PennZzz - an algorithm for estimating behavioral states from wrist-worn accelerometery

Abstract

Sleep is a heterogeneous behavioral state comprised of different stages and interspersed with episodes of wakefulness. Sleep/wake states can be monitored in the sleep laboratory by polysomnography (PSG). However, sleep studies are intrusive, laborious and expensive, and are usually performed over a single night. In contrast, wrist-worn activity-tracking devices (actimeters) are inexpensive, unobtrusive, and can be used to estimate sleep and wake patterns over multiple nights. We designed the PennZzz algorithm to estimate sleep and wake from actimetry data. Results obtained by actimetry-based monitoring in 26 subjects were compared to stages of sleep and wakefulness detected by simultaneous polysomnography. We found that our algorithm identifies PSG-defined wake episodes with a high accuracy (336/431 - 76% of algorithm wake events correspond to true wakefulness). Furthermore, we find that the algorithm is sensitive enough to detect the majority (258/431 - 59%) of true wake episodes occurring after the first NREM1 to NREM2 transition. With correction, algorithm outputs can be used to estimate the total amount of time awake after sleep onset. We further refined this program for application in a high-throughput manner to assess the total amount of sleep, wake, and non-wear during longer recording periods.