AI-Assisted Electrocardiogram Interpretation Improves ST-Elevation Myocardial Infarction Diagnostic Accuracy Among...

Timely and accurate diagnosis of ST-elevation myocardial infarction (STEMI) is critical in military operational environments where evacuation may be delayed. Although artificial intelligence (AI) electrocardiogram (ECG) tools have demonstrated high diagnostic performance, their effectiveness among advanced practice providers (APPs) remains untested. This study evaluated whether AI-ECG interpretation by Queen of Hearts (QoH) AI software by PMcardio improves STEMI diagnostic accuracy, clinician confidence, and time-to-decision among APPs.

This prospective, randomized, multi-reader, multi-case crossover study enrolled 21 certified physician assistant (PA-Cs) at Carl R. Darnall Army Medical Center, Fort Hood, TX. Participants interpreted 50 de-identified 12-lead ECGs (25 STEMI, 25 non-STEMI) under 2 conditions: with and without AI assistance. Each participant served as their own control. The primary outcome was diagnostic accuracy. Secondary outcomes included confidence (5-point Likert scale) and time-to-decision. Data were analyzed using paired t-tests and a random-effects crossover model, with analysis facilitated by the Baylor University Statistical Consulting Center.

AI-ECG interpretation significantly improved diagnostic accuracy (92.9% vs. 79.0%; P < .001; Cohen d = 1.68). With AI assistance, sensitivity was 95.4%, and specificity was 90.5%, compared to 82.5% and 75.6% without AI, respectively. Interrater agreement improved from moderate (κ = 0.58) to excellent (κ = 0.86). Confidence was significantly better with AI (mean Likert total 46.8 vs. 56.9; P < .001; d = -1.02, lower score indicates higher confidence). Cumulative time-to-decision was significantly longer with AI (18.8 vs. 13.0 minutes; P < .001; d = 1.13). The average increase in time per test strip was 14.7 seconds.

AI-assisted ECG interpretation significantly improved STEMI diagnostic accuracy and clinician confidence among PA-Cs. Although AI increased time-to-decision, the mean increase of approximately 14.7 seconds per ECG is unlikely to be clinically meaningful. These findings support the integration of AI-ECG decision-support tools into frontline and operational settings to enhance timely and accurate STEMI recognition.