Study Of Coupling Algorithm Of Dengue Transmission Dynamic Model With Ensemble Kalman Filter And Its Application

Background and objectivesWith the development of the social urbanization,it provides a good opportunity for infectious diseases to spread quickly owing to the dense urban population.Once there is an outbreak happen in urban region,it will cause serious burden to economy and public health.Hence,the early warning and prevention of infectious diseases is acting a more and more important role.In recent years,dynamic model,statistical model and machine learning both have a good performance on prediction of epidemic transmission for infectious diseases,particularly,it is outstanding for the coupling algorithm of compartmental model and data assimilation method.However,for mosquito-borne dengue fever,there still exist great challenges to construction of an accurate prediction model because of multiple features such as the complex transmission mechanism,amounts of influence factors,seasonality and inter-annual variation of outbreak intensities.Consequently,we need to develop and improve the coupling algorithm compartmental model and data assimilation method,under considering the seasonality and peak intensity variability of outbreaks,to establish a methodological foundation for realizing real-time forecast of dengue outbreak.This study aims to construct a coupling algorithm of SIR-EAKF suitable for dengue epidemic in southern China based on seasonal dengue susceptible-infected-recovered model(SIR,a simple dengue compartmental model)and ensemble adjustment Kalman filter(EAKF,a data assimilation method).By doing so,the improved model can simulate the seasonal changes of dengue,and further forecast the occurrence and development of dengue epidemic,to realize accurate real-time forecast of dengue outbreak.Material and methodsIn this study,mosquito density and ambient air temperature information were used to construct a seasonal mosquito-borne dengue SIR-type compartmental model.Then,300 members of dengue SIR model with random initial state and parameters are integrated and coupled with EAKF,where the model assimilates observation of weekly reported new infected dengue cases with simulation and updates unobserved state variables and model parameters.Through updating the states and parameters of compartmental model in this way iteratively,the prediction model can make a more accurate ensemble forecast of dengue epidemic or outbreak.In order to deal with the limitation of inter-annual obvious variation of epidemic scale and outbreak intensity for prediction model,our study proposed a dynamic adjusting and iterative updating estimation method to calculate the effective population size.The effective population scale of infectious diseases compartmental model,which is used to ensure the sufficient supply of susceptible population,is calculated from the historical observation data and the cumulative reported cases in current epidemic season.In doing so,the susceptible population determined by the total population size can be make sure to sufficient and support the forecast of dengue outbreak from compartmental model.By integrating the dynamic adjusting method to effective population and SIR-EAKF model,a combined SIR-EAKF forecast system is constructed,and we provide a methodological foundation for forecast of infectious diseases to adapt with inter-annual variation scale of epidemic and outbreak.ResultsRetrospective forecasts of seven dengue seasons,from the 2011-2012 seasons to the2017-2018 seasons,show that the prediction accuracy of the improved SIR-EAKF model is20% higher than that of the traditional SIR-EAKF model.Specifically,the improved SIR-EAKF model enhances the accuracy of peak time prediction to 85.7% with 6 or more consecutive weeks prior to the observed peak.In addition,the forecast of peak intensity and total dengue cases both have a good performance,leading accurate forecast of 71.4% and57.1% one week ahead respectively,after an accurate peak time is captured.At the same time,the simulation experiment,stability analysis and comparison with GAM forecast model both perform well for the improved SIR-EAKF model.ConclusionsBased on the traditional SIR-EAKF model,we develop a kind of dynamic adjusting and estimating method to effective population size,which is then integrated with data assimilation method to assimilate actual observed dengue cases as well as optimize the ensemble of compartmental model.This dynamic adjustment of effective population size allows the state variables and model parameters of compartment model be adjusted more flexibly and improve the prediction ability for infectious diseases with obvious inter-annual variation of epidemic and outbreak.According to the validation and analysis result of dengue surveillance data in Guangzhou,we have found that the improved SIR-EAKF model proposed in this study shows a good performance on prediction including: first,this method can simulate the seasonality of dengue accurately;second,this method can forecast the tread of dengue outbreak(the number of dengue cases in peak and the peak timing)efficiently.In total,this study proposes an improved SIR-EAKF model,providing a useful scheme in prediction for other infectious diseases.