Aims: Thromboembolic risk predictors have demonstrated limited efficacy in predicting rhythm outcome post-catheter ablation. This study aims to evaluate the performance of the CHA2DS2-VASc score against a refined set of clinical and Heart Rate Variability (HRV) features obtained from implantable cardiac monitors (ICMs) in continuously monitored patients before their first catheter ablation for atrial fibrillation (AF).
Methods: 239 AF patients (age 67 ± 9 years, 40% female) from Optum's de-identified Clinformatics® Data Mart Database (2007-2020) were selected and linked with the Medtronic CareLink database of ICMs. HRV-derived features were extracted from recorded AF episodes and the beat trend preceding their onset, obtained from the last manual transmission before catheter ablation. We compared the AF recurrence predictive performance of the CHADSVASC score with a Support Vector Machine (SVM) classifier. Feature selection was performed to include nine clinical and HRV-derived variables, using the sequential forward floating search algorithm. The F1-score and the area under Receiver Operating Characteristic curve (AUC) were used to compare classification methods.
Results: Among the patients, 182 (76%) experienced AF recurrence, defined as detection of a 2-minute long AF episode outside the 3-month blanking period. We utilized a CHA2DS2-VASc cutoff of 2 to predict AF recurrence. While differences existed in the proportions of patients with high and low CHA2DS2-VASc scores in the recurrence and non-recurrence groups (CHA2DS2-VASc ≥ 2: Recurrence 80%, No Recurrence 54%; p-value < 0.001), the SVM algorithm (F1-score = 0.58, AUC = 0.75) outperformed the CHA2DS2-VASc score (F1-score = 0.47, AUC = 0.64).
Conclusions: The optimized selection of clinical and HRV-derived features demonstrated superior predictive capability for rhythm outcome compared to thromboembolic risk predictors. This advancement could facilitate more effective pre-ablation patient triage strategies, potentially reducing the economic and personal burden associated with AF recurrence.