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

Analysis of influencing factors for neonatal nosocomial infection by Logistic regression model combined with decision tree model

目的通过Logistic回归模型与决策树模型分析新生儿医院感染的危险因素,为降低新生儿医院感染率提供依据。方法将2020年1月至12月邯郸市中心医院新生儿科收治的1552例住院新生儿纳入研究,根据是否发生医院感染分为感染组(n=48)和非感染组(n=1504)。分析新生儿医院感染的危险因素,采用受试者操作特征(receiver operating characteristic,ROC)曲线的曲线下面积(area under the curve,AUC)比较Logistic回归模型与决策树模型的预测效果。结果住院新生儿医院感染率为3.1%(48/1552);医院感染标本共检测出21株病原体(43.8%),革兰氏阴性菌占47.6%(10/21),以肺炎克雷伯杆菌为主;革兰氏阳性菌占47.6%(10/21)。单因素分析结果显示,感染组新生儿胎龄<32周、出生体质量≤2500 g、住院时长>14 d、应用机械通气、联合应用抗生素≥3种、抗生素使用时间≥7 d的比例高于非感染组,差异均有统计学意义(P值均<0.05)。多因素Logistic回归分析结果显示,出生体质量≤1000 g(OR=20.077,P=0.001)、出生体质量1001~1500g(OR=5.673,P=0.020)、出生体质量1501~2500g(OR=7.839,P=0.003)、住院时长>21 d(OR=11.162,P=0.003)、机械通气(OR=5.306,P<0.001)、联用应用抗生素≥3种(OR=10.832,P<0.001)是新生儿医院感染的独立危险因素。Exhaustive CHAID决策树模型分析结果显示,胎龄、住院时长、机械通气、联合应用抗生素≥3种是影响新生儿医院感染的重要变量;联合应用抗生素≥3种是新生儿医院感染的主要因素。ROC比较显示,两种模型的AUC差异无统计学意义(0.951与0.944,Z=0.806,P=0.420)。结论在新生儿医院感染危险因素分析中,Logistic回归模型可将出生体质量低、住院时间长、机械通气、联用应用抗生素≥3种等独立危险因素筛选出来,而决策树模型提示联合应用抗生素对医院感染的患病风险影响最大,两种模型可互为补充,有良好的评估价值。

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 1552 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=1504)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%(48/1552).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 for 47.6%(10/21).The results of the univariate analysis showed that the proportion of gestational age of<32 weeks,birth weight≤2500 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≤1000 g(OR=20.077,P=0.001),birth weight 1001-1500 g(OR=5.673,P=0.020),birth weight 1501-2500 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 tree model 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 comparison showed 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,the Logistic regression model can screen out independent risk factors,including low birth weight,long-term hospitalization,mechanical ventilation,combined antibiotics≥3 types and so on.The decision tree model suggests 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.