Virtual Heart–Based Prediction of Ablation Outcomes in Persistent Atrial Fibrillation Based on Non–Pulmonary Vein Drivers

BACKGROUND: Clinical outcomes of driver-targeted ablation in persistent atrial fibrillation (AF) have been inconsistent, underscoring the need for virtual heart-based approaches to characterize AF driver dynamics and guide ablation strategies. We used virtual heart models from a large persistent AF cohort to define electrophysiological features of extra-pulmonary vein drivers associated with rhythm outcomes after AF ablation. METHODS: We analyzed left atrial virtual heart models integrating cardiac computed tomography and electroanatomical maps in 165 patients with persistent AF from the CUVIA-AF2 study (Clinical Usefulness of Virtual Ablation Guided Catheter Ablation of Atrial Fibrillation 2). During a 34-second virtual AF simulation, stable phase singularities (PSs) were identified by PS lasting ≥2 s within a 10-mm diameter. Dominant frequency (DF) was assessed over 3 consecutive 6-second intervals to evaluate temporal stability. The maximal slope of the action potential duration restitution curve was derived via virtual ramp pacing (200–120 ms). All markers were evaluated across 6 extra-pulmonary vein regions. Associations among stable PS, DF, maximal slope of the action potential duration restitution curve, low-voltage area (<0.5 mV), and 1-year rhythm outcomes were analyzed. RESULTS: Stable PS at extra-pulmonary vein regions was observed in 124 (75%) patients; of these, 43 (34.7%) showed stable DF, and 81 (65.3%) had meandering DF. In regional analysis of the virtual heart models, areas containing stable PS with stable DF showed a higher maximal slope of the action potential duration restitution curve (1.41±0.80 versus 1.09±0.60; P <0.001) and greater local virtual low-voltage burden (2.96±3.23% versus 2.10±2.75%) than regions with stable PS with meandering DF. Stable PS with stable DF correlated with higher AF recurrence at 1-year follow-up (log-rank P =0.002; adjusted hazard ratio, 5.78 [95% CI, 2.18–15.3]). A predictive model including stable PS and DF showed an apparent area under the curve of 0.765 for recurrence prediction. CONCLUSIONS: Stable PS with stable DF identified in virtual heart models was associated with adverse model-derived regional substrate features and higher recurrence rates after ablation. Virtual heart-based identification of these markers may improve prediction of ablation response in persistent AF, warranting prospective validation.