Accelerometer-derived real-world sleep stages and risk of incident diseases: A UK Biobank cohort study and phenome-wide association analysis

BACKGROUND: Previous studies utilizing self-reported data or polysomnography have found that poor sleep quality is associated with increased risk of morbidity and mortality. However, the association between real-world sleep patterns, including sleep stages, duration, and fragmentation, measured objectively with physiological data and the risk of incident disease among middle-aged and older adults has not yet been systematically evaluated. METHODS AND FINDINGS: In this cohort study, we analyzed wrist-worn accelerometer data from 95,559 Biobank participants and derived key metrics of real-world sleep patterns: rapid eye movement [REM], N1, N2, and N3; total sleep duration; sleep irregularity; and wakefulness after sleep onset using the SleepNet algorithm. Participants were followed up from the date of accelerometer wear until the first occurrence of disease, death, or April 1, 2024 with median follow-up length of 8.9 years. Phenome-wide association analysis and restricted cubic spline (RCS) analyses were performed using Cox proportional hazard regression, with adjustment for demographic characteristics, lifestyle factors, and environmental exposures, to map an atlas of associations between real-world sleep patterns and the incidence of 1,049 health outcomes. We found that variations in sleep patterns were associated with 156 incidences of diseases. Specifically, higher amount of REM sleep and deep sleep were associated with lower risks of 83 and 7 diseases, respectively, while greater sleep irregularity and increased WASO were linked to elevated risks of 3 and 6 diseases, respectively. RCS analyses revealed significant non-linear relationships between sleep duration and 86 disease phenotypes (P for nonlinear <0.05), with the minimum-risk hours for 69 phenotypes predominantly concentrated within the 6-8 hours window. In the category-specific analysis, individuals with extreme short sleep (<5 hours) exhibited the most widespread clinical vulnerabilities, accounting for 37 of the 41 identified significant adverse associations compared to the 6-8 hours reference group. The main limitation of this study was its observational design, which remains susceptible to residual confounding and precludes causal inference. CONCLUSION: In summary, this study characterizes real-world sleep patterns and their links to disease risk across multiple systems, highlighting the differential contributions of sleep stages and sleep duration. The findings underscore that maintaining 6-8 hours of sleep is associated with a more favorable sleep architecture and lower disease risk, offering insights for prevention and health promotion.