Combined Effects of Metals and PFAS Exposure on Prevalent Diabetes
TL;DR
Limiting to directly measured exposures, incorporating the full NHANES design in the primary regression, and triangulating across complementary mixture frameworks provide a more rigorous platform than prior single-pollutant or design-naive approaches.
Abstract
Human per- and polyfluoroalkyl substances (PFAS) and metal exposures occur as mixtures, but most studies evaluate them independently. Using data from adults in the 2017–2018 National Health and Nutrition Examination Survey (NHANES) with directly measured PFAS and blood metals (N = 1648), we assessed joint associations of perfluorooctanoic acid (PFOA), perfluorooctanesulfonic acid (PFOS), lead, cadmium, and mercury with prevalent diabetes via survey-weighted logistic regression, Weighted Quantile Sum (WQS) regression, quantile g-computation, and design-aware and naïve Bayesian Kernel Machine Regression (BKMR). Survey-weighted diabetes prevalence was 11.4%. Lead and PFOS emerged as the dominant contributors to the exposure mixture in the BKMR model, with posterior inclusion probabilities of 1.00 and 0.997, respectively. In the WQS model, PFOS and PFOA were the primary drivers of the positive mixture index, consistent with their proposed roles in endocrine disruption and insulin resistance. BKMR exposure-response functions were non-monotonic, indicating that lead and cadmium associations with diabetes are nonlinear rather than uniformly directional across the exposure range, a structure that conventional logistic regression cannot capture. Overall, restricting to directly measured exposures, incorporating the full NHANES design in the primary regression, and triangulating across complementary mixture frameworks provide a more rigorous platform than prior single-pollutant or design-naive approaches.