Development and Internal Validation of a Multivariable Prognostic Model for Sustained Remission of Depressive Symptoms in Middle-Aged and Older Adults: An Analysis of the CHARLS Cohort

Background: Depressive symptoms are common among middle-aged and older adults in China, and formal treatment is uncommon, so the course of symptoms often unfolds without intervention. Prognostic research has concentrated on who develops depression and who relapses after treatment, whereas remission from an already-symptomatic state has rarely been modeled. We developed and internally validated a model to predict sustained remission of depressive symptoms in a large community cohort. Methods: Using the China Health and Retirement Longitudinal Study (CHARLS), we identified adults aged 45 years and older with elevated depressive symptoms, defined as a score of 10 or higher on the 10-item Center for Epidemiologic Studies Depression Scale (CES-D-10), at a 2015 landmark. The outcome was sustained remission, defined as a score below 10 at both the 2018 and 2020 assessments. Fifty-four candidate predictors measured at or before the landmark were entered into a penalized logistic regression, with selection performed by the penalty within household-clustered cross-validation. Missing data were multiply imputed, discrimination was corrected for optimism using a household-cluster bootstrap, and geographic transportability was examined using internal–external validation. Results: Of 4117 participants, 818 (19.9%) achieved sustained remission. The model discriminated moderately, with an optimism-corrected C-statistic of 0.761, and was well calibrated, with a calibration slope of 1.06 and calibration-in-the-large of 0.00. Discrimination was carried principally by the recent symptom trajectory and, to a smaller degree, by pain. Behavioral, psychosocial, and sociodemographic predictors added little independent information. Performance was maintained when whole geographic regions were held out, with a pooled C-statistic of 0.754. A model restricted to information obtainable at a single assessment discriminated less well, with a C-statistic of 0.698, and discrimination was lower among participants with more persistent symptoms. Conclusions: Sustained remission was moderately and consistently predictable from routinely collected information, principally from the recent symptom trajectory and secondarily from pain. The findings support cautious prognostic stratification rather than individual prediction, with pain as a common and readily measured marker of reduced remission.