Comparison of the predictive efficacy of neonatal hypoglycemia risk based on the Logistic model and the CART decision tree model

MENG Xiangqin; LIU Jia; JI Xiuming; CHEN Ping; CHEN Qingqing

doi:10.12102/j.issn.2095-8668.2025.21.006

Abstract：Objective:To analyze the risk factors of neonatal hypoglycemia and establish a risk warning model for neonatal hypoglycemia by using the decision tree CART algorithm.Methods:The clinical data of 235 newborns and their mothers delivered in our hospital from March 2021 to November 2023 were retrospectively analyzed.They were divided into hypoglycemia group and non-hypoglycemia group according to the results of blood glucose test.The prediction model for neonatal hypoglycemia was constructed by Logistic regression model and decision tree CART model.The 5-fold cross-validation method was used for internal validation,and the predictive efficacy of the model was compared.Results:Among the 235 newborns,36 newborns developed hypoglycemia,199 newborns did not develop hypoglycemia,and the incidence of hypoglycemia was 15.32%.Logistic regression analysis showed that gestational diabetes,mode of delivery,premature infant,development,birth weight,and breastfeeding time were independent risk factors for neonatal hypoglycemia(P<0.05).The overall accuracy of the probability prediction model was 81.8%.After 5-fold cross validation,the prediction accuracy was 72.5%.The decision tree model showed that premature infants were the most important influencing factor for neonatal hypoglycemia,with an information gain of 0.40.The AUC value of the Logistic regression model was slightly higher than that of the decision tree(0.886 vs 0.854),and the prediction efficiency of both models was moderate.Conclusions:Gestational diabetes,cesarean section,premature infant,small for gestational age infant,low birth weight infant,and breastfeeding time≥2 h were all influencing factors for neonatal hypoglycemia.Clinically,corresponding prevention and treatment plans could be formulated based on the above factors.