Simple models for complicated situations: Ebola in West Africa

Sarah Rainey, a Biology major at Radford University, studied structural uncertainty in disease models with Dr. John Drake.


Abstract: Mathematical models are an idealization used for disease forecasting and decision making. There is a tradeoff between timeliness and detail for models produced in response to an epidemic. It is important to quantify the level of uncertainty in a model to mitigate inaccuracies that could arise from parameter uncertainty and structural uncertainty. The impact of parameter uncertainty has been heavily investigated; however structural uncertainty is scarcely addressed. The 2014 Ebola outbreak in West Africa was used as a case study to investigate structural uncertainty. Models produced early in the course of an epidemic tend to be simpler due to a lack of information with which to estimate parameters and an abbreviated development time. Complex models are more cumbersome to develop but take into account more complex transmission scenarios reported by in-country observers. The goal of this project is to determine when a simple model can capture the trajectory of an Ebola outbreak generated by a more complex transmission process. The statistical software R was used to solve three mathematical models: a branching-process model (Drake et al., 2015), a modified SEIR model (Legrand et al., 2007), and the classical SIR model (Kermack and McKendrick, 1927). During the 2014 Ebola outbreak in West Africa, the model produced by Legrand et al. was widely used because it included all of the key components to model an Ebola epidemic simplistically. The model published by Drake et al. (2015) included greater detail about contact patterns, interventions, behavior change, and other features that may be necessary to realistically model the Ebola outbreak. The complex branching-process model published by Drake et al. (2015) was used to generate an Ebola epidemic with different scenarios by varying parameter values. The two simpler models were fit to simulated data. The correlation coefficient was calculated to test the fit of the models to observe how well they were able to capture the trajectory of the outbreak. Our findings conclude that the modified SEIR model published by Legrand et al. (2007) was superior to the classical SIR model in representing the disease trajectory of multiple Ebola outbreaks simulated with unique circumstances.

Download (PDF, 3.02MB)