基于联合监测的ARIMA-ANN机器学习模型的实时医院感染率评估方法  

Estimation of real-time healthcare-associated infection incidence based on the ARIMA-ANN machine learning model of integrated surveillance

作　　者：范允舟 曹雄晶 吴艳艳[1] 高芳[1] 邹俊宁[1] 熊莉娟[1] FAN Yun-zhou;CAO Xiong-jing;WU Yan-yan;GAO Fan;ZOU Jun-ning;XIONG Li-juan(Union Hospital,Tongji Medical College,Huazhong University of Science and Technology,Wuhan,Hubei 430022,China)

机构地区：[1]华中科技大学同济医学院附属协和医院医院感染管理科,湖北武汉430022

出　　处：《中华医院感染学杂志》2020年第20期3185-3190,共6页Chinese Journal of Nosocomiology

基　　金：湖北省卫生与计划生育委员会科研基金资助项目(WJ2017M107);华中科技大学同济医学院附属协和医院科学培养基金资助项目(000003737)。

摘　　要：目的探索基于多元化监测指标的时间序列数据对医院感染率进行估算的方法。方法选取监测系统数据库中三个医院感染高发的重症监护病区内,2014-2017年连续157周的监测数据作为研究对象。构建具有不同滞后程度的医院感染率模拟数据,联合一组包含11个指标的抗菌药物数据集的实时监测数据,利用ARIMAANN组合的机器学习模型对实时的医院感染率进行估算,并通过估算值与真实值之间的平均绝对误差百分比(Mean absolute percentage error,MAPE)来验证模型的估算效果。结果在医院感染诊断滞后1周的情景下,当不联合实时抗菌药物监测数据时,各病区ARIMA模型对医院感染率估算的MAPE为15.86%~23.05%;当联合实时抗菌药物监测数据时,各病区ARIMA-ANN模型对医院感染率估算的MAPE为10.75%~16.08%;当医院感染诊断的滞后时间超过1周时,各病区内各模型对医院感染率估算的MAPE均出现具有统计学意义的升高。结论联合实时抗菌药物监测指标集的ARIMA-ANN机器学习模型提高对实时医院感染率的估算精度。OBJECTIVE To explore a method for estimating healthcare-associated infection rate based on time-series data of diversified monitoring indicators.METHODS Totally 157consecutive weeks of monitoring data in three ICU with a high incidence of nosocomial infections in the monitoring system database from 2014to 2017were selected as the research subjects.A set of simulation data with different degrees of lag for healthcare-associated infection was constructed and combined with a real-time monitoring data of antimicrobial drug data sets containing 11indicators.An ARIMA-ANN combined machine learning model was used to estimate the real-time healthcareassociated infection rate.The estimation effect of the model was evaluated by the mean absolute error percentage(MAPE)between the estimated value and the true value.RESULTSIn the scenario where the diagnosis of nosocomial infection lagged 1week,when the real-time antimicrobial monitoring data was not combined,the estimated MAPE of the nosocomial infection rate by the ARIMA model of each ward was 15.86%to 23.05%;when the realtime antimicrobial monitoring data was combined,the estimated MAPE of the nosocomial infection rate by ARIMA-ANN model in each ward was 10.75%to 16.08%.When the lag time of healthcare-associated infection diagnosis exceeded 1week,the MAPE estimated by each model in each ward of the nosocomial infection rate had a significant increase.CONCLUSIONThe ARIMA-ANN machine learning model combined with real-time antimicrobial drug monitoring index set improved the accuracy of real-time estimation of healthcare-associated infection incidence.

关 键 词：医院感染率 机器学习 联合监测 

分 类 号：R183[医药卫生—流行病学]