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

V. Risk Characterization

(Also available in PDF)

Risk characterization integrates information and data acquired during the hazard identification, hazard characterization, and exposure assessment into an estimate of the adverse effects likely to occur in a given population. In this risk assessment, the risk characterization links the probability of exposure to Listeria monocytogenes from consumption of foods with the adverse health outcomes. The primary focus is on the prediction of the relative probability of contracting listeriosis from consumption of a single serving of food in one of the 23 food categories. Additional predictions also consider the extent of annual consumption of the various foods and the predicted contribution of each of the individual food categories to the number of listeriosis cases nationally.

This risk assessment is based on contaminated foods at the retail level. The risk characterization of the overall burden of listeriosis on public health includes both sporadic (i.e., illnesses not associated with a documented outbreak) and outbreak illnesses. Illnesses attributed to documented outbreaks are a small proportion of the total estimated annual cases of listeriosis. At this time it is not possible to separate the risk attributable to specific foods to sporadic and outbreak cases. Outbreaks frequently represent a breakdown in food production, manufacturing, or distribution systems instituted to prevent Listeria monocytogenes contamination. Assessing the likelihood that these systems will fail requires detailed information about the manufacture of individual foods that is beyond the scope of this assessment. However, an important benefit of conducting a risk assessment is the identification of knowledge and data gaps. Continuing research is needed to facilitate future Listeria monocytogenes risk assessment work (see Appendix 11: Reseach Needs).

Simulation Modeling

The model is comprised of two major components--exposure and dose-response. These models are integrated for the risk characterization simulations as shown in Figure V-1.

Figure V-1. Components of the Risk Characterization Model

A separate exposure simulation was constructed for each food category. Results from all the food categories were then carried forward to the dose-response simulations, where a separate simulation was constructed for each of the three subpopulation groups. Details of the various modeling steps are provided in Appendix 3.

The exposure assessment modeled the effect of various factors (e.g., frequency and extent of contamination at retail, consumption patterns, the growth potential of Listeria monocytogenes in foods, length of refrigerated storage, and refrigeration temperatures) that might affect levels of Listeria monocytogenes contamination in a food at the time of consumption. For the exposure assessment, a two-dimensional Monte-Carlo simulation (100,000 variability and 300 uncertainty iterations; total of 30,000,000 iterations) was used to integrate the components of the exposure model for each of the food categories. The result of each exposure simulation is the fraction of servings that occur at designated dose levels (broken out into half-log₁₀ intervals), which are referred to as dose bins. The conversion to dose bins was necessary in order to integrate the exposure simulation, (which evaluated the exposure from individual servings) with the dose-response model (which predicted the number of cases at a population level). The exposure simulations produce 300 distributions (sets of dose bins) of predicted doses for each food category.

The dose-response simulation was carried out in several steps. First, the two-dimensional Monte-Carlo simulation (100,000 variability and 300 uncertainty iterations) was used to integrate the variability and uncertainty of the strain-virulence and host susceptibility functions for each of the subpopulations to provide dose-adjustment factors. The variability dimension for these combined dose-adjustment factors were then grouped into half-log₁₀ bins, which ranged from -5 to +10 logs. Second, a one-dimensional (4,000 uncertainty iterations) dose-response simulation was run for each food category by selecting one of the 300 sets of dose bins from the exposure assessment.

These two sets of distributions (exposure dose bins for each food category and dose-response scaling factors for each subpopulation) consist of a relatively small set of finite values and were combined algebraically by adding the arrays. Although some resolution was lost through the creation of the bins for the distribution, avoidance of the use of random numbers provides greater precision at the tails of the summed distribution. In order to calculate the annual rates of cases of listeriosis, the number of deaths per year were multiplied by factors of 11.3 for intermediate-aged population, 12.7 for neonatal, and 3.7 for the elderly population. To calculate the number of perinatal deaths per year, the neonatal death estimate was multiplied by a factor of 2.5. The 2.5 is the approximate ratio of perinatal (106) to neonatal (41) deaths from the County of Los Angeles Department of Health Services (Buchholz, 2000).

The dose-response scaling factor was adjusted so that the sum of the dose-response function (derived from the mouse model) times the exposure assessment doses equaled the CDC estimates for the annual number of cases of listeriosis. This procedure anchors the overall predicted incidence of listeriosis with the actual incidence of listeriosis. An implicit assumption is that the foods encompassed by the 23 food categories account for all cases of foodborne listeriosis.

The medians of the 4,000 iterations of predicted deaths (per serving) for each food category and each subpopulation were reported. These predictions were ranked from highest to lowest. Because of the variability incorporated into the model (i.e., from differences in consumption of the foods in each categories, pathogen virulence, host susceptibility, and inherent uncertainty), the predicted relative ranking of food categories changes with each of the 4,000 iterations (in some cases significantly). To illustrate the degree of uncertainty associated with the relative risk ranking, the results of each set of the 4,000 iterations was ranked and compared. To this end, the ranking of each food category from 1^st to 23^rd was determined for each set of the 4,000 uncertainty iterations. The number of times each food category was observed to be ranked at each specific position was determined. These data were compiled and presented graphically (see the latitude graphs, Figures V-4a to V-26b in the section below titled "Summaries of the Food Categories").

For a more detailed explanation of two-dimensional Monte-Carlo and the dose-binning process, see Appendix 3.

Results

The results of this risk assessment, the predicted relative risks of listeriosis associated with each food category, are presented first as an initial overview followed by a more detailed consideration of the individual food categories. The individual food category discussions further interpret the meaning and significance of the analyses in relation to the goal of the risk assessment, as well as discuss factors contributing to the variability and uncertainty associated with the predictions.

A significant difference between the FDA/FSIS risk assessment (the 2001 draft and this revised version) approach and prior attempts to evaluate the risks associated with ready-to-eat foods is the complexity of factors considered in the hazard characterization (Lindquist and Westöö, 2000, Buchanan et al., 1997; Farber et al., 1996; Haas et al., 1999; Hitchins, 1995 and 1996; and Teufel and Bendzulla, 1993). In addition to establishing a general dose-response relation, models were developed for three distinct age-based subpopulations and for assessing the full range of virulence potential that is likely to occur among Listeria monocytogenes isolates. It also emphasizes the fact that most exposures to Listeria monocytogenes seldom lead to listeriosis, even among highly susceptible segments of the population.

Medians (the value with 50% of the values above and 50% of the values below) are used to represent the "expected" (central tendency) of the estimated risk values. We used medians rather than means because the distributions have long "tails" (high uncertainty and skewed distributions). Medians are less influenced by these extreme values in the distribution but still allow us to represent the central tendency of the distribution with a single value. For other purposes, such as summing the food categories or additional calculations, the mean values are provided in Appendix 10.

Risk Per Serving

A key value used to assess the predicted relative risk among the 23 food categories is the "per serving" likelihood that consumption of a food will lead to listeriosis. This can be viewed as the risk that individual consumers face when they eat a serving of a food. The risk assessment results indicate that listeriosis could potentially be caused by foods in any of the food categories; that is, no food category is risk-free because almost any food could become contaminated with Listeria monocytogenes. It is equally apparent that there are substantial differences in risk among the different food categories.

