基于移动百分位数法流感预警模型的探讨  

Exploration of a moving percentile-based disease-warning model for influenza

作　　者：杨召[1] 叶中辉[1] 尤爱国[2] 李星[1] 翟羽佳[1] 李岩[1] 王重建[1] 

机构地区：[1]郑州大学公共卫生学院流行病与卫生统计学系,河南郑州450001 [2]河南省疾病预防控制中心传染病所,河南郑州450016

出　　处：《现代预防医学》2014年第8期1345-1347,1353,共4页Modern Preventive Medicine

基　　金：国家自然科学基金资助项目(81001293);郑州大学全国大学生创新创业训练计划项目(2011cxsy173)

摘　　要：目的 建立基于移动百分位数法的流感预警模型，探讨最佳预警界值。方法 采用不同百分位数P作为候选预警界值，对某省2006-2009年流感的周病例报告数进行分析，筛选最佳预警界值；采用2010年流感监测数据验证模型的预警判别效能。结果 流感暴发流行的最佳预警界值为P85，灵敏度为94.12%，特异度为88.89%，ROC曲线下面积为0.915（95%CI：0.804～0.974）。实证研究显示，在P85的预警界值下，该模型的灵敏度为73.68%，特异度为93.94%，阳性预测值为87.50%，阴性预测值为86.11%，ROC曲线下面积为0.838（95%CI：0.710～0.925）。结论 在P85的预警界值下，基于移动百分位数法的模型具有较好的预测判别效能。Objective The study was conducted to establish a moving percentile-based disease-warning model for influenza and to explore the optimum alert threshold. Methods Different percentiles P were adopted as the alert threshold candidates in the analysis of the weekly influenza reports between 2006 and 2009 of a province to determine the optimum alert threshold. Predictive efficacy of the disease-warning model was validated by referencing the influenza surveillance data in 2010. Results The optimum alert thresh- old for influenza epidemic was Pss, with a sensitivity of 94.12% and a specificity of 88.89%; area under the ROC curve was 0.915 （95%CI： 0.804-0.974）. Validation studies showed that at the alert threshold of Pss, sensitivity, specificity, positive predictive value, negative predictive value, and area under of ROC curve of the model were 73.68%, 93.94%, 87.50%, 86.11%, and 0.838 （95%CI： 0.710-0.925）, respectively. Conclusion The moving percentile-based model exhibits good predictive efficacy at the alert threshold of Pss.

关 键 词：流感 预警 移动百分位数法 模型 

分 类 号：R181.2[医药卫生—流行病学]