Impact of major and minor electrocardiographic abnormalities on AI-derived ECG-age: a large-scale population study

Abstract Background Artificial intelligence (AI)-derived electrocardiographic age (ECG-age) and the difference between ECG-age and chronological age (delta-age) are novel biomarkers of cardiovascular ageing that predict adverse outcomes. However, the specific electrocardiographic abnormalities contributing to increased ECG-age remain unclear. Clarifying the role of each ECG alteration may improve insight into cardiovascular ageing and help refine risk assessment tools. Purpose To assess the impact of major and minor electrocardiographic abnormalities on AI-derived ECG-age. Methods This retrospective cross-sectional study included adult patients with ECGs analysed by a large telehealth network from January 2019 to December 2022. ECGs were excluded if they had technical artefacts, inconsistent demographic data, patient age under 18 years, duplicate reports, or were linked to external research projects. Only the first ECG was retained for patients with multiple studies. Outliers were excluded based on predefined criteria: chronological age over 100 years or delta-age greater than 40 years. ECG abnormalities were classified using the Minnesota Code into normal, minor, and major categories. AI-based ECG-age was estimated using a convolutional neural network with a residual architecture. The primary outcome was delta-age (ECG-age minus chronological age). Categorical variables were described as frequencies, and continuous variables as medians and interquartile ranges (IQRs). Associations between ECG category and ECG-age were evaluated using linear regression adjusted for age and sex. Results We analysed 2,045,602 ECGs (median age 54.2 years IQR 40.9–66.3; 59.2% female). Diagnostic categories were: normal (51.8%), minor (35.6%), and major abnormalities (12.5%). Median delta-age was higher in the minor and major abnormality groups (2.08 and 2.44 years, respectively) than in normal ECGs (1.87 years) and in males vs. females (2.61 vs. 1.59 years). In multivariable linear regression, male sex (β = 0.94, p 0.001), younger age (β = –0.34, p 0.001), and minor (β = 3.10, p 0.001) or major (β = 6.63, p 0.001) ECG abnormalities were independently associated with higher delta-age (R² = 0.24, p 0.001). Figure 1 shows delta-age distributions by ECG category, with blue points indicating adjusted β values for minor and major abnormalities relative to the normal category. Conclusions Both minor and major electrocardiographic abnormalities were independently associated with higher AI-derived ECG-age, with a greater impact observed for major abnormalities. These findings suggest that the extent and severity of ECG alterations contribute differently to the cardiovascular ageing patterns captured by AI. Understanding how specific ECG features influence AI-based age estimation may improve the interpretability and clinical utility of this emerging biomarker for cardiovascular risk assessment.Delta-age distributions by ECG category