Longitudinal AI-ECG Biological Heart Age and Cardiovascular Mortality

Background: Biological heart age derived from artificial intelligence-enhanced electrocardiograms (AI-ECG) has emerged as a powerful biomarker for cardiovascular risk stratification. While static ECG age predicts mortality, the longitudinal senescence velocity—the rate of change in delta age (Δ-age = ECG-age – chronological age)—may better capture imminent sudden cardiac death risk by quantifying accelerated biological aging trajectories. Transformer-based deep survival models using self-attention mechanisms offer methodological advantages over traditional convolutional architectures by capturing long-range temporal dependencies in sequential ECG data through the attention operation: Attention(Q, K, V) = softmax(QK^T/√d_k)V, where queries (Q), keys (K), and values (V) enable dynamic weighting of informative time-series features. Methods: We systematically searched PubMed, MEDLINE, EMBASE, and the Cochrane Library from inception to March 2026 to identify relevant studies and conducted a systematic meta-analysis of AI-ECG biological age studies reporting cardiovascular outcomes. Random-effects models pooled hazard ratios (HR) for elevated ECG age (Δ-age ≥ +6 to +8 years) versus age-concordant controls. Heterogeneity was assessed using I 2 statistics and τ 2 . Longitudinal trajectory analyses examined aging velocity as a predictor of mortality. Results: Ten studies encompassing 3,608,747 participants. The pooled HR for elevated biological heart age was 1.78 (95% CI: 1.58-1.99, p < 0.0001), indicating 78% increased risk of adverse cardiovascular outcomes. Substantial heterogeneity was observed (I 2 = 79.7%, τ 2 = 0.0250, Q(9) = 44.40, p < 0.0001), reflecting diverse populations ranging from general health screening cohorts to high-risk kidney transplant candidates. Longitudinal studies demonstrated that accelerated aging velocity (increasing Δ-age over 5-year intervals) conferred higher mortality risk (adjusted HR 1.66, 95% CI: 1.30-2.12) compared to stable trajectories, suggesting that senescence velocity weights imminent risk beyond static age discrepancy. Conclusions: Transformer-based self-attention architectures applied to longitudinal ECG time-series enable precise quantification of biological cardiac senescence velocity. The robust association between elevated AI-ECG age and cardiovascular mortality (pooled HR 1.78) supports clinical integration for risk stratification. Longitudinal aging trajectories may identify patients at highest imminent risk for sudden cardiac death, warranting prospective validation in arrhythmia prediction trials.