Construction of a prediction model for depression risk in elderly patients with chronic obstructive pulmonary disease based on machine learning algorithms and analysis of influencing factors

Patients with chronic obstructive pulmonary disease (COPD) are at a high risk of depression, which not only accelerates disease progression but also significantly reduces patients' quality of life. This study aimed to develop a model for the accurate prediction of depression risk in COPD patients using machine learning techniques. A total of 2234 patients with COPD were enrolled from the China Health and Retirement Longitudinal Study (CHARLS) 2020 wave, and 42 indicators covering behavioral, health, psychological and sociodemographic domains were analyzed. The least absolute shrinkage and selection operator (LASSO) regression was applied to screen predictors, and 9 machine learning models (including light gradient boosting machine (LightGBM), support vector machine (SVM), and multilayer perceptron neural network (MLP)) were constructed to identify the optimal predictive model. In addition, temporal validation was performed using CHARLS 2015 data, and model interpretation was conducted with Shapley additive explanations (SHAP). Among the 2234 included patients, 1007 (44.1%) presented with depressive symptoms. Seventeen key variables were identified and used for model construction. The results demonstrated that the LightGBM model exhibited the best performance in terms of discrimination, calibration and clinical utility, with an area under the receiver operating characteristic curve (AUROC) ranging from 0.76 to 0.81. In the validation dataset, the LightGBM model achieved an accuracy of 75.38%, a sensitivity of 78.39%, a precision of 81.14%, a specificity of 70.51%, an F1 score of 79.74% and an AUC of 0.81. Based on the optimal LightGBM model, this study provides a potentially useful approach for assessing the risk of depression in patients with COPD. The model may support early risk identification and facilitate subsequent clinical evaluation, although further validation is required to confirm its practical applicability. (1) Nine machine learning methods were used to construct the prediction model, and the LASSO method was applied to screen key factors. (2) The model was trained using the latest 2020 cohort of CHARLS and validated using the 2015 cohort. (3) The models were comprehensively compared using ROC curves, precision-recall curves, calibration curves, and DCA curves to identify the optimal model. (1) This study had a retrospective design, and the data lacked key variables such as gold-standard pulmonary function measures. (2) Validation was only performed using different time-series datasets from the same database; external validation should be conducted in future studies.