By Joshua D. Moss, MD

Associate Professor of Clinical Medicine, Cardiac Electrophysiology, Division of Cardiology, University of California, San Francisco

Dr. Moss reports he is a consultant for Abbott, Boston Scientific, and Medtronic.

SYNOPSIS: A large study with virtual enrollment of Apple Watch users helped illustrate the positive predictive value of wearable, pulse-based atrial fibrillation detection technology, as well as the ability to enroll and follow huge numbers of research subjects in a short period.

SOURCE: Turakhia M, Perez M. Results of a large-scale, app-based study to identify atrial fibrillation using a smartwatch: The Apple Heart Study. Presented March 16, 2019, at the 2019 American College of Cardiology Scientific Sessions, New Orleans.

The Apple Watch can measure heart rate and regularity via photo plethysmography, which uses changing absorption of green or infrared light to detect the pulse in the wrist. The goal of the Apple Heart Study was to evaluate the ability of this watch to identify subjects with atrial fibrillation (AF) and help guide subsequent clinical evaluation.

U.S. residents who were at least 22 years of age, did not already have a diagnosis of AF or atrial flutter, and were not using anticoagulation were given the opportunity to enroll virtually via their Apple Watch and iPhone. Over eight months, 419,297 subjects enrolled. Once enrolled, the watch would take periodic, opportunistic measurements of the pulse rate and regularity via generation of a tachogram. A positive finding, suggestive of AF, would trigger more frequent passive measurements. Then, five confirmations would generate a notification to the subject of an irregular pulse. Those subjects were connected to a telehealth doctor who could refer them for additional care or mail them an ECG patch. The primary endpoints were AF confirmed by the ECG patch in subjects ≥ 65 years of age and simultaneous AF noted on the ECG patch and via the watch.

The mean age of enrolled subjects was 41 ± 13 years. Eighty-four percent of subjects were younger than 55 years of age. A majority of subjects were younger than 40 years of age. About 6% of the enrolled population were ≥ 65 years of age. Over the course of the study, 2,161 subjects (mean age, 57 years) received an irregular pulse notification. The notification rate was about 3% in subjects ≥ 65 years of age, about 0.37% in subjects 40-54 years of age, and about 0.16% in subjects 22-39 years of age.

Of these 2,161 subjects, 945 completed a first telehealth visit, and 658 went on to receive an ECG patch. A total of 450 subjects wore the patches and returned them for analysis (0.1% of the original cohort and 21% of those who received irregular pulse notifications). The CHA2DS2-VASc score was ≥ 2 in 13% of the original cohort, 33% of the cohort who received an irregular pulse notification, and 38% of those who wore and returned an ECG patch. The patch yielded a diagnosis of AF for 34% of subjects. The longest episode was one hour or longer in 89% of subjects with AF on the patch. A 90-day survey of subjects who received an irregular pulse notification revealed that 15% of those subjects had received an AF diagnosis prior to study enrollment. While wearing the patch and the watch, an irregular tachogram carried a positive predictive value of 0.71 for true AF (0.60 in patients ≥ 65 years of age). An irregular pulse notification carried a positive predictive value of 0.84 for true AF (0.78 in patients ≥ 65 years of age).


The Apple Heart Study received a great deal of attention both at the American College of Cardiology Scientific Sessions and in the media — and for good reason. It was remarkable in its size and scope, particularly considering that enrollment and data collection occurred over only about eight months. The data accrued add valuable information about the demographics of AF in the United States, although limited to a small subset of the population with the means and desire to purchase and wear an Apple Watch, as well as the willingness to complete the less passive portions of the study (only 21% of the enrolled subjects who received an irregular pulse notification went on to wear and return an ECG patch for analysis).

It is safe to say that the study confirmed the ability of the watch to detect AF via background monitoring. Virtually all cardiologists who treat AF commonly have now met patients who discovered their arrhythmia via a watch notification. However, the actual sensitivity of the algorithm is unknown, considering the lack of gold-standard diagnostic information (or almost any information) about the vast majority of patients who never received an irregular pulse notification. How many of them actually had AF while wearing the watch? A basic analysis of the available data from the tiny subset (0.1% of the overall cohort) who simultaneously wore an ECG patch and the watch suggests that 81 of the 450 must have had true AF on the patch yet received no irregular pulse notification, compared with 72 who had AF and did receive a notification. The calculated “instantaneous” sensitivity is only 47%, although the watch had generated an irregular pulse notification for every one of those patients at some point. False-positive pulse notifications were less common, but about 5% of patients without AF on the ECG patch still received a notification from their watch.

It is interesting that the positive predictive value for watch notifications was slightly lower in older patients compared with the rest of the population. A similar analysis of the cohort of ECG patch wearers who were ≥ 65 years of age suggests an “instantaneous” sensitivity of only 40% and specificity of 94%. While chance could play a role, I suspect older patients simply experience more frequent atrial and ventricular ectopy, the irregularity of which could result in a false-positive tachogram. In a large population, even a relatively small rate of false-positive indications (which will be amplified with longer periods of wearing the watch) may generate a tremendous amount of unnecessary anxiety, testing, and treatment. Therein lies one potential peril. Treatment of AF and anticoagulation for stroke prevention both carry risks, and confirmatory monitoring will undoubtedly reveal other abnormalities that may or may not have become clinically significant. On the other hand, if the true sensitivity of the watch is less than 50%, there also is risk of false reassurance and undertreatment of true disease.

That said, I find it difficult to argue that patients should not be empowered to monitor themselves for conditions that potentially put them at risk. Any motivated patient can check their own pulse many times per day; the watch simply makes that process easier. The added burden to clinicians will come largely from the need for much more discussion with patients about the relative merits and risks of all the potential downstream tests, medications, and procedures.

The healthcare community will need to learn how to build an infrastructure to handle this added burden, which will undoubtedly extend to more disease processes going forward. In the meantime, we will need to use the opportunity and technology to conduct more rigorous studies, even if on a smaller scale, to test the true benefit and risk of continuous pulse screening so that patients can be advised accurately.