δ-Age and the Aging Heart: Electrocardiogram-Based Assessment of Cardiovascular Health

This thesis evaluated δ-age, the difference between an AI-predicted ECG heart age and chronological age, as a novel biomarker of cardiovascular aging and risk. Using data from the population-based Tromsø Study, we examined how δ-age relates to social factors, long-term risk exposures, and cardiovascular outcomes. In Paper I, we tested whether educational level is associated with heart age acceleration. In 4,367 participants free of cardiovascular disease, each higher level of education was linked to a slightly lower δ-age, especially among women. Mediation analysis showed that traditional risk factors (blood pressure, lipids, BMI, smoking) explained most of this association specifically for woman, whereas in men the mediation did not reach statistical significance. This suggests that social inequalities influence heart aging through modifiable factors mainly in women.

In Paper II, we compared three published deep learning algorithms for ECG-age prediction in 7,108 individuals. All models produced fairly concordant ECG-age with similar accuracy, and each model’s δ-age demonstrated comparable predictive ability for incident myocardial infarction, stroke, cardiovascular mortality, and all-cause death during follow-up. The predicted survival probabilities were nearly identical across models, demonstrating that δ-age is a robust risk marker largely independent of the specific algorithm. Higher δ-age consistently indicated increased risk, even after the additional adjustment for traditional risk factors.

In Paper III, we identified four midlife trajectories of systolic/diastolic blood pressure, BMI, and non-HDL cholesterol over 30 years, among 5,244 adults. Clusters characterized by long-term hypertension and/or obesity had the highest cardiovascular event rates. Notably, these adverse clusters also had significantly higher mean δ-age. Each 1-SD increase in δ-age was associated with ~27–30% higher hazards of MI, stroke, and all-cause death. However, causal mediation analysis showed that δ-age explained only a small fraction (~4–9%) of the excess risk from adverse risk-factor trajectories. In other words, δ-age functioned as an independent indicator of cumulative damage of the cardiovascular system, rather than being on the causal pathway of risk.

Together, our findings support δ-age as a meaningful marker of cardiovascular aging with independent prognostic value. It is socially patterned (higher in low-education groups), captures the imprint of lifelong cardiometabolic exposures, and is technically consistent across AI models. Importantly, δ-age identifies individuals whose hearts are biologically older than expected and who are at elevated risk of MI, stroke, and mortality beyond traditional scores. While conventional risk factors remain essential to manage, δ-age provides an easy, non-invasive way to refine risk stratification. In summary, ECG-derived δ-age is a novel, robust biomarker of vascular aging that could guide prevention by revealing hidden elevated heart risk.