4.2.5 Preparation and consumption

The preparation and consumption stage is concerned with the end-users of eggs, the manner in which these end-users store and prepare their eggs, and the effectiveness of practices these end-users apply to destroy S. Enteritidis bacteria in prepared meals. Inputs include pooling practices, serving sizes, pathway probabilities, and cooking effectiveness.

This stage considers eggs following their production, distribution and storage. Therefore the number of bacteria within contaminated eggs and the lag period remaining for these eggs are fixed at the beginning of the Preparation and Consumption stage.

The output of this stage is a distribution of the doses of S. Enteritidis bacteria in servings. This distribution may be refined to reflect the frequency of servings that contain various levels of S. Enteritidis bacteria for specific end users (e.g. homes or institutions), or specific meal types (e.g. pooled or non-pooled egg dishes), or specific cooking practices (e.g. raw versus cooked meals).

Given the multiple pathways within the Preparation and Consumption stage, and the dependency of S. Enteritidis amplification and destruction on the pathway modelled, this part of a risk assessment model is likely to be the most complicated. Complexity is expected because each pathway must be modelled separately, and multiple iterations are necessary per pathway to accurately represent the variability of growth.

Egg pooling and serving size

Pooling refers to the practice of breaking eggs into containers and using the combined eggs to make multiple servings of egg dishes or for use in multiple recipes. Pooling is usually done to save time and control portion size. Pooling does not mean simply combining eggs. As an example of pooling, several dozen eggs could be broken into a large bowl and mixed before a restaurant opens for breakfast. Then as orders for scrambled eggs are taken, portions are ladled from the bowl and cooked. In contrast, mixing a dozen eggs into a cake batter would not constitute pooling because the cake could not be made with less than a dozen eggs. Pooling is essentially exposing consumers to more eggs than they ordered.

As a result of pooling, S. Enteritidis bacteria from a single egg are immediately spread to all eggs in the pool, and the bacteria are given immediate opportunity to grow without needing to wait for a breakdown in yolk membrane integrity.

The likelihood that eggs are pooled probably differs between home use and institutional use, as is the number of eggs that constitute a pool. Following pooling, there is possible storage before cooking. In addition, the likelihood that eggs are undercooked probably differs between eggs cooked at home versus those cooked at institutions.

When eggs are consumed as single eggs, there is a 1:1 correspondence between contaminated eggs and servings. Using eggs as ingredients results in a greater than unitary correspondence between contaminated eggs and servings.

Data

Data are needed to estimate the fraction of all eggs consumed in the home versus institutional settings. Similarly, data are needed that describe the fraction of eggs consumed in pooled dishes, the fraction of meals consisting only of eggs or where eggs are used as ingredients, and the fraction of meals undercooked. These probabilities can be considered fixed in the model; they do not vary but they are uncertain. Additional data are needed to describe how eggs are handled after they are pooled in homes and institutions. These inputs will have both variability and uncertainty associated with them.

Unfortunately, for both the Health Canada and US SE RA exposure assessments, little data were available to estimate these inputs. Therefore distributions were defined based on the opinions of the analysts, with comments from reviewers of the models.

The assumed probability of pooling and number of eggs per pool for the two risk assessments are summarized in Table 4.12. Generally, the @RISKPERT(minimum, most likely, maximum) distribution was used. This distribution is an alternative to the more traditional triangular distribution. The PERT distribution has a smooth shape that assigns less probability weight to the distribution tails than does the triangular distribution. Nevertheless, the value of both these distributions is that the user can define the most likely value and absolute minimum and maximum values, based on opinion. The uniform distribution is another alternative that is less informed. That distribution only requires the user to define minimum and maximum values.

Table 4.12. Pooling inputs used in two exposure assessments.

When eggs are used as ingredients in the Health Canada exposure assessment, the number of servings generated is the same as the number of servings when eggs are pooled. In contrast, the US SE RA analysed a computerized recipe program to determine the number of servings when eggs are used as ingredients in the home. The distribution ranged from 2 to 10 servings, with a mean of 6 servings per egg. In the case of eggs used as ingredients in institutions, the distribution used for pooled servings was doubled.

Figure 4.29. Modelled size of egg pools in homes for Health Canada and US SE RA exposure assessments. Distribution assumptions are as shown in Table 4.12.

Figure 4.30. Modelled size of egg pools in institutions for Health Canada and US SE RA exposure assessments. Distribution assumptions are as shown in Table 4.12.

Methods

Both the Health Canada and US SE RA exposure assessments consider pooling of eggs. Nevertheless, there is considerable disparity in how pooling is modelled and in the subsequent results. In institutions, the Health Canada assessment models larger pools than does the US SE RA. In the home, the pools are larger for the US SE RA than the Health Canada assessment. Figures 4.29 and 4.30 show the results of modelling the pool sizes for homes and institutions for the two exposure assessments.

