Cardiovascular risks of plastic-associated chemicals: Integrating mixture modeling and SHAP-explained machine...

Plastic-associated chemicals (PACs) are widely detected environmental contaminants, yet their cardiovascular relevance in the context of concurrent exposure remains unclear. We evaluated individual and mixture-level associations of urinary PAC biomarkers with cardiovascular disease (CVD) status among U.S. adults.

We analyzed 3831 adults from the National Health and Nutrition Examination Survey (NHANES) 2005-2016. CVD was defined using questionnaire-reported physician diagnoses of congestive heart failure, coronary heart disease, angina pectoris, myocardial infarction, or stroke. Eleven urinary PAC biomarkers were assessed using adjusted single-pollutant models, restricted cubic splines, Weighted Quantile Sum (WQS) regression, Bayesian Kernel Machine Regression (BKMR), linear discriminant analysis (LDA) with SHAP attribution, and lipid mediation analyses.

Most PACs showed limited linear associations with CVD after multivariable adjustment, whereas spline analyses suggested non-linear patterns for selected biomarkers. WQS regression indicated a positive mixture association in the primary split (OR = 1.647, 95% CI: 1.235-2.196), with larger weights for ΣDEHP, MiBP, BPA, MCPP, and MBzP. BKMR highlighted triclosan (PIP = 0.6976) and BPA (PIP = 0.4440) as the leading mixture signals. LDA showed good discrimination in the validation set (AUC = 0.813, 95% CI: 0.780-0.847), and SHAP identified age, hypertension, BMI, and selected PACs as prominent contributors to model output. In lipid mediation analyses, the estimated HDL-C-related mediation proportions were 20.8% for BPA, 39.6% for ΣDEHP, and 14.1% for PrP, while the triglyceride-related mediation proportion was 9.7% for the PrP-CVD association.

Urinary PAC biomarkers and mixture metrics were associated with CVD status among U.S. adults. Prospective studies with repeated exposure assessment and validated cardiovascular outcomes are needed.