Blinded, Randomized Trial of Sonographer versus AI Cardiac Function Assessment

He, B.

doi:10.1038/s41586-023-05947-3

EchoNet-RCT

Blinded, Randomized Trial of Sonographer versus AI Cardiac Function Assessment

Patient / Population Intervention / Exposure Comparison Outcome

In 3,495 echocardiograms undergoing quantification of cardiac function, initial AI assessment of left ventricular ejection fraction was non-inferior to initial sonographer assessment for the proportion of readings the cardiologist substantially changed, and superior.

N

3,495patients

Design

Blinded, randomised non-inferiority trial

Endpoint

Proportion of studies with substantial (>5%) change in LVEF between initial AI or sonographer assessment and final cardiologist assessment

Relevance

1Practice-defining — the trial the guideline rests on.

ResultSubstantial change 16.8% (AI) vs 27.2% (sonographer); difference −10.4% (95% CI −13.2% to −7.7%; P<0.001 for non-inferiority and for superiority). Mean absolute difference from a prior independent cardiologist 6.29% vs 7.23% (difference −0.96%, 95% CI −1.34% to −0.54%); blinding index 0.088.

He B, et al. Nature. 2023;616(7957):520-524. 10.1038/s41586-023-05947-3

Discussion & critique

The first blinded randomised trial of AI in cardiac imaging: cardiologists edited AI-initiated LVEF less often than sonographer-initiated values, with time saved for both, establishing the design template for AI-versus-human reader trials. The endpoint is cardiologist editing rather than patient outcome, 274 poor-quality studies were excluded before randomisation, and a cardiologist still finalised every report, so the trial tested AI-assisted workflow rather than autonomous AI.