Integrating Novel Cardiometabolic Biomarkers With Traditional Risk Factors in SHAP-Based Machine Learning Model to Stratify Cardiovascular Disease Risk

Background: Cardiovascular disease (CVD) remains the leading cause of mortality worldwide.However, conventional risk models often fail to capture its underlying metabolic and inflammatory mechanisms.In this study, we aim to develop and validate an explainable SHapley Additive exPlanations (SHAP)-based machine learning (ML) framework that integrates novel cardiometabolic biomarkers-Visceral Adiposity Index (VAI), Cardiometabolic Index (CMI), and Aggregate Index of Systemic Inflammation (AISI)-with traditional risk factors for cardiovascular risk stratification.Methods: To classify prevalent CVD and perform cardiovascular risk stratification in our study of 7,231 participants from the 2007-2018 cycles of the National Health and Nutrition Examination Survey (NHANES), we incorporated 17 variables spanning the four domains of demographics, lifestyle, comorbidities, and metabolic biomarkers.Five ML models-logistic regression, random forest, gradient boosting, XGBoost, and a soft-voting ensemble-were trained using 5-fold cross-validation.We applied SHAP analysis to facilitate feature-level interpretability.Results: The performances of the logistic regression and the soft-voting ensemble models were comparable (test area under the curve = 0.837; 95% confidence interval [CI], 0.805-0.869).Logistic Regression was selected as the final model based on its high sensitivity (0.786 vs. 0.443) and simple structure.Unlike VAI and AISI, CMI was independently associated with CVD (odds ratio, 1.807; 95% CI, 1.163-2.809;P = 0.009) in surveyweighted regression analyses of the fully adjusted model.The SHAP analysis identified age, hypertension, and CMI as the most influential risk factors.Among the novel biomarkers, CMI showed the strongest contribution, VAI demonstrated a moderate contribution, and AISI showed minimal impact on model outputs.Conclusion: By integrating VAI, CMI, and AISI into the SHAP-based ML framework, it improves the interpretability of the framework with acceptable discriminative performance for CVD risk stratification.In addition to the dominant traditional risk factors, CMI contributes cardiometabolic information to support its role as a complementary marker.Although longitudinal validation is required, our findings highlight the potential of explainable ML to enhance transparent risk assessment.