Quantification of Differences in Sleep Measurement by a Wrist-Worn Consumer Wearable Compared to Research-Grade Accelerometry and Sleep Diaries of Female Adults in Free-Living Conditions.

Abstract

Purpose: The objective of this study is to compare sleep measurements by a consumer-wearable with research-standard actigraphy coupled with sleep diaries in free-living female adults.

Methods: Forty-seven females in the Nurses' Health Study 3 (NHS3) participated in the Sleep and Physical Activity Validation Substudy (SPAVS), where they were asked to concurrently wear a consumer wearable (Fitbit Charge, Models 3 or 5) and a research-grade accelerometer (Actigraph, GT3X+ or Actisleep) on the same wrist and fill out a smartphone-based sleep diary for fourteen consecutive days. We compared measures of total sleep time (TST), time in bed (TIB), and sleep efficiency (SE) from the consumer wearable with actigraphy measures as our research-standard reference for TST and SE and self-reported sleep diary as our reference for TIB. We calculated mean absolute percent error (MAPE) and intra-class correlations (ICC), as well as Bland-Altman analyses to compute mean difference and limits of agreement.

Results: For all three measures, the consumer wearable underestimated sleep parameters relative to research-standard actigraphy, with a mean bias of -16.0 minutes and -11.2 minutes for TST and TIB, respectively, and -1.0% for SE. In terms of agreement, TST (MAPE = 11.18%; ICC = 0.79) and TIB (MAPE = 10.45%; ICC = 0.74) had similar MAPES and ICCs, while and SE (MAPE = 5.09%; ICC = 0.39) had a lower ICC.

Conclusion: In the NHS3 SPAVS, the wearable sleep measurements modestly underestimated wrist actigraphy measures of TST, TIB, and SE from sleep over multiple days; within sleep measures assessed, TST and TIB had greater agreement with research-grade accelerometry than SE.