Transthoracic echocardiographic and artificial intelligence-enabled electrocardiography predictors of atrial arrhythmia recurrence after surgical ablation

TL;DR

This study shows that combining AI-enhanced ECG and echocardiography improves predicting atrial arrhythmia recurrence after surgery.

Contribution

The novel integration of AI-ECG and TTE metrics provides a more accurate prediction model for post-ablation atrial arrhythmia recurrence.

Findings

AI-ECG biomarkers like AF probability and ECG-estimated age were strong predictors of recurrence.

Larger left atrial area and elevated mitral E-wave velocity were associated with higher recurrence risk.

The combined model achieved a concordance index of ~0.67 and better risk stratification.

Abstract

Recurrence of atrial fibrillation (AF)/flutter (AFl) after surgical ablation remains difficult to predict. Integration of novel biomarkers may enhance risk stratification.

This study aimed to assess whether combining preoperative transthoracic echocardiography (TTE) and artificial intelligence-enabled electrocardiography (AI-ECG) scores improves the prediction of AF/AFl recurrence after surgical ablation.

We retrospectively analyzed 1696 patients who underwent surgical AF/AFl ablation from 2006 to 2025 with available preoperative TTE and ECG and postblanking (90-day) ECG follow-up. Clinical, TTE, and AI-ECG variables (AF probability, ECG-estimated age, heart failure with preserved ejection fraction, left ventricular dysfunction, and aortic stenosis scores) were assessed. Cox proportional hazards and random survival forest models (80:20 train-test split) identified predictors of recurrence.

Among 1696 patients (mean age 67.3 ± 10.2 years; 61.7% male), 949 (56%) had AF/AFl recurrence over a median 3.14-year follow-up. Patients with recurrence had larger left atrial area (30.4 vs 24.5 cm2), elevated mitral E-wave velocity (1.015 vs 0.896 m/s), and adverse AI-ECG biomarkers for AF probability, ECG-estimated age, heart failure with preserved ejection fraction, left ventricular dysfunction, and aortic stenosis (all P < .001). In multivariable analysis, independent predictors of recurrence included higher ECG-AF probability (P < .0001), older ECG-estimated age (P = .0002), left atrial area (P = .046), body mass index (P = .036), and diastolic blood pressure (hazard ratio 1.008/mm Hg; P = .010). The final Cox model achieved a concordance index of ∼0.67 and a 3-year Brier score of 0.21, with 3-year freedom-from-arrhythmia rates of ∼85% vs ∼43% for the lowest- vs highest-risk quartiles. Random survival forest modeling yielded a slightly higher concordance index (∼0.69).

Preoperative AI-ECG biomarkers (AF probability, age discordance) and TTE markers of atrial remodeling independently predicted AF/AFl recurrence after surgical AF/AFl ablation. Integration of these metrics improved risk stratification.

Figures