Spatial and Tensor Methods for Estimating the Health Effects of PFAS Mixtures - PROJECT SUMMARY Perfluoroalkyl and Polyfluoroalkyl Substances (PFAS) are synthetic compounds found in the environment and linked to adverse health outcomes. However, understanding the causal effects of PFAS mixtures on various health outcomes remains limited. This project addresses this critical knowledge gap by employing innovative spatial and high-dimensional causal inference methods to enhance our understanding of PFAS health effects. By bringing scientific rigor to observational studies, the project aims to overcome the challenges posed by spatial dependence, various sources of bias, and high dimensionality in PFAS studies. This project leverages two valuable health outcomes data sources relating PFAS in drinking water: (i) a nationwide cohort of Medicare beneficiary data and (ii) a North Carolina electronic health records cohort. These cohorts have been linked to PFAS concentrations in the ground water sources and in the tap water used for drinking. These data of health outcomes contain individual-level information and provide comprehensive resources for enhancing the understanding the effects of PFAS on population health. Based on previous epidemiological analysis of associations with PFAS exposure in the two cohorts we have identified three statistical challenges. First, spatially- dependent outcomes violate standard assumptions of causal inference, necessitating a rethinking of the potential outcomes framework. Second, various sources of bias are present such as preferential sampling and non- random actualization of PFAS exposures. Third, the high dimensionality of the PFAS chemical mixture and health responses introduces computational and stability issues, requiring new dimension-reduction methods. By developing a suite of methods tailored to address these challenges, this project marks a significant advancement in the field of epidemiological research. The proposed methods will enable accurate estimation of causal health effects in observational spatial studies, providing refined insights into the impact of PFAS mixtures on a range of health outcomes. To tackle these challenges, the project outlines three specific aims. Aim 1 focuses on developing spatial causal inference methods for preferentially-sampled data, accounting for bias in sample-location selection. Aim 2 devel- ops spatial causal inference methods for multiple outcomes and exposures that use tensor regression model to explain variation in the health effects by exposure, response, and spatial resolution. Aim 3 develops a structured statistical protocol for spatial causal inference analysis of multiple exposures and outcomes, with application to health effects of PFAS at the state and national levels. These aims not only address the specific challenges in PFAS studies but also have broader implications for epidemiological research dealing with spatial dependence and high-dimensional data. Through the development of these innovative methods, this project aims to transform the analysis of PFAS studies, contributing to advancements in the field and improving our understanding of the health effects of PFAS exposure.
