Journal Logo

A-28 Thematic Poster - Sleep Wednesday, June 1, 2016, 9: 30 AM - 11: 30 AM Room: 110

Validity of Wearable Fitness Trackers on Sleep Measure

106 Board #4 June 1, 9

30 AM - 11

30 AM

Keill, Alyssa K.; An, Hyun-Sung; Dinkel, Danae M.; Lee, Jung-Min

Author Information
Medicine & Science in Sports & Exercise: May 2016 - Volume 48 - Issue 5S - p 10
doi: 10.1249/01.mss.0000485037.05259.dd
  • Free

Sleep and its effect on an individual’s health is becoming recognized as an important aspect of preventative care for many chronic diseases. Wearable trackers that detect sleep offer users a way to track their sleep quality and patterns. However, no studies have tested the validity of these trackers on sleep measure.

PURPOSE: To examine the validity of wearable fitness trackers for estimating total sleep time (TST) with respect to a sleep log as a reference measure.

METHODS: Nineteen healthy individuals (mean ± SD; age = 29.5 ± 13.4 years; body mass index = 25.87 ± 5.03 kg · m2) participated in the study. Participants randomly assigned to one of two groups. Group 1 (n = 10) wore the BodyMedia SenseWear Mini Armband (SWA), Basis Peak (BP), and Fitbit Charge HR (FB). Group 2 (n = 9) wore the ActiGraph Sleep (AG), Jawbone UP3 (JU), and Garmin Vivosmart (GV). Trackers were worn on the non-dominant wrist for one night and a sleep log was completed. Two existing sleep algorithms for the AG (Sadeh and Cole-Kripke) and Fitbit sleep sensitive algorithm were also included for comparison. Pearson correlation was used to examine the linearity of mean TST minutes (TSTM) from each tracker compared to the log TSTM. Mean absolute percentage errors (MAPE) of TSTM from each tracker were calculated against the log TSTM. Lastly, mean differences in average TST between the trackers were examined by a general linear model for repeated measures.

RESULTS: Pearson correlation coefficients were .32, .69, .24, and -.26 for the SWA, FB, and FBs with regard to log TSTM, respectively. Group 2 correlations between the log TSTM and Sadeh, Cole-Kripke, JU, and GV were .34, .65, .54, and .92, respectively. MAPE were 17.1%, 16.3%, 40.2%, and 32.9% for SWA, FB, FBs, and BP, respectively. MAPE of 17.0%, 11.5%, 14.9%, and 10% were observed for Sadeh, Cole-Kripke, JU, and GV, respectively. Bland-Altman Plots showed no systematic bias between all variables for TST compared with log TSTM. ANOVA and post-hoc analysis revealed a significant difference in the Fitbit sensitive TST (p = .001) in Group 1 (F (5, 51) = 8.06, p = .00) and no significant difference between Group 2 (F (4, 40) = 1.27, p = .296).

CONCLUSION: The FB, Cole-Kripke, JU, and GV display the closest estimation of TST compared with log TST minutes. However, further research is needed to validate these monitors with polysomnography.

© 2016 American College of Sports Medicine