FDA/Center for Food Safety and Applied Nutrition
USDA/Food Safety and Inspection Service
Centers for Disease Control and Prevention
September 2003

Quantitative Assessment of Relative Risk to Public Health
from Foodborne Listeria monocytogenes
Among Selected Categories of Ready-to-Eat Foods

Table of Contents

III. Exposure Assessment

(Also available in PDF)

Exposure is a function of the quantity of a food consumed and the level of contamination in that food. While the contamination level in food at consumption is the important parameter in evaluating public health, most of the available contamination data pertain to foods sampled at retail stores. Hence, it was necessary to develop estimates of the frequency and amount of each serving of the contaminated foods likely to be consumed in the United States, as well as the Listeria monocytogenes levels in those foods. Limitations inherent in food consumption data and the paucity of contamination data for certain foods made certain assumptions necessary to develop the estimates. These limitations and assumptions are discussed later in this chapter.

The goal of this risk assessment was to provide information needed to focus risk management strategies among a variety of foods that could be potentially contaminated with Listeria monocytogenes, the purpose of the exposure assessment is to estimate the contamination and consumption of foods that have a potential for Listeria monocytogenes contamination. Therefore, this risk assessment modeled growth of Listeria monocytogenes in foods during post-retail storage and reduction of levels during home cooking or reheating of frankfurters. Growth was also modeled for some contamination data that were collected pre-retail to account for possible growth between manufacture and retail.

Foods that were included in the risk assessment were identified through a comprehensive review of the recall, microbiological and epidemiological literature. Each food was placed in one of 23 food categories. Using distributions of contamination and consumption data, estimates of exposure to Listeria monocytogenes in the various foods were derived. The components of the exposure assessment are provided in Figure III-1, and specific modeling details are provided in Appendix 3.

Figure III-1. Components of the Exposure Assessment Model

Food Category Identification

The first step in the exposure assessment was to consider appropriate foods to include in the risk assessment model. As the risk assessment progressed, foods and food categories were continually reevaluated and modifications were made based on new information, such as the results of growth models or new microbiological or epidemiological literature. Foods that have a significant potential for Listeria monocytogenes contamination were identified. They represent a subset of foods that comprise an individual's total diet. Foods that have not been linked to Listeria monocytogenes contamination were not included, for example, grain products (e. g., bread, cookies, cakes), soft drinks, canned fruits, and cooked mixed dishes (e. g., lasagna, soups). Furthermore, foods that have limited association with Listeria monocytogenes contamination (e. g., cream-filled pastries) were not included because neither contamination level data nor appropriate data to serve as a substitute were available. It was also presumed that some foods that are cooked just prior to consumption (e. g., most meats and seafoods) present a very low likelihood of containing Listeria monocytogenes when consumed and were not included in this risk assessment. Eggs are an example of a food category that was not included in the risk assessment, but could be a vehicle for listeriosis. Although eggs have been implicated in one outbreak with two cases (Schwartz et al., 1988), Listeria monocytogenes has not been isolated from intact eggs and eggs products are typically cooked before consumption (Ryser and Marth, 1999).

A review of the literature was conducted to identify foods that have a significant potential for Listeria monocytogenes contamination. The review concentrated on the following:

• Outbreaks

• Sporadic cases, i.e. individual cases not reported as part of a documented outbreak

• Recalls and regulatory actions

• Literature related to prevalence and incidence of Listeria monocytogenes through analytical testing in North America (the United States and Canada)

• Literature on outbreaks, sporadic cases, and prevalence and incidence studies of Listeria monocytogenes in other countries

The next step in selecting foods for the risk assessment was a review of the available data on contamination and the ability of the food to support growth of Listeria monocytogenes. Food contamination data were compared with the available food consumption data to create food categories.

Foods that are ready-to-eat (RTE) were ultimately selected. Some RTE foods are raw and others receive some processing prior to sale. Still other RTE foods are fully cooked before sale but may be subjected to subsequent handling and storage, thereby increasing the possibility of recontamination.

The identified foods were further sorted into categories based upon food characteristics, use, and the potential for growth of Listeria monocytogenes. For example, Dry/Semi-dry Fermented Sausages were differentiated from other deli meats such as bologna, sliced turkey, and ham. The Cooked RTE Crustaceans food category contains peel-and-eat shrimp, steamed and boiled shrimp, and steamed crabs - foods that may be refrigerated and eaten chilled or allowed to cool after cooking, thus allowing for re-contamination and growth. The Vegetable food category includes many raw vegetables, as well as mixed vegetables such as bagged salads (without salad dressings). Similarly, the Fruits food category includes many raw and dried fruits and mixed fruits such as fruit salads (without salad dressings). In this updated risk assessment, the vegetable and fruit salads with salad dressings are included in the Deli-type Salad food category. While there is a single Deli-type Salad food category for reporting purposes, to model growth of Listeria monocytogenes, salads were segregated into growth and non-growth salads and considered the use of preservatives in salads made in bulk for distribution to retail stores.

In this updated risk assessment, the cheese categories have been reorganized into six categories based on moisture content. Another update to the categories included splitting the Miscellaneous Dairy Products into two categories. The Cultured Milk Products category includes the low pH dairy foods manufactured with lactic acid fermentation. Of this category, yogurt is the most frequently consumed food, followed by sour cream and buttermilk. The High Fat and Other Dairy Products category includes the remainder of the dairy products that generally support growth (including powdered products when reconstituted). Butter, cream and half and half are the most prominent foods in this category, but shakes and chocolate milk made with cocoa or syrup are also included. The frankfurter category has been divided into reheated and not reheated frankfurters to distinguish the impact that reheating before consumption can have on the predicted risk. The number of unreheated frankfurters was represented by a triangular distribution with a minimum of 4%, most likely of 7%, and maximum of 10% of the total frankfurters consumed without reheating. These values were based on surveys conducted by USDA and American Meat Institute.

Table III-1 lists the 23 food categories that were used in this risk assessment. The food categories fall into five general groups: Seafood, Produce, Dairy, Meat, and Combination Foods. (See Appendix 5 for a detailed listing of the foods included in each food category.)

Food Consumption Data

Data from two large-scale, nationwide food consumption surveys were used to provide estimates of exposure to Listeria monocytogenes via distributions of food consumption. The first survey is the Continuing Survey of Food Intakes by Individuals (CSFII 1994-96). This is the latest survey of consumers of all ages conducted by USDA's Agricultural Research Service (USDA/ARS, 1998a, 1998b). The survey consists of the following:

• Two 24-hour recalls of foods eaten during two nonconsecutive days (with the interview for the second day conducted 3 to 10 days after the interview for the first day, but not on the same day of the week).

• Sample weights for weighting the data so that they will more closely reflect consumption by the non-institutionalized United States population.

• A sample of 16,103 respondents, including:

  • Pregnant and/or lactating women (n = 123)

  • Children under 4 years (n = 2,284)

  • People 60 years and older (n = 2,315)

• Over sampling of low income, young children, and the elderly (USDA ARS, 1998a).

• A Population Parameter of 261,897,280, appropriate for 1994-1996.

The second nationwide survey of food consumption is the Third National Health and Nutrition Examination Survey (NHANES III), which was conducted in 1988 to 1994 (US DHHS, 1998). NHANES was conducted by the National Center for Health Statistics in the Center for Disease Control and Prevention (CDC/NCHS), DHHS. The survey consists of the following:

• One 24-hour recall of foods eaten.

