To Construct A Predictive Diagnostic Model Of Diabetic Nephropathy

ObjectiveAt present,the diagnostic gold standard of diabetic nephropathy is renal puncture result,but it is difficult to achieve the high acceptance and economic cost of renal puncture examination for diabetic patients.In this study,a predictive diagnostic model of type 2 diabetic nephropathy was constructed based on different algorithms and its predictive efficacy was evaluated,so as to provide a predictive basis for the early diagnosis of diabetic nephropathy in patients with type 2 diabetes.MethodsThe subjects of the study were patients diagnosed with type 2 diabetes in a third-class a hospital in henan province from March 2015 to March 2017.Relevant examination information,life history and disease history of the patients were collected,and data of 2,197 patients aged 11-89 years were finally included.Then,The overall data is randomly grouped,which is realized through R language programming.The proportion of training set and verification set is about 7:3.In the training set,univariate Logistic regression was used to select predictors of diabetic nephropathy patients.The Logistic Regression,Artificial Neural Network(ANN),Naive NBC Classifier(NBC)and Classification and Regression Tree(CART)models for the diagnosis and prediction of diabetic nephropathy were established.The 10-fold cross method was used for internal validation,and external validation was carried out in the validation set.Area Under the Receiver Operating Characteristic Curve(AUC)and nam-d 'agostino 2 were used to evaluate the predictive performance of the model.Results1.In this study,2,197 patients with type 2 diabetes were studied,including 317 patients with diabetic nephropathy,accounting for 14.4% of the total.2.Single factor in the logistic regression,gender,age,HBALC,FINS,CPL,DBP,SBP,LDL-C,TC,Cr,UA,UTP24,GFR,DF,EH,DTE,DR,DD duration 18 factors such as risk factors in diabetic nephropathy.3.TC,GFR and SBP were included in the five models.Logistic2 included age,DBP,TC,GFR,DF,DR,andDD.Logistic1 increased the level of CPL on the basis of logistic2.ANN and NBC included age,CPL,SBP,LDL-C,TC,UTP24,GFR,DF,EH,DTE,DR,and DD.Four factors including Cr,TC,GFR and SBP were included in CART.4.The NBC model has the worst prediction accuracy and discrimination.The accuracy and discrimination of CART model were the best.AUC of CART model in validation set was 0.7454(95%CI: 0.6896 0.8012),calibration 2=4.2899,P=0.8301.Conclusion1.In the diagnostic prediction model of diabetic nephropathy,the absolute diagnostic prediction of CART model,logistic1 model and logistic2 model was in good agreement with the actual occurrence of diabetic nephropathy.2.Among the 5 models,the calibration degree of CART model in the validation set was within the acceptable range,and the absolute diagnostic prediction of this model on diabetic nephropathy was in good agreement with the actual occurrence of diabetic nephropathy.This model is recommended for the predictive diagnosis of diabetic nephropathy in patients with type 2 diabetes.