Ablation therapy for persistent atrial fibrillation (AF) has consistently yielded poorer outcomes compared to paroxysmal AF, primarily due to the more advanced disease state and the presence of arrhythmogenic substrates beyond the pulmonary veins (PVs) (1). Focal and rotational sources within the left atrial body contribute significantly to the maintenance of arrhythmia (2); however, reliably identifying effective ablation targets outside the PVs has remained a major challenge (3). Recent advances, most notably the TAILORED-AF trial, have shown encouraging results in addressing this issue, demonstrating the potential of artificial intelligence (AI)-driven, real-time software capable of detecting and classifying specific AF electrical patterns, particularly spatio-temporal electrogram dispersion (4). Notably, selective ablation of atrial sites exhibiting spatio-temporal dispersion has already been associated with high rates of acute AF termination and long-term arrhythmia freedom (5, 6) and the concept of spatio-temporal dispersion as an electrical signature of AF drivers was first introduced nearly two decades ago (7).
The TAILORED-AF trial is the first randomized superiority study to compare AI-guided, tailored ablation plus PV isolation (PVI) versus standard PVI-only in patients with persistent and long-standing persistent AF. Conducted across 26 centers in Europe and the U.S. with 370 patients, the trial showed that AI-detected spatio-temporal dispersion areas alongside PVI is superior to PVI only in eliminating AF at 1-year follow-up. The benefit was even greater in patients with longer AF duration (≥6 months). Although the periprocedural AF termination rate was significantly higher in the tailored arm than in the anatomical arm, the procedure and ablation times were twice as long in the tailored arm. While overall arrhythmia freedom after a single procedure was comparable between groups, repeat procedures in the tailored arm were mainly for atrial tachycardia, whereas in the anatomical arm, they were largely for recurrent AF. This difference likely contributed to the higher rate of arrhythmia-free survival at 12 months in the tailored arm, in a combined analysis of recurrences after one or more ablations.
Despite the encouraging results, several limitations and open questions should be acknowledged. The study does not address whether the observed trade-off is sustainable or cost-effective in real-world clinical practice, particularly given the absence of a control group receiving PVI combined with additional ablation sites without AI support. While AI-guided ablation beyond standard PVI shows promise, it also raises concerns about the risk of unnecessary tissue damage, especially in patients with complex atrial substrates. Additionally, the study offers limited discussion on how the system minimizes false positives in identifying spatio-temporal dispersion. Although the multi-center trial involved 51 operators, the variability in operator experience and the potential learning curve associated with using the system are not explored, factors that may have influenced outcomes. All procedures were performed using radiofrequency ablation; however, the applicability of this mapping strategy to other energy sources, such as pulsed field ablation, was not evaluated. Finally, it remains an open question how AI-guided ablation using a broader dataset, including all available electrogram and clinical data rather than solely expert-adjudicated spatio-temporal dispersion, would compare to the approach used in TAILORED-AF.
The AI-guided ablation represents a promising advance in integrating AI into electrophysiology, but significant questions remain about the robustness, interpretability, and long-term utility of the algorithm. Further validation in real-world, diverse patient populations and clearer reporting of performance metrics are essential before broader clinical adoption.
Our mission: To reduce the burden of cardiovascular disease.