Study On Early Warning Technique Of Dengue Fever Based On Imported Cases, Vectors And Climate Factors

BackgroundDengue Fever, an infectious disease caused by dengue viruses and transmitted by Aedes mosquitoes, which is now endemic in Africa, America, South-East Asia, and Western Pacific regions and more than100countries of Europe. There are approximately51million dengue infections worldwide every year and2.5billion people living in dengue-endemic countries. From1978to2008, a total of more than650,000cases were documented in mainland China, resulting in610deaths. Guangdong had the most dengue cases in mainland China and experienced numerous large outbreaks. Until now, dengue fever in Guangdong is still characterized as an imported disease, without recognized evidence of its endemic status. Therefore, evaluating the risk of local dengue fever transmission is critical as well as up-to-date information on prevalence and incidence areas. This will allow for the rapid detection of significant shifts in epidemiology as well as serve as an epidemic early warning system that will provide valuable guidance to local health departments.In light of climate change and increased globalization, there has been growing momentum in the research of socio-ecological effects on infectious diseases such as dengue. The current study finds that the occurrence of disease and epidemic frequency are influenced by the interplay among vectors, climate, environment and socio-economic factors. However, there are several technical difficulties involved with precisely measuring the quantitative correlation between the numerous influencing factors and dengue fever that require overcoming for the development of an early warning model.In this study, we aim to use the dengue fever surveillance data in Guangzhou to improve our understanding of the influencing factors of dengue fever occurrence/outbreaks, and then evaluate our models for their potential as a dengue fever early warning system.ObjectivesTo analyze influence factors of local dengue fever transmission and establish an early warning technique for dengue outbreaks.Data and methods1. Data source1.1Dengue fever casesWe obtained data of dengue fever cases in Guangzhou from2001to2010, from the "Notifiable Infectious Disease Reporting Information System" of Chinese Center for Disease Control and Prevention （CDC）, Guangdong CDC and Guangzhou CDC. All cases were identified to be imported cases or indigenous cases.1.2Dengue fever outbreaksWe obtained data on dengue fever outbreaks in Guangzhou from2001to2010, from the "Public Health Emergency Reporting Information System" of Chinese Center for Disease Control and Prevention, Guangdong CDC and Guangzhou CDC. Based on the definition of dengue fever outbreaks with "Dengue Fever surveillance project in China", we chose the reported events which had3or more related cases.1.3Vector surveillance dataWe obtained data on Breteau Index of vector surveillance data in Guangzhou from2001to2010, from Guangdong CDC and Guangzhou CDC1.4Climate surveillance dataWe obtained climate surveillance data in Guangzhou from2001to2010, from China Meteorological Data Sharing Service System. All data included accumulation precipitation, average atmospheric pressure, maximum atmospheric pressure, minimum atmospheric pressure, average wind velocity, maximum wind velocity, extreme wind velocity, average temperature, maximum temperature, minimum temperature, sunshine accumulation hours, average relative humidity and minimum relative humidity.1.5Population dataWe obtained data on population in Guangzhou from2001to2010, from the "Basis information management system" of Chinese Center for Disease Control and Prevention and "Statistical Yearbook of Guangzhou".2. Choosing influence factorsWe considered the imported cases, vector and climate surveillance data as independent variables. Logistic regression was used to analyze the correlation between the occurrence of indigenous dengue and the independent variables. Time series Poisson regression was used to analyze the correlation between the indigenous dengue cases and the independent variables. We tested different time lags for each variable from lag1up to lag12months, and compared all the regression coefficients in order to select the most significant candidate variables for the multiple regression analysis.3. Multiple regressionWe considered the better significant candidate variables as independent variables. Multiple Logistic regression was used to analyze the correlation between the occurrence of indigenous dengue and the