TrialsgrambêtaRCTs are beautiful
AHA Scientific Sessions 2026Présenté lun. 9 nov. · 10:22 UTC−6LBS.06 Late-breaking Science: Artificial Intelligence in the Diagnosis of Cardiovascular DiseaseTous les essais Late-Breaking Science →

AI echo interpretation RCT

2026

AI-Assisted Comprehensive Echocardiographic Interpretation: A Blinded, Randomized Controlled Trial

In previously acquired echocardiograms from 21 medical centres, AI-generated preliminary reports finalised by blinded cardiologists were non-inferior for the rate of substantial changes by the finalising cardiologist compared with preliminary reports prepared by sonographers and cardiologists.

Résultat

Not public before the presentation
Différence absolue
0,9 points

La différence absolue est dérivée des taux d’événements.

Également rapporté: The result met the criteria for non-inferiority of the AI-assisted approach. Cardiologists spent an average of 4.6 minutes per echocardiogram with AI vs 8.5 minutes without AI assistance, a time saving of about 46%. AI-assisted reports were more consistent with historical records.

Publication originale
Sous-thème
Echocardiography
Schéma
Blinded randomised controlled trial of AI-assisted comprehensive echocardiographic interpretation
Inclus
1 200 patients
Source
Lee KK, et al. Presented at AHA Scientific Sessions 2026, 9 November. American Heart Association news release. Not yet published.

Discussion et critique

In this blinded, randomised comparison, AI-drafted full echocardiogram reports needed substantial changes about as often as human drafts, and cardiologists finished them in less time. The endpoint is how much reports were edited, which is a stand-in for accuracy. The study did not test whether AI improves patient outcomes, and the cardiologists knew they were in a study, which may have changed how they worked. The release does not report a confidence interval, P value or non-inferiority margin, and performance at larger scale and volume has yet to be studied.