Artificial intelligence estimated electrocardiographic age trajectory predicts recurrence after atrial fibrillation catheter ablation

Abstract Aims The application of artificial intelligence (AI) algorithms to 12-lead electrocardiogram (ECG) provides promising age prediction models. We explored whether change in the AI-ECG age between pre-procedural and 3rd month post-atrial fibrillation catheter ablation (AFCA) would predict clinical recurrence of AF. Methods We developed (1,730,222 ECGs from 662,246 participants) and validated a residual network (ResNet)-based model for age prediction on independent multinational datasets. Then, we calculated AI-estimated ECG (AI-ECG) age among those who underwent first AFCA, had pre-procedural and 3rd month post-AFCA sinus rhythm ECGs in a pooled AFCA cohort (2,785 from Yonsei University Health System [YUHS]; 1,821 from Korea University Anam Hospital [KUAH]). We categorized pre-procedural AI-ECG age gap into two groups: aged-ECG (≥10 year) and normal ECG age (<10 year) groups based on the mean absolute ECG age gap error in the validation datasets. Aged-ECG and normal ECG age groups were further categorized as increasing and decreasing ECG age according to 3rd month post-AFCA. Results The ResNet-based AI-ECG age model successfully reproduced the chronological age on the training datasets (959,514 ECGs); CODE-15% (r=0.78) and MIMIC4 (r=0.69) cohorts. Among paroxysmal AF patients with aged-ECG at pre-AFCA, increasing ECG age at 3rd month post-AFCA was associated with significantly increased risk of AF (hazard ratio [HR] 2.34, 95% confidence interval [CI] 1.73-3.17) and significantly increased but attenuated risk among those with decreasing ECG age at 3rd month post-AFCA (HR 1.40, 95% CI 1.08-1.81) when compared to those with normal ECG age at pre-AFCA and decreasing ECG age at 3rd month post-AFCA. In contrast, ECG age at 3rd month post-AFCA was not significantly associated with the risk of AF recurrence (HR 1.16, 95% CI 0.98-1.38) among those with normal ECG age at pre-AFCA. These associations were not apparent among patients with persistent AF. Conclusions Pre-procedural AI-ECG age and the change of ECG age at 3rd month post-AFCA has a prognostic value for AF recurrence among paroxysmal AF patients.