Fig. 4: Seven MRIBAGs mediate the effects of disturbed sleep duration patterns on two LLD subtypes.
From: Sleep chart of biological ageing clocks in middle and late life
We tested whether 7 MRIBAGs mediate the effect of abnormal sleep duration on two artificial-intelligence-derived subtypes of LLD (LLD1 and LLD2)9. The participants were grouped into short (4–6 h), normal (6–8 h; as a reference) and long (8–10 h) sleep categories: sleep duration → MRIBAG → LLD1/2. For each LLD subtype and each contrast (short versus normal, long versus normal), we fit structural equation models using the lavaan package in R. The model included a direct path from the binary-coded sleep category to the LLD subtype (depicted as solid lines with warm colours) and indirect paths mediated through the MRIBAG (shown as dotted lines with cool colours). Only statistically significant pathways (two-sided P < 0.05/7) and estimated path coefficients are displayed. All models were adjusted for relevant covariates, including age, sex, body mass index, and systolic and diastolic blood pressure, among others. Standardized estimates were obtained using 1,000 nonparametric bootstrap resamples. a, Mediational models for short sleep duration → MRIBAG → LLD1. b, Mediational models for long sleep duration → MRIBAG → LLD1. c, Mediational models for short sleep duration → MRIBAG → LLD2. d, Mediational models for long sleep duration → MRIBAG → LLD2. Our primary analyses were based on the temporal ordering of data collection, with sleep duration assessed at the baseline and both MRIBAG and LLD subtypes measured at the second visit. The sample sizes of the analyses are presented in Supplementary Table 8.
