Artificial intelligence—derived electrocardiographic age gap as a predictor of mortality after coronary revascularization: prognostic value and short-term intra-patient variability

The artificial intelligence (AI)-derived electrocardiographic (ECG) age gap—the difference between AI-predicted ECG age and chronological age—is an emerging biomarker of biological ageing linked to mortality. This study assessed its prognostic value for short- and long-term mortality after coronary artery bypass grafting (CABG) or percutaneous coronary intervention (PCI), addressing model bias in ageing cohorts and short-term intra-patient variability.

A residual neural network was trained on 532 301 retrospective ECGs and optimized with a distribution-aware loss function to reduce age-imbalance bias (mean absolute error 6.87 years, R² 0.71). We reproduced known mortality associations in a general cardiology cohort (n = 22 457). Revascularization cohorts included 1354 CABG and 1545 PCI patients, with ECG age gap derived from pre-procedure averages. Multivariable Cox models adjusted for age, sex, Charlson Comorbidity Index, smoking, and obesity assessed 120-day and 3-year mortality. Exploratory analyses quantified intra-patient variability and the change in ECG age gap before and after the procedure. Higher ECG age gap predicted increased mortality: hazard ratios (HRs) per year were 1.05 (95% CI 1.01–1.10; P = 0.01) for 120-day post-CABG, 1.05 (1.02–1.08; P < 0.005) for 3-year post-CABG, 1.03 (0.99–1.07; P = 0.13) for 120-day post-PCI, and 1.04 (1.01–1.06; P < 0.005) for 3-year post-PCI. A 10-year gap corresponded to approximately 60% and 50% higher mortality post-CABG and post-PCI, respectively. Short-term variability revealed a median 8.8-year spread, and CABG patients showed a significant age gap reduction of 1.42-years (P < 0.005).

The AI-derived ECG age gap independently predicts mortality after revascularization, but substantial short-term variability necessitates serial monitoring for reliable clinical use.