Rushlow (EAGLE adoption)

Clinician Adoption of an Artificial Intelligence Algorithm to Detect Left Ventricular Systolic Dysfunction in Primary Care

Patient / Population Intervention / Exposure Comparison Outcome

In 11,573 patients seen by 165 intervention-arm clinicians at 48 practice sites of a US Midwest health system in the EAGLE cluster RCT, high clinician adoption of the AI-ECG low-EF alert (ordering the prompted echocardiogram) increased diagnosis of low ejection fraction among AI-ECG-positive patients compared with low adoption of the alert.

11,573

Design

Secondary analysis of the EAGLE cluster RCT intervention arm, USA (non-randomised comparison of clinician groups)

Endpoint

Diagnosis of low EF among patients with a positive AI-ECG, high vs low adopters

Relevance

ResultAmong AI-ECG-positive patients, low-EF diagnosis 33.9% with high adopters (n=41) vs 16.9% with low adopters (n=124); OR 1.62, 95% CI 1.21–2.17. High adopters were more often advanced practice providers and Family Medicine clinicians, with less complex patients.

Rushlow DR, et al. Mayo Clin Proc. 2022;97(11):2076-2085. 10.1016/j.mayocp.2022.04.008

Discussion & critique

The implementation counterpart of EAGLE: the tool only found low EF when the clinician ordered the echocardiogram the alert suggested, and only 41 of 165 clinicians were high adopters. As a within-arm comparison of clinician groups rather than a randomised contrast it is confounded by clinician type and patient complexity, but it is the clearest demonstration that adoption, not algorithm accuracy, limited the AI-ECG effect.