275 Evaluation of Multiple Wearable Sleep-Tracking Devices Tested Under Ad Lib Home Sleep Conditions

Abstract

Abstract Introduction Consumer wearable sleep-tracking devices are increasingly popular and have performed well versus gold standard sleep measurement techniques (polysomnography and actigraphy) in recent validation studies. However, most validation studies were conducted in laboratories under controlled conditions. We therefore aimed to test the validation performance of multiple consumer wearable sleep-tracking devices under real-world ad lib sleep conditions at home. Methods We tested 21 healthy young adults (12 women, 9 men; 29.0±5.0 years, mean±SD) for 7 nights each. Participants slept at home under ad lib sleep conditions, using a set of consumer wearable sleep-tracking devices and completed daily sleep diaries. Consumer wearables included the Fatigue Science Readiband, Fitbit Inspire HR, Oura Ring, and Polar Vantage V Titan. Participants also wore the Philips Respironics Actiwatch 2, a research-grade actigraphy watch, for comparison. To assess validity of sleep/wake measures, all devices were compared with the previously-validated Dreem 2 electroencephalography-based headband device. Analyses included agreement of epoch-by-epoch sensitivity (for sleep) and specificity (for wake), and sleep summary comparisons of time-in-bed (TIB) and total sleep time (TST). Results Sensitivity and specificity, respectively, were as follows: Actiwatch 2 (0.95, 0.36), Readiband (0.93, 0.43), Inspire HR (0.93, 0.45), Ring (0.94, 0.41), and Vantage V Titan (0.96, 0.33). Device average biases, in minutes±SD, for TIB and TST, respectively, were as follows: Actiwatch 2 (N/A, +0.7±42.4), Readiband (+18.2±34.9, +0.4±49.5), Inspire HR (+7.8±35.0, -5.9±44.4), Ring (+9.2±28.0, +4.4±44.5), and Vantage V Titan (+0.2±50.0, -3.2±46.1). Conclusion The consumer wearable devices had comparable sleep-tracking performance during real-world ad lib home sleep. Similar to prior studies, the devices all had high sensitivity and low-to-medium specificity, indicating a greater ability to accurately detect sleep than wake. Notably, specificity for most consumer wearables was higher than a research-grade actigraph, indicating potentially greater ability than actigraphy to detect wake. Sleep summary outcomes were similar among the wearables, which accurately tracked TIB and TST on most nights. However, on some nights there was still considerable bias and variability. Overall, preliminary findings indicate that consumer wearables are promising for tracking sleep and wake in real-world home conditions. Support (if any) Office of Naval Research, Code 34