Bayesian hidden Markov process model of Salmonella over three generations of flocks in broiler production chain
Jukka Ranta
Risk Assessment Unit, Finnish Food Safety Authority Evira
*Pirkko Tuominen
Risk Assessment Unit, Finnish Food Safety Authority Evira *Antti Mikkelä
Risk Assessment Unit, Finnish Food Safety Authority Evira *Helene Wahlström
The Swedish Zoonosis Center, National Veterinary Institute Full text:
PDF
Last modified: June 11, 2009
Abstract
A bayesian primary production chain model of salmonella in broiler production was developed in 2001 for a Quantitative Risk Assessment [1], QRA. This original discrete time Markov chain model over three generations of flocks (grandparent, parent, and production flocks) was based on data provided by the Finnish Salmonella Control Programme, FSCP. The discrete time stochastic model described the status (true infection state) of each flock at regular time intervals corresponding to the testing times of the FSCP. Although detailed flock specific data were not available, it was possible to use summary data together with auxiliary information about the testing times (ages), number of flock holdings, and the fact that with breeding flocks a positive test result leads to destruction of the flock. Also, informative priors on parameters such as test sensitivity had to be used in a fully Bayesian approach. Based on all available information it was possible to construct a data set in a more detailed manner, regarding life histories, than with the summary data alone. Otherwise, as such, the summary data would be very difficult to use towards estimation of meaningful epidemiological quantities. The model structures closely corresponded to the structure of the production chain, combining models of horizontal infection and vertical infection between flock generations. This leads to a fairly complicated hierarchical Bayesian model with latent variables, accounting for imperfect testing (sensitivity<100%). However, if the number of flocks and/or the number of testing times increase or become irregular, it becomes tedious to accommodate the model to each new situation and it also seriously slows down the computation. Therefore, a continuous time twostate Markov process model was developed for (1) accommodating irregular testing times and hopefully (2) reducing the computing time required. This new model could therefore be more easily applied to different data sets from similar production systems. However, the infection probabilities at testing times require solving a recursive formula which is needed for the likelihood expression as well as in the formulation of vertical transmission probability. Data from Swedish Salmonella Control Programme, SSCP, were used to calculate prevalence estimates for Swedish broiler production chain, based on data from 2007 concerning three generations of flocks: grandparents, parents, and production flocks. Also FSCP data will be analysed. The same model structures apply to egg production chain as well, and could even be applied to other production chains with similar structures. Some results and challenges in Quantitative Microbial Risk Assessment, QMRA, are discussed. The model is implemented in OpenBUGS.
[1] Ranta J., Maijala R. A Probabilistic Transmission Model of Salmonella in the Primary Broiler Production
Chain. Risk Analysis, Vol. 22, No 1, 2002. 4758.


Learn more
about this
publishing
project...

