决策树模型在重症监护病房患者医院感染中的应用

Application of decision tree model in nosocomial infection among intensive care unit patients

目的在回归分析的基础上,应用决策树模型构建重症监护病房(ICU)患者医院感染风险预测模型,探索其医院感染分类规则。方法收集分析某三甲综合医院2020年1月至2022年11月入住ICU>48 h的住院患者病历资料,将logistic回归分析结果中差异有统计学意义的变量作为预测变量构建ICU患者医院感染决策树模型,应用受试者工作特征(ROC)曲线下面积(AUC)评价模型准确度。结果本研究共纳入研究对象1704例,其中医院感染211例,医院感染率12.4%。决策树模型结果显示ICU患者医院感染的危险因素为有创呼吸机使用≥7 d、手术、中心静脉置管≥7 d、入住ICU时间≥10 d、使用抗菌药物,其中有创呼吸机使用≥7 d是ICU患者医院感染最重要的危险因素,决策树模型ROC AUC为0.767(95%CI:0.730~0.805)。结论联合应用logistic回归分析与决策树模型可有效预测不同因素组合下ICU患者医院感染发生风险,为降低ICU患者医院感染率提供理论依据。

Objective To establish the nosocomial infection risk prediction model among the intensive care unit(ICU)patients by the decision tree model under the basis of regression analysis,and to investigate the classification rule of nosocomial infection in ICU.Methods The medical case data of the inpatients admitting to ICU>48 h in a class3A hospital from Jan.2020 to Nov.2022 were collected.The variables with statistical difference in the logistic regression analysis results served as the predictive variables to construct the ICU nosocomial infection decision tree predictive model.The area under receiver operating characteristic(ROC)curve(AUC)was used to evaluate the accuracy of the model.Results A total of 1704 study subjects were included in this study,among them there were 211 cases of nosocomial infection with a nosocomial infection rate of 12.4%.The decision tree model results showed that the risk factors of nosocomial infection among the ICU patients were the invasive ventilator use≥7 d,central venous indwelling catheter≥7 d,time of admitting to ICU≥10 d and using antimicrobial drugs,in which the invasive ventilator use≥7 d was the most important risk factor.AUC of the decision tree model ROC was 0.767(95%CI:0.730-0.805).Conclusion The combination use of logistic regression analysis and decision tree model could effectively predict the risk of nosocomial infection occurrence under different factors combination,which provides the theoretical basis for reducing nosocomial infection rate among ICU patients.