Machine learning model predicts risk of postoperative delirium in total hip replacement patients
ZHANG Xiaoying
LIU Wei
XIE Meiying
ZHOU Jianguo
YANG Jia
Abstract：Objective:To predict the risk of postoperative delirium in patients undergoing total hip arthroplasty using machine learning models.Methods:A total of 622 patients who underwent total hip arthroplasty in Ganzhou People's Hospital were selected as research subjects from January 2020 to December 2024.The Confusion Assessment Method(CAM)was used to assess postoperative delirium.The Boruta algorithm was employed to screen for important feature variables associated with postoperative delirium risk.Patients were randomly divided into training set(442 cases)and testing set(180 cases)at 7∶3 ratio.Nine machine learning models were constructed,trained,and validated using ten-fold cross-validation.The area under the curve(AUC)of receiver operator characteristic was used to evaluate model performance and identify the best machine learning model.Decision curve analysis was used to assess the clinical utility of the model.The SHapley additive explanations(SHAP)method,including bar plots,summary plots,dependence plots,and force plots,was used to interpret and visualize the machine learning models.Results:The incidence of postoperative delirium among the 622 patients undergoing total hip arthroplasty was 30.87%.The Boruta algorithm identified nine important postoperative delirium risk feature variables.Based on the feature importance scores(Z-values),the ranking from highest to lowest was C-reactive protein(CRP),anesthesia duration,albumin(ALB),age,total bilirubin(TB),blood glucose,intraoperative blood loss(IBL),history of diabetes,and cerebrovascular disease(CSD).Multivariate Logistic regression analysis showed that age,ALB,TB,blood glucose,CRP,and anesthesia duration were independent influencing factors for postoperative delirium in patients undergoing total hip arthroplasty(all P<0.05).The XGBoost model demonstrated excellent performance in both the training and test sets,exhibiting the strongest robustness and predictive efficacy for estimating the risk of postoperative delirium in patients undergoing total hip arthroplasty.Interpretation and visualization of the XGBoost model using SHAP revealed that the model could predict postoperative delirium risk in patients undergoing total hip arthroplasty with high accuracy.Conclusions:Age,ALB,TB,blood glucose,CRP,anesthesia duration are independent influencing factors for postoperative delirium in patients undergoing total hip arthroplasty.The XGBoost model demonstrated high predictive value for postoperative delirium in patients undergoing total hip arthroplasty.
Keywords：total hip arthroplastypostoperative deliriuminfluencing factorsmachine learningBoruta algorithmSHapley additive explanationsSHAPXGBoost model
Publication Date：2026-03-25
Online Publishing Date：2026-03-26(First online date of this platform, not the publication date of the document)
Pages：12( 894-905 )