J Affect Disord. 2025 Oct 9:120395. doi: 10.1016/j.jad.2025.120395. Online ahead of print.
ABSTRACT
BACKGROUND: Depressive symptoms and multiple chronic diseases (MCDs) significantly contribute to the global disease burden among middle-aged and older adults, while few studies have considered the long-term dynamics of depressive symptoms or employed machine learning (ML) models to predict the risk of MCDs. We aimed to identify the similarities and differences risk factors of MCDs based on depressive symptom trajectories among Chinese adults aged 45 and older.
METHODS: This cohort study utilized 10-year of national data from the China Health and Retirement Longitudinal Study (CHARLS), with baseline in 2011 and follow-ups in 2013, 2015, 2018, and 2020. Latent class growth modeling (LCGM) and growth mixture modeling (GMM) were employed to identify the long-term trajectories of depressive symptoms. ML algorithms were employed to develop predictive models for MCDs based on these trajectories. Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC-ROC). We also employed SHapley Additive exPlanations (SHAP) to rank the importance of risk factors and provide both global and local explanation. Finally, we developed a web application to input feature values and obtain predicted probabilities of MCDs.
RESULTS: A total of 2552 individuals were analyzed. Four distinct trajectories of depressive symptoms were identified: Stable low symptoms (75.12 %), Persistent high symptoms (6.62 %), New-onset increasing symptoms (12.15 %), and Remitting symptoms (6.11 %). The Random Forest (RF) model performed best for the “Persistent high symptoms” trajectory (AUC-ROC: 0.834 [95 % CI 0.801 to 0.862]), the Extreme Gradient Boosting (XGBoost) model for the “New-onset increasing symptoms” trajectory (AUC-ROC: 0.838 [95 % CI 0.809 to 0.864]), and the Gradient Boosting Decision Tree (GBDT) model for the “Remitting symptoms” trajectory (AUC-ROC: 0.805 [95 % CI 0.771 to 0.838]). Frequently observed risk factors were waist circumference, self-reported health, sleep duration, depressive symptom score, and age. Trajectory-specific risk factors included BMI, grip strength, and nap duration. Sensitivity analyses confirmed the robustness of these findings. The web application is available at: https://chronic-disease-prediction.streamlit.app/.
LIMITATIONS: Depressive symptoms and chronic diseases were based on self-reported data without clinical diagnosis, and findings may not be generalizable beyond the Chinese context.
CONCLUSIONS: This study provides a scientific foundation for personalized interventions and MCDs prevention among middle-aged and older adults. By identifying both frequently observed and trajectory-specific risk factors, our findings suggest that clinicians could more effectively stratify patients according to their depressive symptom trajectories. For example, individuals with persistent high symptoms may particularly benefit from intensified monitoring of waist circumference, age, and self-reported health, while those with remitting symptoms may require targeted attention to nap duration.
PMID:41076160 | DOI:10.1016/j.jad.2025.120395
Recent Comments