Large-scale assessment of a Smartwatch to identify atrial fibrillation

Selected in The New England Journal of Medicine by L. Biasco

The study aimed to evaluate the capabilities of correctly detecting episodes of Atrial fibrillation/flutter (Afib) through a smartwatch-based irregular pulse notification algorithm of the Apple watch.

Reviewer

Luigi Biasco

@BiascoDr

Interventional cardiologist / Cardiologist

Ciriè, Italy

Latest contributions

TAVI for asymptomatic aortic stenosis after EARLY TAVI - TAVI and coronary artery disease after NOTION-3 New insights from the early and delayed aortic valve replacement implant patients: a sub-analysis of the Early TAVR trial PCRonline @ ACC.25 Scientific Sessions

View full profile

My Comment

Why this study – the rationale/objective?

The study aimed to evaluate the capabilities of correctly detecting episodes of Atrial fibrillation/flutter (Afib) through a smartwatch-based irregular pulse notification algorithm of the Apple watch.

How was it executed – the methodology?

Population: 419297 participants were enrolled over a 8 months period.

Inclusion criteria comprised:

• Possession of a compatible Apple iPhone and Apple Watch.
• Age of 22 years or older, United States residency and proficiency in English, as reported by the participant.

Patient’s reported history of atrial fibrillation or current use of oral anticoagulation agents were considered as a exclusion criteria.

Methods:

• Pulse monitoring by means of the irregular pulse detection algorithm was performed.
• When multiple tachograms (pulse wave rhythm strips) compatible with Afib were recorded by the device, the algorithm notified both investigator and participant. Participants were then invited to a first app-based visit to cross check inclusion/exclusion criteria checked and need for urgent intervention.
• Following initiation visit, ECG patches were mailed to eligible participants to confirm by 7 days ECG monitoring the presence of Afib and cross-verify findings of the irregular pulse notification algorithm.
• The study app was used to verify eligibility, obtain participants’ consent, provide study education, and direct participants through the study procedures.
• The study was sponsored and data are owned by Apple while data analysis was performed at Stanford University.

Two co-primary endpoints were considered:

• Afib lasting > 30 seconds on ECG patch monitoring in a participant who received an irregular pulse notification by the Apple watch.
• Simultaneous atrial fibrillation registered at ECG patch monitoring and detected by the irregular pulse detection algorithm of the Apple watch.

What is the main result?

• Out of 419297 participants, 2161 were notified by the algorithm has having irregular pulse compatible with episodes of atrial fibrillation and were invited to a initiate first visit.
• Of them, only 945 were effectively evaluated at initiation visit and 658 received an ECG patch monitor.
• 450 participants returned an ECG patch able to be analyzed.
• Afib was confirmed in 153 participants (34%).
• 86 participants had irregular pulse notifications by the Watch while monitored by ECG patch. In 72 Afib was effectively recorded at ECG resulting in a positive predictive value for the irregular pulse notification of 0.84 (95% CI, 0.76 to 0.92).

Critical reading and the relevance for clinical practice

The study entitled “Large-Scale Assessment of a Smartwatch to Identify Atrial Fibrillation” by Marco Perez represents a turning point in the philosophy of modern health research.

As researcher we are used to building complex, bureaucracy driven study protocols to strengthen the scientific solidity of our outputs. On the other hand, as contemporary citizens, most of us consider smartphone apps and wearable trackers just as funny, expensive gadgets.

The present study clearly smashes those two preconceptions. Firstly, the pragmatic app-based approach used by the authors clearly demonstrated that achieving clinical research is feasible with an innovative, non conventional study design. Secondly, due to the large availability and concrete clinical and research potentials of portable devices, we have to change our minds and start to consider them as companions in clinical investigation.

From a purely clinical perspective, the main message of the present analysis is that the pulse analysis algorithm was able to identify with a positive predictive value of 84% episodes of atrial fibrillation in participants notified with a pulse rhythm alert. A concrete, but limited, output.
From the scientific perspective this study demonstrated that enrolling almost half a million participants in 8 months is nowadays feasible; that app-based clinical research is an option for contemporary scientists and that participants reported clinical data can represent a base to build up research. The possibility to remotely, non invasively and accurately monitor our patients, clearly represents an intriguing opportunity for any health care professional.

As an interventionalist, I’m wondering whether an app-based pulse rhythm monitoring could be of help in patients with a recent PCI and atrial fibrillation with a low arrhythmic burden in order to tailor medical therapy according to their needs. Surely, several different applications of this technology can be implemented and probably the only limits of this field are represented by our ability to imagine newer scenarios. Beyond the scientific relevance of the study, the social implications are worth mentioning. While scientist are changing their mind regarding the app-based approach, confidence and commitment of participants still have to be improved.

More than half of the participants who were notified with an alert for possible Afib (1216/2161) failed to initiate the first visit, and where then excluded. Of 658 participants who received an ECG patch, only 450 effectively returned a device with data able to be analysed. Thus, obtaining and maintaining participants' cooperation in a context where interactions between the study population and researchers are limited to app notifications clearly represents a major limitation of this approach.

In addition, according to study criteria, possession of both an iPhone and an Apple Watch was a prerequisite of study participation. This clearly represents a selection bias able to cut away a large percentage of the general population (e.g. lower income groups, racial minorities, older ages) in whom Afib might have a significant prevalence and that might effectively benefit from monitoring.

And lastly, this study opens the discussion to what extent our society will accept exchanges of health related data and where to draw borders between privacy, health care providers and major companies.

Surely, this experience paths the way for future digital health studies.