As anticipated from the review of the scientific literature that was conducted in conjunction with this risk assessment, five factors have a large influence on the results of the exposure assessment and thus, the characterization of the predicted relative risk. These factors include the following.

• Frequency and extent of Listeria monocytogenes in the food

• Amounts and frequency of food consumption

• Potential for growth of Listeria monocytogenes in food during refrigerated storage

• Duration of refrigerated storage before consumption

• Temperature at which the food was held during refrigerated storage

Any of these factors alone affects the potential contamination level at consumption. Those food categories in which one or more of these factors produce a greater risk of exposure to higher levels of Listeria monocytogenes contamination are more likely to increase consumers' risk of listeriosis. Examination of the food categories shows that certain factors may have a larger role in driving the predictions of higher risk. Food categories that contained foods that have a high growth potential, based on moderate or high growth rates, coupled with moderate or long storage times, were often the categories that had higher predicted relative risk values. These results have to be interpreted being cognizant of the fact that data on actual consumer storage practices were generally not available, so storage times were estimated based on expert judgment and USDA recommended practices. It is likely that the actual consumer storage times of food are longer than USDA recommendations.

As previously indicated in the description of the exposure assessment, other assumptions related to factors that could affect the frequency or extent of contamination could have a significant impact on the predicted relative risk per serving associated with individual food categories. These, in turn, could affect the predicted relative risk rankings of other food categories. For example, during manufacturing frankfurters are fully cooked to temperatures that are lethal for Listeria monocytogenes. However, subsequent recontamination prior to packaging may occur followed by growth of the pathogen. Although frankfurters are usually reheated prior to consumption, a portion of the population consumes them without reheating. To estimate the proportion of frankfurters consumed unreheated, a triangular distribution was used with a minimum of 4%, most likely of 7% and maximum of 10%. The impact of these types of assumptions on the predicted relative risk is considered in the discussion of the individual food categories.

Predicted Cases of Listeriosis per Serving. The results are summarized in Table V-1 as the median number of cases of listeriosis per serving for each of the three age-based subpopulations and the total United States population. Figure V-2 also shows the differential in median predicted risk per serving (the median values on a log scale are represented in the graph as a box) for the total United States population. The figure illustrates the point that elimination of Listeria monocytogenes from any single food will not eliminate foodborne listeriosis; control of listeriosis will require consideration of a variety of foods. However, some foods represent a substantially greater risk per serving and are likely to warrant additional attention from industry and regulators.

In addition to the median values, the 5^th and 95^th percentile values were also calculated for each of the subpopulations and the total United States population (Table V-1). These lower and upper bound values provide a method of estimating the uncertainty associated with the predictions. Figure V-2 shows these lower and upper bounds for the total United States population. In order to more easily present the data in a graph, the cases of listeriosis for each of the food categories is presented in Figure V-2 on a log scale. It is apparent that for some foods, the range covered was substantial. This was largely the result of exposure distributions where either a small percentage of the foods were predicted to have elevated levels of the pathogen or a high degree of uncertainty had to be assumed due to limitations in available data. The predicted relative risk values must be evaluated in relation to observed variability and uncertainty when using them to determine the best course of action for each of the different food categories. This interpretation of the results is discussed in greater depth for each of the individual food categories later in this chapter.

Figure V-2. Predicted Cases of Listeriosis (log scale) Associated with Food Categories
for the Total United States Population on a per Serving Basis

[The box indicates the median predicted number of cases of listeriosis (log scale) and the bar indicates the lower and upper bounds (i.e., the 5^th and 95^th percentiles). The y-axis values are presented on a log scale. For example a log of -6 is equivalent to 1 in a million.]

DM = Deli meats; FNR = Frankfurters (not reheated); P= Pâté and Meat Spreads; UM= Unpasteurized Fluid Milk; SS= Smoked Seafood; CR = Cooked Ready-To-Eat Crustaceans; HFD = High Fat and Other Dairy Products; SUC = Soft Unripened Cheese; PM = Pasteurized Fluid Milk; FSC = Fresh Soft Cheese; FR = Frankfurters (reheated); PF = Preserved Fish; RS = Raw Seafood; F = Fruits; DFS= Dry/Semi-dry Fermented Sausages; SSC = Semi-soft Cheese; SRC = Soft Ripened Cheese; V = Vegetables; DS = Deli-type Salads; IC= Ice Cream and Frozen Dairy Products; PC = Processed Cheese; CD = Cultured Milk Products; HC = Hard Cheese.

Predicted Risk Ranking. The predicted median values for the cases of listeriosis on a per serving basis were used to develop predicted relative risk ranks. The median predicted relative risk ranking among the different food categories is summarized for the three subpopulations and the total United States population in Table V-2. It is apparent that the predicted relative risk rankings of the food categories are similar for the three subpopulations, but not identical.

The uncertainty associated with the risk ranking is described in the latitude ranking graphs that are presented as part of the discussion of each of the individual food categories (see Figures V-4a to V-26b). It is important to note that in a number of instances there are only minor differences separating the rankings of various food categories.

Although the number of iterations in the ranking process was very high (4,000), analysis of variance techniques were used to provide an indication of the statistical certainty of the rankings. Nonparametric analysis of variance technique (i.e. Kruskal-Wallis Test), followed by a multiple comparison procedure, was used to evaluate the differences in the median rankings of risk per serving for the total United States population. The analyses were performed using NCSS (NCSS, 2001) to determine which of the median rankings were not significantly different based on the number of simulation samples (iterations) and an alpha level of 0.05 for the family-wide error rate with respect to all pairwise comparisons of the 23 food categories. The results are shown in Table V-2.

Risk per Annum

A full picture of listeriosis risk requires consideration of the number of servings consumed, as well as the risk per serving. These data were considered for each of the food categories and used to calculate the predicted cases of listeriosis on a per annum basis. If the "risk per serving" is considered the predicted relative risk faced by each consumer, then the "risk per annum" is a measure of the predicted relative risk faced by the country. The risk per annum is greatly affected by the number of servings per year. Thus, a food that has a relatively high risk on a per serving basis but is seldom consumed may have a relatively low per annum risk. Conversely, a food with a relatively low risk on a per serving basis that is consumed extensively is likely to have a higher risk on a per annum basis. Table III-2 shows the wide range in number of annual servings among the food categories. The per annum relative risks inherently have a greater degree of uncertainty than the corresponding per serving relative risk because of the additional uncertainty associated with the number of annual servings. Another factor that affects predicted relative risk on a per annum basis is the size of the subpopulations, in proportion to the total population. They are substantially different, i.e., perinatal, elderly, and intermediate-age groups, represent approximately 2%, 13%, and 85% of the total population, respectively.

The results were generated in a manner similar to that described above for the predicted relative risk per serving. Table V-3 provides the predicted median number of cases of listeriosis on a per annum basis for each of the age-based populations. The upper and lower bounds (5^th and 95^th percentile values) are also provided in Table V-3 to show the range of variability and uncertainty of the estimates. The range in the predicted number of cases of listeriosis is depicted in Figure V-3 for the total United States population.

