Abstract / Summary
Abstract Background. Mental health symptoms, sleep problems, cognitive changes, and limitations in functioning often occur together in later life. Whether the pattern of these connections differs between urban and rural older adults in China is not known. We compared nine-indicator networks and bridge strength between the two settings. Methods. We analysed cross-sectional data from 32,523 community-dwelling adults aged ≥ 65 years in the China Ageing and Health Survey (mean age, 74.2 years; 52.1% female; 62.5% urban). We estimated covariate-adjusted EBICglasso networks comprising PHQ-9, GAD-7, ISI, ESS, SCD, Cognitive-change screening (AD8), PF, GH, and SF. Urban–rural differences in the networks and bridge strength were tested with the Network Comparison Test and label permutations, respectively (1,000 each), with Benjamini–Hochberg correction. Sensitivity analyses excluded the PHQ-9 sleep item, excluded SF, and used ggmModSelect. Results. PHQ-9 had the highest node strength in the overall, urban, and rural networks (overall strength = 1.401) and the highest bridge strength in both residential groups (urban = 0.715; rural = 0.682). Its edge with GAD-7 was the strongest in the network (partial r = 0.699). The urban and rural networks differed in structure (p=.001) and global strength (p=.025), and six edge differences survived false-discovery-rate correction. After correction across nine node-strength tests, AD8, SF, and ESS had higher strength in the urban network, whereas GAD-7 had higher strength in the rural network. AD8 and ESS also had higher bridge strength in the urban network after BH correction. Excluding SF left PHQ-9 as the highest-strength node, yielded high concordance with the primary edge weights (Pearson r=.992–.995), and preserved the structural difference (p=.001). The global-strength difference was no longer significant (p=.232), and only the AD8 and ESS centrality differences survived correction. Conclusions. Depression was the most consistently connected indicator across residential settings. Urban–rural differences were concentrated in several sleep–cognition–emotion connections and in the bridge strength of AD8 and ESS. The SF-exclusion analysis preserved the main edge pattern but not the global-strength or corrected GAD-7 centrality differences. These results identify a small set of cross-domain differences for longitudinal testing.