Foot and mouth disease revisited: re-analysis using Bayesian spatial susceptible-infectious-removed models

The foot and mouth disease (FMD) outbreak in the United Kingdom in 2001 was modeled via the use of Bayesian spatial susceptible-infected-removed (SIR) models. In these models the underlying mean of the incident cases was modeled spatially and in time. Dependence structures at the parish level between previous and current cases were modeled either with individual dependence or with neighborhood dependencies. Additional confounding was modeled via random effects that can have either uncorrelated or spatially correlated prior distributions. The best models found relied on lagged population and infection count within the same parish but neighborhood lagged dependencies overall did not provide a good fit. Models with only a space-time interaction effect were preferred over more complex models. The estimation of 'decline' markers for different areas was considered via difference operators as posterior functionals. These proved to be useful in giving an early indication of the waning phase of the epidemic locally.