AI-guided, targeted screening can detect undiagnosed AF in high-risk adults
An artificial intelligence algorithm applied to existing ECGs in adults at high risk for stroke can identify signatures of atrial fibrillation risk during normal sinus rhythm, researchers reported in The Lancet.
“Atrial fibrillation is common and it is probably underdiagnosed, and there are people for whom the first manifestation of their AF is stroke,” Peter A. Noseworthy, MD, cardiac electrophysiologist at Mayo Clinic, told Healio. “That is a missed opportunity. We would like to find ways to identify people who have subclinical AF who have this impending risk for stroke and try to prevent it. There is only so much we can do after someone has come to clinical attention.”
As Healio previously reported, the U.S. Preventive Services Task Force has previously stated there is insufficient evidence to make a recommendation on screening asymptomatic patients aged 50 years and older for AF.
“Most guidelines do not recommend it,” Noseworthy said in an interview. “It is too costly, the yield is too low and we do not know if screen-detected AF carries the same risk, so we do not know if treating it is actually going to translate to reduced strokes.”
In the prospective, nonrandomized BEAGLE trial, Noseworthy and colleagues analyzed data from 1,003 adults from 40 states with stroke risk factors and no known AF who underwent an ECG during routine practice (mean age, 74 years). Participants wore a continuous ambulatory heart rhythm monitor (MoMe, InfoBionic) for up to 30 days, with the data transmitted in near real-time through a cellular connection.
Researchers applied the AI algorithm to the ECGs to stratify patients as high risk (n = 633) or low risk (n = 370). The primary outcome was newly diagnosed AF.
In a secondary analysis, researchers used propensity-score matching with adults who were eligible for study enrollment who served as real-world controls.
During a mean 22.3 days of continuous monitoring, researchers observed AF in 1.6% of low-risk participants and in 7.6% of high-risk participants, for an OR of 4.98 (95% CI, 2.11-11.75; P = .0002).
During a median follow-up of 9.9 months, AI-guided screening was associated with increased detection of AF compared with usual care for participants at high risk (10.6% vs. 3.6%; P < .0001) and for participants at low risk (2.4% vs. 0.9%; P = .12); however, the difference was significant only for the high-risk group, with an HR of 2.85 (95% CI, 1.83-4.42).
“Perhaps the algorithm will help us determine which patients who’ve had a cryptogenic stroke are likely to have AF, and could anticoagulation prevent a second stroke?” Noseworthy told Healio. “That is a relatively straightforward question to answer. The other question is whether we can use this algorithm to power population-level AF screening and demonstrate a benefit in terms of actual stroke prevention. That is the biggest question.”
Collapse