Independent Retrospective Evaluation of an Artificial Intelligence-Guided Pathway for Rapid Identification of Acute Coronary Occlusion in Emergency Department Patients

Background Identification of patients with acute coronary occlusion requiring emergent intervention remains a diagnostic challenge for emergency department (ED) physicians, with 25-50% of cardiac catheterization lab (CCL) activations resulting in no intervention. Our goal in this study was to determine whether a pathway incorporating artificial intelligence (AI)-based ECG interpretation could rapidly identify a subset of patients who do not have acute coronary occlusion and therefore did not require emergency CCL activation. Methods We retrospectively analyzed patients for whom the CCL was activated by the ED physician for suspected STEMI between 1/1/2020 to 12/31/2023. ECG tracings were analyzed by Queen of Hearts™ AI ECG model (PMcardio, Powerful Medical). Standard care and AI-guided pathways were assessed. The primary outcome was acute coronary occlusion defined by coronary angiography. Sensitivity and specificity of the two pathways for identification of acute coronary occlusion were compared using paired-data methods. Results We studied 495 encounters; median age was 63 (25 th , 75 th 56, 73) years, and 360/495 (73%) were male. Among 260 encounters with acute coronary occlusion, sensitivity was 98.1% (95% CI 95.6%-99.2%) for standard care and 95.4% (95% CI 92.1-97.3%) for AI-guided pathways (difference -2.7%, 95% CI -5.0 to -0.3%). Among 235 encounters without acute occlusion, specificity was 60.9% (95% CI 54.5-66.9%) for standard care and 76.2% (95% CI 70.3%-81.2%) for AI-guided pathways (difference 15.3%, 95% CI 10.3%-20.3%). Conclusion Among patients for whom the ED physician has initiated CCL activation, a pathway incorporating AI-based ECG interpretation increases specificity with a small decrease in sensitivity.