• Sample weights for weighting the data so that they will more closely reflect consumption by the non-institutionalized United States population.

• A sample of 30,818 respondents, including:

  • Pregnant and/or lactating women (n = 399)

  • Children under 4 years (n = 3,979)

  • People 60 years and older (n = 3,919)

• Over sampling of young children, older persons, black persons, and Mexican Americans.

• A United States Population Parameter of 251,097,003, appropriate for 1988-1994.

Consumption data from the CSFII 94-96 survey were used for 21 of the 23 food categories. CSFII data were used preferentially because they are newer and account for up to two days of eating per respondent. Data for unpasteurized fluid milk and unreheated frankfurters were modeled based on CSFII data for pasteurized milk and all frankfurters consumed. NHANES III data were used for two food categories (Raw Seafood and Preserved Fish) for which there are fewer than 30 eating occasions (servings) in the CSFII survey.

The surveys contain consumption data for many foods and each food has an associated food code. Over 640 food codes for RTE foods were matched to one of the 23 food categories. The following information was extracted from the databases for each food category:

• Weighted descriptives (e. g., mean amount eaten in grams, median amount eaten in grams, number of servings) that characterize all eating occasions in two nonconsecutive days of eating (one day for NHANES III).

• Distributions of the amount of food (in grams) eaten in all servings over two days (one day for NHANES III).

• Distributions of the amount of food (in grams) eaten in all servings, expressed as weighted percentiles.

• Weighted descriptives to describe the amount of the food (in grams) eaten per person per day, as well as the number of eaters.

• Per capita estimates of food eaten.

Several limitations of the food consumption surveys had an impact on their use for risk assessment purposes. For some foods, it was a challenge to determine consumption. Surveys listed some particular foods under several food codes, such as ham consumed alone or ham in a ham sandwich. The proportion of a particular food (such as ham) in a mixed ingredient product (such as a ham sandwich) was determined using a generic recipe provided by the survey. The gram amount of the food (ham) consumed was then calculated and added to the intake derived from other food codes for the specific food (ham). For this risk assessment, sandwiches were broken down into individual ingredients. Specifically, for frankfurters, dry semi/dry fermented sausages, deli meats, pâté and meat spreads, and deli salads, the actual consumption of meat or deli salad product consumed alone, as well as the proportion used in sandwiches, was used. In the case of vegetable and fruit salads (in which fruits and vegetables were the major component) and deli-type salads (not included in a sandwich), however, the entire salad was used, rather than the component ingredients.

The consumption surveys do not collect information from consumers to determine whether the milk they drank was pasteurized or unpasteurized (raw). Although federal law requires milk in interstate commerce to be pasteurized, some states allow unpasteurized milk to be sold and consumed within the state. Results of a 1995 FDA/CDC survey of all 50 states, Puerto Rico, and the District of Columbia, showed that 28 states (54%) permit the sale of unpasteurized fluid milk. However, it is estimated that unpasteurized milk accounts for less than 1% of the total volume of milk sold in these states (Headrick et al., 1998). Because consumption surveys did not list "drinking occasions" (servings) of unpasteurized fluid milk, the consumption of this food category was modeled by estimating it as 0.5% of the amount consumed per serving of pasteurized milk (54% x 1%). The consumption surveys did not provide any information on the storage and heating of frankfurters. Estimates for the fraction of frankfurters stored frozen before consumption and those eaten without reheating were obtained from other surveys.

Another limitation of food consumption surveys used is that some food categories have a small number of servings. Estimates based upon small sample sizes may be less statistically reliable than estimates based on larger sample sizes (USDA/ARS, 1998a). Although weighted food consumption data provide a better representation of the United States population, weighting small samples does not provide better reliability. In addition, the surveys do not provide corrections to account for underreporting and over reporting of the amount of a food eaten by consumers.

The food consumption surveys did not collect demographic information delineating consumers who are immunocompromised. Furthermore, the surveys did not measure consumption by the elderly who are living in nursing homes or other forms of assisted living outside of the home, nor did they contain a large enough sample of pregnant women to generalize consumption to all pregnant women. Thus, the available consumption data did not allow the determination of comprehensive estimates of food consumption for each individual susceptible subpopulation. Consumption between the subpopulations was compared. Specifically, nonparametric statistical analyses were conducted to determine if there were significant differences between the distributions of the amount eaten in each serving (expressed as weighted percentiles) for the elderly and the intermediate-age population. Seventeen food categories had sufficient consumption data to permit these analyses. There were no statistically significant differences in consumption patterns for 14 of the examined 17 food categories. Thus, for the purpose of estimating the distribution of serving sizes, the food consumption data representing all eaters were used.

Note: Starting in 2002, CFSII and the dietary component of NHANES were merged into NHANES. The integrated survey will provide two 24-hour recalls of food consumption for 5,000 individuals a year and characterize "What We Eat in America."

Annual Number of Servings of Foods

In order to estimate the number of servings of the foods in each food category eaten in a year, some key data assumptions were necessary. First, it was assumed that the weighted number of servings for one (NHANES III) or two days (CSFII) of consumption of the foods in a specific food category could be extrapolated to the number of servings of those foods eaten by the population on an annual basis. Second, it was assumed that the weighted number of eaters of a food per day would represent the number of eaters of the food over 365 days. Obviously, there are some foods that individuals are more likely to eat each day (e. g., vegetables, milk) and others that they eat frequently (e. g., fruits, deli meats) or occasionally (e. g., frankfurters, cottage cheese). Some foods are seasonal and are not available year round (e. g., some fruits and vegetables), and people may not be likely to purchase more costly items (e. g., shrimp, crabmeat) for regular consumption. Thus, it is important to note that when estimating the consumption of foods on an annual basis, all foods reported in food consumption surveys during a one- or two-day period are not likely to be eaten in the same frequency by the same people over an entire year. To estimate the number of annual servings for each food category, we divided the weighted number of serving consumed in two days by 2 (one-day basis) and then multiplied that value by 365 (annual basis). Table III-2 provides the annual number of servings of food consumed in the United States for each of the 23 food categories.

The annual number of servings associated with the pregnancy exposures resulting in neonatal deaths were estimated using the number of servings in the intermediate-aged group multiplied by the birth rate (1.74%) and a fractional exposure period. A triangular distribution with a minimum of 1 day, a most likely value of 7 days, and a maximum value of 30 days was used to represent the uncertainty in the exposure period. In order to estimate the number of servings in the neonatal group, the annual number of servings in the intermediate-age group was multiplied by the exposure period (triangle distribution) and divided by 365 days to estimate the number of per annum servings consumed by pregnant women. Because the perinatal exposure period is longer than neonatal (the total number of deaths includes prenatal, i.e., stillbirth, cases occurring in the last trimester), perinatal per serving death rates from listeriosis were estimated using an exposure period of 90 days (3/12 yr = 0.25) and a pregnancy rate (2.77%) rather than birth rate.

Serving Size Distributions

Empirical distributions were used to describe the serving sizes (grams of food eaten per serving) in the 23 food categories. These distributions are expressed as a series of population percentiles of the amount of food eaten per serving, weighted to reflect the consumption survey demographics. There were no uncertainties presented for these food categories because empirical distributions were used. The uncertainties associated with the serving size distributions would be relatively small, compared to other uncertainty distributions in this risk assessment for three reasons. First, even the smallest data sets used to characterize the serving size distributions are large relative to other parts of the Listeria monocytogenes risk model. Second, although the data may not be completely representative of the current behavior of the United States population, the data come from surveys that were explicitly designed for that purpose. Third, the variability (range) in serving sizes covers a smaller range (two logs) than many other parts of the model.

