Machine learning-based risk screening model for depressive symptoms in middle-aged and elderly cardiovascular disease patients

Cardiovascular disease (CVD) and depression frequently co-occur, yet systematic depression screening remains underutilized in routine cardiac care. This study aimed to develop and compare machine learning algorithms for identifying concurrent depressive symptoms in middle-aged and elderly CVD patients. Data from 2,204 participants aged ≥ 45 years with physician-diagnosed CVD were obtained from the 2018 China Health and Retirement Longitudinal Study. Depressive symptoms were defined as CES-D10 score ≥ 10 (prevalence: 54.1%). Seven algorithms—logistic regression (LR), artificial neural networks, XGBoost, LightGBM, decision tree, random forest, and support vector machine—were trained and internally validated. Feature selection used LASSO regression. Performance was assessed via AUC-ROC, Brier score, calibration curves, and decision curve analysis, with bootstrap correction for optimism. Among 2,204 participants, 1,193 had depressive symptoms. The LR model showed strong discrimination (AUC = 0.832, 95% CI: 0.804–0.860), though performance differences among the top models were minimal (AUC range: 0.826–0.832). LR was selected for the final nomogram based on its favorable calibration and interpretability. Significant predictors included loneliness, life satisfaction, self-rated health, hearing status, gender, ADL score, sleep, and cognitive function. A web-based dynamic nomogram tool was developed as a research prototype to facilitate individualized risk estimation and illustrate the model’s potential application. An LR-based model using eight accessible variables demonstrated strong discrimination for identifying concurrent depressive symptoms in CVD patients. The nomogram may serve as a preliminary triage tool to prompt confirmatory assessment; external validation is required before clinical implementation.