基于ARIMA模型预测ICU耐碳青霉烯类肺炎克雷伯菌流行趋势  

Application of ARIMA in prediction of prevalence trend of carbapenem-resistant Klebsiella pneumoniae in ICU

作　　者：蒋士藩 张颖杰 陆娟[3] 程进 兰策介[4] 武星[1] JIANG Shifan;ZHANG Yingjie;LU Juan;CHENG Jin;LAN Cejie;WU Xing(Jiangnan University Affiliated Hospital,Wui,Jiangsu 214122,China;不详)

机构地区：[1]江南大学附属医院感染管理处,江苏无锡214122 [2]江南大学附属医院信息处,江苏无锡214122 [3]江南大学附属医院检验科,江苏无锡214122 [4]无锡市疾病预防控制中心消媒与血寄地病防制科,江苏无锡214023

出　　处：《中华医院感染学杂志》2025年第6期933-938,共6页Chinese Journal of Nosocomiology

基　　金：江苏省自然科学基金资助项目(BK20230190);江苏省医院协会医院管理创新研究课题资助项目(JSYGY-3-2024-603)。

摘　　要：目的探讨自回归整合移动平均模型(ARIMA)在重症监护病房(ICU)耐碳青霉烯类肺炎克雷伯菌(CRKP)流行趋势预测中的应用,为医院制定ICU CRKP感染防控策略提供科学依据。方法收集2021年1月-2024年1月无锡市江南大学附属医院ICU CRKP月度检出菌株数,剔除来自同一患者的重复样本,最终纳入分析的CRKP菌株555株。采用R统计软件进行时间序列差分并构建ARIMA模型,利用自相关函数(ACF)和偏自相关函数(PACF)图进行模型参数确定。通过赤池信息准则(AIC)和均方根误差(RMSE)筛选最优模型,并采用Box-Ljung检验评估残差序列的稳健性。选取2023年9月-2024年1月的检出数据作为验证集,评估模型预测精度,并预测2024年2月-4月的CRKP动态趋势。结果2021-2023年医院年度ICU CPKP检出率呈现动态变化(χ^(2)=66.906,P=0.001),痰液和中段尿为主要来源。最优模型ARIMA(8,1,10)的最小赤池信息准则(AIC)为222.1,RMSE为3.67,Box-Ljung(χ^(2)=0.104,P=0.746)检验显示残差序列无自相关性。2023年9月-2024年1月的实际CRKP与预测值均表现出先升后降的趋势,模型预测的平均相对误差为9.62%。模型预测2024年2月ICU CRKP检出数可能达到低谷,随后呈上升趋势,4月可能出现感染高峰。结论ARIMA模型能够有效用于医院ICU CRKP流行趋势的短期预测和动态分析,为医院感染的早期预警和相应防控措施提供理论依据。OBJECTIVE To explore the application of autoregressive integrated moving average model(ARIMA)in prediction of prevalence trends of carbapenem-resistant Klebsiella pneumoniae(CRKP)in intensive care unit(ICU)so as to provide scientific bases for formulating prevention strategies for CRKP infection in ICU.METHODS The number of CRKP strains that were monthly isolated from the ICU patients of Jiangnan University Affiliated Hospital between Jan.2021 and Jan.2024 was collected,the duplicate samples from the same patient were excluded,and totally 555 strains of CRKP were finally enrolled in the study.The time series differencing was performed by using R statistical software,the ARIMA model was established,the parameters of the model were determined by means of autocorrelation function(ACF)and partial autocorrelation function(PACF)images.The optimal model was screened out by Akaike information criterion(AIC)and root-mean-square error(RMSD),the robustness of the residual sequences was assessed by Box-Ljung test.The data that were detected from Sep.2023 to Jan.2024 were assigned as the validation set,the prediction accuracy of the model was assessed,and the dynamic trend of the CRKP strains from Feb.2024 to Apr.2024 was predicted.RESULTS The isolation rate of CRKP strains in ICU showed dynamic change from 2021 to 2023(χ^(2)=66.906,P=0.001),sputum and midstream urine were the major sources.The minimal AIC of the optimal ARIMA model(8,1,10)was 222.1,with RMSE 3.67.Box-Ljung(χ^(2)=0.104,P=0.746)test indicated that there was no autocorrelation among the residual sequences.Both the actual CRKP and the predictive value firstly rose then descended from Sep.2023 to Jan.2024,and the average relative error of the model was 9.62%for prediction.The number of isolated CRKP strains predicted by the model might reached to the lowest point in Feb.2024 and then showed an upward trend,and it might reach to a high peak in Apr.CONCLUSION ARIMA model is effective for short-term prediction and dynamic analysis of prevalence trend of CRKP strai

关 键 词：耐碳青霉烯类肺炎克雷伯菌 自回归整合移动平均模型 重症监护病房 预测 流行趋势 

分 类 号：R378.996[医药卫生—病原生物学]