are used to analyze data from two prospective birth cohort studies in Mexico City.