Forecasting and Uncertainty in Modeling the 2014-2015 Ebola Epidemic in West Africa

The 2014 Ebola epidemic in West Africa is the largest ever recorded, and understanding the interrelated dynamics of surveillance and intervention is a key concern, both for this and future epidemics. Moreover, as transmissibility and mortality are believed to increase as symptoms progress, intervention strategies may depend on individual’s stage of infection. To examine these issues, we developed a stage-structured model of Ebola, which includes a term for fraction of the population at risk, reporting rate, among other factors. We generated short term forecasts for Guinea, Liberia, and Sierra Leone, beginning October 1, 2014, which we have since validated using subsequent data. We examined the relative contributions of the stages of infection, and then expanded the model to consider Ebola treatment unit (ETU) dynamics and interventions, incorporating both stagedependent hospitalization rates and dynamic ETU capacity. We found that a wide range of forecasted trajectories fit well to the data. However, by estimating terms for surveillance and intervention, the best-fit models correctly forecasted the qualitative behavior for all three countries, both individually and for all countries combined. In particular, the models correctly forecasted the slow-down and stabilization in Liberia but continued exponential growth in Sierra Leone through October and November 2014. Because increasing intervention levels lead to improved reporting, interventions and reported cases/deaths can have a seemingly paradoxical relationship, in which increasing intervention levels result in apparent increases in cases and deaths (due to improved reporting), even though there has actually been a significant reduction in underlying total cases/deaths. These simulations suggest that some of the observed reductions in the growth rate of the epidemic are consistent with intervention effects. All three transmitting stages (early, late, and funeral) appeared to contribute significantly to transmission, with intervention on any single stage often insufficient to prevent an epidemic. However, parameter unidentifiability issues impede estimation of the relative contributions of each stage of transmission from incidence and deaths data alone, which poses a challenge in determining optimal intervention strategies, and underscores the need for additional data collection. For the ETU-based scenarios, basic treatment and isolation capacity acted as a prerequisite to other interventions, with early-stage isolation, increased staff and supplies, and reductions in funeral transmission only fully effective once sufficient ETU/isolation capacity was achieved.