A logistic regression ML model predicted arrhythmic recurrence after AF ablation with AUC ~0.7 and major adverse cardiovascular events with AUC ~0.65 across cohorts.
Does a logistic regression-based machine learning model predict long-term arrhythmic recurrence and MACE in patients undergoing catheter ablation for atrial fibrillation?
A logistic regression-based machine learning model using 5 clinical features provides fair and consistent prediction of long-term arrhythmic recurrence and MACE after atrial fibrillation ablation.
Absolute Event Rate: 0% vs 0%
Abstract Background The arrhythmic recurrence and clinical events remain a concern after catheter ablation for atrial fibrillation (AF). Machine learning (ML) offers the potential for prediction using readily available patient characteristics. We investigated the utility of ML, specifically logistic regression (LR), to predict arrhythmic recurrence and major adverse cardiovascular events (MACE) after AF ablation. Methods We retrospectively analyzed an AF patient cohort undergoing catheter ablation at two institutions A and B. Using the Institution A cohort, multivariate LR selected 5 key features for arrhythmic recurrence from 20 patient characteristics. Subsequently, this cohort was divided into derivation and internal validation sets (80%:20%). An LR-based ML model for predicting arrhythmic recurrence was trained on the derivation set and evaluated on both the internal validation set and an external validation cohort from Institution B. The same ML model was applied to predict major adverse cardiovascular events (MACE), defined as all-cause death, heart failure hospitalization, and other cardiovascular hospitalizations. Performance metrics for arrhythmic recurrence and MACE were assessed. Results We included 703 and 153 patients in Institutions A and B, respectively. During 5.9 ± 1.9 years of mean follow-up, 227 (32%) in Institution A experienced arrhythmic recurrence, and 111 (16%) experienced MACE. Procedure year, non-paroxysmal AF, AF duration, brain natriuretic peptide, and left atrial volume index were identified as key features from 20 characteristics. The model for arrhythmic recurrence demonstrated fair performance, with consistent AUCs: 0.708 (derivation), 0.695 (internal validation), and 0.691 (external validation). The same model showed generally applicable performance for MACE prediction, with AUCs of 0.674 (derivation), 0.648 (internal validation), and 0.645 (external validation). Conclusions Our data suggest that an LR-based ML model can provide valuable insights for predicting long-term arrhythmic recurrence after AF ablation. Furthermore, this model demonstrates applicability for predicting MACE as well.
Uchida et al. (Sat,) reported a other. A logistic regression ML model predicted arrhythmic recurrence after AF ablation with AUC ~0.7 and major adverse cardiovascular events with AUC ~0.65 across cohorts.