Research On Assistant Diagnosis Of Type 2 Diabetic Kidney Disease Based On Machine Learning Algorithm

The number of type 2 diabetic patients with kidney disease is large,and the late mortality is high.The early clinical symptoms are not obvious,and the pathogenesis is not clear.At present,the clinical diagnosis mainly depends on the medical history,examination and other indicators.Although renal biopsy is the gold standard for the diagnosis of DKD,it has some disadvantages,such as high cost,low acceptability and ineffective diagnosis of early DKD.This study explored the application of logistic regression model,random forest model and extreme gradient boost(xgboost)model in the individual risk assessment of type 2 diabetic kidney disease through easily available clinical tests and demographic indicators,so as to provide a valuable computer-aided diagnosis method for type 2diabetic kidney disease and explore the potential influencing factors.By collecting 56 clinical data of inpatients in the Department of secretions(nephrology)of the Institute of Medical Data of Chongqing Medical University from January 2014 to May 2020,univariate analysis and Lasso regression were conducted to reduce the dimension,so as to reduce the interaction between variables.R 4.0.2 Logistic regression,random forest and XGBoost were used to establish the classification model of renal disease in type 2 diabetes,and the diagnostic performance of the three models was compared.Results a total of 1768 patients with type 2 diabetes and 1856 patients with type 2 diabetes were included for model analysis.Univariate analysis and lasso regression showed that 25 indexes were significant,which were included in logistic regression classification model(LR),random forest model(RF)and xgboost model.The accuracy of classification was 0.754,0.828 and 0.838,and AUC was 0.831,0.894 and 0.909,respectively.oost has the best performance than LR and RF in all aspects.In this study,statistics and machine learning methods were used to study the kidney disease of type 2 diabetes,and the clinical indicators related to the early diagnosis of kidney disease of type 2 diabetes were screened out,which provided ideas for the follow-up clinical research.The established xgboost prediction model has good clinical auxiliary function for the kidney disease of type 2 diabetes,which needs further clinical test.