Biotracing microbial contaminants in food chains - Salmonella on pork

Abstract

Biotraceability is defined as the ability to use down stream information to point to materials, processes or actions within a particular food chain that can be identified as the source of undesirable agents. We propose a practical framework in which a Monte Carlo model, where sequential events in the chain are simulated, is converted into a Bayesian Belief Network (BBN). A BBN represents the full joint probability of the model which allows interaction between the data and the model parameters and, hence, inferential queries. This makes a BBN a convenient structure for answering biotraceability questions. We discuss a model for biotracing the sources of Salmonella contamination for individual positive carcasses at the re-work station. The model is a BBN, in which Salmonella concentration information at different locations in the slaughterhouse are used in combination with prior knowledge about the dynamics of Salmonella through the slaughter line. Microbial samples were taken in a Dutch slaughterhouse from 118 carcasses directly after exsanguination, from the same carcasses after meat inspection and from the carcass splitter and belly opener. Prevalence and concentration data were used to specify prior beliefs about the model inputs and to iteratively refine the distributions of the parameters in the model so that it optimally describes that specific slaughterhouse. The model results indicate that house flora on or in the carcass splitter was the causative source of contamination for many carcasses, especially for those that carried contamination on the cutting side. Furthermore, the model indicates that the parameter values of the model may be subject to temporal variation and can be used as a tool to provide estimates of such trends. Sensitivity analysis of the model showed that the results of the model were sensitive to the true values of the parameters describing the transfer of Salmonella from the equipment, which could only be estimated with limited accuracy. More realistic estimates of the cross-contamination rates were obtained by a laboratory experiment in which the transfers of Salmonella from contaminated pork meat to a butcher’s knife and from the knife back to pork meat were determined. Recovery efficiency of bacteria from both surfaces was also determined and accounted for in the analysis. A BBN was proposed which allows the combination of uncertainty within one experiment and variability over multiple experiments. The transfer ratio probability distributions were shown to have a large variability, with a mean value of 0.19 for the transfer of Salmonella from pork to the knife and 0.58 for the transfer of Salmonella from the knife to pork. The results of the biotracing model were compared to the results of an epidemiological approach in which Salmonella serotyping data from our sampling experiment were used. The results of both approaches were similar for most Salmonella-positive carcasses at the re-work station, yet, some some differences were observed. The biotracing model can be seen as a step in the process towards an operational biotracing system, by which a stakeholder can initiate immediate responses to hazards imposed in the pork slaughterhouse.