Risk Prediction Of Postpartum Glucose Abnormalities In Gestational Diabetes Mellitus Based On Machine Learning

Objective:Gestational diabetes mellitus refers to the first appearance of abnormal glucose tolerance during pregnancy.The clinical symptoms of gestational diabetes mellitus are not obvious and most postpartum glucose can return to normal,so most patients do not pay enough attention to this disease and do not timely check glucose after delivery.However,gestational diabetes mellitus patients are high-risk groups of type 2 diabetes,so it is particularly important to build a risk recognition model for postpartum abnormal glucose metabolism for gestational diabetes mellitus patients.To some extent,it can help clinicians identify high-risk groups and take early intervention measures to further reduce the prevalence of type 2 diabetes.Methods:According to the inclusion and exclusion criteria,430 patients with gestational diabetes who underwent systematic examination in the Department of Obstetrics,the First Hospital of Shanxi Medical University from August 2021 to January 2023 were selected as the research objects.The basic data,laboratory examination,past history and complications of patients were obtained by retrieving data from the hospital record room and inpatient medical records.Random forest algorithm was used to screen variables,and the selected variables were respectively included in decision tree and random forest algorithm for modeling.The high risk factors of postpartum abnormal glucose metabolism in gestational diabetes patients were obtained.The prediction effects of the two models were compared according to the accuracy,accuracy,recall rate,F1 score and area under the subject operating characteristic curve(AUC)of the models.Then,according to the results of multi-factor Logistic regression analysis,the line graph was constructed and the effectiveness of the model was evaluated.Results:After variable screening by random forest algorithm,19 variables are finally obtained: GLU(2h)at diagnosis of GDM,hyperlipidemia,FT3 in late pregnancy,depression,breastfeeding,D-dimer in late pregnancy,preeclampsia,prepregnancy BMI,delayed insulin release,weight gain during pregnancy,GLU(0h)at diagnosis of GDM,BMI at delivery,Hb A1 c in late pregnancy,diagnosis GLU(1h)at the time of GDM,FT4 in the third trimester,age,obesity,gestational age at diagnosis of GDM,and history of GDM.The above variables were taken as independent variables and whether postpartum glucose was abnormal was taken as dependent variables to construct decision tree and random forest models respectively.The results show that the prediction performance of the model is the best when the maximum depth of the decision tree is 5,the minimum samples leaf is 2,and the minimum sample split is 2.In random forest,when the number of decision trees is 100 and the maximum number of variables contained in each decision tree is 6,the prediction efficiency of random forest model is the best.The prediction efficiency of random forest model was significantly higher than that of decision tree,and the AUC of the two model test sets were 0.976 and 0.939,respectively.Multiariable Logistic regression results show that the ? = 29.984 + 0.157×age + 0.136 ×preconception BMI+0.123×weight gain during pregnancy+ 1.184 × GLU(2 h)+0.879×FT3 + 0.343×D-dimer + 1.493×hyperlipidemia + 1.861×obesity + 1.352×preeclampsia+1.181×insulinotardic.Based on the Logistic regression results,the C-index was 0.955.Conclusion:The decision tree model showed that GLU(2h),D-dimer in the third trimester,weight gain during pregnancy and BMI at birth were risk factors for postpartum glucose abnormalities in gestational diabetes patients.The results of random forest model showed that GLU(2h),depression,hyperlipidemia,breast-feeding and FT3 had significant effects on postpartum glucose metabolism in the diagnosis of gestational diabetes mellitus.The graph showed that age,preconception BMI,weight gain during pregnancy,GLU(2h),FT3,D-dimer,hyperlipidemia,obesity,preeclampsia and insulinotardic at diagnosis of gestational diabetes were risk factors.The prediction accuracy of the three models for postpartum abnormal glucose metabolism in gestational diabetes mellitus patients is relatively high,and all of them can provide certain guidance for clinicians.