Skip to main content
ARS Home » Research » Publications at this Location » Publication #154128
Title: BAYESIAN SYNTHESIS OF A PATHOGEN GROWTH MODEL

Author
item POWELL, MARK - USDA-OCE
item Tamplin, Mark
item MARKS, BRADLEY - MICHIGAN STATE UNIV.
item CAMPOS, DANILO - MICHIGAN STATE UNIV.

Submitted to: International Association for Food Protection
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/1/2006
Publication Date: 4/1/2006
Citation: Powell, M., Tamplin, M.L., Marks, B., Campos, D.T. 2006. Bayesian synthesis of a pathogen growth model. International Journal of Food Microbiology. 109:34-46.

Interpretive Summary: Bayesian synthesis offers one means of combining information from a variety of sources to develop and evaluate predictive microbiology models for use in risk assessment of microbial pathogens in food. The Bayesian synthesis method was first proposed to characterize uncertainty in mechanistic process models (e.g., population dynamics). The basic approach is to generate a distribution on model inputs and outputs and then to develop a post-model distribution on the inputs and outputs after determining the importance of the model inputs. Next, correlations can be estimated among the model parameters. Using this process, the Bayesian synthesis method updates the uncertainty of models with backward and forward propagations of the model. A key feature of Bayesian synthesis is that it assumes that pre-model information for the inputs and outputs is available from independent information, as are the empirical data used to update the model. The present report supports the Bayesian synthesis method by means of an empirical example, applying a pathogen growth model to data from two studies on the growth of Listeria monocytogenes. The results show that Bayesian synthesis can be an effective method of combining information from different sources to produce and evaluate predictive microbiology models for use in risk assessment of microbial pathogens.

Technical Abstract: A variety of approaches have been proposed to evaluate and compare the performance of predictive microbiology models. Each of these methods presumes that an independent benchmark data set is available. In most risk assessment applications, however, ideal data with which to compare model predictions are unavailable. Bayesian synthesis offers one means of combining information from a variety of sources to develop and evaluate predictive microbiology models for use in microbial pathogen risk assessment. In this report, the Bayesian synthesis method for evaluating new information was illustrated by applying a pathogen growth model to data from two studies on the growth of Listeria monocytogenes. The post-model distribution of the Baranyi growth model estimates for L. monocytogenes growth in broth at 5°C, pH 7 showed that the high correlation between the growth rate and lag time parameters in the Baranyi model were inherent to the model, and not just an empirical result. Parameter uncertainty distributions for the Baranyi model fits to two independent data sets were obtained empirically by bootstrap simulation for 1,000 iterations using the least squares regression procedure. Sensitivity analysis indicated that the posterior output of the Bayesian synthesis was insensitive to the prior on the growth model parameters; however, the posterior marginal distributions of the Baranyi model parameters were sensitive to alternative specifications of the stated prior information. This underscores the importance of reasonably good prior information on the model parameters. In this case, specifying the prior for the parameters of the Baranyi growth model was facilitated by intuitive biological interpretation. It also is notable that the input of food microbiology expert opinion may have underestimated the growth potential for L. monocytogenes at 5°C, pH 7. This highlights the need to use a broad uncertainty distribution about "best estimates" based on expert judgment to avoid overconfidence in the prior. This empirical example indicates that Bayesian synthesis can be a useful method of combining information from a variety of sources regarding the inputs and outputs of the model, in addition to the model itself, to develop and evaluate predictive microbiology models for use in risk assessment of microbial pathogens in food.