AI-enabled ECG AF risk score as a predictor of early recurrence and AF burden after ablation

Background An AI-enabled ECG-derived atrial fibrillation (AF) risk score can estimate AF susceptibility from a sinus rhythm ECG.(1) It is hypothesized that the algorithm detects and quantifies atrial cardiomyopathy. Its value for predicting early post-ablation recurrence remains uncertain. Purpose To evaluate the association between an AI-enabled AF risk score and early recurrence of atrial tachyarrhythmias (ERAT) within 8 weeks after AF ablation, applying a 4-week blanking period. Methods We performed a retrospective analysis in a single-center cohort, where smartphone based photoplethysmography (PPG) monitoring was routinely offered after AF ablation. Patients who recorded at least one high-quality measurement within 8 weeks were included. ERAT between 4 to 8 weeks constituted the primary endpoint. The AI-enabled AF risk score (range 1-100) was calculated on a sinus rhythm ECG obtained upon admission or post-procedure. If unavailable the most recent pre-admission sinus rhythm ECG was used. Logistic regression assessed associations with ERAT. AF burden was evaluated using AF load (percentage of time in AF) and SMURDEN (number of days with at least one AF measurement). Because of skewed distribution, AF burden measures were log-transformed for linear modeling. Results Of 789 ablated patients, 511 (64.8%) were offered PPG monitoring based on potential benefit and digital literacy, of whom 349 (68.3%) recorded at least one high-quality measurement. The AI risk score was available in 348 patients included in the analysis with a median risk score of 26.5 (IQR 9-56). ERAT occurred in 65 patients (18.6%), with a median AF load of 12.6% (IQR 2.6-52.9%) and median SMURDEN of 4 days (IQR 1-12). Higher scores were associated with higher odds of ERAT. As a continuous variable, each 10-point increase corresponded to 14% increase in odds of recurrence (OR 1.14, 95% CI 1.04-1.23, p=0.005). Compared with low risk patients (AF risk score ≤ 20), high risk individuals (AF risk score > 50) had significantly higher odds of recurrence (OR 2.70, 95% CI 1.39 – 5.22, p=0.003), while intermediate scores (20-50) showed a nonsignificant trend (OR 1.94, p=0.061). Discriminative ability was modest (AUC 0.616), similar to established clinical risk scores.(2) Higher risk scores were also associated with a greater AF burden. Each 10-point increase corresponded to a 35% higher AF load (p<0.001) and a 26% higher SMURDEN (p<0.001) after log transformation. Conclusions The AI-enabled ECG-derived AF risk score is significantly associated with ERAT and higher AF burden early after ablation. Although its discriminative performance is modest, the score provides clinically meaningful stratification of early post-ablation risk. These findings support its potential role in tailoring follow-up intensity and early management strategies after AF ablation, pending confirmation in broader and prospectively monitored cohorts.