https://doi.org/10.1007/s00392-025-02737-x
1Medizinische Universität Innsbruck Department für Innere Medizin III - Kardiologie und Angiologie Innsbruck, Österreich; 2Institut für Klinische Epidemiologie, Public Health, Medizinische Statistik und Informatik - Medizinische Universität Innsbruck Innsbruck, Österreich
Algorithms using artificial intelligence (AI) in standard 12-lead ECGs provide promising methods for age prediction. The concept of ECG age holds the potential to provide insights into true biological age that extend beyond traditional chronological age. Existing studies have already demonstrated an age-independent association between ECG age and mortality. While prior research primarily focused on general populations, the applicability and prognostic value of ECG age in patients with cardiovascular disease remain underexplored. Understanding these relationships could refine risk stratification and offer novel insights into the biological aging process in this population.
Methods
We employed a validated open-source AI algorithm (3) that was trained on 1,558,415 patients. This algorithm calculates an "ECG age" based on the patient's ECG. This algorithm was applied to our comprehensive ECG database, which contains ECGs from 48,950 unselected patients treated at our cardiology inpatient and outpatient clinics between 2000 and 2021. Pearson's correlation coefficient was used to assess the model's performance in this specific population. In a subsequent step, patients were stratified into three groups based on the difference between their chronological age and ECG age, as previously investigated in related studies. A.) ECG age ≥8 years younger, B.) ECG age ≥8 years older, and C.) ECG age within ±8 years of chronological age. The associations between these groups and all-cause mortality were assessed using a Cox proportional hazards analysis, with adjustments made for age and gender. Subanalyses included various cardiovascular risk factors and ECG parameters in the Cox models.
Results
The analysis included a total of 52% male patients. The mean chronological age was 54 years, while the mean ECG age was 52 years. The Pearson correlation coefficient revealed a low correlation between chronological age and ECG age (r = 0.22, p < 0.001) (Fig. 1), indicating that the AI algorithm captured additional age-related biological information beyond traditional metrics. Patients with an ECG age ≥8 years older than their chronological age exhibited a significantly elevated risk of mortality (hazard ratio [HR]: 1.67 [confidence interval [CI]: 1.50–1.86]; p < 0.001) compared to the neutral group. Conversely, a younger ECG age compared to chronological age was associated with a reduced mortality risk (HR: 0.81 [CI: 0.76–0.86; p<0.001) (Figure 2). Subsequent analysis revealed that these effects remained consistent even when accounting for established cardiovascular risk factors and standard ECG parameters.
Conclusion
The present study demonstrates that the discrepancy between true chronological age and AI-estimated ECG age is an independent predictor of all-cause mortality in patients with cardiovascular disease.
Figure 1: Scatterplot of Chronological Age vs. Predicted Age. The color intensity represents the frequency of data points within each hexagonal bin.
Figure 2: Age-adjusted survival: Classification based on the difference between the chronological age and the AI-ECG age.