Fig. 5: Associations of sleep restriction-recovery period frequencies with all-cause mortality risk.

These analyses quantified the number and type of restriction and post-restriction periods experienced by each participant and examined their associations with mortality risk. a Associations in the overall sample. Model 1 was adjusted for age, sex, ethnicity, and number of days of valid accelerometer wear. Model 2 additionally included employment status, education, BMI, smoking status, alcohol frequency, physical activity, night shift work, and early chronotype. Model 3 further accounted for baseline sleep duration (P values [1 and 2 vs. 0]: \(7.97\times {10}^{-4}\) and \(1.41\times {10}^{-4}\) for SR without recovery; 0.073 and 0.930 for SR with recovery). b Associations by baseline sleep duration. Models were adjusted for the same set of covariates as in Model 2 in (a). P values (1 and 2 vs. 0): for SR without recovery, short (\(9.34\times {10}^{-3}\), \(5.58\times {10}^{-5}\)), medium (0.175, 0.508), and long sleepers (0.040, 0.752); for SR with recovery, short (0.919, 0.243), medium (0.072, 0.628), and long sleepers (0.083, 0.308). Cox proportional hazards regression models with two-sided Wald tests were used to estimate HRs for all-cause mortality. Data are presented as HRs with 95% CIs. SR was defined as cumulative sleep loss of 2.5 h or more relative to individual sleep need in each restriction period, while sleep recovery was defined as an extra sleep duration of more than 0 h relative to individual sleep need in each post-restriction period. Baseline sleep duration categories (short/medium/long) were defined based on the tertiles of average baseline sleep duration in 85,618 participants with sleep restriction-rebound patterns. Source data are provided as a Source data file. SR sleep restriction, HR hazard ratio, CI confidence interval, BMI body mass index, P uncorrected P value. *P < 0.05, **P < 0.01, and ***P < 0.001.