Abstract / Summary
Living arrangements may reflect both social support and care needs in later life, but longitudinal evidence among older adults with mild cognitive impairment (MCI) remains limited. This study examined the longitudinal associations of time-varying living arrangements with all-cause mortality and longitudinal cognitive trajectories among older Chinese adults who were first observed to meet the study-defined mild cognitive impairment (MCI) criterion. This longitudinal cohort study used data from the Harmonized Version D of the China Health and Retirement Longitudinal Study (CHARLS, 2011–2018). A dynamic cohort included participants who were first observed to meet the study-defined MCI criteria at Wave 1, 2, or 3. Living arrangements were classified into five categories and treated as a time-varying exposure. All-cause mortality was analyzed using start–stop Cox models based on directly observed living-arrangement intervals, with wave-updated health and functional covariates. Cognitive trajectories were analyzed using mixed-effects models incorporating quadratic time. Sensitivity analyses examined alternative exposure specifications, time-varying confounding, informative cognitive observation, alternative MCI definitions, and available survey design information. The dynamic cohort included 1,678 participants, of whom 1,634 had positive post-entry survival follow-up. Overall, 233 deaths occurred; 220 deaths contributed to the directly observed primary time-varying Cox analysis. The global test for the association between living arrangement and mortality did not reach statistical significance ( P = 0.051). Compared with living with both a spouse and adult children, living with a spouse was associated with a lower hazard estimate ( HR = 0.690, 95% CI 0.490–0.971), whereas the estimate for living alone was not statistically significant ( HR = 0.718, 95% CI 0.448–1.148). The overall association was attenuated in baseline-fixed, lagged-exposure, post-transition exclusion, and marginal structural model analyses. For cognition, 3,335 assessments were available. Cognitive trajectories were nonlinear, showing early positive change followed by leveling and later negative change. The primary unweighted model showed no overall difference in trajectories across living-arrangement categories (global interaction P = 0.142). Stronger evidence of between-group differences emerged in sensitivity analyses using inverse-probability-of-observation weighting, sampling weights with community-level PSU clustering, and education-specific MCI thresholds. Among older Chinese adults meeting the MCI criterion, the associations of living arrangements with mortality and cognitive trajectories were sensitive to analytic specification. The mortality findings were compatible with health selection and reverse causation, while cognitive findings varied with attrition adjustment and MCI classification. Living-arrangement changes may therefore be better viewed as contextual indicators of evolving health and care circumstances than as evidence that particular household configurations are uniformly beneficial or harmful.