移动百分位数法分地区设定预警阈值对传染病预警效果的影响分析  

Analysis of effect on infectious diseases outbreak detection performance by classifying provinces for moving percentile method

作　　者：张洪龙[1] 孙乔[2] 赖圣杰[1] 任翔[1] 周鼎伦[3] 叶先飞 曾令佳[1] 余建兴[1] 王丽萍[1] 余宏杰[1] 李中杰[1] 吕炜[4] 兰亚佳[3] 杨维中[5] 

机构地区：[1]中国疾病预防控制中心传染病预防控制处传染病监测预警中国疾病预防控制中心重点实验室, 北京102206 [2] 上海市浦东新区疾病预防控制中心 [3] 四川大学华西公共卫生学院 [4]广西壮族自治区疾病预防控制中心 [5]中国疾病预防控制中心

出　　处：《中华预防医学杂志》2014年第4期265-269,共5页Chinese Journal of Preventive Medicine

摘　　要：目的 分析国家传染病自动预警系统（简称预警系统）中移动百分位数法预警模型分地区采用不同预警阈值对传染病暴发探测效果的影响,从而为进一步改进预警系统中预警模型的预警阈值提供依据.方法 采用2011年1月1日至2013年10月31日期间预警系统中移动百分位数法在中国内地31个省份对16种传染病发出的预警信号和响应处理结果,疾病监测信息报告管理系统相关个案数据,以及在突发公共卫生事件报告管理信息系统中报告的相关突发事件作为数据来源.采用灵敏度最高、暴发探测时间最短且预警信号数最少作为最优预警阈值的设定标准,比较分析每种传染病在全国范围内设定统一预警阈值（全国阈值）与分省份设定不同预警阈值（分省阈值）对传染病暴发探测的预警信号数、灵敏度和及时性的差异.结果 16种进行预警的传染病中,痢疾、登革热、甲型肝炎、伤寒和副伤寒、流行性脑脊髓膜炎、流行性乙型脑炎、猩红热、钩端螺旋体病、戊型肝炎、流行性和地方性斑疹伤寒等10种传染病的最优全国阈值和最优分省阈值相同,均为第90百分位数（P90）;其余6种传染病（包括其他感染性腹泻病、流行性感冒、急性出血性结膜炎、流行性腮腺炎、风疹和流行性出血热）的最优全国阈值均为第80百分位数（P80）,而最优分省阈值有所不同.这6种疾病采用最优分省阈值与采用最优全国阈值的预警结果相比,暴发预警灵敏度和探测时间维持不变,信号数分别减少23.71％（45 557条）、15.59％（6 124条）、14.07％（1 870条）、9.44％（13 881条）、8.65％（1 294条）和6.03％（313条）.结论 预警系统中移动百分位数法预警模型通过分地区分病种优化预警阈值,可在维持暴发探测灵敏度和及时性不变的情况下大量减少预警信号,下一步预警系统可考虑增加各省预警阈值设置灵活性,Objective Providing evidences for further modification of China Infectious Diseases Automated-alert and Response System （CIDARS） via analyzing the outbreak detection performance of Moving Percentile Method （MPM） by optimizing thresholds in different provinces.Methods We collected the amount of MPM signals,response results of signals in CIDARS,cases data in nationwide Notifiable Infectious Diseases Reporting Information System,and outbreaks data in Public Health Emergency Reporting System of 16 infectious diseases in 31 provinces in Chinese mainland from January 2011 to October 2013.The threshold with the optimal sensitivity,the shortest time to detect outbreak and the least number of signals was considered as the best threshold of each disease in Chinese mainland and in each province.Results Among all the 16 diseases,the optimal thresholds of 10 diseases,including dysentery,dengue,hepatitis A,typhoid and paratyphoid,meningococcal meningitis,Japanese encephalitis,scarlet fever,leptospirosis,hepatitis,typhus in country level were the 90th percentile （P9o）,which was the same as provincial level for those diseases.For the other 6 diseases,including other infectious diarrhea,influenza,acute hemorrhagic conjunctivitis,mumps,rubella and epidemic hemorrhagic fever,the nationwide optimal thresholds were the 80th percentile （Ps0）,which was different from that by provinces for each disease.For these 6 diseases,the number of signals generated by MPM with the optimal threshold for each province was decreased by 23.71% （45 557）,15.59% （6 124）,14.07% （1 870）,9.44% （13 881）,8.65% （1 294） and6.03% （313） respectively,comparing to the national optimal threshold,while the sensitivity and time to detection of CIDARS were still the same.Conclusion Optimizing the threshold by different diseases and provinces for MPM in CIDARS could reduce the number of signals while maintaining the same sensitivity and time to detection.

关 键 词：传染病 疾病暴发流行 模型 统计学 移动百分位数法 

分 类 号：R181.8[医药卫生—流行病学]