Logistic回归模型和随机森林模型诊断糖尿病周围神经病变效能的比较  

Diagnostic efficacy of Logistic regression model versus random forest model for diabetic peripheral neuropathy:a comparative study

作　　者：桑祎莹 黄仕鑫 易静[3] 曾庆[3] SANG Yi-ying;HUANG Shi-xin;YI Jing;ZENG Qing(Department of Statistical and Policy Research,Chongqing Health Statistics Information Center,Chongqing 401120,China;Yubei District People′s Hospital of Chongqing City,Chongqing 401120,China;School of Public and Health Management,Chongqing Medical University,Chongqing 401120,China)

机构地区：[1]重庆市卫生健康统计信息中心统计与政策研究部,重庆市401120 [2]重庆市渝北区人民医院,重庆市401120 [3]重庆医科大学公共与卫生管理学院,重庆市401120

出　　处：《广西医学》2022年第5期511-515,共5页Guangxi Medical Journal

摘　　要：目的比较Logistic回归模型和随机森林模型诊断糖尿病周围神经病变(DPN)的效能。方法纳入2199例DPN患者作为病例组,2610例健康体检者作为对照组。收集19个实验室指标,包括超敏C反应蛋白、糖化血红蛋白、LDL、HDL、三酰甘油、总胆固醇、总胆红素、总蛋白、白蛋白、ALT、AST、碱性磷酸酶、γ-谷氨酰转肽酶、尿素、尿酸、血红蛋白、钙、钾、钠。使用SPSS 22.0软件构建诊断DPN的多因素Logistic回归模型,使用R 3.6.0软件构建诊断DPN的随机森林模型,采用受试者工作特征(ROC)曲线评价两种模型的诊断性能。结果Logistic回归模型和随机森林模型诊断DPN的正确率分别为81.4%、96.7%,灵敏度分别为72.5%、98.3%,特异度分别为89.2%、95.2%,ROC曲线下面积分别为0.882、0.963。结论随机森林模型对DPN的诊断效能优于Logistic回归模型,同时随机森林模型分析结果给出了各个变量指标的重要性评分,可为DPN的早期诊断提供重要的依据。Objective To compare the efficacy between the Logistic regression model and the random forest model for diagnosing diabetic peripheral neuropathy(DPN).Methods A total of 2199 DPN patients acted as case group,and 2610 healthy check-up individuals acted as control group.Nineteen laboratory indicators were collected,including high-sensitivity C-reactive protein,hemoglobin A1c,LDL,HDL,triacylglycerol,total cholesterol,total bilirubin,total protein,albumin,ALT,AST,alkaline phosphatase,γ-glutamyl transpeptidase,urea,uric acid,hemoglobin,calcium,potassium,and sodium.SPSS 22.0 software was employed to construct a multivariate Logistic regression model for diagnosing DPN,R 3.6.0 software to construct a random forest model for diagnosing DPN,and the receiver operating characteristic(ROC)curve to assess the diagnostic performance of the two models.Results The Logistic regression model and the random forest model yielded the accuracy rates of 81.4%and 96.7%,the sensitivities of 72.5%and 98.3%,the specificities of 89.2%and 95.2%,and the areas under the ROC curves of 0.882 and 0.963,respectively,for diagnosing DPN.Conclusion The random forest model is superior to the Logistic regression model in the diagnostic efficacy for DPN;besides,the results of the random forest model analysis assign the importance ratings of various variable indicators,providing an important basis for the early diagnosis of DPN.

关 键 词：糖尿病周围神经病变 LOGISTIC回归模型 随机森林模型 诊断效能 

分 类 号：R587.2[医药卫生—内分泌] R747.9[医药卫生—内科学]