Table III-3 shows the 50^th (median), 75^th, 95^th and 99^th percentiles of the weighted distributions of serving size. For example, these percentiles for Smoked Seafood are 57, 75, 136 and 142 g/serving, respectively. This distribution indicates that half of the servings were less than 57 g and 95% of the servings were less than 136 g.

Food Contamination Data

Over the last fifteen years, numerous studies have been published that report on foods contaminated with Listeria monocytogenes in a variety of countries and locations. Contamination data included in this risk assessment were reported from the United States and other countries on six continents. Most of the studies were from the industrialized countries of North America and Western Europe. Many studies did not identify the sampling of imported foods or indicate whether imports were excluded from the study. Contaminant serotype information was not considered because the food contamination studies did not usually identify the serotypes.

Data sources included the published scientific literature, published and unpublished official government documents, and data obtained from the private sector. All data and references are available in the docket established for this risk assessment. Two types of data describing the levels of Listeria monocytogenes contamination in food were identified.

• Presence/absence (qualitative) data (i.e., the number of positive samples relative to the total sample collection).

• Enumeration (quantitative) data (i.e., the number of colony forming units (cfu) of Listeria monocytogenes that were measured from a sample). It is conventionally assumed that one cfu is equivalent to one organism.

Both qualitative and quantitative studies were used in the assessment (Table III-4; Appendix 7). Data from presence/absence studies (qualitative data) were converted to numerical data and included in the model by assigning the lowest possible contamination level that can be detected by the laboratory method. For a method that uses a 25-g sample, the lowest detectable level is 0.04 cfu/g of food. Consequently, the qualitative data could be used along with the quantitative data in the construction of the cumulative distribution curves of Listeria monocytogenes levels in food.

Because each food category usually includes many related types of foods, data were collected to represent all the foods in a designated food category. For example, the deli meats include, in part, ham, bologna, and sliced chicken. These deli meats have diverse microbial characteristics and there are relatively few existing studies for each of these foods. Hence, all data available on these products were used with the assumption that the summation of the collected data represented the diverse compositional, geographic, seasonal, home vs. away-from-home, relative frequency of consumption, and other factors that affect the exposure from Listeria monocytogenes in these foods. Where methodologies or designations varied among multiple data sources, the original data were often regrouped or recalculated (particularly for the growth modeling work).

Pairing consumption data with the appropriate contamination data was often imperfect. Dietary intake data were highly specific as to the type of food consumed (e. g., smoked mussels). In contrast, the contamination data reported in the literature were often more generic (e. g., samples may only be described as shellfish).

The analytical methods used in the food contamination studies to determine the presence of Listeria monocytogenes were generally well known and were approximately equal in sensitivity at about 1 cfu per 25 g sample (0.04 cfu/g). However, for enumeration methods of analysis, the sample size was usually less than 25 g and was not as sensitive (typically 20 to 50 cfu/g). Typically, the samples obtained for analysis were from non-composited samples of food. An exception, however, was unpasteurized fluid milk obtained from bulk tanks.

Contamination levels at consumption were modeled with the assumption that contamination distributions for a given food in the United States do not vary significantly from those in other countries, especially Western Europe and other developed countries. Similarly, it was assumed that all foods within a category have a similar pattern of contamination. Furthermore, all Listeria monocytogenes food isolates were accepted as having the potential to cause human illness. No differences in ability to grow or other characteristics between food and clinical isolates were assumed. As will be discussed later, the impact of these assumptions was considered in the uncertainty associated with relative risk determinations.

The available data on Listeria monocytogenes levels had some limitations that affected the distributions for levels of Listeria monocytogenes in foods. First, there are relatively few data points above the limit of detection (0.04 cfu/g). This is because the occurrence of detectable levels of Listeria monocytogenes in food is rare and because most surveys of the occurrence of Listeria monocytogenes in food did not quantify the levels in positive samples. Second, some of the data are not from the United States and this data may not always be representative of food and processing procedures in the United States. To create an estimate of the current United States distribution, the data sets were weighted by the number of samples in the data set, likelihood of the food in that country to be imported to the United States food supply, and the recency of the data. Third, there was a wide degree of variation between studies in the occurrence of high levels of Listeria monocytogenes. The extent to which this variation reflects true variation in a particular food, is not known.

Many of the studies found in the published literature were conducted in the late 1980s and early 1990s. The extent that improved sanitation and other control measures implemented by the food industry have reduced the frequency and level of contamination since 1993 (when the earlier research was conducted) is difficult to determine from published literature. It was felt that some allowance should be made for the age of data and therefore, all data were used but the more recent data were given greater weight (details below). Because some food categories had little data, which would result in a biased estimate, the overall trend in contamination for all the food categories from before 1993 to after 1998 was obtained and applied to these data sets.

The length of time a food was held at retail before it was obtained for microbial sampling was not recorded in the survey studies. It was therefore necessary to assume that foods were sampled without bias and would represent the entire range of post-production and pre-sale conditions for that food.

Growth Data

Growth of Listeria monocytogenes in food is a function of the storage time, storage condition, and rate of growth in specific foods. The storage times were multiplied by the rate of growth to provide an estimate of the amount of Listeria monocytogenes growth occurring between retail purchase of the food and its consumption. The model includes consideration of the interaction of storage time and temperature and maximum growth that specific foods support.

Storage time

Some foods are consumed on the day of purchase whereas others remain in the home refrigerator for lengthy periods of time. This is a major source of variability in the estimate of growth and ultimately, in the numbers of Listeria monocytogenes consumed. Except for frankfurters and deli meats, no data were found on the storage of foods in the home; therefore, storage time, including variation and uncertainty, were estimated based on the expert judgment of the risk assessment team in consideration of recommendations developed by the Food Marketing Institute (2002) and other individuals familiar with the production and use of the various foods. It is recognized that foods may be kept beyond the recommended storage times. This risk assessment modeled estimated consumer food practices, not necessarily the recommended practices. The values were developed by consensus of the risk assessment team and vetted by government subject matter experts and other scientific reviewers including those who submitted comments following the release of the draft risk assessment. The minimum, most likely and maximum storage times used to develop the distribution of storage times for the food categories are presented in Table III-5. These are skewed distributions with relatively few servings at the maximum storage time. For Smoked Seafood, as an example, over 90% of the servings are stored for less than 13 days.

Estimating duration of post-retail storage for Frankfurters and Deli Meats

Preliminary data from a study being conducted for FSIS by Georgetown University (Wachsmuth, 2000) provided information for frankfurters and deli meats used in the draft risk assessment. For frankfurters, 3 of 73 respondents gave 21 days storage and 3 gave 30 days as the maximum time. For deli meats, 2 of 81 respondents gave 21 days of storage, and 2 gave 30 days as the maximum time. FSIS also questioned people who called in to their telephone Meat and Poultry Hot Line about their frankfurter storage and cooking or reheating practices. Of 136 callers, one had kept frankfurters 90 days and one for 180 days (Wachsmuth, 2000).

In response to the need for more comprehensive information on consumer practices for frankfurters and deli meats, the American Meat Institute (AMI) commissioned a consumer survey that asked how long, on average, deli meats and frankfurters were stored before consumption (American Meat Institute, 2001). The responses are shown in Table III-6. These data were used to model storage times for frankfurters and deli meats as described in section "Modeling: Growth Between Retail and Consumption."

