Instinctively, I always want more science and better information. The newly published paper by Brooks-Pollock and her co-authors1 fits an interesting statistical model to data about the occurrence and detection of bovine tuberculosis (bTB) in Great Britain. It presents a mechanistic model of the disease cycle and how it might work and then it challenges this model with the data using Bayesian fitting methods. Fitting the model to the data is a sophisticated process.
In broad terms, the model suggests that the control of the disease is most sensitive to the severity of the control measures applied to cattle and least sensitive to those applied to environmental sources of infection. These “environmental” sources equate to a broad range of factors including badgers, other wildlife species and bacteria that might be present on pasture or elsewhere.
On the face of it, this is not a very profound outcome especially when one considers where the uncertainties might lie within the model. The model itself is provided with most information about cattle-cattle transmission of bTB and, while few would doubt that this is the most important route of infection in cattle, it is not a surprise that if one was to intervene very hard to eliminate the disease by eliminating cattle then the problem (i.e. the control of TB in cattle) converges on a solution. Clearly, if the end point is zero cattle, which could be an outcome of such an approach, then the problem has been solved. It did not need a sophisticated model to tell us this.
In addition, it is important to understand both the strengths and weaknesses of these kinds of models. All models are wrong by different degrees and any observer needs to assess the degree of wrongness. In this case, the model focusses on the disease cycle in cattle (i.e. the way the disease is passed between cattle) and parcels all the multifarious sources of disease, including badgers, in to a single, simple concept. So the model “knows” more about cattle than anything else. It is also wise to look closely at how the model was formulated because this is, in effect, full of prior assumptions about how the disease works. Decisions by the researchers about which fitting procedure to use and how to judge the model outputs statistically can all affect the outcome. Some of these can be more justified than others. For example the decision to represent badgers and all other sources of environmental infection as a single process was based on lack of data. This is fair enough, but such a decision was not tested against how well the model might perform if the data or knowledge for the cattle component of the disease cycle was similarly degraded. Decisions in the construction of these kinds of models, many of which are never fully challenged by examining the counterfactual, can reveal a lot about the prior prejudices of those constructing the models.
So, where does this study take us? What I am trying to square in my mind is the mixture of epidemiological evidence about bTB, some of which is aligned with this study and some not. Putting this very simply, in general, statistical fitting to data as applied in this study gives a broadly consistent picture (cattle-cattle infection is stronger than badger-cattle infection). However, does this mean that badger-cattle infection is unimportant or is it just left relatively unseen in all the noise of uncertain data, model simplification and (possible) selection bias by those carrying out the study? There are other data that suggest another story, such as the genetics of bTB that show an important epidemiological link between badgers and cattle. Data from the pathology and frequency of the disease in badgers combined with a rich knowledge of badger social structure and experimental infection of badgers all support the view that badgers sustain disease at some levels without reinfection from cattle.
A key issue is whether badgers are able to sustain disease without the presence of cattle. This paper suggests not but I am not so sure we have evidence strong enough to support such an inference from a study that made so little effort to model the badger component of the disease cycle.
How should those who shoulder the responsibility for decision-making react to a paper like this? My advice would be that one needs to take great care. The interventions suggested here would affect the lives, careers and livelihoods of many thousands of people and acting on them would require a much higher level of certainty than are present in the paper.
The major science problem in this field is to produce credible management interventions – describe solution, not problems. To be credible, these interventions need to be believable by most of those who are affected. The control of bTB in the UK is not, fundamentally, an epidemiological problem. We actually probably already know enough about bTB epidemiology to control the disease using standard disease control methods. Rather, it is a problem associated with the way in which the costs of different management interventions are divided between different stakeholders.
We need to understand how these interventions might be implemented in ways that are accepted and this is a social science problem. The Brooks-Pollock model does not provide a good fit to the social process and/or the realities of the real world because the epidemiological solution to bTB, including badger control, is only a part of this much larger social problem. Scientists need to have a sharper awareness of where the problem lies if they are to provide workable solutions.
- Brooks-Pollock, E., Roberts, G.O. & Keeling, M.J. (2014) A dynamic model of bovine tuberculosis spread and control in Great Britain. Nature doi:10.1038/nature13529