Interpretable Cardiovascular Disease Risk Prediction for Primary Care Screening Using XGBoost and SHAP

Abstract

Cardiovascular disease (CVD) is currently the leading cause of death among the Chinese population, and the early and accurate identification of high-risk individuals is central to primary prevention of chronic diseases. Traditional risk assessment models often rely on linear assumptions, making it difficult to capture nonlinear relationships and higher-order interaction effects among physiological indicators, which limits their generalization ability in primary screening settings.Based on over 69,000 routine physical examination samples, this study established a standardized preprocessing workflow with strict data privacy protection. Hyperparameter tuning was performed using the Optuna Bayesian optimization algorithm combined with five-fold cross-validation to train an optimized XGBoost risk prediction model. To enhance the model ' s clinical credibility, the SHapley Additive Explanations (SHAP) explainability framework was introduced and combined with feature importance ranking to enable dual-dimensional attribution analysis at both the global and individual levels.Experimental results show that the optimized model achieved an Area Under the Curve (AUC) of 0.7994 on the independent test set, representing a 1.40 percentage point improvement over the logistic regression baseline, with the Brier score reduced to 0.1811. This model enables quantitative risk assessment relying solely on routine non-invasive health examination indicators, providing transparent and reliable technical support for primary healthcare institutions to conduct low-cost, widespread early CVD screening.