independent variables. Time series Poisson regression was used to analyze correlation between the indigenous dengue cases and the independent variables was analyzed by multiple time series Poisson regression. We chose the best-fit model, evaluated the influential factors, and calculated the occurrence probability and count of indigenous dengue fever cases. To analyze the quantitative relationship between indigenous and imported cases in epidemic period, we used data from May to November every year to establish the time series Poisson regression and calculated the indigenous dengue fever cases in the best-fit model.4. Risk assessment and early warning4.1The risk assessment of the indigenous dengue occurrence Using logistic regression, we assessed the risk of indigenous dengue occurrence based on the probability （Pt）. We defined three risk levels:level one （Pt<0.05） is low risk, which means no case was predicted to occur; level two （0.5>Pt>0.05） is middle risk, which means cases may occur; level three （Pt>0.5） is high risk, which means there was a good chance for cases. The count of reported indigenous dengue fever cases in Guangzhou from2001to2010was used to evaluate the accuracy of the risk assessment. There are criteria as follows:there was no case in the level one risk months, case or not in the level two months, or there were cases in the level three risk months.4.2The risk assessment of the dengue fever outbreaksIn the time series Poisson regression, we assessed the risk of dengue fever outbreaks based on the predicted cases （Ct）. We defined three risk levels:level one （Ct=0） is low risk, which means no case was predicted to occur; level two （0<Ct<6） is middle risk, which means cases may be sporadic; level three （Ct≥6） is high risk, which means there is a good chance for outbreaks. The data of reported indigenous dengue fever cases in Guangzhou from2001to2010was used to evaluate the accuracy of the risk assessment. The reported cases were used to define risk level of months in the same criterion, if it is consistent with the predicted results, the prediction was deemed appropriate.Poisson regression, which was in full time series and in different period, was used to assess the risk level of months and evaluate the accuracy respectively.4.3Early warning ranking modelBased on the combination of risk assessment methods, we designed two kinds of early warning technical route.（1） Early warning route I:Type A early warning signal would be generated when outbreak risk becomes level three, Type B early warning signal would be generated when outbreak risk becomes level two, no signal would be generated when outbreak risk remains at level one.（2） Early warning route II:If the indigenous dengue occurrence risk becomes level three, the indigenous dengue outbreak risk should be calculated. Type A early warning signal would be generated when outbreak risk becomes level three, Type B early warning signal would be generated when outbreak risk becomes level two, no signal would be generated when outbreak risk remains at level one.We can establish four kinds of early warning model:Model1was the early warning route I in the risk assessment of full time series data, Model2was the early warning route II in the risk assessment of full time series data, Model3was the early warning route I in the risk assessment of different period data, Model4was the early warning route II in the risk assessment of different period data.4.4Early warning effect evaluationBy calculating evaluation indicator, such as sensitivity and specificity, we compared the performance of four early warning models. Sensitivity is the percentage of the months which had outbreaks and sent early warning signals in total months which had outbreaks. Specificity is the percentage of the months which didn’t have outbreaks and sent early warning signals in total months which had no outbreak.Results1. The influence factors for indigenous dengue occurrence and outbreaksThe best-fit Logistic regression:In (（Pt/（1-Pt））=314.33+2.11CIn（t-1）+0.46CIm（t-1）+0.41ARH (t-2+0.36BI（t-2）-0.01Rain（t-2）0.35AP （t-1）（p<0.001）, where P is probability of indigenous cases occurrence, Cin, Cim, ARH, BI, Rain and AP are the regression coefficients for indigenous cases, imported cases, average relative humidity, Breteau Index, accumulation precipitation and average atmospheric pressure, t-i means i months lags.The result shows indigenous case number （OR=8.25） and imported case number （OR=1.58） before1month, average relative humidity （OR=1.51）, BI （OR=1.43） and accumulation precipitation （OR=0.99） before2months, and average atmospheric pressure （OR=0.71） before1month influence the occurrence of indigenous cases.The best-fit time series Poisson regression in full time series data: Ln（Yt）=-47.03+0.69LnY（t-1）+0.40Tmin（t-1）+0.28ARH（m）+0.25RHmin（t₂）+0.10BI（t-1）-0.61SAF（t）+LnPop （t）（p<0.001）, The best-fit time series Poisson regression in epidemic period data:LnY（t）=-43.89+0.67LnY（t-1）+0.40Tmin（t-1）+0.28RHmin（t.2）+0.25ARH（t-1）-1.38CIm（t-2）-0.55SAF（t）+LnPop（t）（P<0.001）. Where Y, Tmin, ARH, RHmin, CIm,BI, SAF, Pop are the regression coefficients for indigenous cases, minimum temperature, average relative humidity, minimum relative humidity, imported cases, Breteau Index, seasonal adjusted factors and population, t-i means i months lags.The result in best-fit time series Poisson regression in full time series data shows indigenous case number （RR=1.99）, minimum temperature （RR=1.50）, average relative humidity （RR=1.33） before1month, minimum relative humidity （RR=1.29） before2months, Breteau Index （RR=1.11） before1month influence the indigenous cases outbreaks. The best-fit time series Poisson regression in epidemic period data shows the same independent variables such as indigenous case number, minimum temperature, average relative humidity before1month, minimum relative humidity before2months, and the imported cases became the new influence factors.2. The risk assessmentThe risk of indigenous dengue occurrence showed the seasonal characters, there were25months in level three （35.7%）,33months in level two （47.1%）,12months in level one （17.1%） from May to November in2001-2010. And there were3months in level three （6.0%）,1month in level two （4.0%） and46months in level one （92.0%）.The accurate of risk assessment was96.7%.The time series Poisson regression in full time series data and in different had the same assessment result. There were18months in level three （25.7%）,6months in level two （8.6%）,46risk months in level one （65.7%） from May to November in2001-2010. And there were1month in level two （2.0%） and49months in level one （98.0%）. The accurate of risk assessment was86.7%.3. Early warning results3.1Dengue fever outbreaks model in full time series data（1） Model1:26early warning signals were generated in Guangzhou City from2001to2010（including18type A early warning signals and8type B early warning signals）, sensitivity was93.8%and specificity was89.4%. （2） Model2:25early warning signals were generated in Guangzhou City from2001to2010（including18type A early warning signals and7early warning signals）, sensitivity was93.8%and specificity was90.4%.3.2Dengue fever outbreaks model in different period data（1） Model3:26early warning signals were generated in Guangzhou City from2001to2010（including18type A early warning signals and8type B early warning signals）, sensitivity was93.8%and specificity was89.4%.（2） Model4:25early warning signals were generated in Guangzhou City from2001to2010（including18type A early warning signals and7type B early warning signals）, sensitivity was93.8%and specificity was90.4%.By comparing the early warning effects, the outbreaks early warning in different period data had better sensitivity of type A early warning. Model3and Model4had the same sensitivity, and Model4had a better specificity.Conclusions1. Indigenous case number before1month, imported case number, average atmospheric pressure, average relative humid before2months, BI and accumulation precipitation influence the indigenous cases in Guangzhou City, the risk of occurrence in Guangzhou City can be predicted1month in advance by constructing a Logistic regression model.2. Indigenous case number before1month, minimum temperature, average relative humid, BI and minimum relative humid before2months influence the dengue fever outbreak risk in Guangzhou City. In the epidemic period of dengue fever, imported cases number before1month becomes one of the key influential factors on dengue fever outbreak, the risk of which in Guangzhou City can be predicted1month in advance by building a time series Poisson regression model. This study suggests that in high risk regions we can afford technical support for early warning on dengue fever outbreaks by enhancing the monitoring of imported and indigenous cases, BI, and collecting climate data.3.The early warning model of dengue fever outbreak risk was established in this study, taking the differences of epidemic stage into consideration, has a good sensitivity and specificity, which can give reference to the research and application of dengue fever outbreak early warning systems in high risk regions.