Developing reliable epidemic forecasting using branching processes: Ebola as a case study

The ongoing Ebola outbreak in the Democratic Republic of Congo is a reminder of the constant threat to human populations posed by emerging and re-emerging infectious diseases. Data collected during outbreaks is increasingly analysed in real-time to inform policy making during outbreaks. In particular, real-time estimation of transmissibility and incidence forecasting are instrumental to designing, monitoring, and adjusting interventions throughout an outbreak, continuously assist in logistical planning, and for advocacy. Various forecasting approaches have been developed, in particular during and since the West African Ebola epidemic. The approach we developed during that Ebola epidemic, which relies on modelling the epidemic as a branching process, has been shown to perform better than other approaches on simulated data. However, it has never been thoroughly validated using real data. In this work, we propose to formally quantify the performance of a branching process forecasting approach on incidence data from the West African Ebola epidemic. We will develop extensions to the method to account for uncertainty in the data and to ensure it involves solely objective, quantitative and data-driven steps. We will also explore the extent to which relaxing assumptions underlying our method, including homogeneous mixing, constant transmissibility, and absence of super-spreading, improves the performance of our forecasts. Finally, we will implement the best performing method in an open source software. This work will ensure that a reliable, tested and reproducible forecasting tool is available to be used in future epidemics to assist policy makers in a robust and timely manner.