Refrigeration Storage temperature

Data for home refrigerator temperatures were obtained from a 1999 survey conducted by Audits International (Audits International, 1999). Nine hundred thirty nine refrigerators in the United States were included in the survey. Approximately 26% of the refrigerators exceeded 41°F (5°C) and 1.4% exceeded 50°F (10°C) (Table III-7). The refrigeration temperatures were modeled with a discrete distribution where temperature values were randomly sampled from the data provided by Audits International.

Growth Rate

A summary of the growth rate data is presented in Table III-8 and a complete list of the literature data can be found in Appendix 8. Significant differences in composition and processes are present within many of the food categories. Within the Smoked Seafood food category, for example, there were hot and cold smoked fish, various salt levels, both aerobic and vacuum packaging, and different fish species. The modeling process used a cumulative table of the actual data points, not the means and standard deviations presented in Table III-8.

Modeling: Listeria monocytogenes Levels in Food at Retail

The majority of the data collected on the contamination of foods only determined whether or not a sample, typically 25 g, contains Listeria monocytogenes. Compared to the amount of qualitative data on the presence or absence of Listeria monocytogenes in foods, there is relatively little recent quantitative data available. This is due to the additional laboratory effort necessary to enumerate samples, the low frequency of detecting positive samples, the need to test a large number of samples, and regulatory requirements that do not require enumerative data. Therefore, the approach taken was to develop a generic contamination model to describe the distribution of Listeria monocytogenes in food.

A three-step process was used to model levels of Listeria monocytogenes in food at retail.

• Step 1: Characterize the distribution of Listeria monocytogenes across food categories using the contamination data reported in selected quantitative data sets (i.e., create a generic distribution).

• Step 2: Characterize the uncertainty distribution for the frequency of detectable contamination for each food category using prevalence data adjusted to account for study size, age, and country of origin.

• Step 3: Integrate the quantitative data from generic distributions (step 1) with the adjusted prevalence data, specific for each food category (step 2).

The general approach was to assume that the contaminated samples are detectable contaminations arising from a continuous log normal distribution of contamination. The minimum detectable level from presence/absence tests is typically 1 organism in 25 g or 0.04 organisms per gram. A low percentage of samples has contamination at or above this level and the remainder has non-detectable levels (i.e., <0.04 organisms/g). There may be no detectable Listeria monocytogenes in a specific sample (a 25.0 g package), but if 1000 packages from that lot are analyzed Listeria monocytogenes might be found. The average contamination could be one organism in 1000 packages (or a level of 0.00004 organisms per gram), far below the detectable level of 0.04 organisms/g. Therefore, what is observable with the presence/absence and quantitative tests is only the upper tail of the distribution. As shown in Figure III-2, the model fits a curve to the log cfu/g data and the mean and standard deviation are calculated. This curve represents a food category with approximately 10% of the samples positive for Listeria monocytogenes. It also shows that 3.1% of the samples have more than 100 cfu/g.

Figure III-2. Example of the Contamination Curve for a Typical Food Category Showing
Frequencies of Detectable and Nondetectable Samples

Studies with enumerated samples were selected and fitted to a normal distribution. The standard deviations from each of these studies were used to estimate the uncertainty in the distribution. The presence/absence data for each food category were then used to create a frequency distribution of contamination at the 0.04 cfu/g level. A normal curve with the appropriate standard deviation was then fit to the presence/absence data by "sliding" the mean until the percentage of positive samples corresponded to the presence/absence data. A normal curve for the log cfu/g was chosen because studies enumerating spoilage flora that are at sufficiently high levels to observe the curve showed that this distribution was appropriate (Kilsby and Pugh, 1981; Gill et al., 1996).

Step 1: Characterize the distribution of Listeria monocytogenes across food categories

Seventeen studies were selected for quantitative analysis (Table III-9). All of these studies had at least ten samples with enumerated values. The levels of Listeria monocytogenes in the samples were transformed to log scale and the data for each study were fit using a normal distribution (Figure III-3). The mean level of Listeria monocytogenes (log cfu/g) and the standard deviation of the contamination data sets were calculated. This process was repeated for the 17 studies with adequate enumeration data.

Figure III-3. A Lognormal Distribution for Listeria monocytogenes in Smoked Seafood

The standard deviations and mean levels of Listeria monocytogenes (log cfu/g) are summarized in Table III-9. The standard deviations of the distribution for each study ranged from 1.1 to 10.7 although most were less than 5.0.

These standard deviations were used to characterize the variation and uncertainty of the distribution of Listeria monocytogenes concentration in the food categories. The ranges of standard deviations used are given in Table III-10. A default range of 2 to 5 standard deviations was used for all food categories unless additional information was available to refine the uncertainty. Refined standard deviation ranges were used for four food categories (smoked seafood, pâté and meat spreads, deli meats, and deli-type salads) based on information as described in Table III-10. For example, the range of standard deviations assigned to Smoked Seafood is narrower than the default range based on consideration of the standard deviations from four enumeration studies for this food category.

The set of presence/absence studies for each food category was used to generate a discrete uncertainty distribution (a histogram) for the frequency of detectable contamination. First, the presence/absence data were used to generate a single estimate of the fraction of positive samples (i.e., a rate-concentration estimate) for each study. The concentration level was equal to the detection limit of the analysis (typically 0.04 cfu/g; based on 1 organism per 25 g sample). Next, the individual studies were adjusted (weighted) to account for sample size, geographic region of food origin, and date of collection. In addition, some data sets were obtained by sampling at the manufacturer instead of at retail. These data sets were adjusted to allow for growth between manufacture and retail. With this adjustment the data collected at manufacture would then have the same percentage of positive samples but they were assigned higher cfu/g values.

Adjust for sample size, geographic location, and study date

The relevance of a particular contamination data set to represent current United States retail foods for the purposes of this risk assessment was a difficult judgment. If abundant, quantitative, recent and United States data were available, only this data would be used in the risk assessment. However, for most food categories these data were not available. Therefore, all data sources were used and weights were assigned to each data set so that the more relevant sets were given greater importance in this risk assessment. These weights were obtained from a panel comprised of government subject matter experts (Carrington and Dennis, 2001).

The individual studies were weighted by sample size, geographic region, and study date as follows in Equation 1.

Study Weight = n * gw * dw

Equation [1]

Where:

The width of the probability interval assigned to each study was proportional its relative weight as shown in Equation 2.

Study Probability = Study Weight / Total Weight

Equation [2]

where Total Weight is the sum of all the Study Weights for the food category.

Adjustment of older data for food categories without large recent studies

About half of the food categories had large studies that were conducted within the past three years. As a result of the weighting scheme used to weight the studies, these recent studies usually received at least half the probability interval, dominating the analysis. Ten food categories had only older studies and those studies tended to have higher prevalence rates. The higher prevalence ranges may result from higher actual contamination levels or non-representative sampling. In either case, the data may tend to overestimate current Listeria monocytogenes concentrations, thereby biasing these categories compared to categories with recent data. To represent the uncertainty of this bias, the impact of large new studies on prevalence of Listeria monocytogenes was evaluated (Table III-11). Ratios were calculated by dividing the weighted pooled prevalence of 1999 and earlier data (percentage positive samples) by the weighted pooled prevalence of data for all years. A ratio less than 1 indicates that the prevalence of contaminated samples is currently higher than in the past. The reduction ratio values were used to adjust the food categories for which recent, large studies were not available. Specifically, the set of values in Table III-11 were used as an uncertainty distribution to reduce the number of positive values from older studies in categories without newer data. The food categories adjusted with the ratios to account for the lack of newer data include: Preserved Fish, Cooked RTE Crustaceans, Fruits, Hard Cheese, Processed Cheese, and Cultured Milk Products.

