Key Points

• Wearable technologies have optical photoplethysmography (PPG) sensors to measure heart rate coupled with software algorithms that can infer the presence of atrial fibrillation (AF)
• The Fitbit Heart Study sought to evaluate a novel Fitbit device compatible algorithm for detecting AF
• Investigators found that their novel PPG software algorithm for Fitbit wearers had a positive predictive value of 98% for detecting AF amongst patients found to have irregular heart rhythm detection (IHRD) notifications and confirmed AF on subsequent ECG patch monitoring. This was stable for the key subgroup of patients age 65 years and older who are at higher risk of stroke
• The false-positive rate for IHRD alerts was 70%
• Further evaluation of clinical and cost-effectiveness of AF monitoring in the United States population via wearable devices is needed

In an effort to improve the morbidity of atrial fibrillation (AF), consumer-based wearable technology companies have made considerable efforts to create new tools to detect AF. Sensors enabling pulse photoplethysmography (PPG) are commonly embedded within wearable devices in order to analyze PPG output and identify the pulse irregularity that signifies AF. Despite prior studies including the Apple Heart Study and the Huawaei Heart Study which examined the validity of PPG for AF detection, there remains critical need to continually evaluate this novel technology in order to improve AF algorithms and detection parameters.

The Fitbit Heart Study sought to evaluate a novel Fitbit device compatible algorithm for detecting AF, assess continuous monitoring periods for AF detection, and study Fitbit users in the United States. Presented by Steven Lubitz MD, MPH (Massachusetts General Hospital, Boston) at the virtual American Heart Association 2021 Scientific Sessions on November 14^th, the study’s goal was to test the algorithm’s positive predictive value for undiagnosed AF in a large-scale remote trial. The Fitbit Heart study was a single-arm, open-label, prospective clinical trial. Participants were recruited and enrolled remotely and included adults 22 years of age or older who had a compatible PPG-enabled Fitbit tracker or smartwatch and no prior diagnosis of AF. Participants were excluded if they were taking oral anticoagulation or had a pacemaker or defibrillator device. It is important to note that the novel algorithm, developed by Dr. Lubitz and the trial team, required at least 30 minutes of irregular rhythm (determined by 11 out of 11 irregular tachograms) as well as inactivity on the part of the user (detected by accelerometers) to detect AF. The inability to detect AF in active wearers is noted as a limitation of current PPG technology overall and not exclusive to Fitbit. Following recruitment, participants’ data was both retrospectively and prospectively evaluated and in the case of AF detection participants were notified and asked to contact a telehealth physician. They were then mailed a single-lead ECG patch monitor to be worn for 1 week and returned it using pre-paid packaging. At a second telehealth visit a physician then disclosed the patch monitor results to the participant. The primary objective of this study was to determine the positive predictive value of the first irregular heart rhythm detection (IHRD) for concurrent AF during ECG monitoring, which for this study meant the proportion of individuals with confirmed and simultaneous AF (as indicated by the patch) among those who received a device notification while wearing the patch monitor.

Of the 455,669 U.Ss adults enrolled in the study, 4,728 participants (1%) received a IHRD notification over a 5 month period. Of those participants with IHRD notification, 1,671 completed a telemedicine appointment. ECG patches were shipped to 1,409 participants, and 1,162 were returned for analysis. AF was confirmed in 340 (32.2%) on the subsequent ECG patch monitor. In his discussion of the trial Dr. Lubitz notes that 13% of those with AF detected were 65 years or older which is an important subgroup for AF detection due to high thromboembolic risk. The trial’s participants were majority female (71%) and self-identified as white non-Hispanic (73%). The majority of patients with IHRDs however were male and greater than 65 years of age, as is commonly found in the AF population at large. The median AF burden detected was 7% (median 7 hours) which Dr. Lubitz pointed out is much higher than other wearable monitors. IHRD detection for any AF during ECG monitoring had 68% sensitivity (95% CI 62-73%) and 98% specificity (95% CI 97-99%).

This study had several limitations that largely was related to its remote nature, including lack of real-time notification capabilities and drop-off between IHRD notification and study visits which may have been related to participants seeking care at personal healthcare providers. Dr. Lubitz notes in discussion that engagement of provider networks will be important to enhance engagement in future studies. Since the study was limited to existing wearable users the results may not be generalizable. Detection may have been limited due to the algorithm’s 30-minute AF requirement and ECG patch monitoring of only 1 week in duration. Notably this study was not evaluating clinical events or upstream outcomes related to paroxysmal atrial fibrillation detection in this population. Further studies will be needed to assess the population that will derive the greatest benefit from this wearable technology.

Additional Resources:

·       “Rationale and design of a large population study to validate software for the assessment of atrial fibrillation from data acquired by a consumer tracker or smartwatch: The Fitbit heart study” – Lubitz, et al.

·       ClinicalTrials.gov – Validation of Software for Assessment of Atrial Fibrillation From PPG Data Acquired by a Wearable Smartwatch