Apple recently launched the electrocardiogram (ECG) app on the Apple Watch 4, allowing for remote detection of atrial fibrillation. But what does this mean for people with the new smartwatch and the clinicians caring for them?
As a clinician, I am excited about this new frontier in digital health, but I am also cautious about its implications. When a diagnostic test is released to a large population, patients and clinicians must be aware of false positive results (the ECG incorrectly diagnosing atrial fibrillation in patients without it). We must ask the question that matters: If my watch tells me I have atrial fibrillation, what are the chances it is correct?
Answering this question requires knowing the baseline risk of having atrial fibrillation. According to the ATRIA study (JAMA 2001), the prevalence of atrial fibrillation in asymptomatic patients younger than 55 years old was 0.1 percent, 1 percent for patients between 55-64 years old, and 10.1 percent for patients older than 85.
Bayesian statistics can be applied to the prevalence data in combination with the sensitivity and specificity (98 percent and 99.6 percent, respectively) of the Apple Watch ECG algorithm to calculate the probability the watch is correct. In patients less than 55 years old, the prevalence of atrial fibrillation is about 0.1% (one out of 1,000 people). Using Bayesian statistics, the Apple Watch’s positive predictive value is surprisingly only 19.6 percent. Therefore, it is incorrectly diagnosing atrial fibrillation in 79.4 percent of users in this age group. This is because there are four false positives for every one true positive. Try it yourself using this Bayesian calculator. Enter prevalence = 0.001, sensitivity= 0.98, specificity= 0.996 and take a look at the positive predictive value.
It is critical for patients and clinicians to be aware of this, as the population younger than 55 years old constitutes somewhere between 77-91 percent of users.
However, the Apple Watch 4 ECG becomes very accurate as the user’s age increases or they become symptomatic (palpitations, shortness of breath, dizziness) — and both increase the prior likelihood of disease. If the prior probability of having the disease is 1.3 percent (asymptomatic users between 60-64 years old) the positive predictive value increases to 76 percent. It is 91 percent for users between 70-74 years old (4.2 percent baseline risk) and 96 percent for users older than 85 years (10.1 percent baseline risk). See Figure 1 for a breakdown of these values.
Figure 1, above. The positive predictive value (PPV) of atrial fibrillation detection in asymptomatic population stratified by age group.
With all of the data mentioned above, it is necessary for patients and clinicians to interpret the Apple Watch 4 ECG atrial fibrillation data through the lens of the user’s baseline risk. Without this data for perspective, many users will be incorrectly diagnosed with atrial fibrillation and potentially be treated with anticoagulation therapy. For the vast majority of young asymptomatic patients diagnosed with atrial fibrillation via the smartwatch, chances are the watch is wrong. However, the watch will be accurate with its atrial fibrillation detection over 95 percent of the time if the patient’s baseline risk is greater than 7 percent. Ultimately, many factors contribute to this risk including age, prior medical problems and symptoms, so clinician guidance will be essential.
About the Author.
Daniel Yazdi, M.D., M.S., is an internal medicine resident at Brigham and Women’s Hospital/Harvard Medical School who is interested in the intersection of digital health and clinical medicine. He can be followed on Twitter @DanielYazdi.
Send this to a friend