Trends in new diagnoses of atrial fibrillation following release of an ECG-capable smartwatch

On December 6, 2018 an FDA-cleared algorithm capable of detecting atrial fibrillation (AF) was released for the wearable Apple Watch Series, which also included a single-lead ECG feature for Series 4.¹ Given the device’s popularity, with an estimated 53.2 million global shipments in 2018-19,² some predicted a dramatic increase in early AF diagnoses and a disruption in medical practices. Despite ongoing debates on the clinical consequences of AF detection in an asymptomatic population,³ evidence is lacking on whether the release was associated with increased rates of new AF diagnosis.

We used de-identified electronic health record data from athenahealth (Watertown, MA), a nationwide cloud-based healthcare information technology company, to analyze the proportion of outpatient visits of adults aged > 20 years that resulted in a new International Classification of Disease, 10^th edition AF diagnosis claim. The study, which used de-identified health records, was exempt from human subjects review by the institutional review board at Harvard Medical School and individual informed consent was not required. We compared diagnosis trends before and after app release (pre-period, September 1, 2018 to December 5, 2018; post-period, December 6, 2018 to December 5, 2019). To account for seasonal trends in visits for AF, we analyzed diagnosis rates during the identical period in 2016 to 2017. We conducted a difference-in-difference analysis by estimating a linear probability model of the proportion of visits involving AF during pre- versus post-periods in 2018-2019, compared to the difference between the same periods in 2016-2017. In addition to the general adult population, we performed a sub-group analysis restricted to patients who lived in zip codes with a median household income over $70,000, as well as separate analyses for patients over and under 40 years of age, under the assumption that use of the device may vary according to population wealth and age. Specifically, we hypothesized that any increases in AF diagnoses attributable to the app’s release would most likely be found in populations more likely to purchase the watch, those who are younger and/or wealthier. Analyses were performed using Stata, version 9.4 (Stata Corp). The threshold for statistical significance was p≤0.05 in a two-sided test.

A total of 61,423,255 visits in 1,270 practices were included (30,796,599 during September 1, 2018 to December 5, 2019, the pre- and post-periods around app release; and 30,626,656 visits during the same dates two years prior). Among adults, the proportion of visits resulting in an AF diagnosis was 0.39% in the months before and 0.41% after app release; in the control period, proportions were 0.36% before and 0.39% after, a difference-in-difference change of −0.008 percentage points that was nonsignificant (difference-in-difference p-value for interaction=0.13) (Figure 1). Within higher income zip-codes, the proportion of visits resulting in an AF diagnosis was 0.37% before and 0.40% after app release; two years prior, proportions were 0.37% and 0.40% for pre- and post-periods, respectively, a difference-in-difference change of 0.01 percentage points that was nonsignificant (p=0.18). The difference-in-difference changes in age sub-groups 40 years and above (difference-in-difference −0.009%, p=0.39) and under 40 years (difference-in-difference 0.003%, p=0.86) were also nonsignificant. Finally, under the assumption that Apple Heart Study participants could themselves have inflated AF diagnoses before app release, we conducted a sensitivity analysis excluding the trial’s critical enrollment and follow-up period (January 17 to September 26, 2018), which did not change overall results (a nonsignificant difference-in-difference change of −0.004 percentage points).

Figure 1:

Percent of visits with a new diagnosis of atrial fibrillation before and after Apple ECG app release

Despite concerns of over-diagnosis of AF following release of a popular ECG-capable smartwatch, a substantial increase in new AF diagnoses was not detected in the first 12 months of app launch, an early analysis. Potential explanations include low adoption of the app – which had to be manually activated by the user¹ – by December 2019 (the endpoint of our analysis), poor algorithm specificity (including failure to detect irregular rhythms at extreme heart rates), higher rate of false positives in a low risk population, limited confirmatory detection of intermittent AF at the clinic, or a small effect limited to a subset of the population. Findings from the Apple Heart Study may be compatible with the latter: while 44% of participants notified of an irregular rhythm reported a new medical diagnosis of AF, only 0.52% of total participants received this notification.^4,5 Study limitations include analyses of incident AF diagnoses in a large, all-payer, geographically diverse but non-nationally representative sample of physician practices; a limited follow-up period of 12 months after app launch; and the possibility that increases in AF diagnoses may be detected in an individual-level analysis of ECG-capable smartwatch users. While studies about this algorithm’s accuracy to detect AF are beginning to emerge, we have not detected the anticipated surge in new AF diagnoses in the short-term.