论文部分内容阅读
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 To establish a flu early warning model based on mobile percentile method and explore the best precautionary threshold. Methods Using different percentiles P as the candidate precautionary threshold, the number of reported cases of influenza during 2006-2009 in a province was analyzed to select the best precautionary boundary value. The 2010 flu surveillance data was used to verify the discriminant effectiveness of the model. Results The best pre-warning value of influenza outbreak was P85, the sensitivity was 94.12%, the specificity was 88.89%, and the area under the ROC curve was 0.915 (95% CI: 0.804-0.974). Empirical studies showed that the sensitivity of the model was 73.68%, the specificity was 93.94%, the positive predictive value was 87.50%, the negative predictive value was 86.11% and the area under the ROC curve was 0.838 (95% CI : 0.710 ~ 0.925). Conclusion Under the pre-warning threshold of P85, the model based on moving percentile method has better performance of predictive discrimination.