Predicting left atrial appendage thrombus in patients with persistent atrial fibrillation or flutter and no oral anticoagulation using machine-learning models

Résumé fourni par la source

Abstract Background Atrial fibrillation and flutter (AF/AFL) increase the risk of thromboembolic events by promoting the formation of a left atrial appendage thrombus (LAT). We previously developed artificial intelligence (AI) models to predict LAT based on data from the multicentre, prospective LATTEE registry (Left Atrial Thrombus on Transoesophageal Echocardiography, n= 2489) establishing the utility of AI for identifying the presence of LAT. While our model is now available as an online calculator to date there is no data regarding the utility of our AI in guiding cardioversion in haemodynamically stable patients with persistent arrhythmia and no prior oral anticoagulation (OAC). Purpose To assess the performance of AI decision support to perform TOE or readmit the patient after at least 3 weeks of OAC. Methods In this study, we applied the LAT-AI model (based on 39 features) and the LAT-AI reduced model (based on 8 features) to patients with no or insufficient prior OAC, who underwent TOE before direct current cardioversion or catheter ablation (n=311). We excluded patients with paroxysmal arrhythmia, resulting in a final cohort of 150 patients originating from four sites. We investigated the sensitivity and specificity of the respective models to predict LAT based on the score thresholds established in previous studies (0.37 for the LAT-AI and 0.34 for the LAT-AI reduced). Results The median age in the external cohort was 66 (interquartile range 56-73), 59% were male, and 7.3% had AFL. LAT was found in 17 patients (11%). The LAT-AI model achieved 100% sensitivity (confidence interval [CI]: 40-100) and 12% specificity (95% CI: 7-18). The LAT-AI reduced model achieved 100% sensitivity (95% CI: 87-100) and 14% specificity (95% CI: 8-21). By applying the score threshold which has been derived in our prior studies for the the LAT-AI reduced model, the TOE-first approach would be chosen in 18% and no LAT would be found in this group, while the remaining 82% would be recommended for 3 weeks of OAC before TOE. Such approach could be used to identify patients at low risk of LAT to undergo TOE despite the lack of prior OAC, while the remaining patients at higher risk of LAT could potentially avoid repeated TOE thanks to the 3-week course of OAC before the first TOE. Conclusion AI scores allow high sensitivity for prediction of the presence of LAT in patients with AF/AFL irrespective of OAC treatment and have the potential to optimise the management of patients with indications for elective cardioversion and no prior or ineffective OAC.Simulated AI-supported decisionScreenshot of the online app

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