POPULATION-WIDE RISK STRATIFICATION OF CARDIOVASCULAR EVENTS USING AI-AUGMENTED ECG, EPIGENOMIC BIOMARKERS, AND REAL-WORLD EHR DATA

Abstract

Cardiovascular diseases remain the leading global cause of death, underscoring the urgent need for accurate, personalized risk stratification tools. This study presents an AI-driven predictive framework that integrates electrocardiogram (ECG) data, epigenomic biomarkers, and electronic health records (EHRs) to enhance cardiovascular risk assessment across diverse populations. A mixed-methods experimental design was employed, incorporating wavelet-transformed ECG signals, normalized CpG epigenomic profiles, structured EHR variables, and NLP-extracted unstructured notes. The fused dataset was modeled using Random Forests, Gradient Boosting, and Artificial Neural Networks, evaluated through 5-fold cross-validation. Performance metrics demonstrated high predictive accuracy with AUCs exceeding 0.90 and improved F1-scores across risk categories. Feature attribution techniques such as SHAP and LIME ensured model interpretability, while decision support outputs provided personalized intervention suggestions. The results, supported by 9 detailed tables and 12 complex visualizations, revealed significant improvements in risk prediction, especially in early-stage detection. This integrated AI methodology not only improves clinical precision but also offers scalable, real-time insights for population health strategies. The study concludes that AI-powered multimodal integration holds substantial promise for transforming cardiovascular diagnostics and preventive care through personalized, data-driven interventions.