Performance of wearable sleep trackers during nocturnal sleep and periods of simulated real-world smartphone use

Abstract

Goal and aimsTo test sleep/wake transition detection of consumer sleep trackers and research-grade actigraphy during nocturnal sleep and simulated peri-sleep behavior involving minimal movement. Focus technologyOura Ring Gen 3, Fitbit Sense, AXTRO Fit 3, Xiaomi Mi Band 7, and ActiGraph GT9X. Reference technologyPolysomnography. SampleSixty-three participants (36 female) aged 20-68. DesignParticipants engaged in common peri-sleep behavior (reading news articles, watching videos, and exchanging texts) on a smartphone before and after the sleep period. They were woken up during the night to complete a short questionnaire to simulate responding to an incoming message. Core analyticsDetection and timing accuracy for the sleep onset times and wake times. Additional analytics and exploratory analysesDiscrepancy analysis both including and excluding the peri-sleep activity periods. Epoch-by-epoch analysis of rate and extent of wake misclassification during peri-sleep activity periods. Core outcomesOura and Fitbit were more accurate at detecting sleep/wake transitions than the actigraph and the lower-priced consumer sleep tracker devices. Detection accuracy was less reliable in participants with lower sleep efficiency. Important additional outcomesWith inclusion of peri-sleep periods, specificity and Kappa improved significantly for Oura and Fitbit, but not ActiGraph. All devices misclassified motionless wake as sleep to some extent, but this was less prevalent for Oura and Fitbit. Core conclusionsPerformance of Oura and Fitbit is robust on nights with suboptimal bedtime routines or minor sleep disturbances. Reduced performance on nights with low sleep efficiency bolsters concerns that these devices are less accurate for fragmented or disturbed sleep.