Adjustment for growth between production and retail for samples taken at manufacturing/ production

Some studies collected samples at manufacturing/ production prior to the point of retail (see Appendix 7). Since growth can be anticipated between production and purchase, the prevalence of positive samples for those data sets from sampling at manufacture were adjusted with estimates derived from the growth models (see section, "Modeling: Exponential Growth Rates").

The temperature ranges and storage times for the food categories are presented in Table III-12. These values were estimated as likely to be encountered between manufacture and retail. Because the distributions are narrow, rectangular distributions were used for storage time and for the temperature range. The median value from the growth models were used to adjust the contamination level but not the frequency of the presence/absence data. If, for example, the estimated growth was 0.5 logs prior to retail, a study with 5% positive at 0.04 cfu/g (-1.394 log) at manufacture would become 5% positive at 0.13 cfu/g (-0.884 log) at retail [0.5 log + -1.394 log = -0.894 log]. The contamination level was therefore increased from 0.04 cfu/g to 0.13 cfu/g to account for the possible growth of Listeria monocytogenes in food between production and retail.

Step 3: Integration of Prevalence Data and Quantitative Analysis

Frequency distributions for Listeria monocytogenes concentration for each food category were generated by integrating the standard deviation estimates with the rate estimates for detectable Listeria monocytogenes. This was accomplished with a 300 iteration simulation in which pairs of values were randomly selected from a uniform distribution of the standard deviations (Table III-10) and the weighted collection of the presence/absence data sets for each food category (including those at 0.04 cfu/g at retail and those adjusted for pre-retail growth). For each of the 300 pairs of values, a mean of the log cfu/g value was calculated (using the Excel Goal Seek procedure) to find the geometric mean that matches the cumulative frequency of positive samples at the detection limit of the assay (0.04 cfu/g or the adjusted value) with the selected standard deviation. Therefore, for each food category, 300 contamination curves were generated. The average frequency for each contamination level was determined to create the variability of contamination levels. The standard deviation of the frequencies for each contamination level became the uncertainty of the distribution for the contamination data.

Example of the Modeling for Listeria monocytogenes in Food at Retail Using Smoked Seafood

Step 1. Characterize the distribution of Listeria monocytogenes across food categories

Data from NFPA (2002) for Smoked Seafood is used to illustrate this step. As shown in Figure III-3, at the 0.04 cfu/g (-1.4 on log scale) contamination value, 0.958 (95.8%) of the samples (2573/2687) contain less than or equal to that contamination level. Sixty-seven more samples had levels < 0.1 cfu/g and eleven samples were contaminated at less than or equal to 1 cfu/g (0.0 on log scale). Therefore the fraction of negative samples is 0.986 [(2573 + 67 + 11)/2687]. This procedure is repeated for the samples that had higher levels of contamination. A normal curve was fitted to the data points by least-squares and the mean and standard deviation were estimated as -6.7 and 3.1, respectively. This process was repeated for the 17 selected enumeration studies and the resulting means and standard deviations are summarized in Table III-9.

Step 2. Characterize the uncertainty distribution for the frequency of detectable contamination

• Adjust for sample size, geographic location, and study date. The study weight and study probability are calculated as described by Equations 1 and 2 using the total number of samples in the study (n), the geographic weight (gw), and the weight for the date of the study (dw). These values are shown for Smoked Seafood in Table III-13. For example for the Aguado et al., 2001 study, the study weight is 52 (52 x 1 x 1) and the study probability is 0.009 (52/6034.7).

• Adjustment of older data for food categories without large recent studies. This step is not applicable for smoked seafood as recent large studies were available. However an adjustment was made using the range of prevelance ratios given in Table III-11 for Preserved Fish, Cooked RTE Crustaceans, Fruits, Hard Cheese, Processed Cheese, and Cultured Milk Products.

Adjustment for growth between production and retail for samples taken at manufacturing. In Table III-13 the 'collection' column indicates which studies were collected at manufacturing/ product and at retail. For the studies collected prior to retail, the level of Listeria monocytogenes was increased to account for anticipated growth between manufacturing and retail. From Table III-12, the mean exponential growth for smoked seafood of 0.15 logs/day at 5°C was multiplied by a uniform distribution (minimum of 1 day, most frequent of 10 days, and maximum of 30 days of storage) and the median of this resulting distribution was 1.08 logs. The fraction of positive samples (0.04 cfu/g or -1.4 log cfu/g) at manufacture was increased to a fraction of positive samples with a value of 0.48 cfu/g (-0.32 log cfu/g) at retail (-1.4 log + 1.08 log = -0.32 log cfu/g). In Step 3 described below, the procedure for the fitting of the contamination distribution the fraction of positive samples remained the same but the contamination level was now represented by a value of -0.32 log cfu/g for these studies.

Step 3: Integration of Prevalence Data and Quantitative Analysis

The Listeria monocytogenes concentration model for Smoked Seafood is presented in Figure III-4. The model estimates are compared to the prevalence studies and the enumeration data. The median (50^th percentile), lower (5^th percentile) and upper (95^th percentiles) bounds reflect the Listeria monocytogenes concentration model (i.e., the set of Lognormal disitribution parameter values). Each data point in the "Prevalence Studies" data set represents an individual study (weighted for sample size and other study characteristics as described in Step 2). The data points in the "Enumeration Studies" data set are pooled from four different studies as noted in Table III-5. The prevelance studies at the -0.32 log cfu/g level represent the studies collected at manufacturing/ production and were adjusted for potential growth between production and retail.

Figure III-4. Modeled Contamination Data for Smoked Seafood Food Category

Results: Modeled Contamination at Retail

Table III-14 shows the modeled distributions for Listeria monocytogenes contamination for the 23 food categories at retail. The first column of data in Table III-14 provides the median percentage of servings with less than one organism per serving, this estimate is not the same as undetectable values (<0.04 cfu/g) because different foods have different serving sizes. The predicted median of the servings having less than one organism of Listeria monocytogenes per serving ranged from 91.3 to 99.9% for the various food categories. In other words, less than 0.1 to 8.7% of the servings had one or more Listeria monocytogenes per serving, depending on the food category. The 5^th and 95^th percentiles provide information to estimate the uncertainty distributions for each of these median values. Although some servings of all food categories are likely to be contaminated at the retail level, servings of certain food categories (e. g., Smoked Seafood, Raw Seafood, Deli Meats, Dry/Semi-Dry Fermented Sausages, and Deli Salads) were the most likely to be contaminated. Other columns in Table III-14 provide the percentage of servings with higher levels of contamination. Most frequently, the food categories are contaminated with 1 to 1000 cfu/serving. The calculations in the risk assessment model used 0.5 log intervals (referred to as bins) instead of the 3 log intervals shown in Table III-14.

The bar chart in Figure III-5 provides a graphic depiction of the modeled distributions. Most of the servings for each food category are in the <1 cfu/serving level (back row of bars). As the level of contamination per serving rises (moving into the front rows of bars), the fraction of servings decreases markedly for most of the food categories.

