Automated sleep scoring algorithms for activity monitors – a comparison to polysomnography in the Raine Study.

Introduction: Wrist-activity monitors are widely used to assess sleep and activity. Accurate measurement of time at ‘lights out’ (LO) is essential for calculation of several activity-derived measures including sleep latency (SL), sleep duration (SD), wake after sleep onset (WASO) and sleep efficiency (SE). To determine LO some devices use subjective reports (sleep diary) and others use automated proprietary scoring algorithms. This study aimed to compare the sleep measures of two such devices to polysomnography (PSG).

Methods: A community sample of n=50 parents of participants in the Western Australian Pregnancy Cohort (Raine) Study, mean age of 57±5 years, underwent an overnight laboratory based PSG study during which they also wore a Readiband? (v3, Fatigue Science Inc., Canada, analysed using Readiband Sync) (Rb) and an ActiGraph (GTX3+, ActiGraph, FL, USA analyzed using ActiLife) (Act) on the non-dominant wrist. Rb uses an automated proprietary algorithm to determine LO whereas the Act requires participants to complete a diary, noting the time of LO the previous evening. Agreement between devices was assessed by Bland-Altman analyses with limits of agreement (LoA) calculated between PSG, Rb and Act. Differences between devices were assessed using linear mixed models.

Results: Compared to PSG, Rb underestimated LO by 45mins (LoA±150mins), WASO by 85mins (LoA±115mins) and SL by 14mins (LoA±88mins) and overestimated SD by 73mins (LoA±90mins) and SE by 6% (LoA±31%). Compared to PSG, Act overestimated LO by 3mins (LoA±60mins), SD by 45mins (LoA±122mins), SE by 18% (LoA±26%) and underestimated WASO by 58mins (LoA±100mins) and SL by 21mins (LoA±55mins).

Conclusion: Compared to PSG, both the Rb and Act underestimate WASO and overestimate SD and SE. Act underestimates and Rb overestimate SL. The Act device, using a sleep diary, provides a more accurate estimate of LO, although both devices had wide LoA.