Physical activity and sleep associations with episodic memory across four ageing cohorts: a coordinated within-person analysis

Abstract

Longitudinal associations of physical activity and sleep with cognitive ageing can mix between-person differences with within-person change. Consistency in direction and magnitude across cohorts remains unclear, particularly when sleep measures capture different symptoms. We compared within-person associations with episodic memory across four ageing studies. We analysed harmonised longitudinal data from CHARLS, ELSA, HRS, and SHARE. Episodic memory combined immediate and delayed recall scores standardised within cohort. Eligible participants were aged 50 years or older and had at least two waves with jointly observed memory, physical activity, and a cohort-specific sleep indicator; 136,503 participants were included. Primary estimates came from exposure-specific one-stage Mundlak within-between models fitted to complete records for the relevant exposure and covariates. Cohort-specific models, likelihood-ratio tests of cohort-specific slopes, random-effects meta-analysis with Hartung-Knapp-Sidik-Jonkman (HKSJ) inference, weighted fixed-effects models, and SHARE country analyses assessed heterogeneity and robustness. Sensitivity analyses included transition-specific inverse-probability-of-censoring weighting, item-response modelling of partially overlapping sleep items, and lagged indirect-association analyses using exact two-year intervals. Physical-activity coefficients were positive in all four cohorts, although their magnitudes differed ( P < 0.001 for differences in cohort-specific slopes). The one-stage estimate was 0.142 SD (95% CI 0.133 to 0.152), and the four-cohort random-effects summary was 0.122 (95% CI 0.039 to 0.205; I²=95.7%). Sleep-indicator estimates were positive in CHARLS and SHARE, close to zero in ELSA, and negative in HRS ( P < 0.001 for differences in cohort-specific slopes). The one-stage model estimated an average sleep association of 0.017 (95% CI 0.012 to 0.023) across the combined analytical sample; the four-cohort random-effects summary was 0.017 (95% CI -0.041 to 0.075; I²=97.3%). Survey-weighted and transition-specific IPCW analyses produced similar cohort patterns. Within SHARE, physical-activity estimates were positive in all 20 countries included in the country-specific analysis, while sleep estimates varied more widely. Estimates from the item-response analysis of partially overlapping sleep items remained heterogeneous, and lagged indirect associations were very small for sleep and inconsistent for physical activity. Physical-activity estimates were positive in all four cohorts, although their magnitudes varied. Sleep associations differed in direction across cohort-specific measures, and the item-response analysis showed substantial differences in item coverage and item information.