This study investigated the accuracy of wrist-worn accelerometers for sleep stage classification and the association between sleep duration and efficiency with mortality. A self-supervised deep neural network (SleepNet) was developed for sleep stage classification using polysomnography and accelerometry data from 1113 participant nights. SleepNet showed fair to moderate agreement with polysomnography. In the UK Biobank (66,262 participants), short sleepers (<6h) had a higher mortality risk compared to those with normal sleep duration (6-7.9h), regardless of sleep efficiency. The study suggests short sleep duration increases mortality risk, irrespective of sleep continuity.

Hang Yuan, Tatiana Plekhanova, Rosemary Walsmley, Amy C. Reynolds, Kathleen J. Maddison, Maja Bucan, Philip Gehrman, Alex Rowlands, David W. Ray, Derrick Bennett, Joanne McVeigh, Leon Straker, Peter Eastwood, Simon D. Kyle, Aiden Doherty