Peer reviewed AI-assisted

Randomised trial of AI-ECG guided monitoring for undetected atrial fibrillation

In a single-centre trial in Taiwan, 14,665 adults without known atrial fibrillation were randomised to have their past year of sinus-rhythm ECGs analysed by an AI model, with the 1.1% it called high risk offered up to seven days of Holter monitoring, and the composite of newly detected atrial fibrillation, flutter or pre-atrial-fibrillation findings at 90 days was 0.74% against 0.48%, a result the authors describe as statistically fragile.

Model
An institutional AI-ECG model scoring atrial electrical vulnerability from sinus-rhythm ECGs
Field
Medicine
Date
2026-11-07
Human collaborators
Chin Lin, Chin-Chou Lee, Wen-Yu Lin

What was found

Atrial fibrillation often goes undetected until a stroke, and the attraction of an AI-ECG model is that it claims to see a signature of atrial vulnerability on an ordinary ECG recorded in sinus rhythm, when nothing is visible to a reader. Adults aged 40 to 85 with no previous atrial fibrillation or flutter were randomised to an AI-ECG-guided strategy or usual care, 7,337 against 7,328. Within the intervention arm, participants whose ECGs from the preceding year were scored high risk were offered Holter monitoring; 84 participants (1.1%) were flagged and 48 of them (57.1%) were monitored. The design deals with the obvious bias that monitoring finds what monitoring looks for: each monitored participant was paired with an age- and sex-matched control chosen before the control arm's ECGs were analysed, and those controls received identical monitoring, with reporting cardiologists blinded to the AI results and to allocation. The primary composite occurred in 54 intervention participants against 35 controls, an odds ratio of 1.54 (95% CI 1.01 to 2.37, P=0.046P = 0.046), and the effect was concentrated as designed, at 32.1% against 2.6% among high-risk participants and 0.4% against 0.5% among the rest.

Novelty check

Europe PMC, 2026-11-09: TITLE:("AI-ECG" OR "AI-enabled ECG" OR "artificial intelligence-enabled electrocardiogram" OR "artificial intelligence electrocardiogram" OR "artificial intelligence-enabled electrocardiography") AND ("atrial fibrillation") AND (randomized OR randomised) returned 25 records. The closest randomised work is the VITAL-AF trial analysis in JACC in April 2026, which applied AI-ECG models to risk-guide screening within an existing trial rather than randomising an AI-guided monitoring pathway; the rest are retrospective, reviews or other endpoints. This trial's own design paper appeared in Acta Cardiologica Sinica in September 2026. This registry holds 2021-05-06-eagle-ai-ecg-low-ejection-fraction, which is the same institution's family of methods applied to a different target. Despite the shared phrase AI-ECG, the model here is the centre's own and is not the Anumana product tracked separately.

Caveats and known objections

Single-centre and open-label, with the model's developers running the trial, so no independent validation. The result is fragile on the authors' own account: reclassifying a single event as a non-event moves the primary PP value from 0.046 to 0.057. The composite endpoint carries most of the signal in pre-atrial-fibrillation findings, frequent or short runs of premature atrial complexes and non-sustained episodes, whose clinical meaning is unsettled; detection of actual atrial fibrillation or flutter was 0.3% against 0.2% and not significantly different, and initiation of antiarrhythmic therapy at 180 days was 2.3% against 2.4%, essentially identical. No stroke, mortality or anticoagulation outcome was measured, and at 90 days none could be. Only 57.1% of flagged high-risk participants actually underwent monitoring, so the pathway as delivered reached just over half the people it identified. Monitoring itself caused participant-reported harms in 22.9% of monitored intervention participants against 18.8% of matched controls, mostly skin reactions and discomfort. The ClinicalTrials.gov record still lists the trial as recruiting with an estimated completion of 2026-10-01 and an estimated enrolment of 14,726, against the 14,665 randomised in the published report, so the registration has not been updated to match. Autonomy is ai-assisted: the model selected who was offered monitoring, and blinded cardiologists read every recording and made the diagnosis.

Trial registration

ClinicalTrials.govLeveraging AI-ECG Technology for Early Notification and Tracking of AF Development

NCT06847932

That a study was registered, with the outcomes it declared and the date it declared them, so a published endpoint can be checked against the one planned. It does not certify the results, their analysis, or any clinical benefit; registry records are submitted by the sponsor and are not peer reviewed.

No independent check is recorded on this entry yet.

Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.

Or on GitHub: submit a check challenge the grade send a correction or send a pull request

Entry history (1 event)
  1. AddedEntered the registry graded Peer reviewed and AI-assisted.

Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.

Graded peer reviewed for verification and ai-assisted for autonomy. What these mean.

Cite this entry

Plain text
whataifound.org. (2026). Randomised trial of AI-ECG guided monitoring for undetected atrial fibrillation. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-11-07-latent-ai-ecg-af-rct
BibTeX
@misc{whataifound-nationaldefensemedicalcentertaiwan-2026-rct,
  title        = {Randomised trial of AI-ECG guided monitoring for undetected atrial fibrillation},
  author       = {{whataifound.org}},
  year         = {2026},
  howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
  note         = {Result by National Defense Medical Center, Taiwan. Verification: Peer reviewed. Autonomy: AI-assisted.},
  url          = {https://whataifound.org/finding/2026-11-07-latent-ai-ecg-af-rct}
}

Related findings

← All medicine findings in the registry