Comparison of a deep-learning algorithm with p-wave duration for non-pulmonary vein triggers

Résumé fourni par la source

Abstract Background Non-pulmonary vein (non-PV) triggers are commonly targeted in patients with persistent atrial fibrillation (AF) to aid in improvement of clinical outcomes. Current data are limited regarding upfront prediction of non-PV triggers. FactorECG is a deep-learning-based algorithm which is based on 21-explainable factors on the 12-lead EKG that may be able to predict the presence of non-PV triggers based on the pre-ablation sinus rhythm EKG. Purpose To estimate prediction of non-PV triggers using FactorECG from pre-ablation sinus rhythm EKG. Methods We screened patients between 01/2020 and 12/2023 who underwent catheter ablation for AF. Patients who had spontaneous non-PV triggers at the time of catheter ablation or underwent systematic testing (with high dose Isoproterenol infusion +/- rapid atrial pacing) were included. Patients were excluded if they had a prior catheter ablation, prior cardiac surgery, did not have a pre-ablation sinus rhythm EKG, or did not undergo induction testing for non-PV triggers. We recorded baseline comorbidities, echocardiographic data, electrocardiographic data aimed at atrial depolarization (such as P-wave duration and inter-atrial block). Binary logistic regression analysis was conducted to investigate the association between baseline parameters and 12-lead EKG parameters and the likelihood of having non-PV triggers. We used pre-ablation 12-lead sinus rhythm EKGs to apply a machine-learning algorithm (FactorECG, https://decoder.ecgx.ai). Multiple logistic regression under L1 regularization was used to analyze which FactorECG "factors" were predictive of non-PV triggers. Results There were 32 patients who were observed to have non-PV triggers at the time of index ablation for AF and 140 patients did not have non-PV triggers (served as controls). Compared to our control group of patients, those with non-PV triggers at the time of index ablation were older in age and had a higher CHA2DS2-VASc score. On pre-ablation EKG, patients with non-PV triggers had a prolonged PWD (142.3 vs. 128.9, p = 0.002) and a higher incidence of IAB (53.1% vs. 28.6%, p = 0.01). Using binary logistic regression, only PWD was associated with non-PV triggers (Adjusted OR of 1.02 per 1 msec, 95% CI of 1.0-1.04, p= 0.03). Upon considering only the top 10 important factors (23, 10, 31, 25, 32, 5, 8, 1, 6, and 13) from the FactorECG, the area under the receiver operating characteristic curve (AUROC)/c-statistic using FactorECG was 0.72 [95% CI 0.51-0.91] (Figure). The most important factors were 23, 10, 31, and 25. Conclusions Patients with non-PV triggers at the time of AF ablation represent a group of patients who have baseline impaired atrial depolarization; as evident by prolonged PWD and a higher incidence of baseline IAB. FactorECG is a deep-learning algorithm that can effectively predict the presence of non-PV triggers based on the pre-ablation sinus rhythm 12-lead EKG.Baseline parameters ROC of Factor ECG

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.