The predicted relative risk ranking is presented in Table V-4. The uncertainty associated with the ranking is also described using individual latitude ranking graphs based on the rankings for the total United States population (see Figures V-4a to V-26b). These graphs are provided in the discussions of individual food categories. It is important to note that the differences among several of the food categories were very small, so differences between adjacent or closely occurring ranks must only be considered in conjunction with the estimates of uncertainty which are provided as part of the discussion of the individual food categories.

In most instances, the food categories that had high predicted relative risk rankings on a per serving basis also had a high predicted relative risk ranking on a per annum basis. However, there were instances where foods with lower risk per serving rankings had higher risk per annum values and vice versa. For example, Pâté and Meat Spreads had a higher predicted relative risk on a per serving basis than on a per annum basis. This reflects the fact that foods in this category are eaten relatively infrequently and in relatively small amounts. Conversely, Vegetables and Pasteurized Fluid Milk are products where a predicted low or moderate per serving relative risk was elevated on a per annum basis. In these examples, this appears to be a function of two factors. The first is the variability in the data sets available on a worldwide basis (see discussion of individual foods in the section titled "Overview and Discussion of Food Categories"). A wide degree of variability increases the number of predicted exposure values in the "tails" of the distribution. To a large extent, it is these extremes of the distributions that determine the per annum risk. The second is that the numbers of servings consumed annually for Vegetables and Pasteurized Fluid Milk are several orders of magnitude higher than other food categories. Again, this strongly influences the per annum predicted relative ranking for these foods. With both of these food categories, the results of the risk assessment must be interpreted in relation to the uncertainty estimates. The best interpretation may be the need to assure continued vigilance. However, these data do demonstrate how a risk assessment can provide a means of systematically examining risks from different vantage points. The results clearly point out that a relatively low predicted relative risk per serving associated with foods that are consumed extensively (such as Pasteurized Fluid Milk or Vegetables) could lead to a potentially greater impact on the relative risk of listeriosis per annum.

Figure V-3. Predicted Cases of Listeriosis (log scale) Associated with Food Categories
for the Total United States Population on a per Annum Basis

[The box indicates the median predicted number of cases of listeriosis (log scale) and the bar indicates the lower and upper bounds (i.e., the 5^th and 95^th percentiles. The y-axis values are presented on a log scale. For example a log of -3 is equivalent to 1 in a thousand.]

DM = Deli meats; PM = Pasteurized Fluid Milk; HFD = High Fat and Other Dairy Products; FNR = Frankfurters (not reheated); SUC = Soft Unripened Cheese; P= Pâté and Meat Spreads; CR = Cooked Ready-To-Eat Crustaceans; UM= Unpasteurized Fluid Milk; SS= Smoked Seafood; F = Fruits; FR = Frankfurters (reheated); V = Vegetables; DFS= Dry/Semi-dry Fermented Sausages; FSC = Fresh Soft Cheese; SSC = Semi-soft Cheese; SRC = Soft Ripened Cheese; DS = Deli-type Salads; RS = Raw Seafood; PF = Preserved Fish; IC= Ice Cream and Frozen Dairy Products; PC = Processed Cheese; CD = Cultured Milk Products; HC = Hard Cheese.

Overview and Discussion of Food Categories

Because Listeria monocytogenes is ubiquitous in foods and the food-processing environment, a large number of foods needed to be considered in this risk assessment. In order to have a practicable number of food groupings, 23 categories were formed from the more than 640 ready-to-eat foods in the consumption surveys. These categories are sometimes broadly defined to include several distinct but similar classes of food, while in other instances they are quite small and specific. The foods included in this risk assessment are primarily organized into categories based on primary origin of the foods (e.g., seafood, vegetable, dairy, meat), composition and processing (moisture content, raw vs. cooked, pH, salt level), contamination with Listeria monocytogenes, and association with listeriosis. Although generally similar, some characteristics of foods within a single category may vary. For example, within a single food category, consumption may be greater for one food, contamination higher in another, and average rate of growth in a third food. In the future, if further investigations of an individual food category or a particular food within a category are conducted, the model developed in the current risk assessment could be modified to provide a more detailed analysis.

Consumption

Consumption estimates on a per serving basis were determined, as well as the amount of food eaten per person per day. Data indicate that, for the one or two days of the consumption surveys, there were 1.8 x 10⁹ servings consumed of the foods identified in the 23 categories. Extrapolated to an annual basis, there were 3.4 x 10¹¹ servings consumed in a year. The vast majority (96.3% or 2.5 x 10⁸ individuals) of the population reported eating the foods included in this risk assessment. There were a relatively low number of eaters for some of the food categories (e.g., Smoked Seafood, Fresh Soft Cheese, Pâté and Meat Spreads), while other food categories are consumed widely and often (e.g., Pasteurized Milk, Vegetables). Consumption information for each food category is included in the discussion below.

Contamination

Contamination levels at retail ranged from less than 0.04 cfu/g to more than 10⁶ cfu/g in the food data considered in this analysis. The highest levels reported for specifically identified food products were in the range of 10⁵ to 10⁶ cfu/g, although the results of laboratory investigations indicate that contamination levels greater than 10⁶ cfu/g can occur. Studies that were limited to the determination of presence or absence were assigned a contamination value commensurate with the lowest limit of detection possible: 0.04 cfu/g. The highest frequency of contaminated samples was 12.9 % (Smoked Seafood). All food categories demonstrated some contamination, with a range of positive samples from 0.2 % to 12.9 % (see Table III-4). The frequency of occurrence of contaminated samples was lower at higher contamination levels. The contamination studies used in this study were published over a period of seventeen years (1985-2002). Because there was a major effort worldwide to control foodborne listeriosis, the incidence of contamination was evaluated for differences in data published pre-1993, 1993 to 1998, and post-1998. To estimate the current variation in contamination, studies were weighted by number of samples, country, and date of publication as explained in Chapter III: Exposure Assessment. Food categories with no recent data were adjusted by a factor relative to the other food categories.

Growth of Listeria monocytogenes

To predict possible growth between retail sampling and consumption, a growth model was created, based on growth rates from studies of various foods inoculated with Listeria monocytogenes under laboratory conditions. These studies were conducted at a number of temperatures. The reported growth rates were adjusted to give the equivalent growth rate at 5°C. Within each food category, the adjusted Exponential Growth Rate (EGR) from individual studies was used to develop a distribution of growth rate values. As previously mentioned, little data were available that adequately described the distribution of storage times (except for frankfurters and deli meats). Therefore, a modified BetaPert distribution was created for each food category, with minimum, most likely and maximum times (days) to account for the variation in storage times. The minimum time for all food categories (0.5 day) represents food consumed within 24 hours of purchase. For each specific food category, the most likely and maximum values were given an uncertainty range. For frankfurters and deli meats, an empirical data set was used (AMI, 2001). For each iteration of the growth simulation, the model selected a refrigeration storage temperature (that varied from 1 to 11°C) and calculated the EGR (log₁₀ cfu/day) at that temperature. The EGR was multiplied by the storage time to estimate growth from retail to consumption and the estimated growth was added to the initial number of Listeria monocytogenes to calculate the total Listeria monocytogenes. The projected growth was limited by temperature-dependent maximum growth values (stationary phase). The maximum growth was greater at higher storage temperatures than at lower temperatures. In addition, the model contained a negative correlation between storage temperature and storage times. This minimized combinations of long storage times and high temperatures that would most likely result in detectable spoilage from other microorganisms and disposal of the food rather than consumption.