Furthermore, the Health Canada exposure assessment assumes a single destination for pooled eggs - scrambling. Scrambling eggs is highly effective at destroying bacteria in the Health Canada model. The US SE RA allows pooled eggs to be served as egg meals or incorporated as ingredients in recipes. The distinction is important because of the post-pooling storage that is explicitly modelled in the US SE RA. This post-pooling storage gives eggs an immediate opportunity to grow without needing to wait for a breakdown in yolk membrane integrity. In the Health Canada model, pooling has no effect on the number of S. Enteritidis bacteria in a serving. Nevertheless, pooling does increase the likelihood of illness from a single egg because there are more exposures to the bacteria. In the US SE RA model, the number of S. Enteritidis bacteria in a serving is decreased as the pool size increases, but the model also simulates post-pooling growth and assumes that bacteria are able to grow immediately after pooling.

The attributable risk in the modelled output of the two risk assessments differs significantly. Though pooled eggs account for 17.7% of all servings in the Health Canada exposure assessment, only 6% of the risk of S. Enteritidis illness comes from pooled eggs. In contrast, pooled eggs in the US SE RA account for 13.1% of all servings while contributing to 26.8% of the illnesses (Table 4.13).

Table 4.13. Percentage of illnesses attributed to pooled eggs and the proportion of pooled eggs

Pathway probabilities

The Preparation and Consumption stage considers the effect of end user location. For example, eggs consumed in the home are likely to be handled differently from eggs consumed in restaurants or other food service institutions. It seems likely that eggs stored in the home are exposed to different storage times and temperatures compared with those stored in institutions.

Ideally, one would have access to data generated from studies that sought to chronicle the life of a contaminated egg subsequent to its production. If possible, such studies would report the number of bacteria at the beginning of the egg’s life, then describe the effect of time and temperature as the egg moved from the farm to consumption. Having done this for thousands of eggs, we would theoretically have a very good understanding of how eggs are handled during marketing and preparation. Unfortunately, no such data exist. Therefore, understanding the effect of preparation and consumption on S. Enteritidis in eggs requires considering the limited evidence and dividing the problem into elements small enough to make expert opinion meaningful.

Pathway probabilities ultimately are used to determine the fraction of all egg-containing servings consumed by each of the endpoints defined by the risk analyst. The sum of the endpoint probabilities should equal 100% to signify that all servings are accounted for in the model. Nevertheless, the endpoints do not fully describe the risk of S. Enteritidis in eggs. These endpoints serve only to categorize the general pathways that eggs may travel. The consequence of the servings consumed at a particular pathway endpoint is a distribution of number of contaminated servings at different dose levels.

A very simple model might assume that the distribution for number of S. Enteritidis per serving is the same regardless of what it consisted of or where or how the serving was prepared or consumed. In this simple example, the initial contamination level in the egg and the growth dynamics within the egg, as well as the effectiveness of cooking, is always the same for all the pathways. The conclusion of such an analysis would be that the most risk is associated with the most probable pathway, but such a conclusion is trite. Microbial growth dynamics and cooking effectiveness are completely independent of the pathways in this example. Essentially, all that has been done is to apportion consumers into categories and the largest category is where most illnesses occur.

A more complicated, but more rewarding, approach to building an exposure assessment model is to construct pathways and their inputs such that the endpoint is dependent on the pathway. To varying degrees, this was done in the Health Canada and US SE RA exposure assessments.

Data

Opinion - expert and otherwise - plays an important part in defining the shape and content of distributions when data are lacking. Often, expert opinion is based on data that have not yet been sufficiently analysed. In such cases, the exposure assessment helps to document this data.

Absence of data increases uncertainty. Such uncertainty should be reflected in more dispersed inputs and outputs. Uncertainty distributions that are too narrow incorrectly imbue the model output with more confidence than is warranted. Furthermore, the narrow uncertainty associated with the output serves as a disincentive to collect additional information.

In general, little data are available for calculating path probabilities in the preparation and consumption stage. In the Health Canada exposure assessment, a survey of Canadians was used as evidence concerning the probability that single eggs were fried or scrambled. Otherwise, most probabilities were estimated based on opinion.

In the US SE RA, evidence on the probability that pooled eggs are consumed as single egg meals and are undercooked came from a 1996-97 Food Consumption and Preparation Diary Survey. This survey showed that 27% of all egg dishes were consumed as undercooked meals. Another survey estimated that each person consumed undercooked eggs 19 times per year (Lin, Morales and Ralston, 1997). The FDA Food Safety Survey was also cited as evidence for the probability that a pooled egg is used as an ingredient in the home and is not cooked (Klontz et al., 1995). The Lin, Morales and Ralston (1997) study also showed that the average frequency was 0.4 raw eggs consumed per consumer per year.

