Growth Estimation Part MODEL 1. Overview of the E. coli O157:H7 Food Safety

Two-Dimensional Probabilistic Risk Assessment Model 1513 E. coli O157:H7 organisms are estimated to be re- sponsible for some 73,500 cases of infection, 2,150 hospitalizations, and 61 deaths in the United States each year. 27 The model has a two-dimensional prob- abilistic framework and typical characteristics of MFSPR models. 2. MODEL 2.1. Overview of the E. coli O157:H7 Food Safety Process Risk Model The U.S. Department of Agriculture conducted a farm-to-table risk assessment to evaluate the pub- lic health impact from E. coli O157:H7 in ground beef. 8 The E. coli O157:H7 risk assessment includes hazard identification, exposure assessment, hazard characterization, and risk characterization steps. The exposure assessment step consists of three major modules: 1 production; 2 slaughter; and 3 prepa- ration. The preparation module consists of three parts: 1 growth estimation; 2 cooking effect; and 3 serving contamination. The exposure assessment is based upon a probabilistic approach for model- ing the prevalence and the concentration of the E. coli pathogen in live cattle, carcasses, beef trim, and a single serving of cooked ground beef. The model is implemented in Microsoft Excel using Risk as an add-in for defining probability distributions for inputs. The E. coli model has a modular framework. Thus, each key activity is modeled in a separate mod- ule. For example, there are separate modules for estimation of infection of live cattle on farms and contamination of meat trimmings in slaughter plants. Each module may have inputs that are calculated based upon a predecessor module and its outputs may be input to a successor module. In some cases, the out- put of a module is binned into predefined intervals prior to transferring to successor module. For exam- ple, in the slaughter module the contaminant concen- tration in beef trimings is estimated as a continuous output, but subsequently is binned by 0.5 log 10 incre- ments from 0 to 8 log 10 . 8 Each log 10 increment in contamination represents a 10-times increase in the contamination level. Because of binning, there is a loss of one-to-one correspondence between outputs of one module and the inputs of its predecessors. Therefore, binning is an obstacle to global sensitiv- ity analysis of the entire model. Creating one-to-one correspondence in this case would require substan- tial restructuring and recoding of the model. Thus, ANOVA is applied to the individual modules.

2.2. Growth Estimation Part

The growth estimation part in the preparation module is selected for the case studies Fig. 1. This part has a two-dimensional probabilistic simulation framework. Storage temperature and storage time at retail Stage 1, transportation Stage 2, and home Stage 3 influence the E. coli O157:H7 levels in ground beef servings. The output is the growth of E. coli O157:H7 organisms in ground beef servings prior to cooking. 8 The log of growth for each stage is a function of storage temperature T, lag period duration LP, and generation time GT. LP is the time that organ- isms take to become acclimated to a new environment before starting to multiply. GT is the interval from one point in the division cycle of an organism to the same point in the cycle one division later. The log of growth in Stage ‘i’ is calculated as: G i = log 10 2 LPi −Ti GTi . 1 Growth is calculated in each stage only if stor- age temperature is greater than 45 ◦ F, and hence a set of eight potential growth combinations that can occur in Stage 1 through 3 for a single ground beef serving is considered based on the probability that each serving is kept at a temperature above 45 ◦ F at each stage. The lag period in Stages 2 and 3 are ad- justed based on the cumulative lag used in previous stages. At each stage, calculated growth is compared with the theoretic maximum density at refrigeration temperature and the smaller value is considered as the possible growth. Maximum density MD is the maximum possible number of E. coli O157:H7 organ- isms at a specific storage temperature on a serving. More explanation of the modeling approach is given in FSIS. 8 Table I summarizes 13 inputs, including their mean values and 95 probability ranges based on the sampled values of inputs in a comingled anal- ysis of variability and uncertainty. Comingled anal- ysis of variability and uncertainty is explained in Section 3.4.2. Table I also gives the type of variability and uncertainty distributions for each input. 8 Uncer- tainty is incorporated in the parameters of the vari- ability distributions. 1514 Mokhtari and Frey Fig. 1. Schematic diagram of the growth estimation part of the farm-to-table risk assessment model for E. coli O157:H7 in ground beef. 3. MATERIALS AND METHODS 3.1. Overview of ANOVA