Summaries of the Food Categories

Because the risk assessment model is based on many parameters and an extensive amount of both qualitative and quantitative data, it can be difficult to determine the impact of each of the factors considered. Accordingly, sets of qualitative descriptors were developed to aid in the discussion and comparison of these parameters in the food categories. The criteria used to characterize data among food categories as low/moderate/high or short/moderate/long for each parameter are presented in Table V-5a. Table V-5b provides a characterization of each of the parameters for each food category. See Appendices 5, 7 and 8 for the supporting data.

An overview of each of the 23 food categories is provided in this chapter including information for each food category on cases of listeriosis, consumption, contamination, and growth of Listeria monocytogenes, and a summary of the designated parameter levels based on the criteria listed in Table V-5a. In addition, the latitude graphs (Figures V-4a to V-26b) show the uncertainty associated with the predicted relative risk rankings on both a per serving and per annum basis for each food category. These graphs show how frequently a food category ranked 1^st, 2^nd , and so on to 23^rd. A food category that primarily ranked 1^st or 2^nd should be considered a higher risk than a food category that primarily ranked 22^nd or 23^rd. The distribution of rankings shown for a food category is an indication of the certainty of its ranking. The narrower the range, the greater is the certainty associated with the relative risk ranking.

As an initial means of categorizing the results of the risk assessment in order to relate them to the characteristics of the different food categories, the relative predicted risk on a per serving basis was classified as high, moderate, or low. The following criteria was used: high = >5 predicted cases of listeriosis per billion servings; moderate = <5 but 1 predicted case per billion servings, and low = <1 predicted case per billion servings. Based on these criteria, five of the foods were considered to be high risk, four were in the moderate risk group, and the remaining foods fell into the low risk per serving category (Table V-6). The number of predicted cases per annum in the United States for the total population was classified as low (less than 1 case per annum), moderate (>1 to 10 cases per annum), high (>10 to 100 cases), and very high (>100 cases). Based on these criteria, one food category was considered very high, three food categories were considered to cause a high number of cases and five food categories a moderate number of cases, with the remaining considered low. Additional means of grouping the results are considered later in the document (see the cluster analysis in Chapter VII).

Table V-5a. Criteria Used to Designate Parameter Ranges for Listeria monocytogenes Among the Food Categories

Parameter	Designated Parameter Level
	Low/Short	Moderate	High/Long
Number of Annual Servings	1 x109	> 1x109 to < 1 x 1010	1 x 1010
Median Amount Consumed per Serving (g)	40 g	> 40 g to < 90 g	90 g
Contamination Frequency (%)	2%	> 2% to < 5%	5%
Contamination at Retail--Predicted Servings at 103 to 106 cfu (%)	0.1%	>0.1% to < 0.6%	0.6%
Exponential Growth Rate at 5 °C (log10 cfu/day)	0.1	> 0.1 to < 0.2	0.2
Most Likely Storage Time (days)	2 days	> 2 to 5 days	6 to 10 days

Food Category: Smoked Seafood

The foods in the Smoked Seafood category had a high predicted relative risk of causing listeriosis on a per serving basis. This reflects the fact that Smoked Seafood has a high frequency of contamination; high levels of contamination at retail, supports a moderate rate of growth; and is often stored for moderate lengths of time (and occasionally long periods of time). This is offset somewhat by the moderate serving sizes and the low number of servings associated with this food category. These combine to make Smoked Seafood a moderate contributor to the total number of predicted cases of listeriosis per year.

The predicted relative risk per serving for Smoked Seafood is consistent with various smoked seafoods having been associated with listeriosis. Smoked mussels have been linked to outbreaks of listeriosis in Australia and New Zealand, cold smoked rainbow trout to an outbreak in Sweden, smoked salmon to sporadic cases in Australia, and smoked cod roe to sporadic cases in Denmark (Ryser, 1999a; Brett et al., 1998; Ericsson et al., 1997). Contaminated retail packages are regularly identified by regulatory surveillance programs (Ryser and Marth, 1999a). However, the small volume of most production lots and a low number of servings consumed means that outbreaks are unlikely from a contaminated product; sporadic cases would be expected to be the typical consequences of Listeria monocytogenes in this food category.

Foods included in this category from the consumption databases are smoked salmon, trout, herring, oysters, and other smoked fish not identified as to species. Both hot and cold smoked products are included in this category, in part because the consumption databases do not distinguish between these two processes. The predicted median amount consumed per serving for this category is 57.0 g (approximately 2 ounces), and the annual total number of servings in the United States is only 2.0x10⁸ (i.e., less than 1 serving per person per annum, on average).

Data from 30 smoked seafood studies provided the contamination data used for this category. Only six of these studies were conducted in the United States. Quantitative data were available in 10 studies. The contamination database included samples from both hot and cold smoking, but the process or the species was not always specified. Salmon was the most frequent product tested but other finfish and mussels were represented. The smoking process for this category, when specified, was usually cold smoking. The impact of different smoking methods on contamination is not known, but available literature suggests that inactivation resulting from hot smoking is often lost due to recontamination. Cold smoking has no significant effect on Listeria monocytogenes. The percentage of retail samples with detectable contamination was high, about 13% overall. In a few cases, the observed level of Listeria monocytogenes in the enumerated samples was very high. For example, the NFPA (2002) study (Gombas et al., 2003) collected 2,686 samples at retail and found 113 positive for Listeria monocytogenes. Two of these samples were between 10⁵ and 10⁶ cfu/g.

The growth rate data for this category came from 10 studies containing a total of 25 individual growth rates for hot- and cold-smoked salmon, trout, and cod. The average exponential growth rate adjusted to 5°C was a moderate 0.15 logs/day. Home storage times tend to be moderate in most instances but occasionally samples are stored for lengthy periods. The most likely and maximum storage times used were 3 to 5 days and 15 to 30 days, respectively. The estimated number of Listeria monocytogenes consumed per serving was high. The median estimate was 6.7% of servings exceeded 1 x 10³ cfu/serving and 0.2% of the servings exceeded 1x10⁶ cfu/serving.

The predicted median number of cases of listeriosis per serving for Smoked Seafood was 6.2x10^-9. This corresponds to a relative risk ranking of fifth for the Smoked Seafood category for the total United States population. The range for the per serving ranking distribution for Smoked Seafood is clustered in the higher ranks, with a normal distribution with a single mode (Figure V-4a). The level of uncertainty was typical of that observed with most food category rankings. The predicted median per annum relative risk rankings were ninth for the total United States population. The median predicted number of cases per annum of 1.3 for the total United States population was moderate. The relative ranking distribution for the per annum value (Figure V-4b) was shifted slightly to the lower risk ranks, reflecting the lower number of servings per year of foods in this category. Although the uncertainty for the cases per annum was greater than for the per serving value, the uncertainty associated with the per annum value was still typical for those observed with most food categories.

