Application of regression tree in the analysis of factors affecting hospitalization expenses for children infected by respiratory syncytial virus

Qiu-li ZHU,Tao ZHANG,Yun-fang DING,Xue-lan ZHANG,Gen-ming ZHAO

DOI: https://doi.org/10.3969/j.issn.1672-8467.2011.04.003

2011-01-01

Abstract:Objective To analyze and predict the factors affecting the hospitalization expenses for children infected by respiratory syncytial virus (RSV), and to provide information for government decision. Methods There were 959 children randomly collected from all the children with RSV infection from 2005 to 2009 in Soochow Universith Affiliated Children's Hospital. SPSS 18.0 was used in the data analysis procedure, including testing the conditions for the model of multiple linear regression and building the model of regression tree. Results The model of multiple linear regression was not suitable for predicting the factors affecting hospitalization expenses per day. There were 4 factors out of the 10 risk-factors found by the model of regression tree, including the level of C-reactive protein (CRP), age, whether or not having asthma, and whether or not having cardiopulmonary diseases, among which CRP level was the most important one, followed by the age, whether or not having asthma and then whether or not having cardiopulmonary diseases. Mean squared error (MSE) of cross validation was 0.086. When CRP level > 5. 760 mg/L, it was the only factor that affecting hospitalization expenses per day; while when CRP level ≤3.145 mg/L, age began to affect it as well; RSV-infected children with asthma or cardiopulmonary diseases had higher expenditure per day. Conclusions The regression tree may be valuable in epidemiological studies. Factors like CRP level, age, whether or not having asthma and whether or not having cardiopulmonary diseases should be considered when estimating the hospitalization expenditure per day for RSV-infected children.