Machine learning models of depressive symptoms in middle-aged and elderly patients with chronic kidney disease: A cross-sectional study based on CHARLS.

This study aimed to develop and compare the performance of machine learning models in identifying depressive symptom status among middle-aged and elderly patients with chronic kidney disease (CKD), while identifying key factors that influence depressive symptoms. Data from the 2015 China Health and Retirement Longitudinal Study were used to construct the training and validation sets, while data from 2011 served as the independent external test set. The core feature variables were screened through the intersection of the Boruta algorithm, least absolute shrinkage and selection operator regression, and multivariate logistic regression. Seven machine learning algorithms were applied to build models for identifying depressive symptom status: logistic regression, decision tree, random forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine, support vector machine, and artificial neural network. The optimal model was selected based on discrimination ability, calibration accuracy, and clinical applicability. SHapley Additive exPlanations were used to interpret feature contributions to the model. A total of 2088 CKD patients were included to construct the training and validation sets, and 1106 CKD patients constituted the external test set. Ten core feature variables were identified. In the validation set, the XGBoost model performed best in identifying depressive symptom status, with an area under the receiver operating characteristic curve of 0.814 (95% confidence interval: 0.782-0.848), accuracy of 74.92%, precision of 75.94%, specificity of 80.66%, and F1 score of 72.01%. In the test set, the performance of the XGBoost model remained stable. SHapley Additive exPlanations analysis revealed that pain, life satisfaction, physical function, and sleep were the most crucial variables influencing depressive symptoms. The XGBoost model exhibited strong discriminative ability for depressive symptom status in middle-aged and elderly CKD patients and could identify key predictors related to depressive symptoms to inform clinical understanding.