The wrist is not the brain: Estimation of sleep by clinical and consumer wearable actigraphy devices is impacted by multiple patient- and device-specific factors

Clinical actigraphy devices provide adequate estimates of some sleep measures across large groups. In practice, providers are asked to apply clinical or consumer wearable data to individual patient assessments. Inter-individual variability in device performance will impact such patient-specific interpretation. We assessed two devices, clinical and consumer, to determine the magnitude and predictors of this individual-level variability. One hundred and two patients (55 [53.9%] female; 56.4 [±16.3] years old) undergoing polysomnography wore Jawbone UP3 and/or Actiwatch2. Device total sleep time, sleep efficiency, wake after sleep onset and sleep latency were compared with polysomnography. Demographics, sleep architecture and clinical measures were compared to device performance. Actiwatch overestimated total sleep time by 27.2 min (95% confidence limits [CL], 138.3 min over to 84.0 under), overestimated sleep efficiency by 6.8% (95% CL, 34.1% over to 20.5% under), overestimated sleep onset latency by 2.6 min (95% CL, 63.3 over to 58.2 under) and underestimated wake after sleep onset by 50.7 min (95% CL, 162.5 under to 61.2 over). Jawbone overestimated total sleep time by 59.1 min (95% CL, 208.6 min over to 90.5 under) and overestimated sleep efficiency by 14.9% (95% CL, 52.6% over to 22.7% under). In multivariate models, age, sleep onset latency, wake after sleep onset, % N1 and apnea-hypopnea index explained only some of the variance in device performance. Gender also affected performance. Actiwatch and Jawbone mis-estimate sleep measures with very wide confidence limits and accuracy varies with multiple patient-level characteristics. Given these large individual inaccuracies, data from these devices must be applied only with extreme caution in clinical practice.