Analysis of influencing factors for neonatal nosocomial infection by Logistic regression model combined with decision tree model
Song Jia, Tian Jiaran, Ping Lili, et al
(1.Graduate School of Chengde MedicalUniversity, Hebei, Chengde 067000, China; 2.Graduate School of Hebei Medical University, Hebei,Shijiazhuang 050035, China; 3.Department of Neonatology, Handan Central Hospital, Hebei, Handan 056000,China)
Abstract: 【Abstract】 Objective To provide the basis for reducing the nosocomial infection rate of neonates by Logistic regression model and decision tree model analyzing its risk factors. Method A total of 1 552 hospitalized neonates admitted to the Department of Neonatology, Handan Central Hospital from January to December 2020 were included in the study, and they were divided into the infected group (n=48) and the non-infected group (n=1 504) according to whether developed nosocomial infection. The risk factors of nosocomial infection in neonates were analyzed using the receiver operating characteristic (ROC) curve area under the curve (AUC) compared the predictive effect of Logistic regression model and decision tree model. Result Among the neonates, the infection rate was 3.1% (48/1 552). A total of 21 pathogenic organisms were detected in nosocomial infection specimens and detection rate was 43.8%. The Gram-negative bacteria accounted for 47.6% (10/21), mainly Klebsiella pneumoniae. The Gram-positive bacteria accounted for47.6% (10/21). The results of the univariate analysis showed that the proportion of gestational age of <32 weeks,birth weight ≤ 2 500 g, hospital length> 14 d, mechanical ventilation, combined antibiotics ≥3 types,length of the antibiotic use ≥7 d of the infected group were higher than those of the non-infected group. All the differences were statistically significant (all P<0.05). The results of the multivariate Logistic regression analysis showed that birth weight ≤ 1 000 g (OR=20.077, P=0.001), birth weight 1 001-1 500 g (OR=5.673,P=0.020), birth weight 1 501-2 500 g (OR=7.839, P=0.003), length of hospitalization> 21 d (OR=11.162,P=0.003), mechanical ventilation (OR=5.306, P<0.001) and combined antibiotics ≥3 types (OR=10.832,P<0.001) were independent risk factors for nosocomial infection. Exhaustive CHAID decision treemodel analysis results showed that gestational age, length of hospital stay, mechanical ventilation,and combined antibiotics ≥3 types were important variables affecting neonatal nosocomial infection.Combined antibiotics ≥3 types was the main factors of nosocomial infection in neonates. ROC comparisonshowed that there was no significant difference in AUC comparison between the two models (0.951 vs 0.944,Z=0.806, P=0.420). Conclusion In the analysis of risk factors for nosocomial infection in neonates, theLogistic regression model can screen out independent risk factors, including low birth weight, long-termhospitalization, mechanical ventilation, combined antibiotics ≥3 types and so on. The decision tree modelsuggests that the combination of antibiotics has the greatest impact on the risk of nosocomial infections. The two models can complement each other and have a good evaluation value.
宋佳 田嘉然 平莉莉 张瑞敏 翟淑芬. 基于Logistic 回归模型与决策树模型的新生儿医院感染影响因素分析[J]. 发育医学电子杂志, 2023, 11(4): 262-269.
Song Jia, Tian Jiaran, Ping Lili, et al. Analysis of influencing factors for neonatal nosocomial infection by Logistic regression model combined with decision tree model. Journal of Developmental Medicine(Electronic Version), 2023, 11(4): 262-269.
2023, Vol. 11 