Figure V-4a. Rankings of Total Predicted Listeriosis Cases per Serving for Smoked Seafood

Figure V-4b. Rankings of Total Predicted Listeriosis Cases per Annum for Smoked Seafood

Food Category: Raw Seafood

Raw Seafood has a low predicted relative risk per serving of causing listeriosis in the United States. The foods in this category generally were characterized by a low annual number of servings, a low percent of the population consuming the food, and small serving sizes. However, the contamination levels at retail were high and Listeria can grow in these foods at moderate rates. As perishable foods, storage times are typically short which effectively limits the growth and the numbers of organisms likely to be consumed. This combination of factors made the predicted estimates of exposure and illness low. Though the Raw Seafood category has a low predicted relative risk of causing listeriosis in the United States, products in this category have been linked to an outbreak in New Zealand and to a sporadic case in Italy (Farber and Peterkin, 1991).

This category is fairly heterogeneous. Foods for which there were consumption data were flounder, pompano, tuna, sturgeon roe, squid, oysters, and sushi. The median amount consumed per serving is 16.0 g (approximately 0.5 ounce), and the annual total number of servings is low at 1.8 x10⁸.

Forty-six contamination studies (including 11 from the United States) analyzed over 15,500 samples of uncooked seafood and seafood products, primarily to determine the presence or absence of Listeria monocytogenes. Four studies provided quantitative data.

Contamination data were mainly for fresh or frozen whole animals, but products such as cakes, fingers, minces, sushi, and unspecified fish parts are also included. These can be categorized as finfish and non-finfish. Finfish, when specified, included butterfish, catfish, red snapper, trout, and tuna. Both wild caught and aquaculture-reared fish were included. Non-finfish included shellfish and crustaceans. Among the specified foods were lobster, squid, langostino, oyster, shrimp, mussel, clams, and scallops. The percentage of samples with detectable contamination was high (7.0%). Pathogen levels were predicted to be in the high range for the percentage of servings with 10³ to 10⁶ cfu at retail.

Six papers provided Listeria monocytogenes growth rates in these foods. Individual foods were trout, catfish, shrimp, and oysters. The growth rates averaged 0.15 logs per day at 5°C. Storage times were relatively short for these foods; the most likely storage time was 1 to 2 days, and the maximum time was 10 to 20 days.

The predicted median risk per serving for the Raw Seafood category was 2.0x10^-11 and ranked 13th for the total United States population. The range for the per serving ranking distribution (Figure V-5a) is relatively narrow and concentrated in the lower risk ranks. This indicates that there is little uncertainty associated with the predicted per serving relative risk ranking for the Raw Seafood category. The predicted median per annum relative risk ranking was 18^th for the total United States population. The range for the per annum ranking distribution (Figure V-5b) was narrow, indicating that there is also little uncertainty associated with the per annum predicted relative risk for Raw Seafood. This decrease in the per annum ranking compared to the per serving ranking is consistent with the small number of servings consumed per year.

Figure V-5a. Rankings of Total Predicted Listeriosis Cases per Serving for Raw Seafood

Figure V-5b. Rankings of Total Predicted Listeriosis Cases per Annum for Raw Seafood

Food Category: Preserved Fish

Preserved Fish, including pickled, marinated, or dried products, had a low predicted relative risk of causing listeriosis on a per serving basis and a low predicted contribution to the total number of cases on a per annum basis. The foods in this category had a low annual number of servings and a low percent of the population consuming the food, but had moderate serving sizes, high frequency of contamination, and moderate contamination levels at retail. Growth was not modeled for this category, since preserved fish do not support growth. Typically, the inability of a food category to support the growth of Listeria monocytogenes results in a low per serving relative risk. However, in this instance the lack of growth appears to be offset by the frequency of contamination at retail. Moderate level contamination likely occurs because foods in the Preserved Fish category are often prepared using traditional techniques, which require long processing times and occasionally may not meet stringent sanitary standards. This creates the potential for substantial growth of Listeria monocytogenes during initial production steps (e.g., brining) before the product equilibrates to the salt and pH levels that are the basis of preservation. Gravad rainbow trout has been linked to an outbreak of listeriosis in Sweden (Ericsson et. al., 1997).

The Preserved Fish category includes consumption data for pickled or marinated fish, such as ceviche and pickled herring, dried and salted cod, and non-specified dried fish. The median amount consumed per serving for this category is 70 g (approximately 2.5 ounces), and the annual total number of servings is 1.1 x 10⁸.

Contamination data for this food category was from 18 studies. Haddock, gravad trout, ceviche, and unspecified finfish that were pickled, smoked, dried, salted, or preserved were included. Of these studies only one was from the United States. Five studies contained quantitative data. The percentage of samples with detectable contamination was 9.8%, higher than for Raw Seafood, but just slightly less than Smoked Seafood. The predicted percentage of servings contaminated with 10³ to 10⁶ cfu at retail was moderate.

Because these products do not allow growth of Listeria monocytogenes, storage times are not a factor in the levels of Listeria monocytogenes present at the time of consumption. Although not a factor, storage times were also believed to be somewhat shorter than those for Smoked Seafood. The high salt and acidity present in the final products prevent growth of Listeria monocytogenes. However, the microorganism is known to survive these conditions, particularly if held at refrigeration temperatures.

The predicted median risk per serving for the Preserved Fish category was 2.3x10^-11, which corresponds to a relative risk rankings of twelfth for the total United States population. The range for the per serving relative ranking distribution is relatively broad (Figure V-6a) with a bimodal distribution. The wide spread indicates a high degree of uncertainty which likely is due to a combination of the limited quantitative data and broad variability in conditions under which these products are produced. The bimodal distribution may indicate that there are differences among different foods within this food category, and may require that the category be subdivided if additional data become available in the future in order to achieve a more accurate measure of the relative risks associated with the different foods. The predicted median per annum relative risk ranking was low, at less than one case per annum and ranked nineteenth for the total United States population. The range for the per annum ranking distribution was also a bimodal distribution (Figure V-6b), again indicating a substantial degree of uncertainty or variability. Overall this food category is not predicted to make a substantial contribution to the cases of listeriosis in the United States, however, the uncertainty in the risk per serving indicates that it may be a concern for the small population that consumes these products.

Figure V-6a. Rankings of Total Predicted Listeriosis Cases per Serving for Preserved Fish

Figure V-6b. Rankings of Total Predicted Listeriosis Cases per Annum for Preserved Fish

Food Category: Cooked Ready-to-Eat Crustaceans

Cooked Ready-to-Eat (RTE) Crustaceans (crab and shrimp) had a high predicted relative risk of causing listeriosis in the United States on a per serving basis. The foods in this category generally were consumed at a low frequency and with moderate serving sizes. The relatively high growth rate of Listeria monocytogenes in these foods, one of the usual factors that drives listeriosis risk in food, was offset by relatively short storage times. It would be expected that the cooking step in the preparation of these foods would eliminate Listeria monocytogenes. However, foods in this category may often be stored refrigerated after cooking, allowing for recontamination and growth.

