[Paper Review] Forecasting and Uncertainty in Modeling the 2014-2015 Ebola Epidemic in West Africa

Marisa C. Eisenberg, Joseph N. S. Eisenberg+5 more|arXiv (Cornell University)|Jan 22, 2015

Viral Infections and Outbreaks Research35 references11 citations

TL;DR

This study develops a stage-structured Ebola transmission model with a generalized correction term to account for under-reporting, population at-risk, and surveillance limitations. Despite parameter unidentifiability, the model accurately forecasted epidemic trajectories in Guinea, Liberia, and Sierra Leone—correctly predicting Liberia’s slowdown and Sierra Leone’s continued growth—demonstrating that simplified correction terms can improve forecasting in data-sparse outbreaks.

ABSTRACT

The Ebola epidemic in West Africa is the largest ever recorded, with over 27,000 cases and 11,000 deaths as of June 2015. The public health response was challenged by difficulties with disease surveillance, which impacted subsequent analysis and decision-making regarding optimal interventions. We developed a stage-structured model of Ebola virus disease (EVD). A key feature of the model is that it includes a generalized correction term accounting for factors such as the fraction of cases reported and fraction of the population at risk (e.g. due to contact patterns, interventions, spatiotemporal spread, pre-existing immunity, asymptomatic cases, etc.). We generated a range of short-term forecasts for Guinea, Liberia, and Sierra Leone, which we then validated using subsequent data. We also used the model to examine the uncertainty in the relative contributions to transmission by the different stages of infection (early, late, and funeral). We found that a wide range of forecasted trajectories fit approximately equally well to the early data. However, by including the correction factor term the best-fit models correctly forecasted EVD activity for all three countries, both individually and for all countries combined. In particular, the model correctly forecasted the slow-down in Liberia, as well as the continued exponential growth in Sierra Leone through November 2014. Parameter unidentifiability issues hindered 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. Even with these limited data, however, it is still possible to accurately capture and predict the epidemic dynamics by using a simplified correction term that approximately accounts for the complex underlying factors driving disease spread.

Motivation & Objective

• Address the challenge of forecasting Ebola dynamics in West Africa with incomplete and unreliable surveillance data during the 2014–2015 outbreak.
• Overcome limitations in case and death reporting, asymptomatic transmission, and population-level risk factors that hinder accurate modeling.
• Develop a flexible modeling framework that accounts for unknown factors such as reporting fractions, pre-existing immunity, and behavioral changes without requiring detailed data.
• Improve short-term epidemic forecasting for Guinea, Liberia, and Sierra Leone using a minimal but effective correction term.
• Assess the identifiability of transmission parameters across different stages of infection (early, late, funeral) to inform intervention strategies.

Proposed method

• Formulate a stage-structured compartmental model of Ebola virus disease (EVD) transmission with distinct stages for early, late, and funeral transmission.
• Incorporate a generalized scalar correction term to account for under-reporting, fraction of population at risk, asymptomatic cases, and other unmeasured factors affecting transmission.
• Calibrate the model to cumulative incidence and death data from WHO (May 2014–February 2015) using a range of initial growth periods to reduce stochastic noise.
• Use likelihood-based parameter estimation and model validation against independent data to assess forecast accuracy.
• Apply profile likelihood and parameter identifiability analysis to evaluate the reliability of transmission parameter estimates.
• Validate forecasts for individual countries and the combined epidemic by comparing predicted trajectories to observed data through November 2014.

Experimental results

Research questions

• RQ1 Can a simplified correction term effectively account for data limitations such as under-reporting and unknown population-level risk factors in Ebola forecasting?
• RQ2 How accurately can a stage-structured EVD model forecast epidemic dynamics in Guinea, Liberia, and Sierra Leone using early epidemic data?
• RQ3 To what extent are transmission parameters for early, late, and funeral stages identifiable from incidence and death data alone?
• RQ4 Does the inclusion of a correction term improve forecast performance compared to models without such adjustments?
• RQ5 How do parameter unidentifiability and data sparsity affect the reliability of intervention strategy recommendations based on modeling?

Key findings

• The model with the correction term successfully forecasted the epidemic trajectory for all three countries and their combined data, correctly predicting Liberia’s slowdown and Sierra Leone’s continued exponential growth through November 2014.
• Despite high uncertainty, the best-fit models using the correction term produced forecasts that closely matched observed data, demonstrating the method’s robustness in data-sparse settings.
• Parameter unidentifiability was a major challenge: multiple combinations of transmission parameters across stages produced similar fits, making it difficult to estimate the relative contribution of each stage to transmission.
• The correction term was identifiable and played a key role in enabling accurate forecasting, even when underlying transmission parameters were not precisely estimable.
• The model correctly captured the divergence in epidemic trends—Liberia’s decline versus Sierra Leone’s sustained growth—highlighting the value of the correction term in handling heterogeneous surveillance and reporting conditions.
• The study underscores the importance of collecting additional data on behavior, contact patterns, and burial practices to improve parameter identifiability and inform targeted interventions.