基于SMOTE算法的腹膜透析患者合并重度贫血的预测模型构建  

Construction of a predictive model for peritoneal dialysis patients with severe anemia based on SMOTE-algorithm

作　　者：燕雯雯 储杨敏 束永兵[1] 罗寅亮 YAN Wenwen;CHU Yangmin;SHU Yongbing;LUO Yinliang(Department of Nephrology,Lu′an Hospital of Traditional Chinese Medicine,Lu′an 237100,Anhui,China)

机构地区：[1]六安市中医院肾内科,安徽六安237100

出　　处：《皖南医学院学报》2025年第1期29-32,37,共5页Journal of Wannan Medical College

基　　金：安徽省自然科学基金项目(2108086MG246)。

摘　　要：目的:探讨腹膜透析患者合并重度贫血的危险因素,并基于少数类样本合成过抽样技术(SMOTE)算法构建腹膜透析患者合并重度贫血的预测模型。方法:选取2020年1月~2022年10月六安市中医院诊治的212例腹膜透析患者作为研究对象。采用多因素Logistic回归分析筛选腹膜透析患者合并重度贫血的危险因素,应用SMOTE算法构建腹膜透析患者合并重度贫血的预测模型,并对其预测效能进行分析。结果:212例腹膜透析患者中有72例患者发生重度贫血,重度贫血的发生率为33.96%(72/212)。Logistic回归分析结果显示,血清铁降低、营养不良、未使用补血药物及C-反应蛋白(C-reactive protein,CRP)水平升高等是腹膜透析患者合并重度贫血的危险因素,铁蛋白升高、转铁蛋白升高是腹膜透析患者合并重度贫血的保护因素(P<0.05)。基于SMOTE算法的预警模型Logit(P)=1.266血钙+1.702血清铁+1.578营养不良+1.815未使用补血药物-0.115铁蛋白-2.687转铁蛋白+0.511CRP+12.199,H-L检验结果(决定系数R 2=0.187,P=0.924)提示基于SMOTE算法的预警模型拟合度良好。Logistic回归模型的ROC曲线下面积(AUC)为0.902(95%CI:0.860~0.945),基于SMOTE算法的预测模型的AUC为0.936(95%CI:0.911~0.962),基于SMOTE算法的预测模型的ROC曲线下面积高于Logistic回归模型(P<0.05)。结论:血清铁降低、营养不良、未使用补血药物及CRP等是腹膜透析患者合并重度贫血的危险因素,铁蛋白升高、转铁蛋白升高是腹膜透析患者合并重度贫血的保护因素,基于SMOTE算法的预警模型对腹膜透析患者合并重度贫血预测的准确性较高。Objective:To investigate the risk factors for patients with severe anemia undergoing peritoneal dialysis,and create a predictive model for this group of patients based on Synthetic Minority Oversampling Technique(SMOTE).Methods:212 patients undergoing peritoneal dialysis in our hospital between January 2020 and October 2022 were included as the study objects.Multivariate logistic regression model was used to identify the risk factors for the peritoneal dialysis patients with severe anemia,SMOTE-algorithm was used to establish the predictive model for this group of patients,and predictive efficacy of the predictive model was analyzed.Results:In the 212 patients undergoing peritoneal dialysis,severe anemia developed in 72(33.96%).Logistic regression analysis showed that decreased serum iron level,malnutrition,no-use of blood-replenishing drugs and elevated CRP levels were risk factors,yet elevated ferritin and transferrin levels were protective factors for the incidence of severe anemia in peritoneal dialysis patients(P<0.05).Early warning model based on SMOTE-algorithm Logit(P)=1.266 calcium+1.702 serum iron+1.578 malnutrition+1.815,non-use of blood tonic drugs,-0.115 ferritin,-2.687 transferrin+0.511CRP+12.199,and H-L test results(coefficient of determination R 2=0.187,P=0.924)demonstrated that the early warning model based on SMOTE-algorithm had a good fit.The area under the ROC curve was 0.902(95%CI:0.860-0.945)for the logistic regression model,and 0.93(95%CI:0.911-0.962)for the prediction model based on SMOTE-algorithm,and the area under the ROC curve of the prediction model based on SMOTE-algorithm was higher than that of the logistic regression model(P<0.05).Conclusion:Serum iron level reduction,malnutrition,no-use of blood tonic drugs and CRP level are the risk factors for incidence of severe anemia in patients undergoing peritoneal dialysis,and elevated ferritin and transferrin levels are the protective factors for severe anemia in peritoneal dialysis patients.The early warning model based on SMOTE-algorit

关 键 词：腹膜透析 重度贫血 危险因素 少数类样本合成过抽样技术 

分 类 号：R692[医药卫生—泌尿科学] R556[医药卫生—外科学]