Abstract
BACKGROUND: Stroke is a leading cause of death and long-term disability worldwide, particularly among the elderly. Biological age, as a comprehensive indicator of health status during the aging process, can more accurately reflect an individual’s health condition. This study aims to explore the relationship between biological age and stroke risk, and to evaluate the effectiveness of machine learning methods in stroke prediction when incorporating biological age. METHODS: This study utilized the 2011–2015 China Health and Wealth Longitudinal Study (CHARLS) data, including 8,247 adults aged 45 years and older. Multivariate logistic regression models were employed to analyze the relationship between biological age and stroke risk, while Restricted Cubic Spline (RCS) regression was used to examine nonlinear associations between biological age and stroke risk. To further optimize the model, Least Absolute Shrinkage and Selection Operator (LASSO) regression was employed for feature selection, identifying characteristics most strongly associated with stroke risk. To assess the contribution of biological age to stroke prediction, an eXtreme Gradient Boosting(XGBoost) machine learning model was constructed, combined with SHapley Additive exPlanations(SHAP) interpretation to analyze feature importance. Additionally, subgroup analyses explored the moderating effects of comorbid conditions such as hypertension and diabetes on the relationship between biological age and stroke risk. RESULTS: In the fully adjusted model, each additional year of biological age was significantly associated with increased stroke risk (OR = 1.51, 95% CI 1.37–1.68, P < 0.001). RCS analysis revealed a significant linear relationship between biological age and stroke risk (nonlinear P = 0.689). The XGBoost model achieved an Area under the receiver operating characteristic curve(AUROC) of 0.946 on the test set, outperforming other traditional regression models. SHAP analysis further indicated that biological age held greater importance in the model compared to other features. Subgroup analysis revealed no significant interaction between biological age and stroke risk across subgroups (P for interaction > 0.05). CONCLUSION: Biological age plays a significant role in stroke risk assessment, and its integration with machine learning methods can effectively enhance the accuracy of stroke prediction. Future research should further optimize models and expand sample sizes to improve the effectiveness of early stroke screening and intervention.