慢性肾脏病合并心力衰竭诊断模型的建立及验证  

Establishment and validation of diagnostic model of chronic kidney disease complicating heart failure

作　　者：刘真意 姚瀚鑫[1] 宋媛媛[1] 王晶莹[2] 王莹[1] LIU Zhenyi;YAO Hanxin;SONG Yuanyuan;WANG Jingying;WANG Ying(Department of Clinical Laboratory,First Hospital of Jilin University,Changchun,Jilin 130021,China;Department of Clinical Laboratory,China-Japan Union Hospital of Jilin University,Changchun,Jilin 130021,China)

机构地区：[1]吉林大学第一医院检验科,吉林长春130021 [2]吉林大学中日联谊医院检验科,吉林长春130021

出　　处：《检验医学与临床》2024年第12期1714-1720,共7页Laboratory Medicine and Clinic

基　　金：吉林省卫生健康科技能力提升项目(2022LC106)。

摘　　要：目的挖掘、分析慢性肾脏病(CKD)患者实验室检测数据,建立其合并心力衰竭的诊断模型,并评价模型的性能。方法采用横断面研究,选取2021年1月至2023年1月于吉林大学第一医院确诊为CKD的799例患者为研究对象,其中单纯CKD 341例、CKD合并心力衰竭458例。所有患者均进行血糖、肝功能、肾功能、血脂、血常规、心肌损伤标志物及其他指标检测,比较单纯CKD与CKD合并心力衰竭患者的一般资料及各实验室指标水平,排除P≥0.05的指标,再经Lasso回归筛选变量。采用5种机器模型算法,即极值梯度提升(XGBoost)、支持向量机(SVM)、随机森林(RF)、梯度提升机(GBM)和逻辑回归(LR)建立诊断模型,采用受试者工作特征(ROC)曲线筛选最优模型。应用ROC曲线、校准曲线和临床决策曲线评价模型的鉴别能力、拟合优度与临床价值,最后选取吉林大学中日联谊医院75例CKD患者检测结果进行外部验证。结果训练集XGBoost模型在预测CKD合并心力衰竭方面更高效,XGBoost模型的曲线下面积(AUC)明显高于RF、SVM、GBM、LR模型(Z=5.192、5.597、5.597、6.271,P<0.001),依据XGBoost模型确定CKD合并心力衰竭的6个重要性权重变量为尿酸、可溶性生长刺激表达基因2蛋白(ST2)、N末端B型利钠肽前体(NT-proBNP)、血糖(GLU)、γ-谷氨酰转移酶(γ-GGT)和年龄。在内部验证集中XGBoost模型的AUC为0.778,95%CI:0.705~0.850,区分度好。校准曲线显示,XGBoost模型预测有很好的拟合性。临床决策曲线显示XGBoost模型的净获益值较高,临床实用性强(P>0.05)。外部验证显示,XGBoost模型的AUC为0.959(95%CI:0.901~0.989),灵敏度为0.960。结论XGBoost算法建立的CKD合并心力衰竭诊断模型具有高效的诊断效能,能帮助临床医生早期识别、精准预测,为疾病诊断提供决策支持。Objective To dig and analyze the laboratory test data of the patients with chronic kidney disease(CKD),and to establish a diagnostic model for CKD complicating heart failure and evaluate the performance of the model.Methods The cross-sectional study was adopted.A total of 799 patients with diagnosed CKD visiting in the First Hospital of Jilin University from January 2021 to January 2023 were selected as the study subjects,including 341 cases of simple CKD and 458 cases of CKD complicating heart failure.All patients were tested for blood glucose,liver function,kidney function,blood lipid,routine blood test,myocardial injury markers and other indicators.The general data and various laboratory indicators levels were compared between the patients with simple CKD and the patients with CKD complicating hear failure.The indicators with P≥0.05 were excluded.Then the method of Lasso regression was performed and the variables were screened.The 5 kinds of machine model algorithms were used,namely extreme gradient boost(XGBoost),support vector machine(SVM),random forest(RF),gradient booster(GBM)and Logistic regression(LR),to establish the diagnostic models.The receiver operating characteristic(ROC)curve was used to screen the optimal model.The identification ability,goodness of fit and clinical value of the model were evaluated by using the ROC curves,calibration curves,and clinical decision curves.Finally,the testing results in 75 CKD patients from the China-Japan Union Hospital of Jilin University were selected for conducting the external validation.Results The training set XGBoost showed the high efficiency in predicting CKD complicating heart failure.The area under the curve(AUC)of the XGBoost model was significantly higher than that of RF,SVM,GBM and LR models(Z=5.192,5.597,5.597,6.271,P<0.001).Based on the XGBoost model,the six important weight variables for CKD complicating heart failure were determined to be uric acid,soluble growth stimulating gene 2 protein(ST2),N-terminal B-type natriuretic peptide precursor(NT-

关 键 词：心力衰竭 慢性肾脏病 机器模型 XGBoost算法 尿酸 

分 类 号：R446.1[医药卫生—诊断学] R692[医药卫生—临床医学]