Abstract 6871: Electrocardiographic Abnormalities Associated with AI-Derived ECG-Age: A Nationwide Population Study
Abstract
Background: Artificial intelligence-derived electrocardiographic age (ECG-age) has emerged as a potential surrogate marker of biological aging. Increased delta-age, defined as the difference between ECG-age and chronological age, has been associated with higher mortality and adverse cardiovascular outcomes. However, the electrocardiographic abnormalities associated with higher delta-age are not clearly understood. Methods: This retrospective cross-sectional study included ECGs from adults analyzed by a large Brazilian telehealth network (2019-2022). ECGs were interpreted by certified cardiologists, using predefined diagnostic classes. Annotated ECG classes were classified as normal, minor, or major abnormalities. ECG-age was estimated using a convolutional neural network. The primary outcome was delta-age, defined as the difference between ECG-age and chronological age. Associations between ECG abnormalities and delta-age were evaluated using linear regression adjusted for age, sex, and comorbidities. Statistical significance was defined as p < 0.05. Results: A total of 2,045,601 ECGs from individual patients were analyzed. Median age was 54.2 [IQR 40.9-66.3] years; 40.8% were male. Compared with normal ECGs, minor abnormalities were associated with a 2.99-year increase in delta-age, whereas major abnormalities were associated with a 6.44-year increase. Age-stratified analyses demonstrated persistent associations between ECG abnormality and delta-age across all age groups, with the strongest effect among middle-aged individuals. In analyses restricted to individuals without reported comorbidities (984,710), minor abnormalities were associated with a 2.92-year increase in delta-age (95% CI 2.87-2.96) and major abnormalities were associated with a 7.07-year increase (95% CI 6.99-7.14) after adjustment for age and sex. ECG diagnostic categories showed distinct associations with delta-age: pacemaker rhythm was associated with the largest increase (β = 5.72 years, 95% CI 5.47-5.98), followed by ischemia or infarction, arrhythmias, and intraventricular conduction abnormalities. The presence of multiple electrocardiographic diagnostic classes (≥2) was associated with higher delta-age. Associations remained significant after adjustment for ECG measurements. Conclusion: In this large-scale population study, delta-age increased with electrocardiographic abnormality severity. These findings characterize ECG patterns associated with higher delta-age, including abnormalities commonly related to cardiovascular aging, and suggests that delta-age captures additional information beyond conventional ECG intervals. The observed associations contribute to a better understanding of the information captured by ECG-age and support confidence in the method.