TrialsgrambetaRCTs are beautiful

EAGLE (AI-ECG)

2021

Artificial Intelligence-Enabled Electrocardiograms for Identification of Patients with Low Ejection Fraction: a Pragmatic, Randomized Clinical Trial

In 22,641 adults without known heart failure having a routine ECG, clinician access to an AI-ECG low-ejection-fraction prediction increased new diagnoses of low ejection fraction within 90 days compared with usual care without the AI result.

Result

Efficacy: New diagnosis of low EF (≤50%) within 90 days of the ECGOR 1.32 (CI 1.01-1.61) · P = 0.007

Below 1 favours usual care without the AI result, above 1 favours clinician access to an AI-ECG low-ejection-fraction prediction.

Event rates
Absolute difference
0.5 points
NNT
200

Absolute difference and NNT are derived from the event rates.

Also reported: Among AI-ECG-positive patients (6% of the cohort) 19.5% vs 14.5% (OR 1.43, 1.08–1.91; P=0.01); echocardiogram use similar overall (19.2% vs 18.2%, P=0.17).

Original publication
Subtopic
Heart failure & cardiomyopathy
Design
Pragmatic cluster RCT, USA (120 primary care teams)
Relevance
Practice-defining: a trial the field needs to know.
Source
Yao X, et al. Nat Med. 2021;27(5):815-819.doi 10.1038/s41591-021-01335-4

Discussion & critique

First randomised evidence that an AI-ECG decision-support tool changes diagnosis in routine primary care, and the template for the later AI-ECG case-finding trials. The absolute gain was small, the endpoint is a diagnosis at 90 days rather than a clinical outcome, and the trial was open-label by nature. Its gains came only when clinicians acted on the alert: clinician adoption was the limiting step (see the Rushlow entry), and TRICORDER later showed the same mechanism with an AI stethoscope.