Automatic Sleep/Wake Identification From Wrist Activity

TLDR

The study aimed to develop and validate automatic sleep/wake scoring algorithms using wrist actigraphy. Forty-one participants wore wrist actigraphs during overnight polysomnography; algorithms were optimized on a random 20‑subject subsample and then prospectively tested on the remaining 21 subjects across various epoch lengths. The optimized algorithms achieved about 88 % accuracy in distinguishing sleep from wake, with actigraphic sleep percentage and latency correlating 0.82 and 0.90 with polysomnographic measures, demonstrating their usefulness for clinical and research applications.

Abstract

The purpose of this study was to develop and validate automatic scoring methods to distinguish sleep from wakefulness based on wrist activity. Forty-one subjects (18 normals and 23 with sleep or psychiatric disorders) wore a wrist actigraph during overnight polysomnography. In a randomly selected subsample of 20 subjects, candidate sleep/wake prediction algorithms were iteratively optimized against standard sleep/wake scores. The optimal algorithms obtained for various data collection epoch lengths were then prospectively tested on the remaining 21 subjects. The final algorithms correctly distinguished sleep from wakefulness approximately 88% of the time. Actigraphic sleep percentage and sleep latency estimates correlated 0.82 and 0.90, respectively, with corresponding parameters scored from the polysomnogram (p < 0.0001). Automatic scoring of wrist activity provides valuable information about sleep and wakefulness that could be useful in both clinical and research applications.