Application Of Time Series Modeling For Syndromic Surveillance Among The Primary And Middle School Students Of Hangzhou

Traditional disease surveillance system sets up in the hospital diagnostic and laboratory tests, which exists some time lag between the reports or sample collection of symptoms and the final diagnosis of disease. However, we need to quickly identify and respond to biological terrorist incidents or other emergencies of Public Health. Compared to the traditional disease surveillance, syndromic surveillance is monitoring early abnormalities in the outbreak of public health emergencies, through collecting various data related to health events and combining with sophisticated signal detection methods. The data of syndromic surveillance are generally reported directly through the network, which lack of forward-looking, but time series analysis can forecast syndromic surveillance data and achieve the role of early warning.ObjectiveThe purpose of this study is to explore the applications of the time series analysis methods in forecasting the time series of syndromic surveillance. In this issue, trying to establish the statistical prediction model of syndromic surveillance and to forecast the trend of its incidence; Early prediction can provide a scientific and effective basis for development of prevention and control of disease in the recent or long.Data and MethodsThe data of 8 kinds of syndromic surveillance was collected from primary and secondary students in Hangzhou from August 31,2009 to June 31,2010 (44 weeks). With different time series forecasting model, including mixed autoregressive integrated moving average model and exponential smoothing method, establish corresponding prediction model to fit and predict the number of people of fever and cough. Comparing the fitting and prediction accuracy of models and discussing analysis results to identify the more suitable model for such data.Results1 Mixed autoregressive integrated moving average model (ARIMA)Weekly predict model ARIMA(1,0,0) for fever and weekly predict model ARIMA(1,1,1) for cough, that is:2 Exponential smoothing methodThe study also use three exponential smoothing models to predict the series of fever and cough, that is:Simple exponential smoothing method: fever:α=0.99,SSE=24.8319。cough:α=0.45,SSE= 26.4520。Double exponential smoothing method: fever:α=0.27, SSE= 36.2534。cough:α=0.21, SSE=28.9208。Holt-winter's linear exponential smoothing method: fever:α=0.999,γ=0.001, SSE=25.4046。cough:α=0.404,γ=0.001, SSE=26.4351。Conclusions1 Through comparison among those models, ARIMA model and exponential smoothing method all can be used to predict the future series of syndrome. Exponential smoothing method requires to determine the value of a by trails to get the minimum SSE, which only need to know the information of the previous year. The advantage of exponential smoothing is taking the new observed value into account and constantly modifying the model. ARIMA method considers moving average, autocorrelation analysis and the stability of data together, and determine the value of q and p through the analysis of autocorrelation and partial autocorrelation. ARIMA model initially select a trial model based on detailed analysis of time series, and then test the applicability of the model by a series of statistical methods. If the model is not applied, we can make the necessary adjustments, or re-use new trial model. In theory, ARIMA model can be applied to predict all types of time series. ARIMA method is more comprehensive and considering many factors, but sometimes in different application conditions, model selection may be different, this study also confirmed that.2 Results of this study show that exponential smoothing is better than ARIMA models in forecasting, but in the fitting accuracy ARIMA model is better than exponential smoothing.3 The application of time series analysis in syndromic surveillance provide effective support for disease prevention and control in theory. In recent years, traditional disease surveillance based on diagnosis has been difficult to adapt to the demanding requirements of timeliness. Syndromic surveillance is expected to find the incidence of sudden public health incidents and alert earlier by collecting early abnormal data of symptoms and relying on real time data integration and precision in signal detection, which is important for making timely and effectively public health response to reduce morbidity, mortality, and economic losses. It's possible to forecast the abnormalities of symptoms earlier if the appropriate forecast model is established, earlier forecast. Consequently, disease prevention and control will be carried out more timely and more reliable.