Machine learning models of depression in middle-aged and older adults with cardiovascular metabolic diseases

Abstract

Background: The incidence of cardiovascular metabolic diseases (CMD) is increasing, and depression in CMD patients significantly impacts prognosis. Therefore, this study aimed to develop and validate a predictive model for depression in CMD patients using machine learning methods.

Methods: The study utilized data from the Survey of Health, Ageing, and Retirement in Europe (SHARE) for model derivation and internal validation, and data from the China Health and Retirement Longitudinal Study (CHARLS) for external validation. Logistic Regression, K-nearest neighbors, Support Vector Machine, Random Forest, Gradient Boosting Machine (GBM), and Light Gradient Boosting Machine were used to construct depression prediction models. The model performance was assessed mainly using area under the receiver operator characteristic curve (AUC), brier score, calibration plots and decision curve analysis (DCA). Model interpretations were generated using the Shapley additive explanations (SHAP) method.

Results: Among the 14,884 participants in SHARE and 1128 in CHARLS, 5456 and 474 had depression, respectively. The Gradient Boosting Machine (GBM) model demonstrated the best performance in terms of discrimination and calibration, with an AUC of 0.823 in the external validation cohort, and the DCA also verified that the GBM model had the best clinical practicality. The SHAP method revealed that trouble sleep, life satisfaction and loneliness were the top 3 predictors of depression. For the convenience of clinicians, we developed a clinical support system based on GBM model.

Conclusions: We integrated the GBM model into a clinical support system which could assist clinicians in early identifying CMD patients at high risk for depression.

Keywords: Cardiovascular metabolic diseases; Clinical decision; Depression; Machine learning; Predictive model.