Controlling endemic foot-and-mouth disease: Vaccination is more important than movement bans. A simulation study in the Republic of Turkey

Fig. 1. Time and spatial distributions of FMD in Turkey, 2001–2012. The top figure shows the absolute number of recorded ongoing FMD outbreaks by serotype over this period, estimated by taking the recorded start date and end of quarantine date for each outbreak. There are clear epidemics in 2001, 2006, and 2010–12. The bottom figure shows the spatial distribution of these outbreaks, normalised by number of cattle, with each province highlighted according to how many outbreaks per 100k cattle were in each province. Erzurum Province, which we choose to model, is in orange; although it has by far the greatest number of outbreaks, it also has a relatively great population of cattle. Many central provinces are intermediate in outbreaks per 100k cattle, and European Thrace is very low.

Fig. 2. Monthly Cattle Shipments in the Republic of Turkey over the period 2007 to 2012, when data are available. Blue indicates the number of within district shipments, and red the number between districts. There is a clear seasonal pattern of shipments peaking every year for the festival Kurban Bayrami. Within-district shipments make up the clear majority of all cattle shipments recorded.

Fig. 3. Model Schematic. Each farm is treated as an epiunit, and animals within are classified into one of the disease-relevant classes and can proceed through them along the arrows. S, E, I and R, correspond to the basic Susceptible, Exposed, Infectious, and Recovered classes. M indicates Maternally Immune; VS indicates Vaccinated from Susceptible (i.e. no prior immunity), and VR indicates Vaccinated from Recovered (i.e. prior immunity). It is assumed that M, E and I can not be successfully vaccinated. Transition rates between classes are described in Eq. (1), with parameters in Table 1. Between-farm transmission is carried out through a dispersal kernel K(dij) (Eq. (2)) and through animal shipments between farms. Note that for simplicity, arrows for natural birth rates, death rates and sub-classes are not included here.

Fig. 4. Posterior Model Fit. The upper panels show the marginal posterior distributions of the parameters ν (left panel), η (middle panel) and ξ (right panel) after 8 generations of ABC-SMC with 500 particles per generation. Each parameter demonstrates a clearly defined peak, for ν this peak corresponds to a value of 0.5, for η to 2.48, and for ξ to 1.43e-6. Each dotted line indicates this peak for its generation. The lower panel shows a comparison of the observed cumulative outbreak incidence (in red) with the range of the simulated incidence of the final generation (grey). The black line indicates the mean of this distribution, the darker grey area corresponds to the IQR, and the lighter grey the minimum and maximum of the simulated distribution. We see a good fit to the true data.

Fig. 5. Prior and Posterior Dispersal Kernels. Prior and posterior kernels are displayed with their median value as a line and their 95% credible interval as their respective shaded areas. The posterior narrows the kernel credible interval, mainly by reducing the probability of higher probability of transmission.

Fig. 6. Simulated Effect of Different Combinations of Control Policies. The simulated prevalence (number of farms infected) after controls were or were not implemented.To make the main comparisons clearer, only maximum efficacy versions of each policy are included here (i.e. M182 indicates mass vaccination at an interval of 182 days, R10 and S10 a radius of 10 km). Coloured lines indicates mean prevalence for a policy combination. Areas bounded by the same-colour dotted line above and below indicate the IQR of the simulated prevalence.

Fig. 7. Estimated Correlations of Policy Parameters. The estimated correlations between control policy parameters and either total incidence (green), probability of eradication pe (red), or time to eradication TTE (blue). Each panel corresponds to a policy parameter, and each row in each panel is the correlations for a specific combination of control policies indicated using ”—” format (e.g. ”-MS” indicates mass vaccination M and shipment control S are in effect, but not ring vaccination R). Error bars denote the 95% confidence interval. Dotted lines are shown at 0 for ease of analysis.