Table of Contents

IV. EXPOSURE ASSESSMENT

The Exposure Assessment component of a microbial risk assessment is an evaluation of the likelihood of ingesting a pathogenic microorganism via food and the likely level of exposure. In this assessment, the likelihood of exposure to pathogenic V. parahaemolyticus from consumption of raw oysters was evaluated. This risk assessment is a quantitative product pathway analysis in which the key steps from harvest of oysters through post-harvest handling and processing to the point of consumption were modeled. The predicted levels of pathogenic V. parahaemolyticus in oysters were determined at each step in the pathway.

A schematic representation of the Exposure Assessment Module is shown in Figure IV-1. The Exposure Assessment is subdivided into three modules: Harvest, Post-Harvest, and Consumption. The Harvest Module considers the factors influencing the prevalence of total V. parahaemolyticus in oysters up to the time of harvest. The Post-Harvest Module considers factors associated with handling and processing of oysters. The Consumption Module considers factors such as the number of oyster servings eaten per year, the quantity of oysters consumed per serving, and the levels of pathogenic V. parahaemolyticus in the oyster at the time of consumption.

Oysters are harvested throughout the year in the United States from four major regions: the Gulf Coast, Mid-Atlantic, Northeast Atlantic, and Pacific Northwest. Methods and conditions of harvest and handling of oysters after harvest can influence the levels of V. parahaemolyticus in oysters at the time of consumption. These harvest and handling practices and conditions vary considerably in different geographic areas and at different times of year. In the Gulf Coast, the harvest duration (i.e., the time between removal of the oyster from the water to unloading them at the dock) for Louisiana is typically much longer than for other states in that region (Florida, Mississippi, Texas, and Alabama). Therefore, the Gulf Coast was divided into two distinct regions: Gulf Coast (Louisiana) and Gulf Coast (Non-Louisiana). Likewise, the Pacific Northwest was divided into two distinct regions: Pacific Northwest (Intertidal) and Pacific Northwest (Dredged). In the Pacific Northwest, oysters are harvested by two methods: dredging and intertidal. For the intertidal harvest method, oysters are hand-picked when oyster reefs are exposed during the tide cycle and left in baskets until the tide rises to a sufficient depth to allow a boat to retrieve the basket.

The risk assessment modeled six oyster harvest regions [Gulf Coast (Louisiana), Gulf Coast (non-Louisiana), Mid-Atlantic, Northeast Atlantic, Pacific Northwest (Intertidal) and Pacific Northwest (Dredged)] and four seasons [Summer, Fall, Winter, Spring] for a total of 24 region/season combinations. These region/season combinations were separately modeled. Predictions of the number of pathogenic V. parahaemolyticus per serving of oysters at the time of consumption were determined for each of the 24 region/season combinations.

Figure IV-1. Schematic Representation of the Exposure Assessment Component of the Vibrio parahaemolyticus (Vp) Risk Assessment Model
[The boxes with black lettering shaded with light gray show the Harvest Module, the boxes shaded with gray show the Post-Harvest Module, and the boxes with white lettering and shaded in dark grey show the Consumption Module.]

Harvest Module

The Harvest Module considers the factors associated with the likelihood that oysters harvested from specific growing areas and at specific times of the year will contain V. parahaemolyticus (total and pathogenic). Factors which affect the frequency and levels of V. parahaemolyticus in oysters include the routes of introduction, prevalence and persistence of V. parahaemolyticus in the environment. These factors are discussed below.

Routes of Introduction into Oyster-Growing Areas

There are several pathways by which V. parahaemolyticus may occur in oyster growing areas. Vibrio parahaemolyticus may be indigenous to a geographical area. New strains may be introduced naturally by the activities of terrestrial and aquatic animals, or through human activities. Terrestrial and aquatic animals (including plankton, birds, fish, and reptiles) may harbor pathogenic strains of V. parahaemolyticus and may play a role as intermediate hosts and vehicles for its dissemination (Davis et al., 1982; Sarkar et al., 1985). For example, V. parahaemolyticus has been isolated from a number of fish species where it is associated primarily with the intestinal contents (Nair et al., 1980). Vibrio parahaemolyticus can also be introduced into non-contaminated areas by transfer of shellfish from contaminated waters, as would occur during the process of "relaying" shellfish.

