New research shows single-lead electrocardiogram (ECG) tracings from an Apple Watch interpreted by an artificial intelligence (AI) algorithm developed by the Mayo Clinic effectively identified left ventricular (LV) dysfunction in a nonclinical setting.
Findings from the decentralized, prospective study were presented May 1, 2022, at Heart Rhythm 2022.
"We have seen how artificial intelligence has revolutionized the already common ECG into a tool that can be used to identify occult cardiovascular diseases. Our team saw vast potential to expand tracking outside of a physician's office by using popular wearable devices," said presenting author Zachi Attia, PhD, codirector of AI in cardiology, Mayo Clinic, in a press release.
"We set out to create a platform that could not only provide accurate readings, but also would yield high patient engagement with an easy to navigate, user-friendly process that can be completed from the comfort of a patient's home," continued Attia.
To interpret ECG signals generated from the single lead on an Apple Watch, Attia and colleagues modified an established 12-lead algorithm to identify LV dysfunction. The team hypothesized that adaption of this network to a single lead Apple Watch ECG would allow for “massive scaling” of this tool for screening and monitoring persons in nonclinical environments.
To test this hypothesis, investigators invited Mayo Clinic patients with the Mayo Clinic iOS app and an Apple Watch via email to participate in the study. Of invitees, 3884 reported watched ownership, and 2454 consented and downloaded the study app, according to the study abstract. Participants were enrolled from 46 US states and 11 countries and 56% were women and the mean age was 53 years.
The app securely sent all previously acquired ECGs and additional ones as they were recorded by patients to a Mayo Clinic secure data platform for clinician review. App use and engagement were analyzed. Apple Watch ECGs acquired within 1 month of a clinically ordered ECG were analyzed by AI for the presence of ejection fraction (EF) ≤40% using the model researchers adapted for single lead use.
Participants uploaded 125 610 ECGs between August 2021 and February 2022, with 92% of subjects using the app more than once (average, 2.1 times/month). Of the participants, 421 had at least 1 sinus rhythm (NSR) ECG (average, 17 ECGs, with NSR determined by watch algorithm) within 30 days of an ECG, of whom 16 (3.8%) had an EF ≤40%.
Thirteen of the 16 patients with an EF ≤40% were identified by the watch AI ECG, which had an area under the curve (AUC) 0.875, sensitivity 81.2%, and specificity 81.3%, according to the abstract.
"For patients who might unknowingly have this condition, such as those with hypertension, diabetes, advancing age, and people receiving some forms of chemotherapy, the tool could enable early detection and help physicians optimize treatment options," said coauthor Paul Friedman, MD, professor of medicine, Mayo Clinic, in the press release. "This technology has the potential to be scaled and adopted by hospital systems to better serve patients, particularly in remote communities or geographically diverse populations around the world, potentially addressing health care disparities, and enabling physicians to offer more coordinated patient care."
Attia et al are currently seeking FDA approval for the algorithm used in this study and suggest that future studies test additional AI algorithms developed by the team. In addition, researchers plan to expand their algorithm for additional data collection and to screen for other common cardiovascular conditions, such as atrial fibrillation.
The study had no financial or technical support from Apple.
Reference: Attia Z, Dugan J, Friedman PA, et al. Artificial intelligence to identify left ventricular dysfunction from an apple watch ECG: A prospective, decentralized international pragmatic study (abstract LB-736). Presented May 1, 2022, at Heart Rhythm Society 2022 Scientific Sessions.