The ensemble Kalman filter forecast of dengue incidence

Authors: YI, CHUNLIN - Kansas State University; Cohnstaedt, Lee; SCOGLIO, CATERINA - Kansas State University

Submitted to: IEEE Access
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/5/2021
Publication Date: 12/2/2021
Citation: Yi, C., Cohnstaedt, L.W., Scoglio, C. 2021. The ensemble Kalman filter forecast of dengue incidence. IEEE Access. 9:156758-156767. https://doi.org/10.1109/ACCESS.2021.3129997.
DOI: https://doi.org/10.1109/ACCESS.2021.3129997

Interpretive Summary: Dengue fever is a mosquito-borne viral disease of severe public importance and outbreaks are becoming increasingly more frequent in tropical regions during the past few decades. During large dengue outbreaks, public health officials and healthcare facilities often cannot deploy medical personnel, or administer treatment resources and emergency vector control measures in a timely and efficient manner. Therefore, an accurate incidence forecast and advance warning system of dengue outbreaks is needed to reduce morbidity and mortality by optimally managing resources. The classical transition state models susceptible-exposed-infected-recovered (SEIR) for humans and the susceptible-exposed-infected (SEI) for mosquitos were used effectively to forecast outbreaks but estimating the transition parameters between the states has been a source of error in past predictions. Our new approach adds an ensemble Kalman filter to the SEIR-SEI model which allows for the model to optimize the parameters and simultaneously evaluate predictions on a subset of the dengue incidence data. The model was evaluated using weekly dengue incidence during the 2015 outbreak in Kaohsiung, Taiwan. This model tuning resulted in very accurate two-week forecasts which is sufficient time to prepare for a growing outbreak or know when one is waning to allocate the resources to another location.

Technical Abstract: Dengue fever is an increasing problem in tropical areas. This mosquito-borne virus results in in high morbidity and mortality during outbreaks in populations with high mosquito abundance. Forecasting outbreak characteristics to identify the onset, peak, and waning situations will optimize public health responses and the maximize available resources to the locations that most need it before an outbreak rather than in response to an ongoing situation. We developed a model that couples the classic disease state transmission model susceptible-exposed-infected-recovered (SEIR) with an ensemble Kalman Filter (EnKF) to simultaneously predict the transitions parameters and evaluate fit to past case data used to train the model. By combining the EnKF with the SEIR model, the forecasts are very accurate 1-2 time steps into the future. This was tested using a dengue outbreak from 2015 in Kaohsiung, Taiwan. This model tuning resulted in very accurate two-week forecasts which is sufficient time to prepare for a growing outbreak or know when one is waning to allocate the resources to another location.