Ship ballast release is another potential mechanism of introduction of V. parahaemolyticus into a particular geographical area. Most cargo ships must carry substantial quantities (millions of gallons) of ballast water to operate safely when they are not carrying cargo. Cargo ships take on ballast water from the body of water in which the ship originates. Having taken water on board, it is normally retained until the ship is about to load cargo, at which point ballast water is discharged. During de-ballasting, organisms picked up from one port could be introduced into the loading port. It is possible that the non-potable water from a cargo ship could have been the source of V. parahaemolyticus serotype O3:K6 in the Galveston Bay in 1998. This serotype was identified during a large outbreak of culture-confirmed illnesses associated with oysters harvested from this location at this time. Prior to 1998, serotype O3:K6 had not been isolated from either environmental or clinical samples in the United States, but had established an ecological niche in Asia (Arakawa et al., 1999).

Prevalence and Persistence in Oyster-Growing Areas

Prevalence and persistence of pathogenic strains of V. parahaemolyticus in oyster in the environment may be dependent on several parameters. Factors which may determine whether V. parahaemolyticus will become established in a specific area include interactions of environmental conditions, species and physiology of the shellfish, and the genetics of the microorganism. Other factors to be considered in determining the prevalence of V. parahaemolyticus include water temperature (including El Niño and La Niña weather patterns), salinity, zooplankton, tidal flushing (including low tide exposure of shellfish) and dissolved oxygen (Amako et al., 1987; Garay et al., 1985; Kaneko and Colwell, 1978; Venkateswaran et al., 1990).

Environment. Favorable environmental conditions will support the establishment, survival, and growth of the microorganism. Warmer water temperatures and moderate salinities, especially those prevailing during the summer months, favor the growth and survival of V. parahaemolyticus (Covert and Woodburne, 1972; Jackson, 1974; Nair et al., 1980; Zhu et al., 1992). Most of the shellfish-borne illnesses caused by this microorganism occur in the warmer months. In an investigation of the 1998 outbreak, the CDC randomly selected 7 of the 76 existing Texas Department of Health sites for monitoring environmental conditions in Galveston Bay. At these sites, water temperature and salinity levels during May and June, 1998 were found to be significantly higher compared with data recorded over the previous five years for the same months (Daniels et al., 2000b). Elevated water temperatures were also suspected to have played a role in the 1997 outbreak on the West Coast (CDC, 1998).

Vibrio parahaemolyticus often "over-winters" (survives the winter) in the sediment and is absent or below detectable levels in the water column or oysters during the winter months (Joseph et al., 1983; Kaysner et al., 1990a; United States Department of Health and Human Services, Food and Drug Administration, 1995). During the summer, shellfish often have levels of V. parahaemolyticus that are more than 100-fold greater than those in the water (DePaola et al., 1990; Kaysner et al., 1990a). Also, under extreme environmental conditions, Vibrio species, including V. parahaemolyticus, may enter a "viable but non-culturable (VBNC) phase" in marine waters and could be missed by traditional cultural methods (Bates et al., 2000; Colwell et al., 1985; Oliver, 1995; Xu et al., 1982).

The potential influence of nutrients in the water on the prevalence and persistence of V. parahaemolyticus is unclear. Watkins and Cabelli (1985) reported that the densities of V. parahaemolyticus in the water column in Narragansett Bay, Rhode Island were correlated with the densities of fecal coliforms from sewage. The effect of sewage was surmised to be an indirect one, possibly mediated by stimulation of zooplankton with which the V. parahaemolyticus were associated, because laboratory studies showed that nutrients in the sewage did not directly increase V. parahaemolyticus levels. However, another study reported that organic matter does affect growth and survival of Vibrio species (Singleton et al., 1982). In another study, the distribution of V. parahaemolyticus in sediment samples from the Boston Harbor were found to be independent of densities of fecal coliforms (Shiaris et al., 1987).

