Compared with PV Isolation Alone, AI-Guided Ablation Lowers AF at 12 Months

Tailored cardiac ablation guided by artificial intelligence (AI) results in significantly improved long-term outcomes compared with pulmonary vein isolation (PVI) alone for persistent atrial fibrillation (AF), according to the randomized TAILORED-AF trial.

The findings represent significant headway in the search for more effective adjuncts to PVI in this cohort, which other trials like CAPLA have so far failed to demonstrate.

“This is the first large-scale, randomized, multicenter trial in persistent AF to show a significant superiority regarding AF outcome of an AI-guided ‘PVI-plus’ procedure versus PVI alone,” said Isabel Deisenhofer, MD (German Heart Center Munich, Germany), who presented the findings in a late-breaking trial presentation at Heart Rhythm 2024. “We feel that the use of AI for this trial’s result was pivotal because we finally could have an objective, reproducible, and reliable identification of ablation targets and you can have an outlook. The procedures will become more swift and faster due to automatization of the workflow.”

“We all know that the results with PVI leave a lot to be desired,” Jonathan Piccini, MD (Duke University Medical Center, Durham, NC), who was not involved in the study, told TCTMD. There have been “some wins” in trying to improve it, but “it’s really been a loss record in excess of 90%.”

He called the technology used in TAILORED-AF special not only because it showed a positive result, but also because it identifies spatial dispersion using a machine learning-based approach. Importantly, that’s based on the “fundamental concept” that most electrophysiologists already believe in, which is that if you can identify and ablate the important targets, “you can terminate atrial fibrillation and decrease the likelihood that it can return,” he said. “[Previous] trials all demonstrated that the technologies we use to identify those areas were flawed, and so it may be that what we've had difficulty identifying with more traditional computational methods, the machine learning algorithm and the AI approach is able to identify those areas with greater consistency and accuracy that allows us to see the treatment effect that we observe in TAILORED-AF.”

TAILORED-AF Success

For the study, Deisenhofer and colleagues randomized 374 patients with persistent AF (mean age 65.7 years; 79% male) from 26 sites in five countries to traditional anatomical PVI (n = 186) or a tailored procedure (n = 188) involving use of an AI-based algorithm (Volta Medical; Providence, RI) to tag and map dispersion areas for a targeted ablation approach. After a 3-month blanking period, patients were required to follow-up with a minimum of three office visits at 3, 6, and 12 months with 12-lead ECG and 24-hour Holter monitoring, as well as weekly six-lead ECGs using the Kardia smartphone app. In all, the researchers collected more than 17,000 ECG transmissions with 83% compliance.

The trial was designed to meet the US Food and Drug Administration’s investigational device exemption (IDE) regulations for approval.

Safety was seen for both study arms, with no differences observed in the composite endpoint of death, cerebrovascular accident, or serious treatment-related adverse events.

Both rates of acute AF termination by ablation (66% vs 15%) as well as acute sinus rhythm conversion by ablation (53% vs 13%) were significantly higher in the AI-tailored arm compared with anatomical PVI (P < 0.001 for both). Notably, procedures in the tailored group were generally longer than those involving anatomical PVI alone (178 vs 92 minutes; P < 0.001), with longer fluoroscopy and radiofrequency times due to the additional 3D mapping required.

In the modified intention-to-treat analysis (n = 357)—which excluded patients deemed ineligible after randomization, those that did not have any AF ablation, and those lost to follow-up—tailored ablation was associated with greater freedom from AF after a single procedure at 12 months with or without antiarrhythmic drugs compared with PVI alone (88% vs 70%; P < 0.0001).

“This difference of 18%, of course, was even higher than we expected,” Deisenhofer reported. Similar—albeit “more impressive”—findings were observed in the per-protocol analysis (n = 269), which included only patients who were treated successfully without protocol deviations and attended all follow-up visits, she said.

Tailored ablation, however, did not come out ahead of PVI alone in the per-protocol analysis for the secondary endpoint of freedom from any documented atrial arrhythmia after a single index ablation procedure (P = 0.09). AI guidance, however, was superior regarding freedom from any documented atrial arrhythmia after one or two ablation procedures (P < 0.01), with a mean of 1.2 procedures performed in the study.

Also, in a prespecified subgroup of patients who had AF for at least 6 months, the tailored procedure again showed superior outcomes compared with anatomical PVI for the primary and both secondary endpoints.

“One of the most impressive things is that after one ablation, you can see already that in these high burden A-fib patients, the tailored arm is already better after a single procedure, which is mainly also due to the fact that the anatomical group does not profit that much from a single PVI,” Deisenhofer explained. “So, more complex A-fib, less efficacy of PVI.”

Outcomes ‘Can Only Get Better’

Tina Baykaner, MD, MPH (Stanford University, Palo Alto, CA), who discussed the results during the session, called out TAILORED-AF for being the “first large, multicenter trial in a true persistent AF cohort to show any benefit of a PVI-plus approach.” Further, she said, these findings can be directly compared with those of the STAR AF trial, which showed an advantage of adding complex fractionated electrogram ablation to PVI in 2010, “because the PVI-alone arm is almost identical to what was shown there. And the extra benefit of this approach becomes much more obvious compared to the other approaches done at the time.”

The AI-based approach is limited by longer procedure times and the fact that the model is “not static [and] can always change,” Baykaner said. However, “the AI model, I think, can only be optimized to learn better mechanistic signals, better long-term outcomes. And the outcomes, I think, can only get better.”

This is important, she concluded, because “even with the newer ablation modalities, we're not doing too well in these patients with any approach.”

Piccini acknowledged that the addition of AI in the study was not “a small additional thing that was done,” pointing out that the procedure time was almost doubled compared with standard PVI and that there were numerically more minor procedural complications.

Interestingly, the treatment success seen for PVI alone “was actually maybe a little bit better than you would have expected with patients with such advanced atrial fibrillation,” he said. “In some ways, that makes the results even more impressive because if the standard-of-care PVI arm did that well, that to me makes the tailored result even more important.”

Lastly, Piccini pointed out that the study’s primary endpoint, unlike previous studies, only included AF, not flutter. As such, it’s notable that Deisenhofer and colleagues did not show significance of the tailored approach for the secondary endpoint of freedom from any atrial arrhythmia. “It's important to point out if the metric we're judging is any abnormal rhythms from the top chamber of the heart, then the result is maybe not as impactful as we might think,” he said, adding that follow-up clinical trials should help clarify this issue.

Piccini said his institution is “investigating” this technology for use in his lab and he’d be open to using it on patients who meet the entry criteria for the trial, although there remain several logistical issues to work out regarding embedding the plethora of AI tools becoming available. “This is the highest level of evidence—a randomized clinical trial—and these are a group of patients where there haven't been a lot of interventions that have been proven to improve outcomes,” he said.