基于传染病症状监测数据的时间序列和时空聚集性  

Research on predictive analysis methods for symptom monitoring based on time series and spatio-temporal aggregation detection

作　　者：段玮 周晓芳[2] 段丽忠[3] 杨静雯 伏晓庆[2] 王晓雯[4] DUAN Wei;ZHOU Xiaofang;DUAN Lizhong;YANG Jingwen;FU Xiaoqing;WANG Xiaowen(School of Public Health,Kunming Medical University,Kunming 650500,China;Acute Infectious Disease Prevention and Control Institute(Yunnan Academy of Preventive Medicine),Yunnan Center for Disease Control and Prevention,Kunming 650599,China;Baoshan Centre for Disease Control and Prevention,Baoshan 678000,China;STD and AIDS prevention and Treatment Institute,Yunnan Center for Disease Control and Prevention(Yunnan Academy of Preventive Medicine),Kunming 650599,China)

机构地区：[1]昆明医科大学公共卫生学院,昆明650500 [2]云南省疾病预防控制中心(云南省预防医学科学院)急性传染病防治所,昆明650599 [3]保山市疾病预防控制中心急性传染病预防控制科,保山678000 [4]云南省疾病预防控制中心(云南省预防医学科学院)科研教育服务部,昆明650599

出　　处：《中华疾病控制杂志》2025年第3期332-339,共8页Chinese Journal of Disease Control & Prevention

基　　金：兴滇英才支持计划医疗卫生人才专项研究项目(2024年1月-2028年1月);云南省科技重大专项云南省急性传染病综合监测预警体系构建研究(202102AA100019)。

摘　　要：目的探索基于时间序列和时空聚集性探测的症状监测预测方法,为有效分析和利用症状监测数据提供参考依据。方法采用差分自回归移动平均(autoregressive integrated moving average,ARIMA)模型和Holt-Winters模型进行时间序列分析,通过回顾性时空聚集性探测进行聚集区域和时间综合探测分析。结果以在X市监测为例,2023年1月1日―2023年4月30日共监测到34207人次出现与传染病相关症状并前往医疗机构就诊。4月1日―4月19日的模型预测值与实际监测值比较发现,Holt-Winters模型对数据的预测情况优于ARIMA模型,误差更小,几乎所有实际值均在预测值95%CI内。时空扫描分析结果显示,某市就诊人群中具有所监测症状者居住社区涵盖该市9个街道,发热、咳嗽、腹痛和头痛的1类聚集地为D街道、E街道、F街道、G街道和A街道;咽痛和恶心的1类聚集地为D街道、A街道和G街道;腹泻和呕吐的1类集聚地为G街道和D街道。所监测症状的发生时间主要集中在2022年12月―2023年4月。结论Holt-Winters模型对症状数据具有较好的时间趋势预测效果,通过对症状监测数据进行时空扫描分析可及时发现传染病的聚集情况,为防控工作提供重要的空间、时间和时空联合指示。症状监测数据可被用于监测传染病的潜在流行情况,为实现早期预警提供参考。Objective To explore appropriate prediction methods for symptom monitoring based on time series and spatio-temporal aggregation detection,and to provide a foundational reference for the effective analysis and utilization of symptom monitoring data.Methods In this study,ARIMA and Holt-Winters models were used for time-series analysis,and aggregation regions and time-integrated detection analyses were conducted through a retrospective temporal-spatial clustering detection analysis.Results Taking the surveillance in X City as an example,from January 1 to April 30,2023,34207 individuals were monitored for symptoms that may be associated with communicable diseases and visited a healthcare facility.Comparing the model predicted value between April 1 to April 19 with the actual monitoring value found that the Holt-Winters model predicted the data better than the ARIMA model with smaller errors,and almost all the actual values were within the 95%confidence interval of the predicted value.The spatio-temporal scan analysis showed that the residential community of patients with monitored symptoms in the certain city covered 9 clusters of the city.Category 1 included D street,E street,A street,F street,G street and A street;category 1 cluster of sore throat and nausea was D street,A street and G street;category 1 cluster of diarrhea and vomiting was G street and D street.The onset of the monitored symptoms was mainly concentrated between December 2022 and April 2023.Conclusions The Holt-Winters model has a good time trend prediction effect on symptom data,and the aggregation of infectious diseases can be detected in time by analyzing symptom surveillance data through spatio-temporal scanning,which provides important joint spatial,temporal and spatio-temporal indications for prevention and control efforts.Symptom surveillance data can be used to monitor the potential prevalence of infectious diseases and provide reference for realizing early warning.

关 键 词：症状监测 传染病相关症状 时间序列分析 时空聚集性分析 预测 

分 类 号：R51[医药卫生—内科学] R181.2[医药卫生—临床医学]