Lagged symptom-level associations among pain, sleep disturbance, and depressive-affective symptoms in adults aged ≥50 years: an international longitudinal multi-cohort study

Abstract

Background Pain, sleep disturbance, and depressive-affective symptoms commonly co-occur in adults aged ≥ 50 years, but their temporal ordering at the individual symptom level remains unclear. We estimated lagged symptom-level associations across five international aging cohorts and quantified heterogeneity between cohorts. Methods We analyzed harmonized longitudinal data from five aging cohorts with repeated pain and affective symptom measures: ELSA, HRS, KLoSA, MHAS, and SHARE. CHARLS and LASI were excluded due to the absence of repeated direct pain measures or insufficient longitudinal waves. Participants contributed complete adjacent-wave observations when six binary symptom nodes were measured at waves t and t + 1. Cohort-specific weighted logistic cross-lagged panel network models adjusted for lagged symptoms, sociodemographic covariates, wave indicators, participant clustering, survey weights, and inverse probability of attrition weights. Directed network edges were pooled using REML random-effects meta-analyses with Benjamini–Hochberg false discovery rate correction. A five-node sensitivity analysis excluded the low positive affect node. Results The analysis included 123,608 participants contributing 405,472 complete adjacent-wave transitions. Sleep disturbance was the most precise pooled lagged predictor of later depressed mood (OR 1.41, 95% CI 1.36–1.46). Pain predicted later sleep disturbance, fatigue/effort, depressed mood, and loneliness, whereas sleep disturbance and fatigue/effort bidirectionally predicted later pain. Several clinically relevant associations were statistically significant but highly heterogeneous, most notably pain predicting sleep disturbance (I2 = 96.8%), pain predicting fatigue/effort (I2 = 96.9%), and loneliness predicting depressed mood (I2 = 92.5%). These associations remained statistically significant in Hartung–Knapp sensitivity analyses, though prediction intervals indicated substantial between-cohort variation. The five-node sensitivity analysis yielded similar estimates for all main pathways. Conclusion Pain, sleep disturbance, loneliness, fatigue/effort, and depressive-affective symptoms showed consistent prospective lagged associations across five international aging cohorts, though effect sizes varied substantially across populations. These findings can inform joint symptom assessment and identify promising symptom pairs for future within-person and causal-inference research. However, they do not establish direct causal symptom dynamics, actionable intervention targets, or country-level differences.