AI-Derived ECG Age Gap as a Digital Biomarker for Cardiovascular Risk: External Validation in Hospital and Community-Based Prospective Cohorts

ABSTRACT Cardiovascular diseases remain the leading cause of global mortality, and early risk stratification is critical for improving prognosis. Artificial intelligence–derived electrocardiography (AI-ECG) provides a promising approach to derive cardiac biological age as a non-invasive digital biomarker. This study developed an AI-ECG framework based on the ECGFounder foundation model to quantify cardiac biological aging and predict cardiovascular risk. A total of 67,824 ECGs from 63,512 UK Biobank participants were included. The model was trained on a development cohort of healthy individuals (n = 26,871), and evaluated in an independent clinical evaluation cohort (n = 40,953). Cox proportional hazards models were used to assess the association between the AI-ECG age gap and MACCE, as well as other secondary outcomes. External validation was further conducted in an inpatient cohort from Tianjin Medical University Second Hospital (n = 55,860) and the Kailuan community-based prospective cohort (n = 27,065). The model demonstrated good agreement between predicted and calendar age in the development cohort (r = 0.55; MAE = 5.12 years). In the clinical cohort, after adjusting for clinical comorbidities, each 1-year increase in the age gap was associated with a significant 13% higher risk of MACCE (HR = 1.13, 95% CI: 1.11–1.14). Individuals with an overestimated age gap (> 6 years) exhibited substantially elevated risks of MACCE (HR 4.51) and other major cardiovascular outcomes, whereas those with an underestimated age gap ( <− 6 years) showed a significantly lower risk of MACCE (HR 0.46) alongside protective effects across other outcomes. In the external cohort, the AI-ECG age gap remained significantly associated with multiple downstream cardiovascular and cardiometabolic outcomes. The AI-ECG age gap effectively quantifies occult accelerated cardiac aging. As a non-invasive digital biomarker, it has substantial potential for cardiovascular risk stratification in broad populations.