Joint longitudinal trajectories of depressive symptoms and grip strength as predictors of new-onset cardiovascular disease: a multi-center study across five global cohorts

Depressive symptoms and declining muscle strength frequently co-occur during aging, yet their synergistic longitudinal impact on cardiovascular disease risk remains poorly characterized. This study evaluated the predictive utility of joint depression and grip strength trajectories for new-onset cardiovascular disease (CVD) using a multi-center machine learning framework. Data from 35,055 participants across five prospective cohorts (HRS, ELSA, CHARLS, SHARE, and KLoSA) were analyzed. All participants were free of activities of daily living (ADL) disability and CVD at baseline. Group-based trajectory modeling (GBTM) identified joint longitudinal patterns of depressive symptoms and grip strength. Associations were assessed using Cox and Fine-Gray competing risk models. Seven machine learning (ML) algorithms were benchmarked for risk prediction, with clinical utility evaluated via SHAP values and decision curve analysis. Median follow-up was 12, 10, 8, 8, and 6 years in HRS, ELSA, CHARLS, SHARE, and KLoSA, with 1,236, 1,458, 1,535, 2,793, and 277 new-onset CVD events, respectively. Three trajectory patterns were identified by GBTM. The “Low Grip Strength and High Depression” trajectory was associated with a significantly higher CVD risk compared to the “High Grip/Low Depression” reference (adjusted HR: 1.52–2.60; P < 0.05), remaining robust after accounting for competing mortality (sHR: 1.50–2.57). Among the evaluated models, CatBoost showed the best overall performance, with validation AUC-ROC values of 0.670 in HRS, 0.788 in ELSA, 0.751 in CHARLS, 0.761 in SHARE, and 0.667 in KLoSA. SHAP analysis confirmed the joint trajectory as a primary predictive feature. Synergistic deterioration of physical and mental health was associated with increased risk of new-onset CVD. The CatBoost model showed potential for interpretable risk assessment, but further external validation is required before clinical application.