Print PageAn infectious cancer, Tasmanian devil facial tumor disease, is threatening to cause the extinction of the largest surviving marsupial carnivore, the Tasmanian devil Sarcophilus harrisii. In the absence of any effective treatment or vaccine and few indications that any animals may be resistant, one of the few strategies possible to maintain wild populations is to cull infected animals. A trial of the strategy on a semi-isolated peninsula in south-eastern Tasmania did not clearly reduce the force of infection or prevent population decline. Here, we use models to investigate whether other removal strategies may have been more effective. Using an empirical contact network derived from proximity sensing radio collars, we explored whether the observed heterogeneity in contacts substantially altered predictions based on mean field models. We then used age structured models to investigate a range of culling strategies.
Results/Conclusions
The observed contact network had an aggregated degree distribution. However, no age or sex classes could be identified as being especially highly connected, suggesting that removing "superspreaders" is not a viable strategy. An individual-based model showed that, whilst the probability of devil extinction was somewhat lower in models with realistic network structure than in their mean field equivalents, the predicted time until extinction was shorter in the network models. These differences were relatively minor, suggesting that a mean field approximation is adequate. Age structured models based on mean field transmission dynamics showed that a very high removal rate of infected animals was necessary to prevent devil extinction. Removal also needed to be continual or at least monthly, rather than the quarterly strategy employed in the field trial. Only a very narrow range of removal rates permitted the coexistence of the devil and the tumour. Our results have general implications for using culling to manage wildlife disease. First, it is important to test the mean field assumption. Second, frequency dependent transmission makes culling less effective than if transmission is density dependent, but high rates of culling of infected individuals still can eliminate disease. Third, we explored a range of ways of handling time delays and ageing, including distributed delays modeled using a Gamma function, time delayed differential equations and exponentially distributed delays. The way in which time delays were modeled had a major effect on model outcomes, cautioning that it is important to use a time delay representation appropriate to the specific problem in order to obtain useful results.
