| BackgroundIn order to detect early a disease outbreak, and take timely response measures, early-warning methods based on various types of infectious disease surveillance data sources have been developed globally. The Chinese Center for Disease Control and Prevention established the Infectious Disease Automated Early-warning and Response System (temporal model) across the nation in April,2008. Based on the moving percentile method, the early-warning system detects daily the aberration of 19 types of notifiable infectious diseases, which have surveillance data of more than 3 years. For diseases having surveillance data of less than 3 years, which cannot satisfy the data requirements of the moving percentile method, it is imperative to build a detection method based on short baseline data. For a disease which has surveillance data of more than 3 years, there has been little research comparing the outbreak detection effects of methods based on long and short baseline data, and no research has been designed to identify the optimum method in different incidence level regions before. Our study choose four early-warning methods for testing and comparison, which can be internationally applied and are easy to popularize, including two based on short baseline data (exponentially weighted moving average (EWMA) and EARS-C1,C2,C3), and the other two based on long baseline data (the moving percentile method (MPM) and the historical limited method (HLM)).Objectives1. For diseases with only short baseline data, try to establish an appropriate early-warning method.2. For diseases with long baseline data, by comparing the detection effects of methods based on long baseline data and that based on short baseline data, identify the appropriate early-warning method. 3. Compare the effects of early-warning methods by distinguishing different incidence level regions.Data and Methods1. Diseases and regionsA disease was chosen which had short-term surveillance data and received more attention as an objective disease. A representative disease which had more reported cases and outbreaks was selected from the 19 kinds of disease which have long surveillance data in the early warning system (temporal model).Then, the provinces which had the most outbreaks of objective diseases were chosen one-by-one as the objective regions. Hand-foot-and-mouth disease (HFMD) in Shandong Province and epidemic parotitis in Chongqing Municipality were chosen for the study.Shandong Province was divided into low, middle and high incidence level regions according to incidences of HFMD of 7～,150～,250～420 (1 per 100 thousand). Chongqing Municipality was divided into low, middle and high incidence level regions according to incidences of epidemic parotitis of 3～,50～,100～135 (1 per 100 thousand). The methods were then respectively run and tested in the different incidence level regions.2. Data sourcesData included the disease case and outbreak information originating from the "Notifiable Infectious Disease Reporting Information System" and "Emergency Public Health Event Management Information System".Disease case information included reporting date of case and the international code of the case address. Outbreak information included outbreak ID, reporting date of the first case of the outbreak, reporting date of the last case of the outbreak, reporting date of the cases related to the outbreak and the international code of the reporting area.Our study collected data on HFMD in Shandong Province in 2009, and epidemic parotitis in Chongqing Municipality from 2006 to 2009.3. Early warning methods and parameter settingFor HFMD of short surveillance data (less than 3 years), the methods based on short baseline data (EWMA and EARS-C1, C2, C3) were tested and compared in Shandong Province. For epidemic parotitis of long surveillance data (more than 3 years), the methods based on long baseline data (MPM and HLM) and short baseline data (EWMA and EARS-C1,C2,C3) were tested and compared in Chongqing Municipality. All early-warning methods were detected at the county level in the regions.The thresholds of MPM were P50, P55,...P95, P100; the a parameters of HLM were 0.05, 0.1,0.2,...0.9; theλparameters of EWMA were 0.1 to 0.9 (with an internal of 0.1), the thresholds were 0.1 to 5(with an internal of 0.1), the thresholds of EARS-C1,C2,C3 were 0.1 to 5 (with an internal of 0.1).4. Model evaluation index and screening criterionThe evaluation index includes time to detection (TTD), sensitivity and specificity. Sensitivity is the percentage of outbreaks being detected, specificity represents the ability of the method to detect non-outbreaks. The screening criterion is:on the premise of TTD (≤2days) and sensitivity (>90%), choosing the optimized early warning method which has the highest specificity.Results1. For HFMD, in the high incidence level region, the optimized threshold of EWMA is 1.1,λis 0.7, specificity is 83.96%, the best detection method of EARS is the C2, with the threshold 0.4, specificity is 85.29%. In the middle incidence level region, the optimized threshold of EWMA is 1.6,λis 0.6, specificity is 88.45%, the best detection method of EARS is the C2, with the threshold 0.4, specificity is 83.81%. In the low incidence level region, the optimized threshold of EWMA is 1.4,λis 0.2, specificity is 90.72%, the best detection method of EARS is the C3, with the threshold 1.9, specificity is 91.95%.2. For the three different incidence level regions in Shandong Province, we used the optimized methods respectively and got the early warning effect:TTD is 2 days, sensitivity is 92.59%, specificity is 89.80%, with 2895 signals. For the regions without incidence level division, EWMA (threshold is 1.2,λis 0.3) is the optimized method, and TTD is 2 days, sensitivity is 91.76%, specificity is 86.73%, with 3740 signals.3. For epidemic parotitis, the incidence level in Chongqing Municipality had an overall upward trend from 2006 to 2009. In the high incidence level region, the optimized threshold of MPM is P100, specificity is 68.88%, the optimized threshold of HLM is 0.01(a), specificity is 71.64%, the optimized threshold of EWMA is 1.5,λis 0.6, specificity is 89.95%, the best detection method of EARS is the C1, with the threshold 0.9, specificity is 90.28%. In the middle incidence level region, the optimized threshold of MPM is P100, specificity is 72.61%, the optimized threshold of HLM is 0.01 (a), specificity is 74.87%, the optimized threshold of EWMA is 1.6,λis 0.3, specificity is 92.48%, the best detection method of EARS is the C2, with the threshold 1.2, specificity is 92.38%. In the low incidence level region, the optimized threshold of MPM is P100, specificity is 88.12%, the optimized threshold of HLM is 0.01(a), specificity is 90.49%, the optimized threshold of EWMA is 3.5,λis 0.5, or threshold is 3.7,λis 0.6, or threshold is 4,λis 0.9, specificity is 96.82%, the best detection method of EARS is the C1, with the threshold 3.3, specificity is 97.85%.4. For the different incidence level regions in Chongqing Municipality, we used the optimized method respectively and got the early warning effect:TTD is 0 day, sensitivity is 93.75%, specificity is 94.76%, with 335 signals. For the regions without incidence level division, the best detection method is EARS-C1, with the threshold of 1.1, TTD is 0.5 day, sensitivity is 90.63%, specificity is 92.39%, with 459 signals.5. The optimized methods and parameters are different in the different incidence level regions of epidemic parotitis. The specificity of the MPM and HLM based on their optimized parameters are similar. The specificity of the EWMA and EARS based on their optimized parameters are similar. Generally, the specificity of MPM and HLM based on long baseline data is worse, with more signals. Conclusions1. For HFMD which has short baseline data, we can adopt EARS and EWMA as the appropriate early-warning method in Shandong Province.2. For epidemic parotitis which has long baseline data, in Chongqing, the effect of the EARS and EWMA methods based on short baseline data is better than that of the MPM and HLM methods based on long baseline data in Chongqing Municipality.3. Comparing the outbreak detection effects of the optimized methods in different incidence level regions and in the whole area, with sensitivity and specificity improved, more signals were reduced. Therefore, we detect different diseases by different methods and parameters in different incidence levels, in order to get the best detection effect. |
PDF Full Text Request