Wearable monitors are increasingly used to assess sleep. However, validity is unknown for certain monitors and populations. We tested the Garmin Vivofit 4 in a pediatric clinical sample. Participants (n = 25) wore the monitor on their nondominant wrist during an overnight polysomnogram. Garmin and polysomnography were compared using 95% equivalence testing, mean absolute error, and Bland-Altman analysis. On average (mean ± SD), the Garmin predicted later sleep onset (by 0.84 ± 1.60 hours) and earlier sleep offset (by 0.34 ± 0.70 hours) than polysomnography. The resulting difference for total sleep time was -0.55 ± 1.21 hours. Sleep onset latency was higher for Garmin than polysomnography (77.4 ± 100.9 and 22.8 ± 20.0 minutes, respectively), while wake after sleep onset was lower (5.2 ± 9.3 and 43.2 ± 37.9 minutes, respectively). Garmin sleep efficiency averaged 3.3% ± 13.8% lower than polysomnography. Minutes in light sleep and deep sleep (the latter including rapid eye movement) were within ±3.3% of polysomnography (both SDs = 14.9%). No Garmin means were significantly equivalent with polysomnography (adjusted p > 0.99). Mean absolute errors were 0.47 to 0.95 hours for time-based variables (sleep onset, offset, and latency, plus total sleep time and wake after sleep onset), and 8.9% to 21.2% for percentage-based variables (sleep efficiency and sleep staging). Bland-Altman analysis showed systematic bias for wake after sleep onset, but not other variables. The Vivofit 4 showed consistently poor individual-level validity, while group-level validity was better for some variables (total sleep time and sleep efficiency) than others.
Read full abstract