终末期肾病患者骨量异常风险预测模型的建立与验证  

Development and validation of a prediction model for abnormal bone mass in end‐stage renal disease patients

作　　者：陆静 王雨佳 张玉霞 陈志清 李永琪 伍敏[2] 汤日宁 Lu Jing;Wang Yujia;Zhang Yuxia;Chen Zhiqing;Li Yongqi;Wu Min;Tang Rining(Southeast University School of Medicine,Nanjing 210009,China;Department of Nephrology,Zhong Da Hospital,Instisution of Nephrology,School of Medicine,Southeast University,Nanjing 210009,China;Department of Nephrology,Zhong Da Hospital Lishui Branch,Southeast University,Nanjing 211200,China)

机构地区：[1]东南大学医学院,南京210009 [2]东南大学附属中大医院肾内科,东南大学肾脏病研究所,南京210009 [3]东南大学附属中大医院溧水分院肾内科,南京211200

出　　处：《中华肾脏病杂志》2024年第5期345-357,共13页Chinese Journal of Nephrology

基　　金：国家自然科学基金(82370742);江苏省卫生健康委员会重点项目(ZD2022045)。

摘　　要：目的旨在早期识别终末期肾病(end‐stage renal disease,ESRD)患者骨量异常的危险因素,并建立和验证骨量异常风险预测模型及构建动态列线图。方法本研究为回顾性横断面研究,回顾性收集2022年1月至2023年5月就诊于东南大学附属中大医院肾内科的ESRD患者的临床资料。按7∶3的比例将患者随机分为训练组和验证组,根据双能X线吸收测量仪测量的骨密度T值进一步将患者分为骨量正常组和骨量异常组。分别使用向后逐步Logistic回归、最小绝对值收敛和选择算子法(least absolute shrinkage and selection operator,LASSO)构建ESRD患者骨量异常风险预测模型。采用赤池信息准则(akaike information criterion,AIC)、贝叶斯信息准则(bayesian information criterion,BIC)和准确性评估两个模型的性能,进而选取最优模型。采用受试者工作特征曲线(receiver operating characteristic curve,ROC曲线)的曲线下面积(area under curve,AUC)、校准曲线、拟合优度检验(Hosmer‐Lemeshow test)和决策曲线(decision curve analysis,DCA)对最优模型进行评价以判断其区分度、校准度和临床实用性。在最优模型基础上,构建个体化动态列线图。结果最终共纳入254例ESRD患者,其中男性160例(63.0%),血液透析161例(63.4%),骨量异常202例(79.5%)。训练组(n=178)和验证组(n=76)骨量异常患病率差异无统计学意义(79.2%比80.3%,χ^(2)=0.036,P=0.849)。LASSO和逐步回归模型最终纳入的变量及变量参数一致,均为年龄、体重指数(body mass index,BMI)、高血压、糖尿病和骨钙素(osteocalcin,OC)5个变量。两个模型在训练组中的AIC、BIC和准确性也相同,分别为113.45、132.54和0.837。因此,本研究中LASSO模型和逐步回归模型性能一致,可视为同一模型,将其命名为Model。预测模型Model在训练组和验证组的AUC分别为0.923(95%CI 0.884~0.963)和0.809(95%CI 0.675~0.943),在训练组的最佳截断值为0.858,敏感性为0.801,特异性Objective To identify the risk factors,and develop and validate a risk prediction model for abnormal bone mass in end-stage renal disease(ESRD)patients.Methods It was a retrospective cross-sectional study.The clinical and laboratory data of ESRD patients who were hospitalized in the Department of Nephrology,Zhongda Hospital Affiliated to Southeast University from January 2022 to May 2023 were collected retrospectively.The patients were randomly divided into training and validation cohorts at a ratio of 7∶3.They were further divided into normal and abnormal bone mass groups according to the T value measured by dual-energy X-ray absorptiometry(DXA).Then,backward stepwise regression and least absolute shrinkage and selection operator(LASSO)were respectively used to develop the risk prediction model for abnormal bone mass in ESRD patients.Akaike information criterion(AIC),bayesian information criterion(BIC),and accuracy were used to evaluate the performance of these two models,after which the preferable model was selected.Moreover,the receiver operating characteristic(ROC)curve,calibration curve,Hosmer-Lemeshow test,and decision curve analyses(DCA)were applied to evaluate the diagnostic performance of the preferable model.Finally,a dynamic nomogram for individual assessment was constructed based on the preferable model.Results A total of 254 ESRD patients were enrolled,including 160(63.0%)males,161(63.4%)hemodialysis patients,and 202(79.5%)patients with abnormal bone mass.There was no significant difference in the prevalence of abnormal bone mass between training group(n=178)and validation group(n=76)(79.2%vs.80.3%,χ^(2)=0.036,P=0.849).The final variables and variable parameters included in the LASSO and stepwise regression models were the same,which were five variables:age,body mass index,hypertension,diabetes,and osteocalcin.Both models also had the same AIC,BIC,and accuracy in the training group,which were 113.45,132.54,and 0.837,respectively.Therefore,the LASSO model and the stepwise regression model performed

关 键 词：肾功能衰竭 慢性 骨和骨组织 骨质疏松 列线图 终末期肾病 骨量异常 LASSO回归 预测模型 

分 类 号：R692[医药卫生—泌尿科学]