Trajectories of health conditions predict cardiovascular disease risk among middle-aged and older adults: a national cohort study.

Background: Most previous studies have focused on the association between health conditions measured at a single time point and the risk of cardiovascular disease (CVD), while evidence regarding the impact of long-term trajectories of health conditions is limited. This study aimed to construct models of health condition trajectories and to evaluate their association with CVD risk and predictive value. Methods: This study included 2,512 participants aged 45 years and older from the China Health and Retirement Longitudinal Study (CHARLS), who were followed from 2011 to 2018. Trajectories of multimorbidity status, activities of daily living (ADLs) limitations, body roundness index (BRI), pain, sleep duration, depressive symptoms, and cognitive function were identified using latent class growth models (LCGMs). Cox regression models were used to assess associations between these trajectories and incident CVD. Ten machine learning (ML) algorithms were applied to evaluate the predictive capacity of different variable groups for CVD. Additionally, SHapley Additive exPlanations (SHAP) values were used to interpret predictor importance and direction in the machine learning models. Results: Distinct high-risk trajectories of physical and psychological health were independently associated with increased CVD risk. Higher risks of CVD were observed for the moderate-ascending (HR = 1.42, 95% CI: 1.08-1.89) and high-ascending (3.01, 2.16-4.20) trajectories of multimorbidity status; the high-ascending trajectory of ADLs limitations (2.58, 1.87-3.56); the high-stable trajectory of BRI (1.67, 1.03-2.70); the moderate-ascending (1.51, 1.07-2.12) and high-ascending (2.28, 1.56-3.35) trajectories of pain; the moderate-descending (1.51, 1.09-2.10), low-ascending (1.70, 1.22-2.38), and high-posterior-ascending (2.54, 1.69-3.82) trajectories of depressive symptoms; and the low-ascending trajectory of sleep duration (1.33, 1.02-1.74). Notably, the model based on trajectories of health conditions achieved the highest predictive performance among all variable groups (CatBoost AUC = 0.740), with SHAP analysis confirming that the trajectories of multimorbidity status, BRI, and ADLs limitations were the most influential predictors. Conclusion: Long-term deterioration in both physical and psychological health is strongly associated with increased CVD risk, highlighting the importance of early intervention and continuous health monitoring.