Imitation crabmeat has been linked to an outbreak of listeriosis in Canada and shrimp was epidemiologically linked to an outbreak in the United States. (Ryser, 1999a; Riedo et al., 1994). The FDA has also monitored recalls for cooked shrimp and crab.

The Cooked RTE Crustaceans category includes consumption data for steamed, hard shell crab; steamed or boiled shrimp; and cocktail shrimp. The median serving size for this category was

50 g (approximately 1.8 ounces), and the annual total number of servings was 5.5x10⁸.

Eleven contamination studies provided data mainly from cooked crab and shrimp. Four studies were for product in the United States. Two studies, both from the United States, provided quantitative data. The percentage of contaminated samples was moderate at 2.8%. A small number of samples with high contamination levels (greater than 10³ cfu/g) have been reported. The predicted percentage of servings with 10³ to 10⁶ cfu/serving at retail was moderate. Only three papers were found that reported growth rates for pasteurized crab and for cooked shrimp and lobster. This category had the fastest reported growth rates of any food category, averaging 0.38 logs/day at 5° C. Storage times were estimated to be relatively short; the most likely storage time was only 1 to 2 days, and the maximum time was 10 to 20 days.

The predicted median risk per serving for the Cooked RTE Crustaceans category of 5.1x10^-9 corresponded to a relative risk ranking of sixth for the total United States population. The range for the per serving ranking distribution for Cooked RTE Crustaceans (Figure V-7a) is narrow and concentrated in the lower risk rankings (i.e., a higher risk food). This indicates that there is little uncertainty associated with the predicted per serving relative risk for the Cooked RTE Crustaceans category. The predicted median per annum risk is approximately three cases of listeriosis per annum and a relative risk ranking of eighth for the total United States population. The range for the per annum ranking distribution is narrow and generally normally distributed (Figure V-7b), suggesting relatively little variability or uncertainty in the extent to which this food category is consumed.

Figure V-7a. Rankings of Total Predicted Listeriosis Cases per Serving for Cooked Ready-to-Eat Crustaceans

Figure V-7b. Rankings of Total Predicted Listeriosis Cases per Annum for Cooked Ready-to-Eat Crustaceans

Food Category: Vegetables

Foods in the Vegetables category had a low predicted relative risk of causing listeriosis in the United States on a per serving basis. The Vegetables category is difficult to characterize because it encompasses a diverse set of products that are typically consumed without cooking. The annual number of servings of Vegetables is high, while the median serving size, contamination level, and growth rate are low. The storage time and the contamination frequency are moderate.

Both raw and processed vegetables have been implicated in outbreaks. Raw vegetables have been linked to outbreaks of listeriosis in Austria and Western Australia; frozen broccoli, cauliflower, celery, tomatoes, and lettuce in the United States (Ryser, 1999a; Simpson, 1996; Riedo et al., 1994; Farber and Peterkin, 1991; Allerberger and Guggenbichler, 1989). In addition, raw vegetables have been linked to sporadic cases in Australia, the U.K. (English lettuce, vegetable rennet), and Finland (salted mushrooms) (Ryser, 1999a; Farber and Peterkin, 1991).

Foods included in the Vegetables category are raw as well as mixed vegetable salads that contain raw vegetables but not salad dressing. In addition to vegetables typically consumed raw (e.g., spinach, carrots, tomatoes, celery, lettuce, onions), this category includes less frequently consumed vegetables such as artichokes, sprouts, and raw seaweed. However, salads such as cole slaw and potato salads are included in the Deli-type Salads food category because of the creamy dressing base and frequent handling in the retail deli. The median amount consumed per serving for this category is 28 g (i.e., ~ 1 ounce), and the annual total number of servings is 8.5x10¹⁰. The low median serving size most likely reflects the consumption patterns associated with the wide span of vegetable types included in the analysis, though certain vegetables may be eaten in substantially larger amounts (e.g., tomatoes).

Thirty-two contamination studies were found that examined individual raw vegetables or mixed vegetables (without dressing). Of these studies, five were from the United States and eight contained quantitative data. The vegetables analyzed included raw bean sprouts, broccoli, cabbage, carrot, celery, cilantro, cress, cucumber, fennel, legumes, lettuce, mushrooms, parsley, green peppers, onions, radish, scallion, tomato, and watercress. The NPFA (2002) survey collected 2,963 samples of bagged, precut leafy salads and found 2.3% positive, with one sample containing between 10² and 10³ cfu/g. Overall, the percentage of samples with detectable contamination was a moderate 3.6%. The predicted percentage of servings with high contamination levels was low.

Nine papers provided 26 estimates of growth rates for Listeria monocytogenes on vegetables. The vegetables included in these studies were lettuce, cabbage, broccoli, cauliflower, asparagus, tomatoes, and carrots. The average growth rate of Vegetables was slow, 0.07 logs/day at 5°C. Moderate storage times were assumed with the most likely 3 to 4 days and the maximum of 8 to 12 days.

The predicted median risk per serving for the Vegetables category was 2.8x10^-12 and the relative risk ranking was eighteenth for the total United States population. The range for per serving distribution for Vegetables (Figure V-8a) is similar to what was observed with most food categories and clustered in the lower risk rankings. This indicates that there is relatively little uncertainty associated with the predicted per serving relative risk for the Vegetables category. The predicted median per annum risk was less than one case and the corresponding relative risk ranking was twelfth for the total population. The per annum ranking distribution (Figure V-8b) had a relatively broad range, indicating substantial uncertainty. The distribution was shifted to the higher risk ranks compared to the per serving distribution. These results presumably reflect the large number of servings of Vegetables consumed, as well as the variability in the products encompassed in this highly diverse category. The broad range suggests that this food category and its ranking could benefit from additional investigations and the possible subdivision of the food category into several smaller groupings.

Figure V-8a. Rankings of Total Predicted Listeriosis Cases per Serving for Vegetables

Figure V-8b. Rankings of Total Predicted Listeriosis Cases per Annum for Vegetables

Food Category: Fruits

Foods in the Fruits category had a low predicted relative risk of causing listeriosis on a per serving basis. Fruits have not been linked to outbreaks or sporadic cases of listeriosis, and this might explain why there is little contamination data in the published literature available for this category. The annual number of servings, median serving size and contamination frequency of Fruits are high. These factors lead to a high risk for fruits on a per annum basis even though the growth of Listeria monocytogenes during storage would be low. The high level of uncertainty indicates a need for more information and data for this food category. This is a diverse food category that includes acidic fruits (such as pineapples) and pH neutral fruits (such as cantaloupes).

The Fruits category includes consumption data for many types of raw and dried fruits, as well as fruit salads (with fruits as the main ingredient without salad dressing). This category is simplified from the 2001 draft risk assessment in that fruit salads containing salad dressing were moved to the Deli-type Salad food category. The median amount consumed per serving for this category is 118 g (i.e., slightly over 4 ounces), and the annual total number of servings is 4.9x10¹⁰.

Only four contamination studies, two of which were from the United States were available. None of these studies included quantitative data. Fruits specified in these studies included apples, blueberries, cantaloupes, pears, pineapples, and fruit products. The percentage of samples with detectable contamination was 11.8%, a high contamination frequency. The contamination levels were estimated from the presence/absence data assuming the standard deviation of the frequencies of contamination levels. The high frequency of contamination would indicate that high levels of contamination could also occur.

