糖尿病视网膜病变伴发糖尿病肾病诊断预测模型的构建与验证分析  

Construction and validation analysis of a diagnostic prediction model for diabetic retinopathy complicated with diabetic nephropathy

作　　者：游金龙 刘艳丽 王雪 吕康 刘少东 姜敏 张文君[3] 李媛媛 王敏珍[1] 郑山[1] YOU Jinlong;LIU Yanli;WANG Xue;LYU Kang;LIU Shaodong;JIANG Min;ZHANG Wenjun;LI Yuanyuan;WANG Minzhen;ZHENG Shan(Institute of Epidemiology and Statistics,School of Public Health,Lanzhou University,Lanzhou 730000,China;Department of Nephrology,The People's Hospital of Wuwei,Wuwei 733000,Gansu,China;Department of Nephrology,The Second Hospital of Lanzhou University,Lanzhou 730030,China)

机构地区：[1]兰州大学公共卫生学院,流行病与卫生统计学系,甘肃兰州730000 [2]甘肃省武威市人民医院肾内科,甘肃武威733000 [3]兰州大学第二医院肾内科,甘肃兰州730030

出　　处：《兰州大学学报(医学版)》2025年第2期49-59,共11页Journal of Lanzhou University(Medical Sciences)

基　　金：甘肃省杰出青年基金资助项目(23JRRA1019);武威市市级科技计划项目(WW2202RPZ037)。

摘　　要：目的构建并遴选糖尿病视网膜病变伴发糖尿病肾病诊断预测模型,分析评价诊断敏感指标,为临床决策提供依据。方法基于中国人民解放军总医院的糖尿病并发症预警数据集,将数据集按照7∶3随机拆分为训练集和验证集,在训练集上采用LASSO回归、支持向量机和随机森林3种方法分别筛选预测因子,并结合Logistic回归模型分别构建3种糖尿病视网膜病变伴发糖尿病肾病预测模型;在训练集和验证集上分别通过十折交叉验证评估模型的内部一致性,通过Hosmer-Lemeshow检验、受试者操作特征曲线及曲线下面积评估模型区分度,通过Calibration校准曲线评估模型校准度,通过决策曲线分析评估模型临床有效性;采用受试者操作特征曲线评价糖尿病视网膜病变伴发糖尿病肾病的诊断敏感指标,分析各敏感指标的诊断价值。结果LASSO回归筛选出4个变量;支持向量机筛选出5个变量;随机森林筛选出8个变量。3个预测模型均具有较好的预测性能,以随机森林-Logistic回归预测模型最优,其中训练集Hosmer-Lemeshow检验结果为P>0.05,受试者操作特征曲线下面积为0.875(95%CI:[0.849,0.902],P<0.001),特异度为0.825,灵敏度为0.809,准确度为0.815,Kappa值为0.614,阳性预测值和阴性预测值分别为0.712和0.890,Calibration校准曲线结果表明模型校准度良好,决策曲线分析表明模型具有良好的临床使用价值;验证集中结果相似。对敏感指标分析发现,尿微量白蛋白肌酐比值、血肌酐、血清白蛋白及糖化血清蛋白是糖尿病视网膜病变伴发糖尿病肾病的主要敏感指标,其受试者操作特征曲线下面积均大于0.75(P<0.001)。结论采用随机森林构建的随机森林-Logistic回归模型具有良好的预测性能;糖尿病视网膜病变伴发糖尿病肾病诊断的主要敏感指标为尿微量白蛋白肌酐比值、血肌酐、血清白蛋白和糖化血清蛋白。Objective To construct and select a prediction model for the diagnosis of diabetic retinopathy with diabetic nephropathy,as well as analyze and evaluate the sensitive indicators for such a diagnosis,in order to provide a basis for clinical decision-making.Methods Based on the early warning dataset of diabetic complications in the Chinese People's Liberation Army General Hospital,the dataset was randomly split into a training set and a validation set in accordance with 7∶3,and the three methods of LASSO regression,Support Vector Machine and Random Forest were used to screen the predictors in the training set and to construct three diabetic retinopathy with diabetic nephropathy prediction models with the combination of the Logistic regression model;the internal consistency of the models was assessed by 10-fold cross-validation in the training and validation sets,respectively.The internal consistency of the models was assessed by 10-fold cross-validation on the training and validation set,the model differentiation assessed by the Hosmer-Lemeshow test,the receiver operator characteristic and the area under the curve;the model calibration was assessed by the calibration curve,and the clinical validity of the models assessed by the decision curve analysis;the receiver operator characteristic curve was used to evaluate the diabetic retinopathy with diabetic nephropathy prediction model;and the receiver operator characteristic curve was used to evaluate the diabetic retinopathy with diabetic nephropathy prediction model.Receiver operator characteristic curves were used to evaluate the sensitive indicators of diabetic retinopathy with diabetic nephropathy,and the diagnostic value of each sensitive indicator was analyzed.Results LASSO regression screened four variables,support vector machines screened five variables,and random forests screened eight variables.Three prediction models had good prediction performance,and the Random Forest-Logistic regression prediction model was optimal,in which the training set Hosmer-Leme

关 键 词：糖尿病肾病 糖尿病视网膜病变 诊断试验评价 预测模型 敏感指标 

分 类 号：R587.2[医药卫生—内分泌] R692.9[医药卫生—内科学]