[A study of factors associated with neonatal necrotizing enterocolitis]

Objective: To explore the related risk factors of neonatal necrotizing enterocolitis (NEC) by constructing and comparing nine regression models. Methods: All NEC patients admitted to the neonatal internal medicine department, neonatal surgery department, and neonatal intensive care unit of Shanxi Provincial Children's Hospital (Shanxi Provincial Maternity and Child Health Center) from 2020 to 2022 were included as the case group. A control group consisted of children admitted during the same period based on the inclusion and exclusion criteria. The NEC data collected were used for feature selection by using the Boruta algorithm. Logistic regression, multi-decision tree gradient boosting, efficient gradient one-sided sampling, random forest, decision tree, gradient boosting decision tree (GBDT), neural network, support vector machine, and K-nearest neighbor models were constructed. The optimal model was selected through rigorous comparison and Shap explainable analysis was performed on the GBDT model. Results: Thirteen key factors were identified through screening for nine regression models construction. After strict comparison and analysis, the GBDT model showed higher stability compared with other eight regression models. In the validation set, the area under the receiver operating characteristic curve of the GBDT model was 0.958, with an accuracy of 0.925, and sensitivity and specificity of 0.827 and 0.950, respectively. Shap explainable analysis on the GBDT model revealed that suffering from anemia, non-invasive ventilator use, procalcitonin use, premature birth, and low birth weight increased the risk for NEC, while breastfeeding and probiotics decreased the risk for NEC. Conclusion: This study identified the risk factors and protective factors for NEC by using the GBDT model, which provided evidnce for the prevention and treatment of NEC.