Shellfish Physiology. Vibrio parahaemolyticus is frequently found on marine particulates, zooplankton and other chitin sources (Amako et al., 1987). Microorganisms are internalized by shellfish through shellfish filter feeding. Factors that favor active filter feeding by shellfish increase the probability that shellfish in a given area will take up the pathogen (Murphree and Tamplin, 1991). Shellfish species and physiology (e.g., sexual maturity, immune function, and metabolic state) can affect survival and growth of disease-causing Vibrio spp. within shellfish. There is evidence that the immune status of the shellfish may play an important role in the prevalence and persistence of the microorganism (Fisher and DiNuzzo, 1991; Kothary et al., 1997; LaPeyre and Volety, 1999; Ordás et al., 1998; Volety et al., 1999). There also appear to be seasonal differences in the oyster's cellular defense system. A study by Genthner et al. (1999) showed that the bactericidal activity of hemocytes (oyster blood cells) was greater in summer than in winter. Other factors such as spawning or adverse environmental conditions play a role in the incorporation of V. parahaemolyticus in the oyster by reducing or stopping filter feeding or changing oyster physiology. For example, the presence of the oyster parasite, Perkinsus marinus, influences the ability of oyster hemocytes to kill the internalized microorganisms (Kothary et al., 1997; LaPeyre and Volety, 1999; Tall et al., 1999). The presence of chemicals in the environment (e.g., tributyltin oxide, polycyclic aromatic hydrocarbons, wood preservative leachates) may reduce filter feeding (Sujatha et al., 1996; Weinstein, 1995; Wendt et al., 1996).

Genetics of the Microorganism. It is not known whether the prevalence and persistence of pathogenic and non-pathogenic strains are affected in a similar fashion by environmental factors. However, the presence of a pathogenicity island (a physical grouping of virulence-related genes) in V. parahaemolyticus may foster rapid microevolution, promote growth and survival, and result in transmission of factors, such as those responsible for virulence, to other strains (horizontal gene transfer) (Frischer et al., 1990; Ichige et al., 1989; Iida et al., 1998). Bacteriophages may genetically alter vibrios (Baross et al., 1978; Ichige et al., 1989).

Effect of Intertidal Harvest Practices.

The practice of intertidal harvest is used extensively in some of the estuaries of the Pacific Northwest region. Typically, after the tide recedes from an intertidally harvested area, the shellfish are hand picked and placed into large baskets, which are left in the harvest area until the tide rises to a sufficient depth to permit a vessel to retrieve the baskets and transport them to the processing plant. Alternatively, harvesters may transport the harvest by truck after collection, depending upon the location of the harvest area. In either case, intertidal harvest potentially exposes oysters to favorable conditions for growth of V. parahaemolyticus, especially on sunny summer days.

The effect of intertidal harvest practices has been shown to have a significant impact on V. parahaemolyticus densities in the harvested oyster. Vibrio parahaemolyticus levels were reported to increase (>100-fold) in oysters from the Puget Sound during intertidal exposure (Herwig and Cheney, 2001). In another study, oysters were analyzed before and after being submerged on a beach for 24 hours (DePaola et al., 2002). Vibrio parahaemolyticus levels were found to be below or near the minimum detectable level (10 cfu/g) when they were first removed from the water and after 5 hours exposure to ambient temperature and sunlight. After 24 hours, V. parahaemolyticus levels were approximately 500 cfu/g in oysters harvested on a sunny day and approximately 100 cfu/g in oysters harvested on a cloudy day. With respect to oysters collected from commercial reefs, the overall mean V. parahaemolyticus densities were found to be as much as 8-fold higher after maximum exposure compared to samples exposed for less than 1 hour, but there was considerable variation among sites (DePaola et al., 2002).

Data Selection and Criteria for the Harvest Module

A number of factors were identified that potentially affect the levels of V. parahaemolyticus in oysters at time of harvest. Modeling these factors required that both sufficient quantitative data were available and that the data permit consideration of regional and temporal variation. Due to the relatively low prevalence of pathogenic V. parahaemolyticus and limitations of current methods of detection, most quantitative studies have focused on the levels of total V. parahaemolyticus. Salinity can influence the prevalence and growth of V. parahaemolyticus in oysters, and preliminary modeling included a consideration of that parameter (see 2001 draft risk assessment at www.foodsafety.gov/~dms/fs-toc.html). However, subsequent consideration of the model indicated that water salinity is not as strong a determinant of V. parahaemolyticus levels in the regions that account for essentially all of the commercial harvest and was overshadowed by the impact of water temperature (Appendix 5). Accordingly, salinity was not included as a variable in the model.