Although both risk assessments described their uncertainty in path probabilities as distributions (usually PERT distributions), the average probabilities assumed by each model are summarized in Tables 4.14 and 4.15.

Table 4.14. Summary of average pathway probabilities assumed in Health Canada exposure assessment.

Table 4.15. Summary of average pathway probabilities assumed in US SE RA.

Methods

The Health Canada and US SE RA exposure assessments model eggs through distinct pathways. The Whiting and Buchanan (1997) risk assessment considers a single product (mayonnaise) consumed in the home. Therefore only the methods used in the Health Canada and US SE RA models are here compared.

The Health Canada exposure assessment models twelve combinations of location (home ([H] or food service facility [F]), use (egg meal [M] or an ingredient in a recipe [R]), and type of cooking (boiled [B], scrambled [S] or fried [F] for egg meals; raw [R], lightly cooked [L] or well cooked [W] for recipes). Each of these paths is replicated for three types of growth: none, some or maximum growth. A total of 36 distinct pathways are thus modelled (Figure 4.31).

Figure 4.31. Schematic diagram of pathways modelled in the Health Canada exposure assessment. R is raw, L is lightly cooked, W is well cooked, F is fried, S is scrambled and B is boiled.

The Health Canada exposure assessment considers growth to be independent of pathway. Thus, regardless of the eventual location or use of the egg, there is a 0.962 probability that no growth will occur, 0.036 probability that some growth will occur and 0.002 probability that maximal growth will occur. Table 4.16 lists the twelve pathways and the associated endpoint probability for each.

Table 4.16. Summary of average endpoint path probabilities in the Health Canada exposure assessment. Codes are explained in the text.

The US SE RA (Figure 4.32) models sixteen combinations of location (home [H] or institution [I]), pooling (pooled [P] or not pooled [N]), use (single egg meal [E] or ingredient in a recipe [I]), and type of cooking (thorough [T] or lightly cooked [L] for egg meals, cooked [C] or raw [R] for ingredients). This model considers growth to be dependent on pathway. In other words handling of eggs in homes may differ from handling of eggs in institutions. Modelling this dependency avoids situations where the model depicts results of particular time and temperature inputs that should not occur in a particular setting. This can be done by collecting output from each of the pathways and then integrating the results by weighting them by the pathway probabilities. Table 4.17 lists the sixteen pathways and the associated endpoint probabilities for each. As can be seen in the table, one path accounts for nearly 34% of all eggs (HNET) while another for only 0.01% (HPIR).

Figure 4.32. Schematic diagram of pathways modelled in the US SE RA. C is cooked and R is raw. T is thoroughly cooked and L is lightly cooked.

Table 4.17. Summary of average endpoint path probabilities in US SE RA Codes are explained in the text.

From the distribution and storage chapter, it can be recalled that the probability of eggs being consumed in the home is about 75%. That information and the information in Table 4.15 can be used to illustrate how final path probabilities in Table 4.17 can be calculated. For example, the HNET pathway represents the fraction of eggs that are consumed in the home in non-pooled single-egg meals that are thoroughly cooked. The fraction of all eggs consumed that travel the HNET pathway can be calculated as the product of the following terms: the probability that eggs are consumed in the home (75%), the probability that home eggs are not pooled (97%), the probability that pooled eggs in the home are consumed as single egg meals (70%), and the probability that these meals are thoroughly cooked (67%). The product of these probabilities equals 34%. This is approximately the same as that shown in Table 4.17 for the HNET pathway. Differences are partly due to rounding error and the fact that the probabilities are actually skewed distributions for which the mean of the product does not precisely equal the product of the mean.

Cooking

Data

Data are available for d-values and z-values for S. Enteritidis in eggs. Unfortunately, these values are not helpful unless information on cooking times and temperatures is also available. Inputs to both exposure assessments are thus based on results of direct measurements of log reduction for different types of cooking when applied to single egg meals.

Some data pertaining to expected log reduction when eggs are undercooked was cited in the US SE RA (Humphrey et al., 1989b). This evidence provided estimates of the effectiveness of boiling, frying or scrambling eggs at suboptimal temperatures.

Methods

The Health Canada and the US SE RA exposure assessments use almost identical inputs to model the log reduction predicted from various types of cooking (Table 4.18).

Table 4.18. Comparison of distributions used to model the log reduction from cooking different single-egg servings in the Health Canada and US SE RA exposure assessments.

