Epidemiological inference for partially observed epidemics: The example of the 2001 foot and mouth epidemic in Great Britain

Abstract

This paper develops a statistical framework for a retrospective analysis for well-observed livestock epidemics during which intervention policies may conceal cases, thus potentially biasing naively derived parameter and final size estimates. We apply the methods to the 2001 foot and mouth epidemic (FMD) in Great Britain, during which a large number of farms (about 7500) were pre-emptively culled as part of the control effort without ever being diagnosed as being infected. We infer farm-level infectivity and susceptibility parameters, a distribution for the delay from infection to report, together with a time varying farm infectivity profile for farms. Hidden infections among proactively culled farms were accounted for using a data augmentation approach utilising reversible jump MCMC methods. Simulated epidemics derived using the parameter estimates obtained reproduced the 2001 epidemic well. Our analysis demonstrates that time-varying infectivity profiles fit the 2001 data better than naive assumptions of constant infectiousness. We estimate that around 210 (or 2.8%) of the farms proactively culled in the 2001 epidemic were infected. However, for the parameter estimated obtained, preliminary simulation results indicate that had contiguous culling not been applied in 2001, the epidemic might have been substantially larger.

New model diagnostics for spatio-temporal systems in epidemiology and ecology

“…Predicted dynamics of epidemiological and ecological systems can be extremely sensitive to the choice of model, with consequent implications for the design of control strategies [1,[5][6][7]. For example, choices of temporal distributions influence both epidemic final size and the persistence of disease, and have important implications for devising effective control strategies which target symptomatic subjects and the timing of infectiousness [6,12,13].…”

“…We refer to the residualsr 1 so that individual j becomes exposed due to contact with individual i, and p ij ¼ p (s 0 ) . The inclusion of the ordering operation in this process is motivated by our aim of designing tests that may be sensitive to mis-specification of the spatial kernel function in the model equation (2.1).…”

“…To assess our model-testing framework, we fit to the simulated data y a model with the correct structure (Case A), and three further models in which the spatial kernels (Cases B and C) and the latent period distribution (Case D) have been mis-specified, respectively. Specifically, we consider -Case A: Kðd ij , k 1 Þ ¼ expðÀk 1 A Weibull infectious period is fitted in all cases. Uniform priors, which should be constrained to bounded regions to ensure a proper posterior distribution, are specified for all model parameters estimated in the following analyses.…”

A Bayesian Inference Framework to Reconstruct Transmission Trees Using Epidemiological and Genetic Data

“…Several attempts to reconstruct transmission pathways have tried to combine genetic and other epidemiological data, many by adding spatial or temporal information to the process of phylogenetic reconstruction [6]–[11]. However, Jombart et al point out that a “phylogenetic” approach attempts to infer hypothetical common ancestors among the sampled genomes, and may not be appropriate for a set of genomes containing both ancestors and their descendants [12].…”

Estimation of the Infection Window for the 2010/2011 Korean Foot-and-Mouth Disease Outbreak

“…However, not all animals suspected of being infected were tested for the presence of antibodies. Our protocol remains reliable, however, because animals showing the most severe clinical lesions were purposefully chosen for specimen collection to detect the reactor [24].…”