Environmental chemical factors associated with human biological age acceleration: an interpretable machine learning study.

This study utilized an interpretable machine learning (ML) framework to analyze data from 1693 adults in the 2015-2016 National Health and Nutrition Examination Survey (NHANES), investigating the association between 78 environmental chemicals and two biological age acceleration (BA-Accel) measures: KDM-BA-Accel and PhenoAge-Accel. The ML models demonstrated moderate predictive performance, with the best models achieving R² values between 0.20 and 0.27. Including chemical exposure data did not statistically improve model performance compared to baseline models with only demographic and clinical covariates. The analysis identified clinical factors like diabetes and body mass index (BMI) as the most significant predictors of BA-Accel. Concurrently, Shapley Additive Explanations (SHAP) analysis identified blood toluene (B_TOL), a urinary benzene metabolite (U_DHB), and urinary benzophenone-3 (U_BP3) as key chemical contributors. Furthermore, the framework revealed complex nonlinear relationships and potential synergistic interactions between these chemicals and BA-Accel. A sensitivity analysis confirmed a significant interaction between diabetes status and blood toluene. In conclusion, while single cross-sectional measurements of chemical exposures have limited utility for predicting individual aging rates, the applied ML approach provides a method to identify key chemicals within mixed exposures and characterize their complex modes of action, which can inform public health risk management.