Apple Heart Study Raises Myriad Questions

Daniel M. Keller, PhD

March 27, 2019

LISBON — Contrary to the adage that an apple a day keeps the doctor away, the Apple Heart Study rallied doctors on both sides of the Atlantic to consider its results and implications.

The American College of Cardiology (ACC) and the European Heart Rhythm Association (EHRA) held their annual meetings simultaneously, one in New Orleans and the other here in Lisbon.

The Apple Heart Study results were first presented at ACC on March 16, followed by presentation at EHRA 3 days later, including European commentary on the study.

Andrea Russo, MD, professor of medicine at Cooper Medical School of Rowan University, president-elect of the Heart Rhythm Society, and a member of the steering committee of the trial, came to Lisbon to tell EHRA delegates the outcomes of the study.

In brief, this innovative study used an Apple Watch to detect heart rhythm using photoplethysmography (PPG), a noninvasive optical technique using light to detect through the skin blood volume changes in the microvascular bed, calculating time intervals between pulses (tachogram). Five tachogram readings suggestive of atrial fibrillation (AF) triggered a notification to the wearer via an iPhone. Then the user could contact a telehealth doctor to receive an electrocardiogram (ECG) patch to wear and record heart rhythms. After mailing the patch back for analysis, the user could discuss the results with a study doctor.

"The overall goal of the study was to evaluate the ability of the irregular pulse identification algorithm to identify atrial fibrillation and guide subsequent clinical evaluation," Russo told delegates here. Participants were 22 years or older, had an Apple Watch series 1 to 3 and an iPhone 5s or higher with a downloaded study app on it, could not have known atrial fibrillation or atrial flutter, and could not be using an anticoagulant. These earlier models of watch did not have the ability to measure the ECG themselves.

The primary end points of the trial were simultaneous AF on the ECG patch and the individual tachogram or, if the participant was 65 years or older, AF for more than 30 seconds on the patch. The study was prospective, single group, and open-label.

It enrolled 419,297 participants, of whom 24,626 were 65 years or older. During the study, there were 2161 individuals notified (0.5%), of whom 945 (44%) took part in a first (telehealth) visit. ECG patches were shipped to 70% of them, of whom 68% returned the patches for analysis.

For the cohort as a whole, including all ages, "there were 86 positive notifications" for irregular pulse, Russo said, "and out of those, 72 had atrial fibrillation on the external patch, giving a positive predictive value of 0.84," (95% CI, 0.76 - 0.92).

Surveys were sent out at 90 days and at the end of the study. Of the participants who had received notifications, 64% completed the 90-day survey, and 43% completed the one at end of study. Of the 99.5% of participants who never received a notification of possible AF, 70% completed the survey.

Of the 2161 participants who received a notification on their iPhones and completed the 90-day survey, 57% said they contacted a nonstudy physician for follow-up. Of these, 28% reported they started on a new medication, 33% were referred to a specialist, and 36% were sent for additional testing.

Limitations of the study were that it relied on self-assessment of enrollment criteria and outcomes, had a higher than anticipated drop-off rate after notification, had fewer ECG patches analyzed than planned, which decreased study precision, and it used a virtual study design, which was large, quick, and a pragmatic design.

Designated discussant Hein Heidbuchel, MD, PhD, president of EHRA and professor of medicine and chair of cardiology, Antwerp University, Belgium, said: "Even without results, this is a landmark trial." He explained that it is a harbinger of a new era in medicine. Whereas doctors have traditionally ordered tests for patients, "this trial shows that the future may be the patients test themselves and send the [results] to us.... That's a real game changer."

He said another major change in process was that the patients were managed with a "virtual" first encounter and were not seen in person.

The trial was also of interest because the medical profession has for years considered the issue of screening for atrial fibrillation "before strokes strike," but what has been unknown is the minimum duration of silent AF that increases the risk for stroke, which may be avoidable by intervening earlier.

Heidbuchel also asked about the optimal duration needed to monitor patients at risk to find AF, and how to do it cost-effectively. He also pointed to the issue of diagnostic accuracy in terms of sensitivity and specificity, potentially creating false negatives and false positives. As with any new technology, one has to decide which part of the population to use it with to optimize sensitivity, specificity, and, therefore, positive and negative predictive value in a cost-effective way. He said that he was impressed with the 84% positive predictive value for AF shown in this study.

The algorithm used by the device requires five of six tachogram measurements to be positive for it to make a notification, which enhances the specificity. As expected, notifications occurred among a higher proportion of users in the 65 years and older age group. He said he thinks that the newer, current generation of the watch, which can record an ECG, will result in "dramatically improved" accuracy if ECG recordings can be triggered by a positive signal from the PPG monitoring. "This, for me, is the real strength of the trial."