Two studies (orange juice and fresh apple slices) were found that characterized the rate of Listeria monocytogenes growth in fruits. When the pH was less than 4.8, Listeria monocytogenes did not grow. At pH 5.0, growth was slow, at 0.05 logs/day. Moderate storage times were assigned for this category, with a most likely time of 3 to 4 and a maximum time of 8 to 12 days.

The predicted median risk per serving for the Fruits category was 1.9x10^-11 which corresponds to a relative risk ranking of fourteenth for the total United States population. The range for the ranking distribution for Fruits (Figure V-9a) is broad. The predicted median risk per annum is approximately 1 case per year and the relative risk ranking is tenth for the total United States population (Figure V-9b). This increase in relative risk compared to the per serving value reflects the large number of servings consumed annually. The range for the ranking distribution was broad indicating substantial uncertainty in the predicted relative risk ranking. This likely reflects the limited data available, the diversity of the products that fall within this food category, and the variability in the frequency and extent of contamination rates among the data that were evaluated. The bimodal nature of the distribution suggests that the food category may need to be subdivided when additional data become available. Overall, the Fruits category is a broad category with varied consumption and contamination, and few data were available to characterize this category. Thus, there is a high degree of uncertainty associated with this category.

Figure V-9a. Rankings of Total Predicted Listeriosis Cases per Serving for Fruits

Figure V-9b. Rankings of Total Predicted Listeriosis Cases per Annum for Fruits

Food Category: Fresh Soft Cheese

Fresh Soft Cheese had a low predicted relative risk of causing listeriosis on a per serving basis. These cheeses are high moisture (>50%) fresh cheeses consumed shortly after manufacture. This category includes traditional Hispanic-style soft cheese (sometimes made from raw, unpasteurized fluid milk) such as panela, Queso de Crema, Queso Fresco,and Queso de Puna. The 2001 draft risk assessment included Queso Chihuahia and Queso Asadero attributed to this category, but these cheeses were moved from this category because they are not fresh, high moisture cheeses. The contamination level at retail, contamination frequency, growth rate during storage, and the annual number of servings are all low.

Fresh Soft Cheese (suspected to be made from unpasteurized milk) has been linked to both outbreaks and sporadic cases of listeriosis in the United States (Ryser, 1999a; Linnan et al., 1988; CDC, 2001), including an outbreak in Los Angeles in 1985 and one in North Carolina in 2001. The 1985 outbreak in Los Angeles was the incident that convincingly established Listeria moncytogenes as an important serious foodborne pathogen. In 2000/2001, an outbreak in the Carolinas associated with homemade cheese made from unpasterurized milk resulted in 12 cases of serious listeroisis.

Consumption data was only available for one type of Fresh Soft Cheese, Queso fresco. The median amount consumed per serving for this category is 31 g (just over 1 ounce), and the annual number of servings is 7.1x10⁷. Data are not available to estimate the proportion of Fresh Soft Cheese that is consumed in the United States made from unpasteurized milk; however, since the initial outbreak there has been a concerted effort to reduce the consumption of soft fresh cheeses made from unpasteurized milk. Fresh soft cheese made from unpasteurized milk does not meet FDA standards for interstate commerce.

Data from eight contamination studies were used to model the frequency of contamination for the Fresh Soft Cheese category. Cheeses in these studies were described as Hispanic-style, Queso Fresco, panela, requesoy, and fresh cow and goat milk cheeses. The most recent study was the NFPA (2002) survey and the contamination levels found in this study were much lower than those previously observed. In that study, 5 contaminated samples out of 2,936 total samples were positive, all at a level of less than 100 cfu/g. The samples from the NFPA study were collected in retail stores and were most likely made from pasteurized milk. Products made outside the retail system (including those made from unpasteurized milk) were not reflected in the NFPA survey. A 'what if' scenerio test was conducted to allow a comparison of the expected estimate of the risk per serving for fresh soft cheese made from pasteurized vs. raw, unpasteurized milk (see below).

Only one growth rate study with these cheeses was available. That study reported a low growth rate of 0.082 logs/day when adjusted to 5° C. The assumed storage times for Fresh Soft Cheese were 1 to 5 days and 15 to 30 days for most likely and maximum times, respectively.

The median risk per serving for the Fresh Soft Cheese category of 1.7x10^-10 corresponds to a relative predicted risk ranking of tenth for the total United States population. The range for the predicted per serving risk ranking distribution for Fresh Soft Cheese (Figure V-10a) is relatively narrow and concentrated in the middle of the risk rankings. This indicates that there is little uncertainty associated with the per serving predicted relative risk for the Fresh Soft Cheese category. The predicted median per annum risk was less than one case per year and the relative risk ranking was fourteenth for the total United States population. The range for the per annum ranking distribution is concentrated in the higher risk rankings (Figure V-10b) indicating a lower risk. The breadth of the range indicates that there was somewhat more uncertainty associated with the per annum predicted relative risk ranking for the Fresh Soft Cheese category. This is likely associated with variability in the number of servings and the serving sizes.

An area of uncertainty associated with this food category that is not captured in this risk assessment is the consumption of "homemade" soft cheeses made from raw, unpasteurized milk. Raw milk soft cheeses are not produced and marketed through typical commercial means and have in the past been illegally brought into the United States. Data on such cheeses are not captured in the contamination data base used to develop this risk assessment. However, we recognize that a substantial portion of soft cheeses consumed in the United States may be made from unpasteurized milk.

Scenario Testing: Fresh Soft Cheese Made From Contaminated Unpasteurized Milk

Unlike the 2001 draft risk assessment, the revised risk assessment indicates that the risk from Fresh Soft Cheese is low. This change is largely attributable to the inclusion of additional new data indicating a very low prevalence rate in this food category. However, in the past there has been a strong epidemiological correlation between Hispanic-style fresh soft cheese (Queso Fresco) and listeriosis. A likely explanation for this discrepancy is that the data collected for this category is not representative of the cheese linked to the disease (i.e., fresh soft cheese made from raw, unpasteurized milk). In particular, although most commercial sources of fresh soft cheese are manufactured from pasteurized milk, some sources of queso fresco are made from raw milk. Many of these sources appear to be restricted to specific local areas and have not had the benefit of FDA oversight.

To characterize the risk from highly contaminated queso fresco an exposure model was constructed using the same analog as in the 2001 draft risk assessment - soft unripened cheese made from raw milk (Loncarevik, et al., 1995), where 50% of the samples tested were positive. The tested 'high prevalence' scenario increased the predicted risk on a per serving basis approximately 40-fold for the perinatal and elderly subpopulations. (For additional details, see Chapter VI 'What-If' Scenarios.)