NOTES: Min. = minimum; ML = most likely; Max. = maximum.

When eggs are used as ingredients in recipes, however, the Health Canada and the US SE RA exposure assessments differ markedly in how they model the log reductions. Figure 4.33 shows that the Health Canada exposure assessment uses a bimodal distribution with peaks around 3 and 10 logs, while the US SE RA has an equal likelihood of any log reduction from 0 to 8 logs.

Table 4.19 summarizes the pathways and events within pathways including cooking for the two exposure assessments. Although average values are shown in this table, it is important to realize that specific values can vary from one egg to the next. The table is organized to associate similar pathways defined in the two exposure assessments. Hence the pooled egg pathways, "_P_ _", in the US SE RA are associated with the scrambled cooking pathways, "_ _S", of the Health Canada assessment. Those USDA pathways modelling eggs consumed in either homes or institutions that were prepared as single-egg meals and not thoroughly cooked did not neatly fit within one of the Health Canada pathways. These pathways actually would fit in all three of the cooking types (i.e. HMF, HMB and HMS) modelled in the Health Canada assessment. For simplicity, these pathways are simply separated in Table 4.19. Other associations were similarly made to simplify this presentation. Nevertheless, direct comparisons for path probabilities are problematic based on these associations.

Figure 4.33. Comparison of predicted effectiveness of cooking meals containing eggs as ingredients for two risk assessments.

Table 4.19 also illustrates some of the similarities between these two analyses. Average initial contamination levels are the same for all pathways within each exposure assessment, and these are similar between the assessments. Average logs of growth are the same for all pathways in the Health Canada assessment (i.e. 0.14 logs per egg), but vary by location (e.g. home or institution) and pooling practices (e.g. pooled or not pooled) in the US SE RA. The logs of growth for institutional users of non-pooled eggs is similar for both assessments (i.e. 0.14 versus 0.24 logs per egg). For all other pathways, the USDA model predicts more growth. In particular, the USDA pooled pathways average one log of growth more than the Health Canada pathways. Such results reflect the explicit modelling of growth after pooling in the USDA model. Cooking effectiveness is also similar between both models. The only dramatic difference is the average log reduction for well cooked meals containing eggs as ingredients (i.e. 10.2 log reduction in the Health Canada assessment).

A quantitative farm-to-table model of S. Enteritidis will contain the components shown in Table 4.19. Using a Monte Carlo approach, the initial logs of bacteria are added to the logs of growth to determine the pre-cooking exposure. Log reduction from cooking is then subtracted to determine the exposure dose remaining. These calculations are completed for each of the pathways. Because the inputs (i.e. initial logs, logs of growth and logs reduction) are random variables, Monte Carlo simulation does the calculations iteratively to determine a final distribution for each pathway. Pathway probabilities and number of servings then serve to weight each distribution. In this manner, these multiple distributions can be integrated into a single exposure distribution.

Table 4.19. Summary of average values predicted by two exposure assessments for all pathways modelled.

NOTES: (1) Pathway codes are explained in the text. (2) HC = Health Canada Exposure Assessment (Health Canada, 2000). (3) USDA = US SE RA (USDA-FSIS, 1998). (4) N.A. = not applicable

4.2.6 Summary

Absence of data should increase the uncertainty in an exposure assessment. If one can imagine that replacing a triangular distribution based on expert opinion with an empirical distribution based on limited test data would increase uncertainty, then the original distribution must be too narrow.

Careful attention should be directed to those areas in exposure assessments in which the product changes form or the units change. Pooling eggs into a container creates a product distinctly different from shell eggs. This product is able to support immediate bacterial growth and its storage must be modelled as a unique event.

Not all available data are necessarily useful to an exposure assessment. Detailed information on certain processes often can not be used without more information. In the case of S. Enteritidis, knowing the d-values does not help in construction of a model unless cooking times and temperatures either are known or can be modelled. Thus, the high degree of uncertainty and variability in cooking effectiveness inputs noted in this comparison of models emphasizes the need for more research on these inputs.

Given the dearth of published evidence on relevant egg consumption and preparation practices among populations of end users, the preparation and consumption component of an exposure assessment is the most difficult to accurately model. Unfortunately, even with perfect information, this component is very complicated. Multiple pathways reflecting multiple end users, products, practices and cooking effectiveness levels guarantee that assessing the preparation and consumption component is fraught with difficulties. Nevertheless, the advances inherent in both the Health Canada and USDA models provide reasonable starting points for subsequent analyses.

Neither model includes the possibility of re-contamination of egg meals following cooking. These models also do not account for possible cross-contamination of other foods from S. Enteritidis-contaminated eggs. These limitations might be addressed in future models.