The influence of empirical contact networks on modelling diseases in cattle

We present two stochastic models of the passage of an SEIR (susceptible-latent-infected-resistant) disease through herds of cattle. One model is based on a contact network constructed via continuously recorded interaction data from two herds of cattle, the other, a matching network constructed using the principles of mass-action mixing. The recorded contact data were produced by attaching proximity data loggers to two separate herds of cattle during two separate recording periods. The network constructed using the principles of mass-action mixing uses the same number of contacts as the recorded network but distributes them randomly amongst the animals. The recorded networks had a greater number of repeated contacts, lower closeness and clustering scores and greater average path length than the mass-action networks. A lower proportion of simulations of the recorded network produce any disease spread when compared to those simulations of the mass-action network and, of those that did, fewer infected animals were predicted. For all parameter values tested, within the sensitivity analysis, similar differences were found between the recorded and mass-action network models.