Predictive Power of Machine Learning Algorithms in Stratifying Atherosclerotic Risk

Background: Atherosclerotic cardiovascular disease remains the leading cause of morbidity and mortality worldwide, yet current risk prediction models, such as the Framingham Risk Score and ASCVD pooled cohort equations, are limited by population-specific biases, static variables, and poor performance in diverse patient cohorts. Machine learning (ML) algorithms offer a data-driven alternative capable of integrating large-scale clinical, biochemical, imaging, and genetic data to enhance risk stratification. By learning complex, nonlinear relationships between predictors and outcomes, ML has the potential to identify high-risk individuals earlier and with greater accuracy. Methods and Results: This study evaluated the predictive performance of multiple supervised ML models, including random forest, XGBoost, support vector machines, and deep neural networks, using a cohort of 45,000 individuals from a multi-ethnic biobank with longitudinal follow-up. Input features included traditional cardiovascular risk factors, laboratory values, coronary artery calcium scores, polygenic risk scores, and lifestyle metrics. Model training and validation were conducted using nested cross-validation and SHAP (Shapley Additive Explanations) to assess feature importance. ML models significantly outperformed traditional risk calculators in predicting incident myocardial infarction, stroke, and composite atherosclerotic events, with area under the curve (AUC) values exceeding 0.87 across models. Integration of coronary calcium and genetic risk notably improved early prediction in low-to-intermediate risk patients. Decision curve analysis demonstrated superior net clinical benefit of ML-guided risk stratification over guideline-based thresholds. Conclusion: Machine learning algorithms offer substantial improvements in the accuracy and granularity of atherosclerotic risk prediction by incorporating multidimensional data and capturing nonlinear interactions missed by traditional models. These findings support the integration of ML tools into clinical workflows to enable personalized prevention strategies and better allocation of therapeutic resources in cardiovascular care.