2011-2013年国家传染病自动预警系统中时间模型和时空模型应用效果比较

Comparing the performance of temporal model and temporal-spatial model for outbreak detection in China Infectious Diseases Automated-alert and Response System ,2011-2013, China

目的 比较国家传染病自动预警系统（简称预警系统）中时间模型与时空模型的预警效果,为预警模型的进一步改进提供依据.方法 2011-2013年中国CDC通过预警系统在20个省份的208个试点县（区）同时应用时间模型和时空模型,根据发病水平对16种传染病分两类进行预警分析,结合疾病监测信息报告管理系统报告的16种传染病个案数据、突发公共卫生事件报告管理信息系统报告的暴发事件,采用预警信号数、灵敏度、错误预警率和及时性等指标,比较两个模型的暴发探测效果.结果 对于16种传染病整体而言,时间模型与时空模型的灵敏度分别为96.23％（153/159）和90.57％（144/159）,差异无统计学意义（Z=-1.604,P=0.109）;时间模型的错误预警率（1.57％,57 068/3643 279）高于时空模型（0.64％,23 341/3643 279）（Z=-3.408,P=0.001）;两者的暴发探测时间中位数分别为3.0d和1.0d,差异无统计学意义（Z=-1.334,P=0.182）.对于发病水平较低的6种Ⅰ类疾病（流行性出血热、流行性乙型脑炎、登革热、流行性脑脊髓膜炎、流行性和地方性斑疹伤寒、钩端螺旋体病）,时间模型和时空模型的灵敏度均为100％（8/8,8/8）,错误预警率均为0.07％（954/1 367437,900/1 367 437）,二者的暴发探测时间中位数分别为2.5和3.0d,时空模型比时间模型减少2.29％（23条）预警信号.对于发病水平较高的10种Ⅱ类疾病（流行性腮腺炎、痢疾、猩红热、流行性感冒、风疹、戊型肝炎、急性出血性结膜炎、甲型肝炎、伤寒和副伤寒、其他感染性腹泻病）,时间模型和时空模型的灵敏度分别为96.03％（145/151）和90.07％（136/151）,时空模型比时间模型减少59.36％（56 656条）预警信号,各病种信号数和错误预警率均低于时间模型,时空模型暴发探测时间中位数（1.0 d）短于时间模型（3.0 d）.结论 总体上时空模型比时间模型预警效果较好,但对于不同发病水平的传染病,时空模型和时间模型的暴发探测效果有所差别,预警系统应根据具体的病种来调整和优化时间模型与时空模型.

Objective For providing evidences for further modification of China Infectious Diseases Automated-alert and Response System （CIDARS） by comparing the early-warning performance of the temporal model and temporal-spatial model in CIDARS.Methods The application performance for outbreak detection of temporal model and temporal-spatial model simultaneously running among 208 pilot counties in 20 provinces from 2011 to 2013 was compared; the 16 infectious diseases were divided into two classes according to the disease incidence level; cases data in nationwide Notifiable Infectious Diseases Reporting Information System was combined with outbreaks reported to Public Health Emergency Reporting System,by adopting the index of the number of signals,sensitivity,false alarm rate and time for detection.Results The overall sensitivity of temporal model and temporal-spatial model for 16 diseases was 96.23% （153/159） and 90.57% （144/159） respectively,without significant difference （Z =-1.604,P =0.109）,and the false alarm rate of temporal model （1.57%,57 068/3 643 279） was significantly higher than that of temporalspatial model （0.64%,23 341/3 643 279） （Z=-3.408,P =0.001）,while the median time for detection of these two models was not significantly different,which was 3.0 days and 1.0 day respectively （Z =-1.334,P =0.182）.For 6 diseases of type Ⅰ which represent the lower incidence,including epidemic hemorrhagic fever,Japanese encephalitis,dengue,meningococcal meningitis,typhus,leptospirosis,the sensitivity was 100% for both models （8/8,8/8）,and the false alarm rate of both temporal model and temporal-spatial model was 0.07% （954/1 367 437,900/1 367 437）,with the median time for detection being 2.5 days and 3.0 days respectively.The number of signals generated by temporal-spatial model was reduced by 2.29% compared with that of temporal model.For 10 diseases of type Ⅱ which represent the higher incidence,including mumps,dysentery,scarlet fever,influenza,rubella,hepatitis E,acute hemorrhagic conjunctivitis,hepatitis A,typhoid and paratyphoid,and other infectious diarrhea,the sensitivity of temporal model was 96.03% （145/151）,and the sensitivity of temporal-spatial model was 90.07% （136/151）,the number of signals generated by temporal-spatial model was reduced by 59.36% compared with that of temporal model.Compared to temporal model,temporal-spatial model reduced both the number of signals and the false alarm rate of all the type Ⅱ diseases ; and the median of outbreak detection time of temporal model and temporal-spatial model was 3.0 days and 1.0 day,respectively.Conclusion Overall,the temporalspatial model had better outbreak detection performance,but the performance of two different models varies for infectious diseases with different incidence levels,and the adjustment and optimization of the temporal model and temporal-spatial model should be conducted according to specific infectious disease in CIDARS.