Prediction Of Different Stages Of Type 2 Diabetes Mellitus By Machine Learning

Purpose Objective To establish a predictive model of type 2 diabetes complications based on machine learning theory using clinical data and biochemical tests in patients with type 2 diabetes mellitus(T2DM).Secondly,the binary logistic regression was used as a supplement to initially screen the related factors and influences of complications in patients with type 2 diabetes.A well-established model was used to predict whether patients with type 2 diabetes at different stages of the disease had complications,and a preliminary analysis of the association between related factors and type 2 diabetes and complications.Method The electronic medical record information of 420 inpatients with type 2diabetes in a hospital in the northwestern region of July,2017-July 2017 was collected,and the diagnostic data and biochemical examination data of diabetes were collected.After data pre-processing,77 variables were found that may have implications for type 2 diabetes and complications.The data was normalized and the database was established.Using the MATLAB software,the artificial neural network(ANN)algorithm in machine learning was used,and the error back propagation(BP)neural network was used as the fitting model.The sample was divided into a training group and a verification group,wherein the training group contained 358 samples(85% of the total sample)and established a BP neural network model.The structure is 77-50-1,the learning rate is 0.1,the training error is 0.2,the maximum number of training steps is set to 1000 steps(epoeh),the transfer function is logarithmic Sigmiod function,and the network is trained by Levenberg-Marquardt optimization method.There were 62 samples in the validation group(15% of the total sample size),verifying the predictive performance of the BP neural network model and predicting whether it had complications.Secondly,the application of binary logistic regression based on epidemiological study design,using chi-square test and t-test to screen statistically significant variables,and then the selected variables were included in the binary logistic regression analysis.All data were processed using SPSS 18.0 software to screen for association factors for type 2 diabetes complications.Combined with the influencing factors of BP neural network model and binary logistic regression,the related factors were analyzed from actual clinical study and literature,and the relationship between it and type 2 diabetes and complications was discussed.Result The accuracy of BP neural network model predicted by validation group is about 93.55%,and the sensitivity and specificity are 95.6% and 88.2% respectively.Among 77 variables,basophil ratio,mean corpuscular volume,corpuscular volume,distribution width of corpuscular volume,waist circumference,diastolic blood pressure,systolic blood pressure,total bile acid,age and family history of diabetes ranked first.Age and mean corpuscular volume in binary logistic regression are two relevant factors for the development of type 2 diabetes mellitus to complications.Conclusion The prediction accuracy of BP neural network model is about 93.55%,the sensitivity and specificity are 95.6% and 88.2% respectively,it can be seen that the BP neural network model has high prediction performance.Among them,basophilic granulocyte ratio,mean corpuscular volume,erythrocyte,corpuscular volume distribution width,waist circumference,diastolic pressure,systolic pressure,total bile acid,age and family history of diabetes were ranked first.Age and mean corpuscular volume are two relevant factors for the development of type 2 diabetes mellitus to complications in binary logistic regression.The variables predicted by BP neural network also include those obtained by logistic regression.In this study,BP artificial neural network algorithm is used to establish a prediction model of complications of type 2 diabetes mellitus,which can be used to predict whether patients with type 2diabetes mellitus are at the stage of complications at different stages.It is expected to screen more complex and diverse indicators and obtain ranking of impact degree,so as to provide a scientific basis for early diagnosis of type 2 diabetes mellitus from its development to complications stage.