Fitbit Heart Study: Novel algorithm for wearable devices may identify undiagnosed AF
Click Here to Manage Email Alerts
An algorithm used in Fitbit wearable devices showed the capability to identify asymptomatic atrial fibrillation, according to results of the Fitbit Heart Study presented at the American Heart Association Scientific Sessions.
About one-third of participants who had an irregular heart rhythm detected by their Fitbit device and subsequently wore an ECG patch had AF confirmed by the ECG patch, Steven Lubitz, MD, MPH, associate professor of medicine at Harvard Medical School and cardiac electrophysiologist at Massachusetts General Hospital, said during a press conference.
“A novel photoplethysmography software algorithm for Fitbit wearables enabled large-scale identification of undiagnosed atrial fibrillation,” Lubitz said. “The algorithm operates in the background during inactivity and doesn’t require the user to prompt it to initiate sampling. Individuals with irregular heart rhythm detection tended to have elevated likelihood of atrial fibrillation on subsequent ECG patch monitoring and had a considerable burden of atrial fibrillation.”
Novel software algorithm
Researchers developed the algorithm with frequent overlapping photoplethysmography (PPG) pulse tachogram sampling and tested its positive predictive value for undiagnosed AF in a large-scale remote clinical trial.
The cohort included 455,669 participants who were at least 22 years old, had a compatible Fitbit device (Ionic, Charge 3, Charge 4, Inspire HR or Inspire 2 trackers or Versa, Versa Lite, Versa 2, Versa 3 or Sense smartwatch), had an Android or iOS smartphone with the Fitbit app, and had no prior diagnosis of AF or atrial flutter, use of oral anticoagulation or implantation of a pacemaker or defibrillator. Enrollment occurred over 5 months during the COVID-19 pandemic.
“The algorithm continuously samples the pulse data in 5-minute blocks, which by design are overlapping by 50%,” Lubitz said. “If 11 out of 11 consecutive blocks, or tachograms, are irregular, an irregular heart rhythm detection occurs. By design, that means that at a minimum, the algorithm requires at least 30 minutes of an irregular rhythm to detect atrial fibrillation. The algorithm only operates if the participant is inactive, which is judged by accelerometers on the device. The algorithm resets with a normal 5-minute tachogram period. Unanalyzable periods, perhaps due to activity, are skipped by the algorithm.”
High positive predictive value
Among the cohort, 4,728 participants had an irregular heart rhythm detection. Those participants were prompted to have a telehealth visit, and those who completed the visit were sent an ECG patch monitor (ePatch, BioTel) to wear for 1 week. Lubitz said 1,162 ECG patches were returned, of which 1,057 had data able to be included in the analysis.
Among the entire cohort, 1% of participants had an irregular heart rhythm detection, which was more frequent in men than in women (2.1% vs. 0.6%) and in individuals aged 65 years or older than in those who were younger (3.6% vs. 0.7%), Lubitz said.
Among those with ECG readings, 32.2% had AF confirmed (men, 36.3%; women, 27.8%; people aged 65 years or older, 33.4%; people aged younger than 65 years, 31.3%). This is a much higher rate than seen in previous studies of AF detection using ECG patch monitors in which PPG sensors to pre-screen for irregular heart rhythms weren’t used, Lubitz said, Lubitz said.
The primary endpoint of positive predictive value of the algorithm for concurrent AF on the ECG patch monitor was 98.2% in the overall cohort, 98.4% in men, 98% in women, 97% in participants aged 65 years or older and 99.2% in those aged younger than 65 years, Lubitz said.
Those with confirmed AF on the patch monitor had a median AF burden of 7%, in contrast to the approximately 1% seen in previous studies of AF detection with ECG patch monitors that did not use PPG sensors to pre-screen for irregular heart rhythms. The median longest AF episode was 7 hours, he said.
The requirement for sampling to be done during periods of inactivity is similar to that in other studies and “is more or less a limitation of the current technology,” Lubitz said.
“One can imagine potential algorithms in the future where one can interpret data during periods of activity, to sample over a longer period of duration,” he said.
However, Lubitz said, “the [positive predictive] values that we saw in our study are higher than had been previously reported by other software algorithms, including the Apple Heart Study, and were stable in an elderly population, which is an important subgroup of individuals, because if they have atrial fibrillation, they are at the highest risk of having an ischemic stroke. We were very pleased with the high predictive value of the algorithm in that particular subgroup.”
The algorithm is currently being reviewed by the FDA for clearance and widespread use, according to an AHA press release.
Perspective
Back to Top
Roopinder K. Sandhu, MD, MPH
The use of wearable technology for the detection of AF is rapidly rising. It is critical that these technologies are validated, and for this reason the Fitbit Heart Study provides important information.
The study found the positive predictive value of the novel algorithm in Fitbit wearables used to detect an irregular pulse that is confirmed on an ECG patch monitor as AF is very high. The algorithm requires at least 30 minutes of irregular rhythm to detect AF, which means shorter episodes will not be detected. Interestingly, only one-third of the patients with an irregular heart rhythm detection completed a virtual visit with a physician.
This is helpful information for clinicians who may have patients who present with an irregular heart rhythm detection from a Fitbit wearable, but there are still many unanswered questions. Specifically, what is the clinical significance of these AF episodes, how should clinicians manage these events and does treatment, specifically with oral anticoagulation therapy for stroke prevention, provide a net clinical benefit?
We should encourage the use of wearables once we have data to clearly define the optimal patient population, clinical care pathway and management strategy. We have not done that yet.
We need to have more randomized clinical trials that compare wearable technologies to usual care, assess clinical outcomes and potential harms.
Perspective
Back to Top
Sana M. Al-Khatib, MD, MHS, FHRS
The usefulness of these results to people with Fitbit devices depends on why those users have a Fitbit. If they have it to detect irregular rhythms, then these results suggest that the positive predictive value of the algorithm for that purpose is very high.
For doctors, the results are somewhat useful, but we certainly need more data on this. What is not clear is how the AF that was detected was managed and how this management affected outcomes.
We need to know whether these results are reproducible, which patients are most or least likely to have an arrhythmia identified by this algorithm and how might detecting AF and managing it change outcomes.
There is a need for AF screening studies that examine AF burden, duration and pattern in relation to multiple outcomes. And there is a need for randomized clinical trials that directly compare screening with usual care and that assess both health outcomes and harms.
Collapse
Lubitz SA, et al. LBS.04. Information Overload? Striving to Improve Care Delivery Through Digital Health and Automated Data. Presented at: American Heart Association Scientific Sessions; Nov. 13-15, 2021 (virtual meeting).
Read more about
Click Here to Manage Email Alerts