Integrating social, lifestyle, and genetic profiles to predict epigenetic age acceleration in older adults

Abstract Epigenetic age acceleration (EAA) is linked to social, lifestyle, and behavioural exposures, yet their relative importance when considered together remains unclear. To address this gap, we take an integrative, multi-domain approach to compare the relative importance of demographic, socioeconomic, psychosocial, health-behavioural, clinical, and genetic factors for EAA prediction. In this cross-sectional study, we analysed 4018 Health and Retirement Study participants and used LASSO, Random Forest, and XGBoost to predict EAA across different epigenetic clocks. To identify key predictors and quantify domain contributions, we fitted all-predictor models and trained domain-only, ablation, and permutation models. In all-predictor models, GrimAge showed the highest predictability (5-fold CV R² = 0.41), followed by DunedinPoAm (R² = 0.20), while Hannum, Horvath, and PhenoAge showed little signal. Overall, predictive signal for GrimAge and DunedinPoAm EAA was concentrated in health behaviours, largely reflecting smoking-related signal, with smaller contributions from physical activity, gender, household mean income, and African American ancestry.