2型糖尿病视网膜病变临床预测模型的构建与评价  

Establishment and evaluation of a clinic prediction model of diabetic retinopathy in patients with type 2 diabetes mellitus

作　　者：杨洪燕 夏淼 刘赞朝[2] 邢玉微[2] 赵维丽[2] 李洋 赵增毅 王敏珍[1] YANG Hong-yan;XIA Miao;LIU Zan-chao;XING Yu-wei;ZHAO Wei-li;LI Yang;ZHAO Zeng-yi;WANG Min-zhen(Department of Epidemiology and Statistics,School of Public Health,Lanzhou University,Lanzhou,Gansu Province 730000,China;不详)

机构地区：[1]兰州大学公共卫生学院流行病与卫生统计学研究所,甘肃省兰州730000 [2]河北省糖尿病基础医学研究重点实验室石家庄市第二医院,河北省石家庄050051

出　　处：《中国慢性病预防与控制》2023年第1期2-7,共6页Chinese Journal of Prevention and Control of Chronic Diseases

基　　金：河北省卫生厅科研基金项目重点科技研究计划(20190156)。

摘　　要：目的构建糖尿病视网膜病变预测模型并进行验证,为糖尿病视网膜病变的筛查和治疗提供依据。方法于2016年3月—2021年6月,选取石家庄市第二医院5900例2型糖尿病患者为研究对象。收集患者基本信息、体格检查和实验室检查资料,将数据集按7∶3随机分为训练集和验证集。采用SPSS 22.0软件进行t检验、χ^(2)检验和秩和检验。使用R 4.1.2进行最大相关-最小冗余算法及随机森林算法筛选最优单因素子集,将筛选出的变量集引入logistic回归分析,并以此建立糖尿病视网膜病变列线图预测模型;使用Bootstrap自抽法进行内部验证,验证集进行外部验证。通过C指数和受试者工作特征曲线评价模型区分度,绘制Calibration校正曲线并进行Hosmer-Lemeshow检验评估模型一致性,通过决策曲线分析评估模型临床有效性。结果5900例2型糖尿病患者中,糖尿病视网膜病变904例,占15.32%。列线图包括糖化白蛋白、乳酸脱氢酶、腰围、糖尿病病程、糖尿病神经病变、糖尿病肾病和降糖药物使用情况7个变量。经验证,该模型具有中等预测能力,训练集C指数为0.816(95%CI:0.802~0.823),受试者工作特征曲线下面积为0.818(95%CI:0.806~0.830),验证集两者分别为0.822(95%CI:0.803~0.841)和0.827(95%CI:0.809~0.844)。Calibration校正曲线显示一致性较好(P>0.05)。决策曲线分析表明,训练集风险阈值在2%~80%之间以及验证集风险阈值在2%~72%之间时,该模型可产生较大净获益,具有临床使用价值。结论基于糖化白蛋白、乳酸脱氢酶、腰围、糖尿病病程、糖尿病神经病变、糖尿病肾病和降糖药物使用情况构建的列线图对2型糖尿病患者发生糖尿病视网膜病变有良好的预测效能和临床使用价值。Objective To establish and validate a predictive model for the risk of diabetic retinopathy,and provide the basis for screening and treatment of diabetic retinopathy.Methods A total of 5900 outpatients and inpatients with type 2 diabetes mellitus(T2DM)from March 2016 to June 2021 in the Second Hospital of Shijiazhuang were selected as the subjects.The investigation was performed by collecting the data of basic information,physical examination and laboratory test.The subjects were randomly divided into the training set and the validation set by 7∶3.The t test,χ^(2)test and rank-sum test were used to analyze the data,the used software was SPSS 22.0.Optimal single factor set were selected by the maximum relevance-minimum redundancy algorithm and random forest algorithm,and then these variables were introduced into logistic regression analysis to establish the nomogram prediction model,the used software was R 4.1.2.Bootstrap sampling was used for internal validation,and validation set was used for external validation.The C-index and receiver operating characteristic(ROC)curve were used to evaluate the discriminative ability of the nomogram and the calibration-ability was evaluated by Calibration plot and Hosmer-Lemeshow test.Decision curve analysis was used to assess the clinical value of the model.Results Among 5900 T2DM patients,904 cases(15.32%)had diabetic retinopathy.The nomogram model included 7 variables(glycated albumin,lactate dehydrogenase,waistline,duration of diabetes,diabetic neuropathy,diabetic kidney disease,hypoglycemic drugs).After validating,the nomogram model displayed the moderate prediction ability with C-index 0.816(95%CI:0.802-0.823)and the area under the ROC curve of 0.818(95%CI:0.806-0.830)in the training set,and 0.822(95%CI:0.803-0.841)and 0.827(95%CI:0.809-0.844)in the validation set.The adjusting curve displayed the good consistency(P>0.05).The decision curve analysis showed that the risk threshold is between2%-80%in the training set and the risk threshold is between 2%-72%in the valida

关 键 词：糖尿病视网膜病变 机器学习 列线图 预测模型 

分 类 号：R587.1[医药卫生—内分泌]