广东医学2026，Vol.47Issue(7)：1003-1014,12.DOI:10.13820/j.cnki.gdyx.20261276
慢性肾脏病患者骨质疏松危险因素与多机器学习临床预测列线图构建
Risk factors for osteoporosis in patients with chronic kidney disease and development of a clinical nomogram based on multiple machine learning algorithms
摘要
Abstract
Objective To develop and validate a clinical prediction model for secondary osteoporosis in patients with chronic kidney disease(CKD)using multiple machine learning algorithms and to construct a nomogram for individu-alized risk assessment.Methods A retrospective analysis was performed on the clinical data of 242 patients with CKD admitted to the General Hospital of Southern Theater Command of the Chinese People's Liberation Army between January 2016 and December 2025.Among them,69 patients(28.5％)were diagnosed with osteoporosis by dual-energy X-ray absorptiometry(DXA).Candidate predictors were jointly selected using least absolute shrinkage and selection operator(LASSO)regression and the Boruta algorithm.Seven machine learning models were developed,including logistic regres-sion(LR),decision tree(DT),extreme gradient boosting(XGBoost),gradient boosting machine(GBM),support vec-tor machine(SVM),neural network(NN),and k-nearest neighbor(KNN).The dataset was randomly divided into training and test sets at a ratio of 7∶3.Model performance was evaluated using 10-fold cross-validation and leave-one-out cross-validation(LOOCV),with optimism-corrected area under the receiver operating characteristic curve(AUC)reported.LR and XGBoost were predefined as the primary analytical models,whereas the remaining algorithms were con-sidered exploratory.Calibration curves and decision curve analysis(DCA)were used to assess model calibration and clini-cal utility.SHapley Additive exPlanations(SHAP)were applied to interpret the optimal model,and a nomogram was sub-sequently established.Results Four key predictors were consistently identified by both LASSO regression and the Boruta algorithm:body weight,alanine aminotransferase(ALT),alkaline phosphatase(ALP),and low-density lipoprotein cholesterol(LDL-C).In the test sets cohort,the LR model achieved an AUC of 0.790 4,with an accuracy of 0.726,sensitivity of 0.842,and specificity of 0.685.After optimism correction using LOOCV,the AUC increased to 0.804(95％CI:0.755-0.853).The XGBoost model yielded a validation AUC of 0.700 3,with an optimism-corrected AUC of 0.751.The remaining models(DT,GBM,SVM,NN,and KNN)demonstrated varying degrees of overfitting or inferi-or predictive performance.Calibration analysis indicated systematic compression of predicted probabilities in the LR model(calibration slope=0.32),suggesting that although the model exhibited satisfactory discrimination,its absolute risk esti-mates should be interpreted with caution.DCA demonstrated a favorable net clinical benefit of the LR model across a clini-cally relevant range of threshold probabilities.A nomogram incorporating body weight,ALT,ALP,and LDL-C was estab-lished based on the LR model and achieved an AUC of 0.792,comparable to that of the complete model while substantially improving clinical applicability.SHAP analysis identified body weight,LDL-C,and ALP as the most influential positive predictors.Conclusion The logistic regression model and nomogram incorporating body weight,ALT,ALP,and LDL-Cdemonstrated good discriminative ability for predicting osteoporosis in patients with CKD.Owing to its simplicity and in-terpretability,the model may serve as a practical tool for risk stratification,particularly in primary healthcare settings.Nevertheless,given the limited sample size and imperfect calibration,the predicted absolute probabilities should be inter-preted cautiously.Further prospective multicenter studies are warranted to externally validate and refine the model.关键词
慢性肾脏病/骨质疏松/机器学习/预测模型/神经网络/列线图Key words
chronic kidney disease/osteoporosis/machine learning/prediction model/neural network/nomo-gram分类
医药卫生引用本文复制引用
杨柳,朱凯旋,李佳..慢性肾脏病患者骨质疏松危险因素与多机器学习临床预测列线图构建[J].广东医学,2026,47(7):1003-1014,12.基金项目
国家重点研发计划项目(2021YFC2501701) （2021YFC2501701）
国家卫生健康委能力建设和继续教育中心2025年度慢病管理研究课题(GWJJMB202510024006) （GWJJMB202510024006）
