Diagnostic accuracy of a smartphone application for artificial intelligence-based interpretation of 12-lead ECG in primary care (AMSTELHEART-1)

Abstract

Abstract Background The use of 12-lead electrocardiogram (ECG) is common in routine primary care, however it can be difficult for less experienced ECG readers to adequately interpret the ECG. Objective To validate a novel smartphone application as a stand-alone interpretation tool for 12-lead ECG in primary care. Methods We recruited consecutive patients who underwent 12-lead ECG as part of routinely indicated primary care in the Netherlands. All ECGs were assessed by a smartphone application which analyses a photographed image of 12-lead ECG for automated interpretation, installed on an Android platform and an iOS platform. We validated the application for detecting: 1) any major ECG abnormality (MEA, primary outcome) defined as atrial fibrillation/flutter (AF), markers of (past) myocardial ischemia or clinically relevant impulse and/or conduction abnormalities; or 2) AF (key secondary outcome). The reference standard was a blinded expert panel. Results We included 290 patients from 11 Dutch general practices with median age 67 (IQR 55-74) years, 48% were female. On reference ECG, 71 patients (25%) had MEA, 35 (12%) had AF. The app's sensitivity and specificity for MEA were 86% (95%CI:76-93) and 92% (95%CI:87-95), respectively. For AF, sensitivity and specificity were 97% (95%CI:85-100) and 99% (95%CI:97-100), respectively. Performance was comparable between Android and iOS platform (Kappa=0.95, 95%CI:0.91-0.99 and Kappa=1.00, 95%CI:1.00-1.00 for MEA and AF, respectively). Conclusions A smartphone application that interprets photographed 12-lead ECG images had good diagnostic accuracy in a primary care setting for major ECG abnormalities, and near-perfect properties for diagnosing AF.