Forecasting the spatial spread of an Ebola epidemic in real-time: comparing predictions of mathematical models and experts

Ebola virus disease outbreaks can often be controlled, but require rapid response efforts frequently with profound operational complexities. Mathematical models can be used to support response planning, but it is unclear if models improve the prior understanding of experts. We performed repeated surveys of Ebola response experts during an outbreak. From each expert we elicited the probability of cases exceeding four thresholds between two and 20 cases in a set of small geographical areas in the following calendar month. We compared the predictive performance of these forecasts to those of two mathematical models with different spatial interaction components. An ensemble combining the forecasts of all experts performed similarly to the two models. Experts showed stronger bias than models forecasting two-case threshold exceedance. Experts and models both performed better when predicting exceedance of higher thresholds. The models also tended to be better at risk-ranking areas than experts. Our results support the use of models in outbreak contexts, offering a convenient and scalable route to a quantified situational awareness, which can provide confidence in or to call into question existing advice of experts. There could be value in combining expert opinion and modelled forecasts to support the response to future outbreaks.