Development and cross national external validation of a model for incident depressive symptoms in adults aged 50 years or older

Depressive symptoms in older adults coexist with chronic disease and functional limitations, yet cross-national longitudinal prediction remains limited. We developed and externally validated a model to estimate the 4-year risk of screening-defined incident depressive symptoms in adults aged 50 years or older. The China Health and Retirement Longitudinal Study (CHARLS) was used for development and internal validation, and the Health and Retirement Study (HRS) and English Longitudinal Study of Ageing (ELSA) for external validation. Nine approaches were compared using repeated nested cross-validation with multiple imputation. The cohorts included 5,816, 11,337, and 5,818 participants, respectively, with 4-year event proportions of 36.1%, 21.2%, and 19.8%. Among the nine candidate approaches, logistic regression was selected for the final model, with an area under the receiver operating characteristic curve (ROC AUC) of 0.675 (95% confidence interval [CI], 0.661–0.689) in internal validation. External ROC AUCs were 0.728 (95% CI, 0.717–0.739) in HRS and 0.745 (95% CI, 0.730–0.761) in ELSA. Brier scores were 0.1505 and 0.1399, and observed-to-expected event ratios were 0.814 and 0.844, respectively. A web-based risk calculator was developed. The model showed moderate external discrimination but overestimated absolute risk. Local recalibration and prospective impact evaluation are needed before individual risk assessment in new populations.