Predicting Positive CT Findings Of Pneumonia In Fever Clinic Patients Based On Logistic Regression And Decision Tree Model

Objective CT examination at fever clinics has been widely used in China,and this study aims to construct a logistic regression and decision tree model to improve the prediction rate of the positive CT findings of pneumonia in patients of fever clinic,so as to improve its clinical application efficiency.Methods A retrospective analysis was conducted to study the general data,clinical manifestations,and auxiliary examinations of 1478 patients at fever clinic of the First Affiliated Hospital of Zhejiang Chinese Medical University(The first central)and the First Affiliated Hospital of Zhejiang University School of Medicine(The second central)from October 2021 to March 2022,and divided into pneumonia group and non-pneumonia group according to chest CT results.Taking the first central case(1229 cases)as the training group and the second central case(249 cases)as the verification group,univariate,multivariate logistic regression and multivariate gradual logistic regression were used to analyze and screen out independent risk factors,and finally constructed a logistic regression model and decision tree model for predicting chest CT pneumonia at fever clinic.Logistic regression model and decision tree model were visualized by Nomo and decision tree diagrams,respectively.The diagnostic performance of different models was evaluated by the area under the receiver operating characteristic curve(AUC).Calibration curves and clinical decision curves were used to evaluate model consistency and efficiency.Results The AUCs of clinical model,laboratory model,logistic combined model and decision tree combined model established by logistic regression and decision tree algorithm were 0.708,0.675,0.759,0.761 in the training group and 0.699,0.697,0.734,0.762 in the validation group,respectively.The decision curve analysis plots of the four groups of models show that the decision tree model has been further improved compared with the logistic model,so it is more clinically practical.In addition,the feature importance graph of the decision tree model shows the first six important features in the decision tree model,namely CRP,age,cough,white blood cell count,sputum production,and maximum body temperature,which greatly contribute to the prediction results.Conclusion Based on decision tree machine learning,it has good predictive energy efficiency for chest CT pneumonia positivity at fever clinic,which can promote clinicians to selectively perform chest CT examination for febrile patients,and also assist in triaging patients at fever clinic.In addition,the logistic combination model can also predict the possibility of positive chest CT findings of pneumonia at fever clinic patients,but it is slightly inferior to the decision tree model.