Study of LASSO-BN Model for Necrotizing Enterocolitis in Newborns
10.11969/j.issn.1673-548X.2024.11.012
- VernacularTitle:新生儿坏死性小肠结肠炎的LASSO-BN模型研究
- Author:
Qiyue YANG
1
;
Xiaoyun JIA
;
Xinhua ZHANG
Author Information
1. 030001 太原,山西医科大学公共卫生学院流行病学教研室;030001 太原,山西医科大学附属儿童医院
- Publication Type:Journal Article
- Keywords:
LASSO regression;
Logistic regression analysis;
Bayesian network model;
Necrotizing enterocolitis;
Relevant factors
- From:
Journal of Medical Research
2024;53(11):57-63
- CountryChina
- Language:Chinese
-
Abstract:
Objective To screen variables through LASSO regression,conduct multifactor Logistic regression analysis based on the screening results,and construct a Bayesian network model using max-min hill-climbing(MMHC)algorithm to explore the related fac-tors of necrotizing enterocolitis(NEC)in newborns and the complex network relationships among factors.The study also aimed to compare the two models to find the optimal modeling tool.Methods All NEC patients admitted to the Department of Neonatology,Department of Neonatal Surgery,and NICU of Shanxi Children's Hospital(Shanxi Maternal and Child Health Hospital)from January 2020 to December 2023 were retrospectively studied.NEC investigation data were collected and variable screening was conducted using LASSO regression.Multifactor Logistic regression analysis was performed based on the screening results.The MMHC mixed algorithm was employed for struc-ture learning,and the maximum likelihood estimation method was used for parameter learning to construct the NEC Bayesian network model.Results After variable screening,10 factors including prematurity,low birth weight,feeding method,intrauter distress and post-natal asphyxia history,anemia,non-invasive ventilator,probiotics,gestational diabetes,C-reactive protein(CRP),and procalcitonin(PCT)were included in the model construction.The area under the receiver operating characteristic(ROC)curve of the Bayesian net-work model in the modeling group and validation group were 0.825 and 0.817,respectively,with accuracies of 89.78％and 90.43％,respectively.The AUC of the multifactor Logistic regression analysis in the modeling group and validation group were 0.777 and 0.741,respectively,with accuracies of 70.01％and 69.44％,respectively.The performance of the Bayesian network model was superior to that of multifactor Logistic regression analysis.Furthermore,the Bayesian network model showed that low birth weight,feeding method,probi-otics,and PCT were directly related to NEC,prematurity and non-invasive ventilator were indirectly related to NEC through low birth weight,and CRP was indirectly related to NEC through PCT.Conclusion By comparing the two models,it was found that the Bayesian network model is an effective tool for in-depth study of NEC and the network relationships among related factors.Through this model,the association strength between NEC and various factors can be accurately evaluated,providing a scientific basis for the prevention and treat-ment of NEC.
