“Given the broad and growing use of smartwatches and ready accessibility of downloadable mobile applications, this approach may ultimately be applied to perform [atrial fibrillation] detection at large scale, ultimately leveraging common wearable devices to guide [atrial fibrillation] management and rhythm assessment,” the researchers wrote.
The study included three cohorts, each with a different study purpose. The first was made up of 9,750 people in the Health eHeart Study, which enrolls people with an Apple Watch and for this study have the Cardiogram mobile application, according to researchers. Individuals enrolled in the Health eHeart study were sent an invite to participate in this research.
This cohort was used for algorithm development and training, with data being collected remotely. Researchers built a neural network that transformed raw sensor measurements like heart rate and step counts into a sequence of scores that related to the probability a participant was in atrial fibrillation at each time interval, according to the study.
The network was then trained to detect atrial fibrillation when a user was in workout mode.
In the next cohort, researchers enrolled 51 patients with atrial fibrillation. These participants underwent an electrical cardioversion. Then researchers applied a 12-lead ECG and an Apple Watched with the Cardiogram app to participants' wrist for at least 20 minutes when they were in workout mode, according to the study. The Apple Watch data was then fed into the neural network. However, all rhythm diagnosis was determined by the ECGs and over seen by cardiac electrophysiologists.
Researchers also recruited a smaller group from the remote cohort who had self-reported atrial fibrillation to perform exploratory analysis for detecting the condition from ambulatory data. Researchers looked at the self-reported data against the smartwatch data.
The authors of the study found that in the second cohort the deep neural network had a C statistic of .97 to detect atrial fibrillation against the reference standard, 12-lead ECG-diagnosed atrial fibrillation. Among the 51 patients with atrial fibrillation, the deep neural network's sensitivity was 98 percent and specificity was 90 percent.
“This external validation demonstrates that the neural network can passively detect [atrial fibrillation] from smartwatch data with excellent performance characteristics obtained in sedentary individuals captured at high temporal resolution (ie, Workout mode),” researchers wrote. “Even within these constraints, public health implications for [atrial fibrillation] screening may be broad because periods of sleep can provide long, uninterrupted periods of sedentary data, and it is technically feasible to enable high temporal resolution data collection at scheduled periods.”
In the third cohort, where people self reported their atrial fibrillation, the sensitivity was 67 percent and specify was 67 percent.
Researchers noted that one of the limitations of the study was that many people did not finish the survey and link to the Cardiogram account. In addition, the fact that all participants were required to own an Apple Watch before the study could leave out certain individuals and introduce bias.
However, the researchers were positive about using this technology in the future.
“Atrial fibrillation is the leading cause of stroke, and its detection is difficult because of its often asymptomatic nature and paroxysmal frequency,” authors of the study wrote. “Readily accessible means to detect and screen for silent [atrial fibrillation] are needed. Even though monitors with automated capabilities, such as implantable loop recorders, can be used to detect [atrial fibrillation], they are invasive, expensive, and inconvenient. The ideal instrument for [atrial fibrillation] detection would be noninvasive and provide real-time, accurate [atrial fibrillation] detection in a passive fashion — specifically, not requiring the user to remember to perform some action and not limited to any one snapshot in time. Smartwatches are well positioned to accomplish these goals in a cost-efficient and re- source-efficient fashion.”
Cardiogram helped fund the study.