The RAPIDD Ebola Forecasting Challenge: Synthesis and Lessons Learnt

Cecile Viboud, Kaiyuan Sun, Robert Gaffey, Marco Ajelli, Laura Fumanelli, Stefano Merler, Qian Zhang, Gerardo Chowell, Lone Simonsen, Alessandro Vespignani


Infectious disease  forecasting is gaining traction in the public health community; however,  limited systematic comparisons of model performance exist. Here we present  the results of a synthetic forecasting challenge inspired by the West African  Ebola crisis in 2014-2015 and involving 16 international academic teams and  US government agencies, and compare the predictive performance of 8  independent modeling approaches. Challenge participants were invited to  predict 140 epidemiological targets across 5 different time points of 4  synthetic Ebola outbreaks, each involving different levels of interventions  and "fog of war" in outbreak data made available for predictions.  Prediction targets included 1-4 week-ahead case incidences, outbreak size, peak  timing, and several natural history parameters. With respect to weekly case  incidence targets, ensemble predictions based on a Bayesian average of the 8  participating models outperformed any individual model and did substantially  better than a null auto-regressive model. There was no relationship between  model complexity and prediction accuracy; however, the top performing models  for short-term weekly incidence were reactive models with few parameters,  fitted to a short and recent part of the outbreak. Individual model outputs  and ensemble predictions improved with data accuracy and availability; by the  second time point, just before the peak of the epidemic, estimates of final  size were within 20% of the target. The 4th challenge scenario - mirroring an  uncontrolled Ebola outbreak with substantial data reporting noise - was  poorly predicted by all modeling teams. Overall, this synthetic forecasting  challenge provided a deep understanding of model performance under controlled  data and epidemiological conditions. We recommend such "peace time"  forecasting challenges as key elements to improve coordination and inspire  collaboration between modeling groups ahead of the next pandemic threat, and  to assess model forecasting accuracy for a variety of known and hypothetical  pathogens.

Related publications