Thus, for the Smoked Seafood category, the fraction of servings at <1, 1 to 10³, 10³ to 10⁶, 10⁶ to 10⁹, and >10⁹ cfu/serving are about 93.6, 5.8, 0.8, 0.1, and 0.0% of servings, respectively. The sum of the fractions of servings for a food category do not necessarily equal 100% because of rounding and because adding medians is not mathematically correct.

 

Modeling: Growth Between Retail and Consumption

Most of the contamination data used in this risk assessment were from samples collected at retail. Because Listeria monocytogenes can grow slowly at refrigeration temperatures, a growth module was incorporated into the exposure assessment to account for the potential growth of the organism in the food during storage in the home, prior to consumption. The growth model provides an estimate of the numbers of Listeria monocytogenes in the food at the time of consumption.

The growth model included the initial level of Listeria monocytogenes in the foods at retail where the food is purchased, the storage temperature in the home refrigerator, the exponential growth rate of Listeria monocytogenes in a food stored at a specific temperature, the storage time in the home and the maximum growth (stationary phase). Inoculated food studies, where growth of Listeria monocytogenes inoculated into a food was measured, showed that maximum growth at low refrigeration temperatures (<5°C) was often less than growth in the same foods at higher temperatures. It was also concluded that refrigeration temperature and storage time are not independent factors. High storage temperatures and long storage times would not be likely to occur because this combination would lead to obvious spoilage and the food would not be consumed. The output from the growth model was a frequency distribution of the log cfu/g for each food category at the time of consumption.

Exponential Growth Rates

The square root model for exponential growth rate (EGR) was chosen because of its simplicity and general acceptance as indicated by the documented use in the microbiology literature (Ratkowsky et al., 1982). A straight line results when the square root of the EGR is graphed for different growth temperatures. The equation for the model is:

Equation [3]

where EGR is the exponential growth rate (log₁₀ cfu/day), T is the growth temperature (°C), T₀ is the extrapolated minimum notational growth temperature (°C), and a is the slope parameter for Listeria monocytogenes in the specific food. T₀ values were estimated from four sources (Alavi et al., 1999; Duh and Schaffner, 1993; USDA, 1997 Pathogen Modeling Program; Wijtzes et al., 1993) and an average of these values (-1.18°C) was used in the model.

Different storage temperatures were used in the studies from the published literature that reported growth of Listeria monocytogenes in various foods. Therefore, using the data from these studies, equivalent EGRs (log₁₀ cfu/day) at 5°C were calculated. The equation, presented as Equation 4, is a ratio and rearrangement of Equation 3. The slope factor (a) is a constant and cancels out in the equation.

Equation [4]

where:

EGR₅ is the converted growth rate at 5°C,

EGR_lit is the growth rate from the inoculated pack study,

T₅ is set to 5°C to standardize the EGRs, and

T_lit is the temperature used in the literature.

If a category had five or more data points, variation was modeled by fitting statistical distributions to the resulting values (using the software program ParamFit). Paramfit employs ten different distribution models: Beta, Cauchy, Exponential, Gamma, Logistic, Lognormal, Normal, Rectangular (Uniform), Triangular, and Weibull. There is no theoretical support for any one distribution to be more appropriate than any other distribution. Therefore, the range of values generated by each of the ten statistical distributions reflects the uncertainty.

The 10 distribution models are used to construct a probability tree for the predictive model. Within an iterative simulation, the frequency of use of each model is allocated according to its relative model weight which is calculated as follows:

Model Weight = (((1 + n / Pn) ^O) x ((1 - gof) ^H)

Equation [5]

where

n = number of observations
Pn = number of model parameters
gof = Goodness-of-Fit
O = an arbitrary constant to describe parameter penalty, a value of 19 was used
H = An arbitrary constant to modify and provide a better fit, a value of 141 was used

ParamFit uses least residual squares for the predicted percentiles as the optimization criteria. The ratio of the sum of residual squares to the sum of total squares for the predicted percentile is used as a goodness-of-fit statistic. This approach fits the middle of the distribution, so that outliers have less impact on the shape of the distribution.

In some food categories (such as Dry/Semi-dry fermented sausages and Deli-type Salads), the Listeria monocytogenes levels decline at a slow rate. The rate of decline was modeled with the same square root model (Equation 3) as for growth with a negative slope (a) and a negative EGR. Negative EGR values from the literature were combined with positive data to create one distribution, which was fitted to the growth models as explained earlier. The rate of decline was adjusted for temperature, after being converted to a positive value, by the same ratio method of Equation 4. Increasing the storage temperature above 5°C increases the rate of decline and conversely temperature decreases below 5°C decrease the rate of decline. This approach agrees with the USDA Pathogen Modeling Program (USDA, 1997), which predicts faster rates of decline at higher storage temperatures. This relatively simple approach to modeling growth versus decline (survival) sufficiently accounted for the relatively slow rates of declines encountered in this risk assessment.

If all of the growth values were positive, the data were fit with all ten distributions and the four with the highest weights were used in the probability tree. If some of the growth values were negative (reflecting a possible decline in Listeria monocytogenes numbers), then the data were only fit with the Beta, Cauchy, Normal, Triangular, and Rectangular distributions as these are the only distributions of the ten that will accept negative values. Of these five distributions those with the three highest weights were used.

Several of the food categories had only two or three data points. Under this circumstance, probability trees were constructed with equiprobable rectangular or normal distribution. The maximum and minimum values were used as the parameters for the rectangular distribution. A standard algebraic formula was used to calculate the mean and standard deviation of the normal distribution.

Details on the variations and uncertainties used in the risk assessment for each food category are provided in Appendix 5. A value of zero for the EGR at all refrigeration temperatures is assigned to food categories that did not support growth (such as ice cream) and in which the pathogen levels remained stable over an extended period.

As an example, data from the Smoked Seafood food category (see Appendix 5) will be used to illustrate how the exponential growth rate of Listeria monocytogenes was calculated. Briefly, the data sets of EGR values at 5 °C are placed in order of ascending magnitude. Figure A5.1.2 (see Appendix 5) titled 'Cumulative Distribution for the Exponential Reference Growth Rate (EGR) at 5 °C,' is a cumulative frequency graph where the x-axis is the EGR in log₁₀ cfu/day and the y-axis is the fraction of data points from the literature with that value or lower (values are from Appendix 8). Different statistical distributions are fitted to the cumulative frequency distribution with the residual sums of squares for each frequency distribution used to weight the distributions. The probability column from Table A5.1.6 (see Appendix 5) indicates the weights for the four best-fitting distributions. In this example, the Lognormal and Gamma distributions have 40 and 31% of the weight, respectively. The Beta and triangular distributions had poorer fits and carried relatively little weight (16 and 13%, respectively). The probability of each growth model dictates the frequency of selection of each distribution for use in each uncertainty iteration during a Monte Carlo simulation (Cassin, et al., 1998; Vose, 1998). The variation predominantly reflects the shape(s) of the most heavily weighted statistical distribution.

Post-Retail Storage Times

The distribution of storage times were multiplied by the EGR to provide an estimate of the amount of Listeria monocytogenes growth occurring between retail purchase of the food and its consumption. Some foods are consumed on the day of purchase whereas others remain in the home refrigerator for lengthy periods of time. This is a major source of variability in the estimate of growth and ultimately, in the numbers of Listeria monocytogenes consumed. The variation in storage time was described using a modified BetaPert distribution (Figure III-6). A BetaPert distribution is defined by minimum, most likely, and maximum values. The distribution was modified by increasing the weight for the central value from 4 to 7. This modification reduced the frequency of values in the extended tails. The storage times were not used in the modeling for foods where Listeria monocytogenes does not grow. The uncertainty was described using a ±20% uniform distribution for the most frequent value and a ±50% uniform distribution for the maximum value, with a 100% correlation between the two distributions.

