Comparison of the predictive performance of two predictive models based on decision tree and logistic regression algorithms for the risk of pulmonary multidrug-resistant organism infection in intensive care unit patients
10.3969/j.issn.1671-8348.2025.04.029
- VernacularTitle:基于决策树和logistic回归算法的两种预测模型对ICU患者肺部多重耐药菌感染发生风险的预测性能比较
- Author:
Yurui XIAN
1
;
Jing WANG
;
Min ZHANG
Author Information
1. 电子科技大学医学院附属绵阳医院/绵阳市中心医院手术室,四川绵阳 621000
- Keywords:
decision tree model;
logistic regression model;
neurointensive care;
multidrug-resistant or-ganism infection;
predictive model
- From:
Chongqing Medicine
2025;54(4):954-959
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop predictive models based on decision tree and logistic regression algo-rithms for pulmonary multidrug-resistant organism(MDRO)infection risk in intensive care unit(ICU)pa-tients,and compare their predictive performance to provide references for pulmonary MDRO infection preven-tion.Methods A total of 350 ICU patients admitted to the hospital from January 2021 to December 2023 were enrolled and divided into an infection group(n=52)and a non-infection group(n=298)based on the pres-ence of pulmonary MDRO infection.Clinical data from both groups were collected and analyzed to identify risk factors for pulmonary MDRO infection in ICU patients.Using the risk of pulmonary MDRO infection in ICU patients as the dependent variable and the indicators with statistically significant differences in univariate anal-ysis as the independent variables,predictive models for pulmonary MDRO infection risk were developed by u-sing decision tree algorithms and logistic regression analysis.The predictive performance of these two models was compared.Results There were statistically significant differences(P＜0.05)between the infection and non-infection groups in terms of age,history of diabetes,hospital stay≥2 weeks,indwelling urinary catheter,chlorine dioxide disinfection,prolonged bedridden status,and disturbance of consciousness.Logistic regression analysis with pulmonary MDRO infection risk in ICU patients as the dependent variable showed that age ≥60 years,indwelling catheter,hypoproteinemia,and disturbance of consciousness were risk factors for pulmonary MDRO infection in ICU patients,while chlorine dioxide disinfection was a protective factor.Based on these risk factors and regression coefficients,the original warning model was established as:logit(P 1)=0.856 × age+0.928 × indwelling catheter+0.916 × chlorine dioxide disinfection+0.866× hypoproteinemia+0.986 × disturbance of consciousness-4.371.The Hosmer-Lemeshow test indicated good model fit(coefficient of de-termination R2=0.579,P=0.531).The decision tree model for pulmonary MDRO infection risk in ICU pa-tients constructed using decision tree algorithm comprised 4 layers and 9 nodes,selecting four clinical features as nodes:presence of consciousness disturbance,indwelling catheter,hypoproteinemia,and age≥60 years,with consciousness disturbance being the most significant predictor.The decision tree model showed an area under the curve(AUC)of 0.892(95％CI:0.835-0.949)on receiver operating characteristic(ROC)analy-sis,compared to 0.862(95％CI:0.812-0.912)for the logistic regression model.Delong test revealed no sig-nificant difference in predictive performance between the two models(Z=1.148,P=0.095).Conclusion Both the decision tree model and logistic regression model established for predicting the risk of pulmonary MDRO infection in ICU patients demonstrate high predictive performance with comparable efficacy.
