Abstract Body (Do not enter title and authors here): Background: The use of Artificial Intelligence (AI) to estimate age from 12-lead electrocardiogram (ECG) data has shown promise as a biomarker for predicting future cardiac disease risk. However, the long-term outcomes based on the temporal variations in the AI-ECG age gap remain uncertain.
Hypothesis: This study aims to longitudinally track the differences between AI-predicted age and actual chronological age using ECG data and evaluate whether significant clinical differences in atrial fibrillation (AF) occurrence exist among various patient groups.
Methods: From a single-center cohort not previously used for AI-ECG age prediction (121,702 individuals, 522,261 ECGs), 5,317 individuals with 24,368 ECGs who had an initial ECG and a follow-up ECG in the fifth year, along with at least one intervening ECG, were included for analysis. The absolute age gap, defined as the absolute difference between ECG-predicted and chronological age, was constrained to ±10 years for this analysis. Latent class trajectory modeling was employed to categorize groups based on the AI-ECG age gap trajectories.
Results: The analysis identified four distinct trajectory groups based on age gap patterns: Consistently Young ECG (28.7%), Reversed ECG Aging (35.5%), Accelerated ECG Aging (8.0%), and Consistently Old ECG (27.9%). Both the Accelerated ECG Aging group (adjusted hazard ratio (HR) 1.73, 95% confidence interval (CI) 1.07-2.79, p-value = 0.025) and the Consistently Old ECG group (adjusted HR 1.78, 95% CI 1.27-2.49, p-value < 0.001) had a significantly higher cumulative incidence of AF compared to the Consistently Young ECG group. Four independent predictors of these trajectories were identified: baseline age, chronic kidney disease, diabetes mellitus, and hypertension.
Conclusion: The study identified four distinct AI-ECG age gap trajectory groups, with the Accelerated ECG Aging and Consistently Old ECG groups showing a significantly increased risk of AF occurrence.