遵义医科大学学报2025，Vol.48Issue(9)：919-931,13.
基于机器学习的老年髋部骨折术后严重并发症预测模型构建和验证
Construction and verification of predictive model for severe complications after hip fracture in elderly patients based on machine learning
摘要
Abstract
Objective To use machine learning method to construct a predictive model of severe postoperative complications in elderly patients with hip fractures to assist clinical decision-making.Methods A total of 685 eld-erly patients with hip fracture surgery admitted to Chongqing University Three Gorges Hospital from January 2018 to January 2024 were selected as the research subjects,including 548 cases in the non-severe complication group after surgery and 137 cases in the severe complication group after surgery.Data preprocessing characteristics se-lect common clinical indicators for surgical patients(general data,surgery-related information and laboratory tests),and all patients are randomly divided into training sets(n=480)and test sets(n=205)according to a 7∶3 ratio.After balancing the dataset using the SMOTE(Synthetic Minority Oversampling Technique)algo-rithm,a risk prediction model for postoperative outcomes in elderly hip fracture patients was developed based on 12 machine learning algorithms,including Random Forest,Gradient Boosting Decision Tree(GBDT),and Cate-gorical Boosting(CatBoost),etc.The model was evaluated by area under the ROC curve(AUC),sensitivity,specificity,accuracy,and F1 score.CatBoost with excellent performance was selected for SHAP(Shapley addi-tional explanations)characteristic variable analysis and clinical decision curve(DCA)verification.Results LightGBM,CatBoost,and XGBoost have high prediction accuracy,and their AUCs are 0.916,0.904 and 0.898,respectively.The top 10 characteristic variables that are predictive for severe postoperative complications are:fracture type,combined with other severe chronic diseases,coronary heart disease,surgical type,hemoglo-bin,age,gender,blood potassium,surgery timing,and arrhythmia.Decision curve analysis suggests that active clinical intervention within the threshold range of 0.1-0.8 may have significant benefits.Conclusion The predic-tive model constructed by machine learning using data from a hospital can help predict the probability of serious complications after hip fracture surgery in the hospital,and is conducive to guiding preoperative decision-making and personalized disposal.关键词
髋部骨折/机器学习/术后并发症/风险因素/老年人/预测模型Key words
hip fracture/machine learning/postoperative complications/risk factor/aged/prediction model分类
医药卫生引用本文复制引用
刘王卫,杨勇..基于机器学习的老年髋部骨折术后严重并发症预测模型构建和验证[J].遵义医科大学学报,2025,48(9):919-931,13.基金项目
重庆市科卫联合医学科研青年项目(NO:2022QNXM039). （NO:2022QNXM039）
