Reducing the burden of inconclusive smart device single-lead ECG tracings via a novel artificial intelligence algorithm

• Published in: Cardiovascular digital health journal 2024-02, Vol.5 (1), p.29-35
• Main Authors: Weidlich, Simon, Mannhart, Diego, Kennedy, Alan, Doggart, Peter, Serban, Teodor, Knecht, Sven, Du Fay de Lavallaz, Jeanne, Kühne, Michael, Sticherling, Christian, Badertscher, Patrick
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Atrial fibrillation; Digital health; Original; Single-lead electrocardiogram; Smart device; Smartwatch
• ISSN: 2666-6936; 2666-6936
• DOI: 10.1016/j.cvdhj.2023.12.003

Multiple smart devices capable of automatically detecting atrial fibrillation (AF) based on single-lead electrocardiograms (SL-ECG) are presently available. The rate of inconclusive tracings by manufacturers’ algorithms is currently too high to be clinically useful. This is a prospective, observational study enrolling patients presenting to a cardiology service at a tertiary referral center. We assessed the clinical value of applying a smart device artificial intelligence (AI)-based algorithm for detecting AF from 4 commercially available smart devices (AliveCor KardiaMobile, Apple Watch 6, Fitbit Sense, and Samsung Galaxy Watch3). Patients underwent a nearly simultaneous 12-lead ECG and 4 smart device SL-ECGs. The novel AI algorithm (PulseAI, Belfast, United Kingdom) was compared with each manufacturer’s algorithm. We enrolled 206 patients (31% female, median age 64 years). AF was present in 60 patients (29%). Sensitivity and specificity for the detection of AF by the novel AI algorithm vs manufacturer algorithm were 88% vs 81% (P = .34) and 97% vs 77% (P < .001) for the AliveCor KardiaMobile, 86% vs 81% (P = .45) and 95% vs 83% (P < .001) for the Apple Watch 6, 91% vs 67% (P < .01) and 94% vs 82% (P < .001) for the Fitbit Sense, and 86% vs 82% (P = .63) and 94% vs 80% (P < .001) for the Samsung Galaxy Watch3, respectively. In addition, the proportion of SL-ECGs with an inconclusive diagnosis (1.2%) was significantly lower for all smart devices using the AI-based algorithm compared to manufacturer’s algorithms (14%–17%), P < .001. A novel AI algorithm reduced the rate of inconclusive SL-ECG diagnosis massively while maintaining sensitivity and improving the specificity compared to the manufacturers’ algorithms.