Statistical Methods for Precision Environmental Health with Mixture Exposures

Project Summary/Abstract Precision environmental health focuses on individual risk assessment to inform targeted disease prevention strate­ gies. Identifying individuals with increased sensitivity to environmental exposures is especially challenging with mixture exposures. The health effects of exposure to mixtures are likely to depend on the composition of the mixture, characteristics specific to the espoused individual including individual­ and neighborhood­level factors, and the developmental stage at which an individual is exposed. We propose to develop statistical methods for precision environmental health with mixture exposures. The proposed methods will estimate mixture­exposure­ response relationships that are individualized based on multiple candidate modifying factors. The framework we develop will allow for data­driven discovery of novel combinations of individual­ and neighborhood­level factors that define susceptible subgroups. We will address three specific data settings. In Aim 1 we propose a general framework for effect heterogeneity using established mixture methods including Bayesian multiple index models. This will include heterogeneous versions of Bayesian kernel machine regression and linear index models. In Aim 2 we develop methods to identify critical windows of susceptibility to mixtures that are assessed longitudinally. The methods will allow for identification of individualized windows of susceptibility to a mixture and estimation of individualized mixture­exposure­time­response functions. In Aim 3 we develop heterogeneous mixture methods for multiple outcomes. The multiple outcome methods will apply to trajectories defined by repeated measures of common endpoint or pathway as well as shared information across multiple related endpoints, such as multiple measures of a common pathway. In Aim 4 we will develop software to implement the methods, along with vignettes and tutorials. We will use the methods developed to analyze air pollution mixtures in a large administrative birth cohort and in a Northeastern United States longitudinal perinatal cohort drawing from multiple source populations. We will estimate individualized mixture­exposure­response functions for birth weight and multiple neurodevelop­ mental endpoints assessed at multiple times. The methods we develop will allow for new avenues of precision environmental health to better identify individuals at increased risk of adverse effects of the environment, which will better inform targeted disease prevention strategies.