预防医学论坛

2026, v.32(08) 600-606+614

[打印本页] [关闭]
本期目录(Current Issue) | 过刊浏览(Archive) | 高级检索(Advanced Search)

基于开环门限自回归模型的学生流感发病阈值识别
Identification of influenza incidence thresholds among students based on an open-loop threshold autoregressive model

王增伟,薛秀丽,赵虹
WANG Zeng-wei,XUE Xiu-li,ZHAO Hong

摘要(Abstract):

目的 评估基于开环门限自回归模型(open-loop threshold autoregressive model, TARSO)所识别的学生流感发病阈值在县区流感防控中的适用性。方法 收集太原市某区2010年1月至2026年5月24日流行性感冒(流感)报告发病数据及人口普查数据,构建周报告发病率时间序列。采用交叉相关函数和Granger因果检验分析学生与社区发病的时序关系。经平稳性检验和一阶差分处理后,以TARSO模型识别学生报告发病率阈值,通过赤池信息准则、贝叶斯信息准则和似然比检验确定滞后阶数,采用Wald检验、Hansen检验和块自助法评估门限效应的稳健性。以向前链式分层验证评估阈值对社区高风险周的区分能力,并结合滚动窗口模拟和敏感性分析探讨模型的动态应用策略。结果 学生发病领先社区约1周(Granger因果检验P<0.001),假期及新型冠状病毒流行期间学生/社区发病比显著下降。TARSO模型识别出学生报告发病率的阈值为1.946 3/10万,低于和高于该阈值时,原始社区报告发病率均值分别为0.410/10万和11.220/10万。差分后社区报告发病率时间序列高低机制AR(1)系数差的95%CI为0.063 8~0.807 6,不包含0。门限效应显著(Wald P<0.001,Hansen P=0.035 0)。向前链式分层验证结果提示阈值需动态调整。滚动窗口模拟结果表明,以最近7个流感季为训练窗口,每4个流感季更新一次阈值可获得最佳分辨效能。敏感性分析显示纳入学生病例年龄口径调整为5~14岁对阈值区分能力的影响大于单纯排除新冠病毒流行期。结论 学生流感报告发病率存在明确阈值,超过该阈值后社区流感进入高水平流行期。采用滚动窗口更新阈值可有效维持模型的区分能力。TARSO模型在县域流感防控中具有潜在的应用价值。
Objective To evaluate the applicability of influenza incidence thresholds among students, identified via an open-loop threshold autoregressive(TARSO) model, in county/district-level influenza prevention and control. Methods Weekly reported influenza incidence time series were constructed using influenza surveillance data and census data from a district in Taiyuan City between January 2010 and May 24,2026.Cross-correlation functions and Granger causality tests were employed to analyze the temporal relationship between student and community incidence.Following stationarity tests and first-order differencing, a TARSO model was applied to identify the threshold for student-reported incidence.Lag orders were determined using the Akaike information criterion(AIC),Bayesian information criterion(BIC),and likelihood ratio tests.The robustness of the threshold effect was assessed using Wald tests, Hansen tests, and block Bootstrap methods.A forward chaining cross-validation was conducted to evaluate the threshold′s discriminatory power for high-risk community weeks.Furthermore, rolling window simulations and sensitivity analyses were performed to explore dynamic application strategies for the model. Results Student incidence led community incidence by approximately one week(Granger causality test, P<0.001),and the student-to-community incidence ratio decreased significantly during holidays and the coronavirus epidemic period.The TARSO model identified a student-reported incidence threshold of 1.946 3 per 100 000.Below and above this threshold, the mean raw community-reported incidences were 0.410 and 11.220 per 100 000,respectively.The 95%CI for the difference in AR(1) coefficients between the high and low regimes of the differenced community-reported incidence time series was 0.063 8-0.807 6,excluding zero.The threshold effect was statistically significant(Wald P<0.001,Hansen P=0.035 0).Forward chaining cross-validation indicated that the threshold requires dynamic adjustment.Rolling window simulations demonstrated that optimal discriminatory performance is achieved using a training window of the most recent seven influenza seasons,with threshold updates every four seasons.Sensitivity analyses revealed that adjusting the age definition for student cases to 5-14 years had a greater impact on the threshold's discriminatory power than simply excluding the COVID-19 epidemic period. Conclusion A distinct threshold exists for student-reported influenza incidence;exceeding this threshold signals the transition of community influenza into a high-level epidemic phase.Updating thresholds via rolling windows can effectively maintain the model′s discriminatory power.The TARSO model holds potential application value in county/district-level influenza prevention and control.

关键词(KeyWords): 流行性感冒;学生;报告发病率;阈值开环;开环门限自回归模型;向前链式分层验证
Influenza;Students;Reported incidence rate;Threshold;Open-loop threshold autoregressive model(TARSO);Forward-chaining stratified validation

Abstract:

Keywords:

基金项目(Foundation):

作者(Author): 王增伟,薛秀丽,赵虹
WANG Zeng-wei,XUE Xiu-li,ZHAO Hong

DOI: 10.16406/j.pmt.issn.1672-9153.2026.08.08

参考文献(References):

扩展功能
本文信息
服务与反馈
本文关键词相关文章
本文作者相关文章
中国知网
分享