Joint Trajectories of Depression and Cognition and New-onset Osteoporosis Risk: Evidence From Three Large Population-Based Cohorts

Background Depressive symptoms and cognitive decline frequently co-occur in older adults. However, the impact of their joint longitudinal trajectories on bone health remains unclear. This study aimed to identify synchronized trajectories of depression and cognitive function across three major global aging cohorts and to examine their associations with the risk of new-onset osteoporosis in middle-aged and older adults without diabetes. Methods Data were derived from the Health and Retirement Study (HRS), the English Longitudinal Study of Ageing (ELSA), and the China Health and Retirement Longitudinal Study (CHARLS). Group-based trajectory modeling (GBTM) was used to identify joint trajectories of depression and cognition over the first three waves. Associations between trajectory membership and new-onset osteoporosis were evaluated using Cox proportional hazards models, Fine–Gray competing risk models, and discrete-time logistic regression. In addition, seven machine learning models were developed within the MLR3 framework. Feature contributions were interpreted using SHAP values, and clinical utility was assessed using calibration curves and decision curve analysis (DCA). Results During a median follow-up of 10 years (HRS), 8 years (ELSA), and 8 years (CHARLS), 3,196, 865, and 2,003 non-diabetic participants developed osteoporosis, respectively. GBTM identified trajectories including “low depression + high cognition” (reference group), “high depression + moderate cognition”, and “moderate depression + declining cognition”. After adjustment for confounders, the “high depression + moderate cognition”, and the “moderate depression + declining cognition” trajectory was significantly associated with an increased risk of new-onset osteoporosis compared with the reference group: HRS (HR = 1.30, 95% CI: 1.17–1.45), ELSA (HR = 1.55, 95% CI: 1.29–1.86), and CHARLS (HR = 1.64, 95% CI: 1.45–1.85). These findings were consistent across competing risk and discrete-time models. Among the machine learning approaches, CatBoost demonstrated the best predictive performance across all cohorts (AUC-ROC 0.60–0.80). SHAP analysis further indicated that joint depression–cognition trajectories were key predictors of osteoporosis risk. Conclusions Worsening depressive symptoms accompanied by cognitive decline strongly predict new-onset osteoporosis in non-diabetic middle-aged and older adults. Integrating machine learning algorithms such as CatBoost into early screening may help identify high-risk individuals and support targeted intervention strategies.