PFAS exposure and hypertension in US adults: A multi-method, cross-sectional analysis of NHANES 2005 to 2014.

Per- and polyfluoroalkyl substances (PFAS) are persistent environmental contaminants, but evidence linking PFAS exposure to hypertension remains limited and inconsistent. We analyzed 5 contiguous National Health and Nutrition Examination Survey cycles (2005-2014), including 6644 adults. Associations with hypertension were estimated by multivariable logistic regression. Restricted cubic splines and piecewise threshold models assessed nonlinearity. Mixture effects were examined using weighted quantile sum and quantile g-computation (Q-gcomp). Eight machine-learning classifiers were compared using cross-validated AUC. XGBoost was selected as the final model and interpreted with SHAP. In continuous multivariable models, PFOA (OR, 1.04, 95% CI, 1.00-1.08, P = .036) and PFOS (OR, 1.04, 95% CI, 1.01-1.07, P = .007) were positively associated with hypertension; Me-PFOSA-AcOH showed a marginal positive association (OR, 1.07, 95% CI, 1.00-1.14, P = .045). Quartile analyses showed elevated odds for PFOA and for PFOS. Spline and threshold models identified compound-specific inflection points with differential associations below versus above cutpoints. Weighted quantile sum positive-direction index was modestly associated with hypertension (OR, 1.13, 95% CI, 1.03-1.26, P = .01) with PFUA contributing most; q-gcomp results were concordant. XGBoost achieved the highest discrimination (AUC = 0.814); SHAP ranked PFOS, PFNA and PFOA as leading predictors by mean SHAP. Specific PFAS were associated with prevalent hypertension, and PFAS mixtures showed a modest joint effect. Machine-learning and SHAP analysis highlighted specific congeners driving predictive signal.