A two-step Bayesian clustering approach to relate PFAS mixture profiles and dietary patterns in early pregnancy.

Diet is a major source of exposure to per- and polyfluoroalkyl substances (PFAS). However, few studies have investigated dietary patterns in relation to differences observed in the PFAS exposure patterns. We conducted a cross-sectional analysis within the Project Viva cohort of 1383 pregnant women enrolled at their first prenatal visit (1999-2002), who completed a validated food frequency questionnaire (158 items) and had plasma concentrations of six PFAS. We implemented a Bayesian repulsive Gaussian mixture model (BRGM) to identify PFAS exposure patterns. These patterns defined our PFAS subpopulations for a subsequent analysis, where we derived dietary patterns that accounted for differences amongst these subpopulations. Using a robust profile clustering (RPC) model, we detected dietary patterns both across the entire cohort and within specific PFAS subpopulations. The BRGM model identified six PFAS subpopulations with distinct PFAS exposure levels. Across the entire cohort, we identified six dietary patterns, plus one additional local dietary pattern per PFAS subpopulation identified by the RPC model. Notably, we observed clear consumption differences between high- and low-PFAS exposure subpopulations. Subpopulations with higher PFAS concentrations showed greater consumption of sugar-sweetened beverages, packaged and processed condiments, decaf coffee (during pregnancy), skim milk, and poultry, while those with lower PFAS concentrations tended to show greater consumption of vegetables, fruits, regular beer (before pregnancy), hot cereal, and sweet baked goods. Our approaches provided novel insights into the relationship between prenatal exposure to PFAS and dietary intake by identifying dietary patterns unique to different PFAS subpopulations.