Microbial risk assessment is emerging as a new discipline in risk assessment. A systematic approach to microbial risk assessment is presented that employs data analysis for developing parsimonious models and accounts formally for the variability and uncertainty of model inputs using analysis of variance and Monte Carlo simulation. The purpose of the paper is to raise and examine issues in conducting microbial risk assessments. The enteric pathogen Escherichia coli O157:H7 was selected as an example for this study due to its significance to public health. The framework for our work is consistent with the risk assessment components described by the National Research Council in 1983 (hazard identification; exposure assessment; dose-response assessment; and risk characterization). Exposure assessment focuses on hamburgers, cooked a range of temperatures from rare to well done, the latter typical for fast food restaurants. Features of the model include predictive microbiology components that account for random stochastic growth and death of organisms in hamburger. For dose-response modeling, Shigella data from human feeding studies were used as a surrogate for E. coli O157:H7. Risks were calculated using a threshold model and an alternative nonthreshold model. The 95% probability intervals for risk of illness for product cooked to a given internal temperature spanned five orders of magnitude for these models. The existence of even a small threshold has a dramatic impact on the estimated risk.

No abstract

The abilities of cells of a particular type of bacteria to leave lag phase and begin the process of dividing or surviving heat treatment can depend on the serotypes or strains of the bacteria. This article reports an investigation of serotype-specific differences in growth and heat resistance kinetics of clinical and food isolates of Salmonella. Growth kinetics at 19 degrees C and 37 degrees C were examined in brain heart infusion broth and heat resistance kinetics for 60 degrees C were examined in beef gravy using a submerged coil heating apparatus. Estimates of the parameters of the growth curves suggests a small between-serotype variance of the growth kinetics. However, for inactivation, the results suggest a significant between-serotype effect on the asymptotic D-values, with an estimated between-serotype CV of about 20%. In microbial risk assessment, predictive microbiology is used to estimate growth and inactivation of pathogens. Often the data used for estimating the growth or inactivation kinetics are based on measurements on a cocktail--a mixture of approximately equal proportions of several serotypes or strains of the pathogen being studied. The expected growth or inactivation rates derived from data using cocktails are biased, reflecting the characteristics of the fastest growing or most heat resistant serotype of the cocktail. In this article, an adjustment to decrease this possible bias in a risk assessment is offered. The article also presents discussion of the effect on estimating growth when stochastic assumptions are incorporated in the model. In particular, equations describing the variation of relative growth are derived, accounting for the stochastic variations of the division of cells. For small numbers of cells, the expected value of the relative growth is not an appropriate "representative" value for actual relative growths that might occur.

Great uncertainty exists in conducting dose-response assessment for microbial pathogens. The data to support quantitative modeling of dose-response relationships are meager. Our philosophy in developing methodology to conduct microbial risk assessments has been to rely on data analysis and formal inferencing from the available data in constructing dose-response and exposure models. The probability of illness is a complex function of factors associated with the disease triangle: the host, the pathogen, and the environment including the food vehicle and indigenous microbial competitors. The epidemiological triangle and interactions between the components of the triangle are used to illustrate key issues in dose-response modeling that impact the estimation of risk and attendant uncertainty. Distinguishing between uncertainty (what is unknown) and variability (heterogeneity) is crucial in risk assessment. Uncertainty includes components that are associated with (i) parameter estimation for a given assumed model, and (ii) the unknown "true" model form among many plausible alternatives such as the exponential, Beta-Poisson, probit, logistic, and Gompertz. Uncertainty may be grossly understated if plausible alternative models are not tested in the analysis. Examples are presented of the impact of variability and uncertainty on species, strain, or serotype of microbial pathogens; variability in human response to administered doses of pathogens; and effects of threshold and nonthreshold models. Some discussion of the usefulness and limitations of epidemiological data is presented. Criteria for development of surrogate dose-response models are proposed for pathogens for which human data are lacking. Alternative dose-response models which consider biological plausibility are presented for predicting the probability of illness.

No abstract