In many countries epidemiological models are used as tools to inform preparedness and response plans for potential disease outbreaks. Policy decisions about epidemics can have substantial impacts on economics and public health. When there are multiple quality models for any given disease that provide differing and sometimes inconsistent results it can be difficult for policy makers to select a single model on which to base high-stakes decisions. Ensemble modeling methods provide a standardized and transparent way of producing a single, interpretable projection from multiple model outputs. This methodology has seldom been applied to epidemiology though it is frequently used in weather forecasting and climate change predictions and has increased the accuracy of predictions in those fields. The Bayesian Reliability Ensemble Average (BREA) method has recently been adapted for use in epidemiology and provides a powerful framework for developing ensemble methods for outbreak response and planning. Here we discuss an exploration of the BREA ensemble method as a potential tool for outbreak response, using foot and mouth disease as a case study. Foot and mouth disease (FMD) outbreaks have serious consequences both in agriculture and economics for the infected country. Therefore, many FMD-free countries have invested in models that can help create robust FMD response and preparedness plans. This study is part of an international collaboration that includes six FMD models that are frequently used in policy decisions around the world. Using these models, and data from the initial weeks of the 2001 United Kingdom FMD outbreak we explored whether the BREA method could improve the accuracy of the model predictions early in an epidemic before the outcome of the outbreak is known.
Results/Conclusions
We demonstrate that the BREA method is capable of capturing the observed outbreak data and performs better than any single model alone. Our preliminary results show that this holds even when the models are provided with only the first two weeks of data, which is important because outbreak data are often limited. These results suggest that the BREA method could be a powerful tool for epidemiological applications. Additionally, ensemble models produce present a single, interpretable prediction and therefore, can reduce the confusion caused by multiple, sometimes inconsistent predictions about the same outbreak. Ensemble modeling therefore has the potential to improve our ability to make epidemiological forecasts both for preparedness plans and outbreak response, which would be highly beneficial for animal and public health worldwide.