Bluetongue model for forward prediction of disease spread

The novel model retains the strongly data-driven elements of the baseline model (Szmaragd et al., 2009), representing the on-farm transmission of bluetongue, and in particular the temperature dependence of the relevant midge infectivity and mortality. However, the following novel aspects are added:

• A flexible (directed) network structure of between-farm transmission probabilities

• A long-range transmission parameter describing the probability of transmission between farms regardless of their spatial locations

• A non-homogenous spatial process describing a distribution of midge population density and biting rates at a spatial scale relevant to the local transmission of bluetongue between farms.

The most substantial part of the bluetongue case study work involved developing a new modelling framework, within which the highly complex baseline model and the derived novel model could be implemented within a simpler and more flexible user interface. The development of the proposed framework was influenced by the following requirements:

1. Usability. The interface to the model must be as simple as possible, and preferably implemented using R, which is a stated preference for EFSA.

2. Flexibility. The model must be capable of accepting different inputs and running using different assumptions, so that models of bluetongue spread can be run for different countries and/or time-points, rather than being limited to the autumn 2007 UK simulation. This is to ensure that the model meets the future needs of EFSA.

3. Modifiability. It must be possible to modify elements of the model (such as the between-farm transmission model) in isolation, so that the required novel aspects of the model can be added within the same framework as the baseline model. It would be highly desirable for the different aspects of the model to be self-contained, and with framework implementation code separated from the code modelling the biological system, so that individual aspects of the model are more amenable to future modification by EFSA without the need to understand the implementation of the framework itself.

4. Portability. The model must be installable and runnable on a variety of different systems, preferably without a requirement for commercial software. This is to ensure that the model is accessible to EFSA in the future.

5. Speed. The model is mechanistic and highly complex, and must be capable of simulating the spread of disease throughout the entire UK livestock population of over 120,000 farms. The computationally intensive elements of the model must therefore be implemented in a compiled language.

6. Robustness. Programming errors are a particularly relevant problem for these types of highly complex mathematical models that lack a method to assess the statistical fit or adequacy of the model predictions. The model code must therefore make use of frequent logic checks to guard against coding errors, whilst still satisfying the requirement for speed.

An additional objective of the novel model is to assess the importance of the precise temperature and livestock demographic data, in order to gauge the importance of these data sources for modelling future outbreaks. As a specific aspect of requirement 2 (as given above), the temperature, farm locations and cattle and sheep populations must all be easily modifiable so that the impact of changing these can be assessed.