Introduction: As part of a large digital medicine study investigating the identification of asymptomatic atrial fibrillation (AF) using a continuous single-lead ECG patch, we evaluated the usability and the efficacy of a wrist wearable device for the detection of AF.
Hypothesis: A wearable, non-ECG sensor will allow for rhythm monitoring for longer periods of time and therefore potentially provide the evidence to motivate future large studies on photoplethysmography (PPG) based AF detection.
Methods: We provided via mail an Amiigo wrist worn PPG device (Wavelet Health) to 237 subjects randomly selected, along with instructions on how to use device and mobile app. Due to battery constraints, the devices collected PPG data for 30 seconds every 20 minutes during day and every 7 minutes by night. On a subset of the available data, the intervals of heart rate data were further analyzed by computing the Shannon entropy on the differences between consecutive RR intervals, and detecting AF event with a threshold based algorithm.
Results: More than 6000 hours of PPG data were collected from participants during this digital health study. 137 patients wore their wristband at least 1 day. Average device usage was 99 (min=1, max=459) days per person, and a wear time of 10.7 (min=1, max=17) hours per day. The average worn duration was 8.4 hours at night. Part of the data cannot be analyzed due to motion artifacts, as expected for a PPG device. We considered 3 subject (2 of them with at least 1 AF event), and 2950 selected intervals of 30 seconds, of which 1475 with an AF event detected by the ECG based Xio Patch (iRhythm Tech. Inc.), which represents the ground truth in this study. Preliminary results show accuracy of 91% (2674/2950), specificity of 96%, and sensitivity of 85% in the detection of an AF event in a 30 second interval using a wearable PPG device.
Conclusions: This first of its kind study shows feasibility of wearables data collection through self-enrollment of participants. We highlighted some issues due to the high variability in the device usage and the characteristics of the sensor limiting the amount of useful data for the analysis. Preliminary results indeed show high promise for future studies on the detection of AF events with wearable PPG devices.