暴发规模和持续时间对CUSUM预警模型效能的影响

采用数据模拟方法,在一段服从Poisson分布的数据序列中插入11个不同暴发规模、不同持续时间的暴发事件,研究暴发规模和持续时间对CUSUM(Cumulative Sum)预警模型效能的影响.结果发现,在现有模拟条件下,灵敏度为9.1％～100.0％,特异度≥98.6％；灵敏度与暴发规模有明显的关系,暴发强度越大,灵敏度越高,而暴发持续时间并不影响灵敏度；灵敏度达到100.0％,暴发规模需≥2.6；暴发规模越大,滞后时间越短；当暴发规模≥1.8时,滞后时间控制在＜1d.可见CUSUM预警模型的预警效能受暴发规模的影响,与暴发持续时间关系不明显.该模型对规模≥2.4的暴发具有较好的探测能力.

Effect of magnitude and duration on the performance of Cumulative Sum

To explore the effect of magnitude and duration on the performance of Cumulative Sum (CUSUM), with simulation method used on the subject after the insertion of 11 outbreak events into baseline data with Poisson distribution. Sensitivity fluctuated from 9.1% to 100.0% with specificities higher than 98.6%. Sensitivity was significantly correlated with magnitude, and increased along with the increase of magnitude. However, no significant correlation was observed between sensitivity and duration. A magnitude which was at least 2.6 times higher than that of the mean daily baseline could result in the sensitivity of 100.0%. Time-lag would be improved along with the increase of magnitude. Time between onset and detection of an outbreak was no longer than one day when magnitude was more than 1.8 of the mean daily baseline. In summary, the performance of CUSUM was influenced by magnitude, but not by duration. CUSUM had the advantage of good time-lag and high sensitivity when the outbreak magnitude was more than 2.4 time over the baseline data.