基于移动百分位数法流感预警模型的探讨

Value of precaution model of influenza based on moving percentile method

目的探讨基于移动百分位数法流感预警模型的最佳预警界值。方法采用不同百分位数P作为移动百分位数法的候选预警界值,对深圳市宝安区2009～2011年流感的周病例报告数进行分析,评价模型的灵敏度和特异度,确定最佳预警界值,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P80,ROC曲线下面积AUC=0.925(95%CI:0.804～0.963),灵敏度为91.29%,特异度为90.31%。进一步分析发现,在P80的预警界值下,该模型的灵敏度为77.78%、特异度为85.29%、约登指数为63.07%、阳性预测值为73.68%、阴性预测值为87.88%,在P80界值时预警模型具有较好的预警判别效能。结论基于移动百分位数法预警模型能够有效的预警流感的暴发流行,最佳的预警界值为P80。

Objective To explore the optimization alert threshold of the precaution model of influenza based on the moving percentile method. Methods The different percentiles of P were respectively adopted as the candidates of alert thresholds on the moving percentile method. Data weres analyzed based on the reported cases of influenza in Baoan district of Shenzhen from 2009 to 2011. The sensitivity and specificity of the precaution model were evaluated in different percentiles of P, and the optimization alert threshold was determined. In addition, the discrimination efficacy of precaution model was validated. Results The optimization alert threshold of precaution model was 80 percentile, where the area under the ROC curve was 0.925（95% confidence internal： 0.804 to 0.963）, and the sensitivity and specificity were 91.29% and 90.31%, respectively. Further analysis showed that the better discrimination efficacy of precaution model was found （sensitivity= 77.78%, specificity=85.29%, Youden＇ s index=63.07%, positive predictive value=73.68%, and negative predictive value= 87.88%, respectively）. Conclusion The precaution model of i,fluenza outbreak can be, based on the moving percentile method, used for prediction of influenza outbreak and the optimization alert percentile was 80 percentile.