Constructing a depression risk prediction model for elderly people with activities of daily living impairment based on machine learning algorithms
HUANG Siying
MA Yan
ZHANG Yuan
WU Ruikai
HAN Zhengfeng
Abstract：Objective Explore the influencing factors of depression in elderly people with activities of daily living(ADL)disorders,and construct a depression risk prediction model for elderly people with ADL disorders in China based on machine learning(ML)algorithms.Methods Based on the fifth round of data from the China Health and Retirement Longitudinal Study(CHARLS)project,the Boruta algorithm and Lasso regression algorithm were used to screen depression risk factors in elderly people with ADL disorders.The 9 ML methods of random forest,light gradient boosting machine,extreme gradient boosting,Logistic regression,K-nearest neighbor,support vector machine,artificial neural network,decision tree,and Elastic Net regression algorithm were used to construct a depression risk prediction model,and SHAP algorithm was used to explain the final model.Results A total of 3,167 elderly individuals with ADL disorders were included,with a depression detection rate of 60.69％.The random forest model has the best predictive performance,with AUCs of 0.804[95％CI(0.788,0.820)]and 0.779[95％CI(0.752,0.806)]on the training and testing sets,respectively.The SHAP algorithm results showed that the impact of life satisfaction,pain and discomfort,self-rated health status,satisfaction with child relationships,gender,whether any falls have occurred since 2018,the number of outpatient visits to medical institutions in a month,age,education level,and whether the internet has been used in the past month on the depression risk prediction model for elderly people with ADL disorders decreased in order.The calibration curve indicates that the predicted performance of the model is basically consistent with the actual results,and the decision curve shows that the model has good clinical applicability.Conclusion Early age,female,low educational level,declining self-evaluation of health status,low life satisfaction,low child relationship satisfaction,increased number of outpatient clinics in medical institutions,falls,physical pain,and inability to use the internet significantly increased the risk of depression among the elderly with ADL disorders.Among the depression risk prediction models constructed based on ML algorithm,the random forest model achieved the optimal prediction performance.
Keywords：Activity of daily living abilityDepressionRisk prediction modelMachine learningSHAP algorithm
Publication Date：2026-01-25
Online Publishing Date：2026-03-13(First online date of this platform, not the publication date of the document)
Pages：10( 37-46 )