Evolving latent chronic disease states and episodic memory trajectories among middle-aged and older adults in China: evidence from CHARLS, 2011–2020

Background Chronic conditions tend to cluster, yet little is known about how those clusters change over time or whether they track with memory ageing. We mapped transitions between latent chronic disease states and examined their association with episodic memory in middle-aged and older adults in China. Methods We analysed five waves (2011, 2013, 2015, 2018, and 2020) of the China Health and Retirement Longitudinal Study. Eligible participants were aged 45 years or older and had chronic disease and episodic memory data from at least two waves. A time-heterogeneous latent Markov model was fitted to 12 self-reported physician-diagnosed conditions. Linear mixed-effects models related time-varying latent states to within-wave standardised episodic memory, adjusting for age, sex, education, and time-varying cancer history. Results The latent Markov analysis included 18,891 participants; 18,884 participants and 69,093 person-wave observations contributed to the primary mixed-effects model. The eight-state solution identified low-burden, hypertension-dominant, hypertension–arthritis, diabetes–metabolic, arthritis-dominant, digestive–arthritis, respiratory–musculoskeletal, and multisystem high-burden states. From 2011 to 2020, the low-burden state fell from 37.25% to 23.07%, while the multisystem high-burden state rose from 4.98% to 10.82%. Memory trajectories differed by state (time × state: F = 4.995, P < 0.001), although the contrasts were small. The hypertension-dominant versus low-burden difference accumulated to about 0.12 SD over nine years. High-confidence classifications, a seven-state solution, and raw-score analyses restricted to 2011–2018 gave similar evidence of trajectory heterogeneity. Conclusions Chronic diseases followed structured hypertension/metabolic, arthritis-related, respiratory–musculoskeletal, and multisystem pathways. These states were associated with different relative episodic memory trajectories, but the between-state effects were modest and sensitive to outcome scaling and cross-wave comparability. The findings support population-level risk stratification rather than deterministic predictions for individual patients.