Figure III-6. Example of a Modified BetaPert Distribution

Frankfurters and Deli Meats

The survey sponsored by the American Meat Institute (AMI, 2001) asked for the average time consumers keep frankfurters and deli meats in the refrigerator. For example, 4% of the survey responders indicated that they stored frankfurters for an average of 11 to 14 days (Table III-6). This means that those responders consumed some individual servings of frankfurters after shorter storage times and others were kept longer than 14 days. While this is helpful information, what was needed for the model was the likely distribution of storage times for individual servings of frankfurters and deli meats. Thus, AMI data estimates inter-household variation. To get information on intra-household variation, consumers could be asked how long they stored the product the last time it was consumed. In order to introduce intra-household variation to the AMI data set, a log normal distribution was applied to the empirical AMI data points. The magnitude of the intra household variation, expressed as the Geometric Standard Deviation (GSD), ranged from 0.4 to 0.6 to be consistent with the data from the USDA/FSIS hotline study. The USDA hotline study asked for the 'last storage time' (Wachsmuth, 2000).

Figure III-7 shows a comparison of the USDA/FSIS hotline data (used in the draft risk assessment) and the AMI survey (indicated in the figure as 'individual average' data). The uncertainty bounds (GSD 0.2 to 0.6) are also shown in Figure III-7.

Figure III-7. Storage Time Distribution for Frankfurters and Deli Meats

Deli-type Salads

The data and assumptions behind growth estimates in deli salads were reexamined after the 2001 draft risk assessment. Data provided by Johnson (1993) and studies conducted in FDA's laboratories (Eblen, 2002a) showed that Listeria monocytogenes populations decline during the refrigerated storage of most deli foods. This is a consequence of processor-made salads having sufficient acidity and other preservatives to prevent growth. Locally- or store-made salads may not have these ingredients. The FDA studies indicated that growth only occurred in the shrimp and crab seafood salads. With the assistance of industry production data (Mitchell, 2001) the split between store-made and processor-made salads was estimated to be 15:85. It was also estimated that shrimp and crab salad are less than 10% of the total salad sales. Therefore, a triangular distribution of (1, 1.5, 3) was used to represent the fraction of deli salads that support growth and the uncertainty in that estimate. The growth rate at 5°C averaged 0.122 logs/day in the salads that supported growth and the declining rate averaged 0.143 logs/day in the majority of salads that did not support growth.

Modeling: Interaction of Storage Times and Temperatures

Increases in the levels of Listeria monocytogenes were calculated as the product of the EGR (which is dependent on the refrigeration temperature) and storage time. The Monte Carlo simulation program randomly selects different values from each calculated distribution. Both temperature and time distributions are concentrated toward the center of their ranges, 4°C and 8 days, respectively for Smoked Seafood. As a result, the most frequent estimates of growth would reflect these conditions. The simulation process would also select, at a lower frequency, the combination of low refrigeration temperatures and short storage times. Such combinations would result in relatively little growth. Similarly, the process could select high refrigeration temperatures and long storage times, 10°C and 45 days, which would result in extensive growth. However, this combination of temperatures and times would likely result in the food showing obvious spoilage and hence would not be consumed. Modeling the refrigeration temperature and storage time distributions as independent distributions was not believed to be appropriate. Therefore, the uncertainties in the mode and maximum storage times were negatively correlated to the temperature. For example, for Smoked Seafood, this means the mode ranged from 6 to 10 days. When refrigeration temperature was 10°C, the time was 6 days and when the temperature was 0°C the time was 10 days. The maximum storage time similarly ranged from 15 to 45 days for 10°C and 0°C storage, respectively. This means that at higher temperatures the distribution for storage times is much compressed relative to the distribution at lower temperatures.

Maximum Growth Levels

Growth is the product of the EGR (at a specific temperature) and the storage time. For each iteration of the Monte Carlo simulation, the logarithm of growth estimated during storage was added to a contamination level at retail. No lag phase was calculated; it was assumed that the Listeria monocytogenes cells were already in the food and adjusted to the food's environment during the period before retail purchase. The only change made from retail to storage was to a new refrigerator temperature. This relatively small change would take several hours for a packaged food and the cells would effectively adjust as the temperature changes.

The populations for the stationary phases of Listeria monocytogenes in foods were obtained from the published literature and were used to establish limits for the maximum calculated growth levels for each food category (Appendix 8). If the calculated level for Listeria monocytogenes exceeded the maximum level, the maximum value was used. The literature also indicated that the maximum growth level is dependent upon the storage temperature. However, there were only a few studies in the literature that provided for the growth in a food to the stationary phase at more than one temperature to permit accurate estimation of this behavior.

Duffes et al. (1999) showed maximum levels (cfu/g) in smoked salmon to be less than 10⁵ at 4°C and 10^8.1 at 8°C. Pelroy et al. (1994a) found maximum levels in smoked salmon to be 10⁵ and 10^6.5 at 5 and 10°C, respectively. Maximum populations were reported in cream as 10⁷ and 10^7.5 at 4 and 8°C, respectively (Rosenow and Marth, 1987); in butter it was reported as 10^5.5 and 10⁶ at 4 to 6 and 13°C, respectively (Olsen et al, 1988); and in lettuce it was reported as 10 ⁵ to 10 ^5.5 and 10^6.5 to 10^7.5 at 5 and 10°C, respectively (Beuchat and Brackett, 1990b). In addition to direct comparisons, a collection of individual growth studies also showed this tendency to grow to higher population levels at higher temperatures.

The maximum growth levels (cfu/g) used were applied across all food categories with 10⁵, 10^6.5 and 10⁸ used as maximums for temperatures of <5, 5 to 7 and >7°C, respectively. For milk, sufficient data were found in the literature for growth levels of 10⁷, 10^7.5 and 10⁸, to use as the maximums for the three temperatures, respectively. A uniform range of one logarithm was used to represent the uncertainty for each of the maximum growth levels. The calculated growth levels were added to the retail contamination levels during each iteration of the model, and these new levels of Listeria monocytogenes contamination in food were compared to the maximum growth level. If the calculated growth level exceeded the maximum growth level in any iteration, the amount of growth was reduced to the maximum growth level.

Modeling: Thermal Inactivation

Frankfurters have been implicated in outbreaks of listeriosis although they are generally reheated before serving. Because they are precooked during manufacturing, frankfurters are considered to be a RTE food. Reheating will kill Listeria monocytogenes in food. Frankfurters are usually, but not always, reheated before consumption. Therefore, a thermal inactivation step was included in the model for frankfurters. The frequency of insufficient reheating and the extent of inactivation of Listeria monocytogenes when not properly reheated were estimated in this step of the model.

No data describing the prevalence or extent of under-reheating of frankfurters has been published. However, the Georgetown survey (n=90) found approximately 1% of the respondents did not reheat their frankfurters (Wachsmuth, 2000). In an FSIS Hotline survey, 14% of the respondents indicated that at least one household member has eaten frankfurters directly from the package (Wachsmuth, 2000).

