The Apple Watch 4 can now do something that was once available mainly in your doctor’s office or a hospital: take an electrocardiogram. This recording of your heart’s rhythm can detect rapid or skipped heartbeats. The watch can also check your heart’s rhythm in the background and notify you if an irregular rhythm appears to be atrial fibrillation.

That’s a useful task. A heart in atrial fibrillation beats faster than normal, and the upper and lower chambers don’t work together. Atrial fibrillation can go unnoticed. It can also cause fatigue, dizziness, heart palpitations, or chest pain. Noticed or not, it can increase the risk of having a stroke.

As a physician, I’m excited about this new frontier in digital health. But I’m also cautious about its implications. Diagnostic tests like this one are rarely perfect. They usually generate false positives: incorrectly diagnosing a condition in people who do not have it.

advertisement

According to data that Apple submitted in its petition asking the Food and Drug Administration to give clearance to the heart monitoring app, the app accurately detects atrial fibrillation 99 percent of the time it gets a good reading.

That sounds amazing. But it leaves out an important little something: the likelihood of having atrial fibrillation is low among younger individuals and increases with age. I did some calculations to answer the question, “If my watch tells me I have atrial fibrillation, what are the odds it is correct?” The answer depends on the watch wearer’s age.

I first needed to know the baseline risk of having atrial fibrillation. According to the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study, which included nearly 1.9 million Americans, the percentage increases with age: 0.1 percent of those younger than age 55 have atrial fibrillation, rising to 9.1 percent among those over age 85.

Next I needed to know the sensitivity (the probability of having atrial fibrillation if the watch detects it) and specificity (the probability of not having atrial fibrillation if the watch does not detect it) of the Apple Watch app. Apple helpfully supplies those numbers — 98 percent sensitivity and 99.6 specificity — in its user’s guide.

I applied Bayesian statistics to this information. This branch of statistics uses prior information, such as disease prevalence, to interpret the likelihood of an outcome, in this case having atrial fibrillation. The result of the calculation is the positive predictive value: the probability of having atrial fibrillation when it is detected by the watch.

In people younger than 55, Apple Watch’s positive predictive value is just 19.6 percent. That means in this group — which constitutes more than 90 percent of users of wearable devices like the Apple Watch — the app incorrectly diagnoses atrial fibrillation 79.4 percent of the time. (You can try the calculation yourself using this Bayesian calculator: enter 0.001 for prevalence, 0.98 for sensitivity, and 0.996 for specificity).

The electrocardiogram app becomes more reliable in older individuals: The positive predictive value is 76 percent among users between the ages of 60 and 64, 91 percent among those aged 70 to 74, and 96 percent for those older than 85.

Here’s the bottom line: For the vast majority of individuals under age 55 whose Apple Watches tell them they have atrial fibrillation, the odds are high that the watch is wrong. But it is more accurate for the aging population that is becoming a part of the wearable generation.

Does this mean younger folks should ignore a notification about atrial fibrillation?

No. Even younger individuals whose Apple Watch signals atrial fibrillation should contact their physicians to discuss any symptoms or medical conditions that heighten the risk for this disease. The Apple Health app lets users download a PDF of the watch’s electrocardiogram tracing to send to a physician. Efforts to integrate data from this app with electronic health records are currently underway.

Daniel Yazdi, M.D., is a second-year internal medicine resident at Brigham and Women’s Hospital in Boston. A version of this article was published by Medtech Boston.

Leave a Comment

Please enter your name.
Please enter a comment.

  • The key here is not the fidelity of the ECG sensor (it has a terrific sensitivity/specificity), but rather what happens when a very large population is screened for an illness. There are bound to be false positives, and when the prior likelihood of disease is low (those under 55) the false positives will outweigh the true positives making misdiagnosis common.

    • I’m not sure it can be “fixed”. It may already be at the limit of this sensor. Bear in mind it uses dry electrodes, which are sensitive to electrode movement and contact pressure. There’s only so much you can do, given the fundamental design of the sensor.

  • This article highlights what happens when engineers / techies drive clinical apps. 20% accuracy rate may be OK in beta software or other industries, but this level of error will result in sending MDs already drowning in data more useless or bad info, and cause patients undue anxiety.

    Is this really helpful to anyone because Apple is “cool” ?

  • I am 75 years old and have a history of two episodes of A.fib 10 years ago. Almost coincident with the release of the ECG app I started having symptoms again. Fortunately I work in the healthcare field and got to cardiogram machine right away. Since then I have been using the app. It has “diagnosed” AF multiple times since then. Even with that history, I am reluctant to overwhelm my cardiologist with frequent ECG/EKG PDF’s. What I have done is text the PDF’s to myself and save them till my visit with the doctor or ask him if he wants to see them between visits. I even e-mailed them to my home computer and printed them out – full page 8X11″ paper. The cardiologist was impressed and found it easy to confirm the diagnosis.

Sign up for our Daily Recap newsletter

A roundup of STAT’s top stories of the day in science and medicine

Privacy Policy