A Random Survival Forest Model Incorporating Dual Left Atrial Diameters for Predicting Atrial Fibrillation Recurrence After Left Unilateral Thoracoscopic Ablation

Background: Left atrial anteroposterior diameter (LAAP) is commonly used to assess left atrial remodeling and recurrence risk after thoracoscopic atrial fibrillation (AF) ablation, but may incompletely reflect atrial enlargement, whereas volumetric assessment has equipment- and operator-related limitations. Our study aimed to develop a prediction approach for AF recurrence by incorporating complementary left atrial diameters. Methods: This single-center retrospective study included 563 patients undergoing left unilateral thoracoscopic ablation, with 430 patients from January 2018 to December 2022 assigned to the training set and 133 patients from January 2023 to December 2024 assigned to the testing set. Model development was performed in the training cohort. Model performance was evaluated using Harrell’s C-index and receiver operating characteristic curves. The model was further converted into a risk score system (LATA) for clinical application. Results: LAAP and left atrial superoinferior diameter (LASI) were significant predictors of 1-year AF recurrence, with cutoffs of 4.7 cm and 7.0 cm, respectively. An RSF model incorporating six selected variables, including LAAP and LASI, achieved AUCs of 0.834 in training set and 0.795 in testing set. The LATA score showed good discrimination and effectively stratified recurrence risk. Conclusions: Our machine learning model and score system (LATA) combining LAAP and LASI showed favorable predictive performance and potential clinical utility.