Early Detection of Cardiovascular Disease Risk Using Multi-Parameter Biomarker Analysis and Machine Learning

Abstract

ABSTRACT Background: Cardiovascular disease (CVD) remains the leading cause of global mortality, with most events occurring in individuals without prior diagnosis. Early risk stratification using accessible biomarkers could enable timely intervention. We evaluated a machine learning (ML) system for predicting CVD risk 4-6 weeks before clinical manifestation in asymptomatic adults. Methods: In this prospective cohort study conducted January 2024-January 2025, 500 employees (300 males, 200 females; ages 35-50) without known cardiac disease underwent weekly screening measuring BMI, blood pressure, heart rate, single-lead ECG, and random glucose. A supervised ML algorithm generated cardiovascular risk scores. Monthly comprehensive cardiac evaluations (12-lead ECG, echocardiography, laboratory testing) performed by blinded cardiologists served as clinical validation endpoints. Results: Over 26,000 screening sessions were completed (98.4% adherence). During follow-up, 38 participants (7.6%) developed early-stage CVD. The ML model achieved 96.0% accuracy (480/500 correct classifications), 71.05% sensitivity (27/38 true positives; 95% CI: 54.1-84.6%), and 98.05% specificity (453/462 true negatives; 95% CI: 96.4-99.0%). Positive predictive value was 75.0% and negative predictive value was 97.6%. AUROC was 0.923 (95% CI: 0.884-0.962). Males with BMI 28-30 kg/m.sq, pulse pressure 60-74 mmHg, ECG abnormalities (ventricular ectopy, ST-segment changes), and random glucose 156-164 mg/dL demonstrated highest risk (81.5% detection rate). Performance remained consistent across demographic subgroups. Conclusions: Integration of routine physiological parameters with ML algorithms demonstrates high specificity and acceptable sensitivity for near-term CVD risk detection in asymptomatic working-age adults. The system's high negative predictive value suggests utility for population-level screening, though modest sensitivity indicates complementary clinical assessment remains essential. Further validation across diverse populations and demonstration of impact on clinical outcomes are needed before widespread implementation. Funding: XpertFlow, Singapore Competing Interests: RA and SA are employees and equity holders of XpertFlow. AHB served as paid consultant to XpertFlow during the study. Clinical Trial Registration: Not applicable (observational study)

Links & Resources

Authors

Cite This Paper