Simulation modeling for microbial risk assessment.

Quantitative microbial risk assessment implies an estimation of the probability and impact of adverse health outcomes due to microbial hazards. In the case of food safety, the probability of human illness is a complex function of the variability of many parameters that influence the microbial environment, from the production to the consumption of a food. The analytical integration required to estimate the probability of foodborne illness is intractable in all but the simplest of models. Monte Carlo simulation is an alternative to computing analytical solutions. In some cases, a risk assessment may be commissioned to serve a larger purpose than simply the estimation of risk. A Monte Carlo simulation can provide insights into complex processes that are invaluable, and otherwise unavailable, to those charged with the task of risk management. Using examples from a farm-to-fork model of the fate of Escherichia coli O157:H7 in ground beef hamburgers, this paper describes specifically how such goals as research prioritization, risk-based characterization of control points, and risk-based comparison of intervention strategies can be objectively achieved using Monte Carlo simulation.

Quantitative risk assessment for Escherichia coli O157:H7 in ground beef hamburgers.

Quantitative Risk Assessment (QRA) is a methodology used to organize and analyze scientific information to estimate the probability and severity of an adverse event. Applied to microbial food safety, the methodology can also help to identify those stages in the manufacture, distribution, handling, and consumption of foods that contribute to an increased risk of foodborne illness, and help focus resources and efforts to most effectively reduce the risk of foodborne pathogens. The term Process Risk Model (PRM) is introduced in this paper to describe the integration and application of QRA methodology with scenario analysis and predictive microbiology to provide an objective assessment of the hygienic characteristics of a manufacturing process. The methodology was applied to model the human health risk associated with Escherichia coli O157:H7 in ground beef hamburgers. The PRM incorporated two mathematical submodels; the first was intended to described the behaviour of the pathogen from the production of the food through processing, handling, and consumption to predict human exposure. The exposure estimate was then used as input to a dose-response model to estimate the health risk associated with consuming food from the process. Monte Carlo simulation was used to assess the effect of the uncertainty and variability in the model parameters on the predicted human health risk. The model predicted a probability of Hemolytic Uremic Syndrome of 3.7 x 10(-6) and a probability of mortality of 1.9 x 10(-7) per meal for the very young. These estimates are likely high for all hamburger meals, but may be reasonable for the home-prepared hamburgers described by this model. The efficacy of three risk mitigation strategies were evaluated by modifying the values of the predictive factors and comparing the new predicted risk. The average probability of illness was predicted to be reduced by 80% under a hypothetical mitigation strategy directed at reducing microbial growth during retail storage through a reduction in storage temperature. This strategy was predicted to be more effective than a hypothetical intervention which estimated a plausible reduction in the concentration of E. coli O157:H7 in the feces of cattle shedding the pathogen and one aimed at convincing consumers to cook hamburgers more thoroughly. The conclusions of this approach are only accurate to the extent that the model accurately represents the process. Currently, uncertainty and ignorance about the hygienic effects of the individual operations during production, processing, and handling limit the applicability of a PRM to specify HACCP criteria in a quantitative manner. However, with continuous improvement through stimulated research, a PRM should encompass all available information about the process, food, and pathogen and should be the most appropriate decision-support tool since it represents current knowledge.

Quantitative risk assessment of human listeriosis from consumption of soft cheese made from raw milk.

Microbial hazards have been identified in soft cheese made from raw milk. Quantification of the resulting risk for public health was attempted within the frame of the Codex Alimentarius Commission, 1995 approach to quantitative risk assessment, using Monte Carlo simulation software. Quantitative data could only be found for Listeria monocytogenes. The complete process of cheese making was modeled, from milking to consumption. Using data published on the different sources of milk contamination (environment and mastitis) and bacterial growth, distributions were assumed for parameters of the model. Equations of Farber, J.M., Ross, W.H., Harwing, J. (1996) for general and at-risk populations were used to link the ingested dose of L. monocytogenes to the occurrence of listeriosis. The probability of milk contamination was estimated to be 67% with concentration ranging from 0 to 33 CFU ml-1. The percentage of cheese with a predicted concentration of L. monocytogenes greater than 100 CFU g-1 was low (1.4%). The probability of consuming a contaminated cheese serving was 65.3%. Individual annual cumulative risk of listeriosis, in a population each consuming 50 servings of 31 g, ranged from 1.97 x 10(-9) to 6.4 x 10(-8) in a low-risk sub-population and 1.04 10(-6) to 7.19 10(-5) in a high-risk sub-population. The average number of expected cases of listeriosis per year was 57 for a high-risk sub-population and one for a low-risk healthy sub-population. When the frequency of environmental milk contamination was reduced in the model and L. monocytogenes mastitis was eliminated, the expected incidence of listeriosis decreased substantially; the average number of expected cases was reduced by a factor of 5. Thus the usefulness of simulation to demonstrate the efficiency of various management options could be demonstrated, even if results should be interpreted with care (as many assumptions had to be made on data and their distributions.