Figure V-10a. Rankings of Total Predicted Listeriosis Cases per Serving for Fresh Soft Cheese

Figure V-10b. Rankings of Total Predicted Listeriosis Cases per Annum for Fresh Soft Cheese

Food Category: Soft Unripened Cheese

The Soft Unripened Cheese category has a moderate predicted relative risk of causing listeriosis on a per serving basis. The cheeses in this category have moderate frequency and levels of contamination and can have a long storage time. However, they support only a low rate of growth. Serving sizes are typically low, whereas the annual number of servings and contamination levels at retail are moderate. There was a sporadic listeriosis case in the United States linked to the consumption of a highly contaminated ricotta cheese (Ryser, 1999a). There are no reported cases of listeriosis associated with consumption of cottage and cream cheese, but there have been FDA recalls of cream cheese products.

The category represents high moisture (>50%), white curd varieties such as cottage cheese, baker's cream, and American-type Neufchatel cheese. Milk to be manufactured into soft unripened cheese is coagulated through the production of acid by the starter culture (or by direct acidification of milk) rather then by addition of a coagulant. Unlike fresh soft cheese, the refrigerated shelf-life is typically up to 60 days.

Consumption data available were available for cottage, cream, and ricotta cheeses. The median amount consumed per serving for this category is 29 g (about 1 ounce), and the annual total number of servings is 4.4 x 10⁹.

There were eight studies with contamination data for these cheeses, with two from the United States. Three quantitative studies provided quantitative data. Cheeses in the contamination database included Anari, Halloumi, farmer, gournay, Quark, and cottage cheese. Of the 32 positive samples, four samples contained over 500 cfu/g and four samples over 10⁶ cfu/g. The percentage of positive samples was 3.9%.

Twenty-nine data sets provided data on the growth or survival of Listeria monocytogenes in these cheeses. Nine of these studies showed a decline in levels over time. The research literature indicates that growth or decline of Listeria monocytogenes in these low salt cheeses is largely dependent upon pH. For example, ricotta cheese (pH=5.9 to 6.1) permitted rapid growth, whereas declines were observed in some cream cheeses (pH=4.8). The growth rates were standardized to 5 °C and a distribution fitted to the data to allow growth or decline (i.e., negative growth) in proportion to the available data. The average growth rate was 0.09 logs/day. Storage times were relative long, with the most likely 6 to 10 days and the maximum 15 to 45 days.

The median risk per serving for the Soft Unripened Cheese category of 1.8x10^-9 corresponds to a relative predicted risk ranking of eighth for the total United States population. The range for the predicted per serving risk rankings for Soft Unripened Cheese (Figure V-11a) is bimodal but concentrated in the higher risk rankings. This indicates some uncertainty associated with the per serving predicted relative risk for this category. The median per annum risk was predicted as approximately 8 cases per year and the relative risk ranking was fifth for the total United States population. The range for the per annum ranking distribution is concentrated in the lower risk rankings, which corresponds to a higher risk (Figure V-11b). However, the broad ranges in uncertainty likely result from the differences of the products in this food category to support growth or cause a decline in levels of Listeria monocytogenes. Based on these results, this food category could benefit from subdivision.

Figure V-11a. Rankings of Total Predicted Listeriosis Cases per Serving for Soft Unripened Cheese

Figure V-11b. Rankings of Total Predicted Listeriosis Cases per Annum for Soft Unripened Cheese

Food Category: Soft Ripened Cheese

The cheeses in the Soft Ripened Cheese food category had a low predicted relative risk of causing listeriosis in the United States on a per serving basis. This food category includes high moisture (>50%), ripened cheeses such as mold surface-ripened cheeses (Brie, Camembert), pickled (white brined) cheeses, feta, and soft Italian-style cheeses (mozzarella). There are a moderate number of annual servings and small serving sizes. Growth rates were low but, contamination frequencies and levels at retail were moderate and storage times were long. Soft Ripened Cheeses including mold-ripened cheeses have been linked to outbreaks of listeriosis in Denmark, France and Switzerland and linked to sporadic cases in Belgium, Canada, and the U.K (Ryser, 1999a; Riedo et al., 1994; Art and Andre, 1991; Farber and Peterkin, 1991). There have not been any confirmed reports of sporadic cases or outbreaks associated with these cheeses in the United States.

The median amount consumed per serving for this category is 28 g (~1 ounce) and the annual number of servings is 1.9x10⁹. Data are not available on the proportion of United States or imported cheese that is made from unpasteurized fluid milk. Market data indicate that the United States imports approximately 50% of the Camembert and Brie Cheese and 20% of the feta cheese sold in the United States (National Cheese Institute, 1998).

Contamination data was obtained for 17 studies with three being from the United States. Five studies provided quantitative data. Brie, Camembert, Feta, and Taleggio are some of the cheeses represented in the contamination data. Of the 17 studies, 6 contained quantitative contamination data. In the 2001 NFPA study, two samples were positive for Listeria monocytogenes with levels less than 10 cfu/g. The frequency of contamination was 3.8%.

Listeria monocytogenes populations were reported in the research literature to both increase and decrease in these cheeses. Of 17 studies, 7 showed declines, one no change, and 9 indicated growth. Therefore, the growth rate distribution used with this food category (-0.013 logs/day) included both growth and decline, with the 'average' response being a slow rate of decline. Storage times for this food category were long, with a maximum of 15 to 45 days.

The median risk per serving for the Soft Ripened Cheese category of 5.1x10^-12 corresponds to a relative predicted risk ranking of seventeenth for the total United States population. The range for the predicted per serving risk ranking distribution for this category (Figure V-12a) is broad but concentrated in the higher risk rankings (low predicted risk). This indicates substantial uncertainty associated with the per serving predicted relative risk for this category resulting from the ability of some of these cheeses to support growth of Listeria monocytogenes and other cheeses to cause a decline. The median per annum risk was predicted as less than one case of listeriosis per year and the relative risk ranking was sixteenth for the total United States population. With the wide range for the per serving rankings, the resulting range for the per annum ranking distribution is quite broad (Figure V-12b) indicating high uncertainty associated with the per annum predicted relative risk ranking.

Figure V-12a. Rankings of Total Predicted Listeriosis Cases per Serving for Soft Ripened Cheese

Figure V-12b. Rankings of Total Predicted Listeriosis Cases per Annum for Soft Ripened Cheese

Food Category: Semi-soft Cheese

The Semi-soft Cheese food category has a low predicted relative risk of causing listeriosis on a per serving basis. Semi-soft Cheese has a moisture content that ranges between 39% and 50%. The cheeses in this food category include blue, brick, Edam, Gouda, havarti, Limburger, Monterrey jack, Muenster, and provolone. The serving sizes are small, the annual number of servings, and contamination frequency are moderate, and the levels at retail are low. Although the storage times are long, the growth rates are low. Blue cheese has been linked to an outbreak of listeriosis in Denmark (Jensen et al., 1994) and Monterrey jack cheese made from raw milk to a sporadic case in the United States (Ryser, 1999a). FDA has monitored recalls of several semi-soft cheeses because of the presence of Listeria monocytogenes.

The median amount consumed per serving for this category is 28 g (1 ounce), and the annual number of servings is 1.8x10⁹. Data are not available to describe the proportion of United States or imported cheese that is made from unpasteurized fluid milk. Market data indicate that the United States imports approximately 20% of the blue cheese (including Gorgonzola) sold in the United States (National Cheese Institute, 1998).

There were eleven studies with contaminati