There have been a number of studies conducted over a wide range of geographic locations showing the relationship of environmental factors and total V. parahaemolyticus levels in water and oysters. These studies were reviewed and evaluated for their utility for estimating an appropriate predictive relationship between pathogenic V. parahaemolyticus densities in oysters and environmental conditions. The studies are discussed in detail in this chapter and a summary of the key results of the studies is provided in Appendix 5. Most of the studies do not provide sufficient information with respect to a quantitative relationship, primarily because these studies were either limited to specific seasons with little variation of environmental parameters, measured V. parahaemolyticus levels in water or sediment rather than oysters or reported little quantitative data on densities per se.

The selection of data for use in the Harvest Module considered the availability of data and limitations of the data sources. Tables IV-1a, IV-1b, and IV-1c provide a summary of the criteria used to select the studies for the Harvest Module. Data used in this module include the following:

1. water temperature distribution for each region/ season combination
2. the relationship between total V. parahaemolyticus in oysters and water temperature
3. the ratio between pathogenic and total V. parahaemolyticus in oysters

Water Temperature. Criteria for selecting studies used to describe the water temperature distributions for each region/season combination is summarized in Table IV-1a. The data set must include long-term historical data so that the extent of year-to-year variation can be determined. Also, because of the large number of records needed to characterize the distribution of water temperatures across regions and seasons, the data must be available electronically. See Table IV-1a for details.

In comparison to the NBDC sites, STORET and NERR are more specific to estuaries as opposed to open coastal waterways. Some NBDC sites such as Thomas Point Lighthouse (Chesapeake) are located within estuaries but similar sites could not be identified for the Gulf Coast and Northeast Atlantic within the NBDC database. Comparison of NERR data for Weeks Bay, AL, versus that of the Dauphin Island NBDC buoy suggests that shallow water estuaries may be slightly warmer than open coastal waters but that the difference is not substantial (i.e., ~1 °C (1.8 °F) difference on average). An additional consideration is the availability of enough long-term historical data to determine extent of year-to-year variation. As already indicated, data are available from most NBDC buoys from 1988 to the present. The NERR program started data collection in 1995. Although STORET has considerable long-term historical data associated with monitoring of water quality dating back to 1964, access to STORET records is not readily available. Also, STORET records do not necessarily correspond to fixed locations, as is the case for NBDC and NERR. Additional data on water temperature measurements specific to oyster harvesting areas were made available to the FDA by State agencies in Texas, Alabama, New York, and Connecticut. The state data were not substantially different from the NBDC data selected for each region.

Relationship of Water Temperature and Total Vibrio parahaemolyticus in Oysters. Criteria for selecting studies to define the relationship between water temperature and total V. parahaemolyticus in oysters is summarized in Table IV-1b. A quantitative method must have been used to determine the levels of V. parahaemolyticus in oysters (enumerated, not presence/absence). Also, data would ideally be available over multiple years and regions. See Table IV-1b for details.

The Ratio of Pathogenic to Total Vibrio parahaemolyticus in Oysters. Criteria for selecting studies to define the percentage of pathogenic V. parahaemolyticus in oysters relative to the levels of total V. parahaemolyticus is summarized in Table IV-1c. Ideally, the study design should include analysis of individual oysters for the percentage of the total V. parahaemolyticus that are pathogenic (i.e., TDH⁺) such that the variation across individual samples can be accounted for in the model. Two different studies, DePaola et al. (2002) and Kaufman et al. (2003) were conducted in the summer of 2001. Both studies utilized a gene probe technique for enumeration of total and pathogenic V. parahaemolyticus in replicate aliquots from all samples collected. See Table IV-1c for details.

Assumptions Made for Modeling the Harvest Module

1. Individual oysters comprising a serving at time of consumption are harvested at the same time and location.
2. Levels of V. parahaemolyticus in oysters (log basis) at the time of harvest are normally distributed with mean proportional to water temperature.
3. The variability in water temperatures is adequately summarized by the mean and variance of daily noon-time temperatures at selected sites considered typical of each region/season.
4. Pathogenesis is based on the presence of the most characterized virulence factor of the microorganism, thermostable direct hemolysin (TDH).
5. Variation of the relative abundance of pathogenic versus total V. parahaemolyticus across collections of oysters is distributed as a Beta distribution.
6. The relationship between pathogenic and total V. parahaemolyticus is temperature independent (i.e., percentage pathogenicity is constant throughout the year).
7. The relationship between pathogenic and total V. parahaemolyticus is the same for the Gulf Coast, Northeast Atlantic, and Mid-Atlantic harvest regions.
8. Intertidal harvesting consists of ~75% of Pacific Northwest harvest.
9. For the Pacific Northwest (Intertidal) region, a range of exposures of between 4 to 8 hours before the oysters are collected was assumed for intertidal harvesting.

