采用北京市麻疹的暴发模拟数据比较几种预警模型的功效

Comparison between early outbreak detection models and simulated outbreaks of measles in Beijing

目的优选出北京市麻疹的最佳预警模型及其参数，为其自动预警提供技术支持。方法利用暴发模拟软件生成一系列不同性质的暴发信号，将其添加到北京市2005-2007年麻疹的实际日报数中。综合采用指数加权移动平均（EWMA）、C1-MILD（C1）、C2-MEDIUM（C2）、C3-ULTRA（C3）及时空重排扫描统计等预警模型识别加入的暴发信号，比较各模型不同参数的约登指数（YD指数）和检出时间（DT），优选出最佳模型参数，进而比较各模型最优参数下的预警功效，优选出最佳预警模型。结果EWMA模型的最优参数为λ=0.6，κ=1.0；C1的最优参数为κ=0.1，H=3σ；C2的最优参数为κ=0.1，H=3σ；C3的最优参数为κ=1.0，H=4σ；时空重排扫描模型的最优参数为时间聚集性最大值为7d，空间聚集性最大值为5km：各模型的预警功效评价结果：EWMA的YD指数为90.8％，DT为0.121d；C1的YD指数为88.7％，DT为0.142d；C2的YD指数为92.9％，DT为0.121d；C3的YD指数为87.9％，DT为0.058d；时空重排扫描的YD指数为94.3％，DT为0.176d：结论5种模型中，时空重排扫描的预警功效最优。

Objective Using simulated outbreaks to choose the optimal model and its related parameters on measles so as to provide technical support for developing an Auto Warning System （AWS）. Methods AEGIS-Cluster Creation Tool was applied to simulate a range of unique outbreak signals. Then these simulations were added to the actual daily counts of measles from the National Disease Surveillance System, between 2005 and 2007. Exponential weighted moving average （EWMA）, C1-MILD （CI） , C2-MEDIUM （C2） , C3-ULTRA （C3） and space-time permutation scan statistic model were comprehensively applied to detect these simulations. Tools for evaluation as Youden＇ s index and detection time were calculated to optimize parameters before an optimal model was finally chosen. Results EWMA （λ=0.6,κ=1.0）,C1（κ=0.1,H=3σ）,C2（κ=0.1,H=3σ）, C3（κ=0.1,H=4σ） and space-time permutation scan statistic （maximum temporal cluster size=7 d, maximum spatial cluster size = 5 km ） appeared to be the optimal parameters among these models. Youden＇ s index of EWMA was 90.8% and detection time being 0.121 d. Youden＇s index of C1 was 88.7% and detection time being 0.142 d. Youden＇s index of C2 was 92.9% and detection time being 0.121 d. Youden＇s index of C3 was 87.9% and detection time being 0.058 d. Youden＇ s index of space-time permutation scan statistic was 94.3% and detection time being 0.176 d. Conclusion Among these five early warning detection models, space-time permutation scan statistic model had the highest efficacy.