Automated atrial fibrillation detection with a smartwatch and smart-ring in individuals with cardiovascular disease

Abstract

Abstract Funding Acknowledgements Type of funding sources: None. Introduction Wearable technology is increasingly used to identify episodes of atrial fibrillation (AF). However, most validation studies using single-lead electrocardiograms (ECGs) only included patients with either sinus rhythm or AF and, therefore, likely overestimate the sensivities and specificities of AF detection algorithms. Purpose We sought to evaluate the diagnostic accuracy of the Apple Watch and CART-I ring AF algorithms in a broad cohort of patients with pre-existing cardiovascular disease. Methods In this single-centre prospective study, individuals with known cardiovascular disease underwent simultaneous 12-lead ECG recordings with a 30-second single-lead ECG from an Apple Watch Series 6 and a CART-I Ring. The order of devices was randomly assigned. Two cardiologists independently evaluated the 12-lead ECGs for the presence of AF; any disagreements were resolved by a third cardiologist. The wearable devices’ algorithm labelled ECGs as "AF", "Not AF" or "Unclassified". Diagnostic accuracy was determined by calculating the sensitivity and specificity for each device and comparing them using McNemar’s test. The primary analysis included Unclassified ECGs, which categorised as false-positive (no AF on the 12-lead ECG) or false-negative (AF on the 12-lead ECG). A separate analysis excluded all Unclassified ECGs. Results We enrolled 400 consecutive patients. Seven patients had suboptimal 12-lead ECGs and were excluded from the analysis. The median age was 63 (58-75) years, and 73% of the population was male. Baseline demographics are depicted in Table 1. A total of 1573 ECGs were analysed. Forty-two percent of patients were in AF, 9.9% were in an atrial tachycardia, 3.3% had sinus rhythm with frequent ectopy. If all ECGs were included, the CART-I ring's AF sensitivity was significantly higher than the Apple Watch, at 86.6% (95% CI: 80.6%-91.1%) and 67.4% (95%CI: 60.0%-74.1%), respectively (Fig 1). Similarly, AF specificity was higher with CART-I ring (89.4% [95%CI: 84.7%-92.8%]) when compared to the Apple Watch (74.9% [95%CI 68.9%-80.1%]). The Apple Watch had more Unclassified ECGs than CART-I ring (78 vs. 2, respectively). They were divided into three categories: 14.1% as HR 50 bpm, 10.21% as HR > 120 bpm, and 76.9% as Inconclusive (Table 2). When Uclassified ECGs from the Apple Watch were excluded, AF sensitivity and specificity improved by 16.33% (p=0.002) and 13.38% (p<0.001), respectively. Conclusion In patients with cardiovascular disease, the diagnostic accuracy of AF detection from a single-lead ECG was lower than previously reported. The CART-I ring's AF sensitivity and specificity outperformed those of the Apple Watch. One in three patients with AF went undetected by the Apple Watch.