Some frankfurters are frozen by the consumer when they are brought home from the retail store. Information on the proportion of frozen frankfurters from the AMI survey and FDA Food Safety survey (Lando, 2003) led the risk assessment team to assign a uniform distribution from 3.0 to 8.7 % to represent this proportion and its uncertainty. To the frozen portion of frankfurters, the growth of Listeria monocytogenes would be set to zero, that is, the bacteria don't grow or die during the frozen storage. The time of storage would be irrelevant. It was assumed that all of the frozen frankfurters would be reheated before consumption. Therefore, the distribution of Listeria monocytogenes inactivation used for part of the non-frozen frankfurters was applied to all of the frozen frankfurters.

The final distribution of Listeria monocytogenes consumed per serving in reheated frankfurters is the summation of the respective proportions of the frankfurters stored frozen and reheated and the frankfurters stored refrigerated and reheated. The number of cases per annum was calculated from the total number of frankfurter servings and the proportion of the total in these two groups. The distribution of Listeria monocytogenes consumed per serving in non-reheated frankfurters represents the remaining proportion, represented by a triangular distribution of (4, 7, 10) percent of the non-frozen frankfurter servings (uncertainty distribution).

It was recognized that frankfurters are reheated in boiling water and microwave ovens more frequently than grilling, and that frankfurters are more likely to be contaminated on the surface than the interior. The Georgetown survey showed that 20% of the frankfurters were microwaved; the percentage of all responses for the FSIS Hotline was 19.4% with 4.7% microwaved less than 1 minute (Wachsmuth, 2000). In a preliminary experiment conducted by FDA/CFSAN, the heating of frankfurters by microwave ovens was measured with low (600 W) and high (1,100 W) powered microwave ovens (Buchanan, 2000). Four types were tested, including chicken frankfurters, low salt frankfurters, and two different size diameter frankfurters. Using various combinations of the two microwave power settings and four types of frankfurters, it was shown that the surface temperature increased faster than the center temperature. Heating for 25 seconds in the high power oven (1,100 W) and 40 seconds in the lower power oven (600 W) raised the surface temperature to at least 59 °C and, in some cases, raised the surface temperatures to over 70 °C. There is no information on what combinations of heating times and temperatures are actually realized by consumers, but this preliminary experiment suggests that microwave heating is likely to be sufficient to cause substantial inactivation of any Listeria monocytogenes that might be present.

Inadequate data were found with which to directly model thermal inactivation in the frankfurters that receive some heating by microwaving, boiling, frying, grilling, broiling or other means. Therefore, data from inactivation of E. coli O157:H7 in hamburgers were adapted (Juneja et al., 1997). These authors determined that survival of E. coli O157:H7 after cooking to maximum temperatures ranging from 54 to 77°C (129 to 171 °F) may be estimated by this equation:

log₁₀ survivors = 20.53 - 0.12 T

Equation [6]

The maximum cooking temperature to calculate the decrease (T) is in degrees Fahrenheit. Because the initial contamination was 6.6 logs, the equation can be rearranged to calculate the decrease in contamination and applied to any initial level of contamination. The temperature was also converted into degrees Celsius:

log₁₀ reduction = 0.216 (T - 46.4)

Equation [7]

A standard deviation of 0.5 logs was used to represent the uncertainty in the estimated reduction. This value reflects the sampling error from a similar experiment with E. coli O157:H7 (Jackson et al., 1996) where a 4.1 log₁₀ reduction was observed after cooking to 68.3°C.

Reductions in Listeria monocytogenes levels were calculated by estimating a distribution of cooking temperatures with a triangular distribution having a minimum of 54 °C, most frequent temperature in the range of 69 to 73 °C, and a maximum of 77 °C. The four-degree range for the most frequent temperature represents uncertainty in the cooking temperature distribution. Table III-15 contains the resulting cumulative distribution in log reductions for the frankfurters that were given some reheating. The remainder had no reduction in Listeria monocytogenes after the growth modeling.

Results: Modeled Listeria monocytogenes Levels in Food at Consumption

The estimated levels of Listeria monocytogenes at consumption are presented on Table III-16. This table has the same format as the table for Listeria monocytogenes contamination at retail (Table III-5), and may be directly compared to it to observe the shift in levels of Listeria monocytogenes after storage and/or heating. The median percentage of servings that fall within designated ranges of Listeria monocytogenes levels per serving are presented. The actual simulation modeling was at narrower levels (every logarithm and half-logarithm cfu/serving). The 5 and 95% values for the distributions for Listeria monocytogenes levels in each food are also given. These distributions indicate the uncertainty in the value for each median. The distribution observed with the values across a row gives the variation in Listeria monocytogenes levels expected for each food category. Because these medians are from skewed uncertainty distributions and because of rounding errors, a row may not sum to exactly 1.00.

As shown previously with the retail contamination estimates, every food category has some fraction of servings with at least 1 cfu/serving. The food categories range from 0.1% in hard cheeses to 8.7% in raw seafood. The column in Table III-16 showing 10⁶ to 10⁹ Listeria monocytogenes per serving is the level where the occurrence of listeriosis would be expected to be most likely. Smoked Seafood, Cooked RTE Crustaceans, Frankfurters not reheated, Deli Meats, and Pâté and Meat Spreads categories comprise a group of foods estimated to have the greatest likelihood of containing 10⁶ to 10⁹ Listeria monocytogenes per serving. These levels are illustrated in Figure III-8. The row in the rear represents the fraction of servings with <1.0 cfu Listeria monocytogenes. All of the food categories have more than 90% of their servings in this contamination range. The rows have increasing levels of contamination toward the front of the figure.

Comparing corresponding values in Tables III-14 and III-16 shows the predicted effect of storage on the levels of Listeria monocytogenes at consumption. Cooked RTE Crustaceans, Frankfurters (not reheated), Deli Meats, and Pâté and Meat Spreads have some of the most dramatic changes. For example, at retail, 1.0% of Deli Meat servings would be in the 10³ to 10⁶ cfu/serving group. This increases to 1.6% at the time of consumption. In addition, the reduction in Listeria monocytogenes from reheating frankfurters is evident by comparing the <1, 1-1000 and 10³ to 10⁶ cfu/serving groups in Table III-16. The levels of Listeria monocytogenes in foods that do not permit growth, such as ice cream, do not show a change in comparing the values in Table III-14 (at retail levels) and Table III-16 (at consumption levels).

An approximation of the overall frequency of consumption of Listeria monocytogenes by the United States population can be made by multiplying the fraction of servings in each food category-dose bin (Table III-16) by the annual number of servings in each food category (Table III-2). The numbers of servings are then summed for each dose for all of the food categories. Table III-17 shows that most of the servings have less than one Listeria monocytogenes and the number of contaminated servings decreases with increasing levels of contamination. If the number of contaminated servings is divided by the United States population (2.6 x 10⁸), an approximation of how frequently the "average person" would encounter these levels of Listeria monocytogenes each year can be calculated. This "average" person would consume a serving with 10³ to 10⁶ microorganisms 2.4 times per year, 10⁶ to 10⁹ microorganisms once every two years and more than 10⁹ microorganisms once every three years.

Description of Image

Figure III-1: Components of the Exposure Assessment Model

This diagram shows the components of the exposure assessment model. It begins with contamination levels at retail (cfu/g). Four factors used to determine the contamination levels are retail--enumeration data, listeria monocytogenes distribution in food, percentage frequency data and an adjustment for study age, region, and size. Next the post-retail growth is calculated from four factors: exponential growth rate, storage time, home refrigerator temperature, and maximum growth. Next the contamination level at consumption (log cfu/g) is calculated, from this level the dose at consumption (log cfu.serving) is calculated in consideration of the impact of cooking (frankfurters) and serving size.

Return to image