An app on a smartwatch can be used to detect atrial fibrillation, providing a boost for the role of digital technology in healthcare.
Mintu Turakhia and Marco Perez (Stanford University School of Medicine, CA, USA) presented the findings from the Apple Heart Study at the American College of Cardiology’s 68th Annual Scientific Session (ACC 2019; 16–18 March, New Orleans, USA).
They said: “The notification positive predictive value of 0.84 supports the ability of the app to correctly identify atrial fibrillation among those notified, and these findings may inform further clinical evaluation after notification.”
The Apple Heart Study was a prospective single-arm pragmatic study. It aimed to evaluate the ability of the watch’s irregular pulse notification system to identify atrial fibrillation and to measure the rate of initial contact with a healthcare provider. Inclusion criteria were age ≥22 years, and possession of an iPhone 5s or higher, as well as an Apple, watch series 1–3. Those currently using anticoagulation, or with atrial fibrillation or atrial flutter were excluded from the final analysis.
The watch operates using optical sensors that detect pulse waveforms passively to measure heart rate. Atrial fibrillation is identified by the detection of pulse irregularity. Following five confirmations of an algorithm that is suggestive of atrial fibrillation, the user is notified.
Participants who received a notification based on the algorithm were advised to consult a telehealth doctor, following which they were either referred to urgent care or sent an electrocardiogram (ECG) patch for further monitoring. The results from the patch were reviewed by a telehealth doctor, with a 90-day follow-up survey.
Primary and secondary endpoints were measured for those who received notifications. The primary endpoint was atrial fibrillation >30 seconds as measured on the ECG patch for patients who received an irregular pulse watch notification. A co-primary endpoint was simultaneous atrial fibrillation on ECG patch and when the spot tachogram was positive for an irregular pulse among those who received a notification. Secondary endpoints were simultaneous atrial fibrillation on ECG patch with notification based on multiple tachograms and self-reported contact with a health care provider within three months following an irregular pulse watch notification.
The study had an enrolment goal of up to 500,000, with the assumption of 503 patches for each group among those aged ≥65 and those <65 years. It was designed to have sufficient provision to estimate the yield of atrial fibrillation on subsequent ECG patch with 97.5% confidence intervals (CI) no wider than 10% in age ≥65 and <65, and to have sufficient provision to estimate the positive predictive value (PPV) of an individual tachogram with a simultaneous ECG patch—97.5% CI lower bound >0.70 and upper bound ≥0.75.
Enrolment commenced in November 2017, and finished in July 2018; 419,297 participants entered the study. Of these, 2,161 (0.5%) received a pulse notification, and 945 (44%) took part in a study visit. Following exclusions, an ECG patch was sent to 70% (n=658), of whom 68% (n=450) returned it for analysis.
Mean age of those whose patch was analysed was 59 years, and 23% were female; 38% had a CHA2DS2VASc score ≥2, 44% had hypertension, and 43% had obesity.
Among the patches analysed, the mean time to hook up was 13 days, and mean wear time was 6.3 days. The atrial fibrillation yield was 34% (97.5 CI 29–39). The PPV of irregular tachograms was 0.71 (97.5% CI 0.69–0.74), and PPV of irregular pulse notifications was 0.84 (95% CI 0.76–0.92).
Among the 2,161 participants who received a pulse notification, 1,376 (64%) completed a 90-day survey, of whom 161 (15%) had atrial fibrillation before they enrolled in the study; 787 (57%) contacted a non-study healthcare provider—218 (28%) started new medication, 262 (33%) were referred to a specialist, and 287 (36%) underwent additional testing.
There were 71 serious adverse events among the entire cohort of 419,297, none of which were related to the app.
Turakhia and Perez pointed to the limitations of the study, which included a failure to achieve the pre-specified precision of the co-primary endpoints, and the virtual study design, which meant that participants self-assessed for enrolment criteria and self-reported the outcomes. And, they said: “The number of drop-offs after notification was higher than anticipated, and there were fewer telehealth visits and ECG patches returned for analysis than we expected.”
However, they highlighted the operational success of the study, and the low rates of pulse notification across a diverse population, as well as the safety of the app.
They concluded: “The future direction is for rigorous investigation of this technology to assess its potential role in a clinical setting. The Apple Heart Study provides a solid foundation on which further research in digital care can be provided.”