A Systematic Review of Machine-Learning Models for Cardiovascular Risk Prediction

This systematic review evaluates machine learning (ML) models for cardiovascular disease (CVD) risk prediction compared to traditional regression-based scores. Following PRISMA-۲۰۲۰ guidelines, we conducted comprehensive searches in MEDLINE, Embase, Web of Science, Scopus, and IEEE Xplore from January ۲۰۱۰ to May ۲۰۲۵. Two independent reviewers screened ۴,۳۷۲ records, selecting ۱۸ studies meeting predefined PICOTS criteria. Data extraction focused on model architecture, validation strategies, performance metrics (discrimination, calibration, net reclassification improvement), and interpretability methods. Risk of bias was assessed using PROBAST, and evidence certainty was graded using a modified GRADE framework. Our analysis revealed that ML models achieved modest but consistent discrimination improvements (ΔAUROC +۰.۰۲-۰.۰۸) over conventional scores, with gradient-boosted trees and deep survival networks performing best. However, calibration drift occurred in external validation, and clinical implementation remains limited by infrastructure requirements and unproven impact of interpretability tools. These findings suggest ML models offer incremental rather than transformative improvements in CVD risk prediction.

Keywords:

Machine Learning , Cardiovascular Diseases , Risk Assessment , Systematic review , Prognostic models