Bootstrap在流感监测预警中的应用

Application of bootstrap method in the early warning for influenza surveillance

目的:应用bootstrap方法估计流感样病例(influenza-like illnesses,ILI)发病水平95%置信区间(95%CI)作为预警阈值,评价该方法在ILI预警中的价值。方法:以2005—2014年(除2006、2009年)的周监测数据为基线,采用历史2年流感病毒最高阳性检出率的30%作为流感流行标准,确定2006年和2009年流感流行期金标准,通过回顾性分析,比较bootstrap法、移动平均法和金标流行期的一致性;并以2012—2015年的周监测数据为基线,探讨bootstrap法在实际预警工作中的应用。结果:江苏省ILI流行曲线有2个高峰,冬春季高峰分布于11月—次年2月,夏季高峰分布于6—8月;bootstrap法、移动平均法和金标流行期一致性均较好;与移动平均法相比,bootstrap法对2009年初的流感流行相对更敏感,与金标流行期的一致性相对更高(kappa=0.728)。采用bootstrap法对2016年ILI%开展预警,与金标准结果接近。结论 :bootstrap法作为一种不依赖于ILI数据分布的计算密集型方法 ,可作为传统移动平均控制图预警方法的替代和补充,用于流感流行的监测和预警。

Objective： To investigate the application of bootstrap method in the early warning for influenza surveillance with 95%CIs of weekly influenze-like illnesses（ILI）% estimated by bootstrap method specified as the early warning thresholds. Methods： The influenza surveillance data from 2005 to 2014（not including 2006 and 2009） were taken as baseline. Thirty Percent of the highest positive rate of influenza viruses in two years was taken as the gold standard of influenza pandemic. Consistency of the early warning results was compared among bootstrap method, moving average method and the gold standard. We also discussed the application of bootstrap in practice in 2016 based on the surveillance data from 2012 to 2015. Results： Influenza viruses circulated throughout the year and peaked during cold winter and spring season from November to following February and hot summer from June to August. The result of bootstrap method has high consistency with moving average method and gold standard. The bootstrap method was more sensitive in the early warning of influenza pandemic in 2009 compared with moving average method and more consistent with gold standard（kappa=0.728）. Conclusion： Bootstrap was a compute-intensive method which was not rely on the distribution of ILI% data, and it can be applied as an alternative and complementary method in the early warning for influenza surveillance.