Abstract 11284: Predicting Thermal Injury by Machine Learning Analysis of Luminal Esophageal Temperature Patterns in Atrial Fibrillation Ablation

Abstract

Listen

Translate article

Introduction: Luminal esophageal temperature (LET) monitoring during catheter ablation for atrial fibrillation (AF) is widely used to reduce the incidence of esophageal thermal injury (ETI). We sought to assess whether specific patterns of LET variation are associated with ETI. Methods: The CIRCA-S multi-sensor probe was used to record LET in AF patients undergoing radiofrequency (RFA) or cryoballoon ablation (CBA) ( Fig A ). Patients underwent upper endoscopy one day after ablation. LET data was analyzed for patterns associated with ETI ( Fig B ). A random forest classifier with 5x stratified cross validation was trained to predict ETI, and SHapley Additive exPlanations (SHAP) analysis was used to assess feature importance in the prediction of ETI. Results: A total of 90 patients (63% paroxysmal AF; 27% female) were included. Among them, 68 patients underwent RFA, and 22 patients CBA. ETI was detected in 11 patients (10 RFA and 1 CBA). In case of RFA, the highest LET recorded was similar between patients who had an ETI and those who did not (40.4 [39.1, 41] vs 40.6 [40.1, 41], p=0.4); however, the rate of LET rise for the highest recorded peak was higher (0.08 [0.03, 0.12] vs 0.02 [0.01, 0.05]; p=0.02) and area under the curve (AUC) for highest peak was smaller (412.5 [206.8, 634.1] vs 545.5 [369.9, 855.1]; p=0.066) in patients who had ETI. In case of CBA, the patient with ETI had a faster LET decline (0.12 vs 0.07 [0.02, 0.14]), and a smaller AUC for the lowest trough (2491.3 vs 2629.3 [1712.6, 5283.2]) compared to patients who had no ETI. SHAP analysis revealed that for the overall population, rate of LET change was the most important feature in the prediction of ETI ( Fig C ), and that a rate of LET change >0.1°C/s is most predictive of ETI ( Fig D ). Conclusion: Rate of LET change and AUC for the recorded temperature are more important predictors of ETI compared to absolute peak temperatures. Prospective studies guided by these parameters are needed to demonstrate their efficacy in reducing ETI.