Two more strengths, he said, is that the study investigators were able to keep the results completely independent of Apple, the study sponsor. And the study design implemented a system for follow-up of notifications by virtual clinic contacts online, further work-up using the patch, and further referral to in-person doctors as appropriate.

Heidbuchel said a main limitation of the study is that there are no data on the sensitivity of the algorithm; essentially, how many people's AF in the study went undetected. He noted that studies using other technologies missed one-quarter of cases because of noisy PPG readings that could not be interpreted. So for such a study, it is not possible to say that it did not present harm to the participants if, for example, they were falsely reassured that they did not have a problem.

Another concern was the high drop-out rate from the trial. Only 44% of people who received a notification contacted the virtual clinic and received patches. Only 68% of them returned the patches for analysis. Combining these numbers resulted in a 69% drop-out rate.

"A rigorous clinical trial would never accept a 69% drop-out rate," he said. On the upside, the people who received and sent back their patches had the same clinical characteristics as the study cohort overall, indicating that their results may be generalizable to the entire cohort.

Russo pointed out other limitations. "This was not designed to be a screening study. We didn't have outcomes, we didn't have a lot of the things we'd need for a screening study," she said. "This is not to be used to replace the gold standard: ECG monitoring. The study was really to evaluate the ability of these algorithms to detect A fib."

Also, participants without notifications never wore ECG patches, so there is no information on false negatives.

She had noted that the algorithm used by the watch is proprietary, and Heidbuchel said it would be essential to know in the future what is the algorithm, such as how often it samples data. He also noted that the population was self-selected, so it was not a full population screening trial.

He said for the future, the technology, along with improvements, such as incorporating the ability to measure ECG, will put more healthcare screening ability in the hands of consumers. But real outcomes data will be needed based on systematic care delivery as was incorporated in this trial. As such, there will need to be organizational changes for care delivery to deal with this sort of patient-centered information.

Haran Burri, MD, professor at the University Hospital of Geneva, Switzerland, commented to | Medscape Cardiology that as artificial intelligence gets better and machine learning develops, the algorithm should improve. As it is, he thinks the algorithm is performing well with few false positives because of premature beats, noise, or other artefacts. Two European companies are developing similar devices, and they, too, have shown very good specificity.

"People are getting more and more empowered with their own healthcare," he said, citing home blood pressure devices, diabetes patients monitoring glucose levels and changing insulin doses, and patients adjusting doses of vitamin K antagonist anticoagulants.

Given the high drop-off rate, he cautioned about interpreting this figure. "I think what's important to state is these were not patients. These were subjects. There is a whole load of difference between someone who feels fine" and a patient with a diagnosed problem, he said.

He said asymptomatic individuals may be notified of an AF event and not act on it, but another potential source of harm is if someone feels palpitations but does not get a notification by the device and is therefore falsely reassured. But with continuous monitoring by the watch, true AF may be picked up even if a couple of episodes are missed.

"As time goes on, I think that sensitivity is going to increase simply by the virtue of the fact that it's continuously monitoring," Burri said.

In a statement, the EHRA provided further reactions and perspectives on the study from some of its leaders.

Isabelle Van Gelder, MD, PhD, chair of EHRA's National Cardiac Societies Committee, said: "My main concern about the study [is] that it brings a lot of anxiety among the people wearing such devices."

Gerhard Hindricks, MD, editor-in-chief of EHRA's scientific journal EP Europace and a past president of EHRA, said: "From an innovative technological point of view, this is a fantastic new approach. But are we ready for these technological innovations? Are the citizens ready? Is the healthcare system ready? I have my doubts about whether we are ready to implement this approach now."

Hindricks further wondered whether the technology would have an impact on outcomes, and where the data will end up or how they will be used. "Will these data potentially bypass the traditional healthcare system? Are there legal issues? Has all that been clarified and really thought through? I've got my doubts."

Emma Svennberg, MD, PhD, EHRA m-health coordinator, pointed out that the detection rate of atrial fibrillation in this population was about 0.2%. "This should be compared to other screening studies in elderly populations of much shorter duration where they found approximately 15 times more atrial fibrillation, with numbers approximating 3%."

AF, she added, "is not a benign condition in the elderly, nor in those with cardiovascular risk factors. But in this trial of more than 400,000 people, only 6% were above the age of 65."

Apple Inc. sponsored the study. Russo disclosed that she is on the steering committee of the Apple Heart Study and has received honoraria, speaking, and consulting fees from Apple Inc., Medtronic, and UpToDate, and research grants from Boehringer Ingelheim and Boston Scientific. Neither Heidbuchel nor Burri reported any disclosures. Hindricks, Van Gelder, and Svennberg had no disclosures to report.

European Heart Rhythm Association (EHRA) 2019 Congress. Presented March 19, 2019.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.