ICU住院患者耐碳青霉烯类肺炎克雷伯菌感染危险因素及其BP神经网络模型的预测价值  

Risk factors for carbapenem-resistant Klebsiella pneumoniae infection in ICU-hospitalized patients and its predictive value of BP neural network

作　　者：贾辰 王璐 李海峰 JIA Chen;WANG Lu;LI Hai-feng(General Hospital of the Northern Theater Command,Shenyang,Liaoning 11oo16,China)

机构地区：[1]北部战区总医院疾病预防控制科,辽宁沈阳110016

出　　处：《中华医院感染学杂志》2024年第12期1778-1783,共6页Chinese Journal of Nosocomiology

基　　金：全军医院感染监测体系及综合防控平台的建立与应用(145AHQ080020005X)。

摘　　要：目的 探讨ICU住院患者耐碳青霉烯类肺炎克雷伯菌感染危险因素及其BP神经网络和Logistic回归模型的预测价值.方法 收集2019年4月1日-2023年4月1日人住某三甲级医院ICU的患者信息,包括人口学特征、基础疾病史、抗菌药物应用史、抗菌药物种类、抗菌药物累计使用天数、侵入性操作史及真菌感染等;其中发生耐碳青霉烯类肺炎克雷伯菌(CRKP)医院感染的137例,定义为CRKP组,采用病例对照研究方法,按照1:3选取对碳青霉烯类抗菌药物敏感的肺炎克雷伯菌(CSKP)感染患者381例,定义为CSKP组;利用人工神经网络和Logistic回归算法构建CRKP感染预测模型,利用受试者工作特征(ROC)曲线评价数据和预测模型的准确性.结果 2019年4月-2023年4月ICU共分离肺炎克雷伯菌1 507株,其中CRKP 338株,总的分离率为22.43%;每年的分离率依次为26.83%、22.55%、19.30%、19.05%;每年CRKP医院感染发生率分别为3.79‰、3.64‰、2.91‰、3.84‰;Logistic多元回归分析结果显示:恶性肿瘤、住院期间进行手术、留置导尿管、应用碳青霉烯类抗菌药物、应用甘氨酰四环素类抗菌药物、呼吸机使用时间是CRKP医院感染的危险因素(P<0.05);BP神经网络模型结构为{30-6-1},其预测准确率为80%;Logistic回归模型ROC曲线下面积为0.781,BP神经网络模型ROC曲线下面积为0.786,模型预测准确率为80%.结论 CRKP检出率与全国CRKP检出率总体走势表现一致,两种预测模型准确度较好,BP神经网络模型的预测性能略优于Logistic回归模型.OBJECTIVE To investigate the risk factors for carbapenem-resistant Klebsiella pneumoniae infection in ICU-hospitalized patients and its predictive value of BP neural network and logistic regression models.METHODS Information on patients admitted to the ICU of a military tertiary hospital from Apr.1,2019 to Apr.1,2023 was collected,including demographic characteristics,history of underlying diseases,history of antimicrobial drug application,types of antimicrobial drug,cumulative days of antimicrobial drug use,history of invasive operation,and coexisted fungal infection.Among them,137 cases of CRKP nosocomial infection were defined as CRKP group,and 381 patients with Klebsiella pneumoniae(CSKP)infections who were sensitive to carbapenems antimicrobial drugs were selected according to the case-control study method in l:3 ratio,and were defined as CSKP group.The CRKP infection prediction model was constructed using artificial neural network and logistic regression algorithm, and the accuracy of the data and prediction model was evaluated using the receiver operating characteristic(ROC) curve. RESULTS From April 2019 to April 2023, a total of 1 507 strains of Klebsiella pneumoniaewere isolated from ICU, of which 338 were CRKP, with an overall isolation rate of 22. 43%. The annual isolationrates were 26. 83%, 22. 55%, 19. 30%, and 19. 05% in order. The annual incidences of CRKP nosocomial infectionwere 3. 79%o, 3. 64%o, 2. 91%o, and 3. 84%o respectively. Logistic multivariate regression analysis showed thatmalignant tumor, surgery during hospitalization, indwelling urethral catheter, use of carbapenem antibiotics, useof glycyl tetracycline antibiotics, and ventilator use time were the risk factor for CRKP nosocomial infection (P<0. 05). The BP neural network model structure was (30-6-1), and its prediction accuracy was 80%. The area underthe ROC curve of the logistic regression model was O. 781, and the area under the ROC curve of the BP neuralnetwork model was 0. 786, with a model prediction accuracy of 80%. CONCLUSIO

关 键 词：ICU 耐碳青霉烯类肺炎克雷伯菌 医院感染 BP神经网络 LOGISTIC回归 风险预测 

分 类 号：R378[医药卫生—病原生物学]