Modeling the Harvest Module

The various model inputs and output for the Harvest Module are illustrated in Figure IV-2 and discussed in detail below.

Figure IV-2. Schematic Depiction of the Harvest Module of the Vibrio parahaemolyticus (Vp) Exposure Assessment Model

Studies and Data Sources Used for the Harvest Module

1. Water temperature: Data from the National Buoy Data Center (NBDC), 1984 to 1998 was used for all regions except the Pacific Northwest region. Data from the Washington State Department of Health (1999) were used for the Pacific Northwest region.
2. The relationship between water temperature and levels of total V. parahaemolyticus in oysters: Data from FDA/ISSC, 2001 (data were also reported by Cook et al., 2002b and DePaola et al., 2003a) and DePaola et al. (1990) were used for all regions except the Pacific Northwest. Data from Washington State Department of Health (2000; 2001) were used for the Pacific Northwest.
3. Ratio between pathogenic and total V. parahaemolyticus in oysters: Data from Kaufman et al. (2003) was used for the Gulf Coast, Northeast Atlantic, and Mid-Atlantic regions. Data from DePaola et al. (2002) was used for the Pacific Northwest region.
4. Pacific Northwest Intertidal Harvest. See description of growth rate model in the Post-Harvest module.

Water Temperature Distributions

Regional and seasonal distributions of water temperatures were estimated based on accumulated records of coastal water buoys from the National Buoy Data Center (NBDC) for all regions except for the Pacific Northwest. Seasons were defined by calendar month; winter: January through March, spring: April through June, summer: July through September, and fall: October through December. The available data for most buoys contain hourly air and water temperatures from 1984 up to the present, with occasional data gaps due to instrumentation malfunction. Representative buoys were identified for the Gulf Coast, Mid-Atlantic and Northeast Atlantic regions. For each region a buoy site was selected for which both water and air temperature data were available because air temperature was identified as a relevant parameter needed with respect to post-harvest effects and examination of the NBDC data indicated a correlation between air and water temperature for shallow water areas.

For the Pacific Northwest, there were no buoys in the NBDC database that could be taken to be representative of the temperature conditions of the shallow water estuaries where oysters are harvested. Water temperature distributions for this region were therefore estimated based on temperature measurements taken during routine monitoring of selected oyster harvesting sites (Washington State Department of Health, 1999).

Based on the observation that oyster harvesting generally commences early in the morning and ends mid or late afternoon, the daily water temperature recorded at noon was taken to be representative of the average temperature determining V. parahaemolyticus densities at harvest. A single average daily temperature was used because examination of the NBDC data indicated that diurnal temperature variations were relatively minor relative to temperature variations occurring across different days or weeks. This is discussed in more detail in Appendix 5.

Within a given season, region, and year, the midday water temperature data from the NBDC buoys was generally found to be unimodal. For simplicity, a normal distribution was fit to the empirical water temperature data (for each region, season, and year). The mean(μ) and standard deviation (σ) of the distribution of water temperatures within any particular year for different region and season combinations are shown in Table IV-2. The extent of year-to-year variation of these distributions is summarized by the mean and the variance of the parameters μ and σ. The mean and variance of these parameters are denoted in the table as mean(μ), variance(μ), mean(σ) and variance(μ), respectively. The correlation between μ and σ is denoted by corr(μ, σ). A positive correlation between parameters μ and σ can be interpreted as indicating that when the mean water temperature is higher than normal the variation in temperatures from one day to the next is generally greater than that observed when the mean temperature is lower than normal. Similarly, a negative correlation summarizes the observation that temperatures are less variable when the mean water temperature is higher than normal.

The NBDC buoy located at Dauphin Island, Alabama was chosen as representative of water temperatures for the Gulf Coast. This buoy has recorded water temperatures beginning in 1987. For the spring season, the distribution of midday water temperature was found to vary from year to year with an average mean of 24.5 °C (76.1 °F). The variance of the mean from one year to the next was 0.98, which corresponds to a standard deviation of 0.99 °C. Similarly, for the standard deviation of the within year temperature distributions, the central tendency across different years was an average of 3.5 °C with a variance of 0.27, which corresponds to a standard deviation of 0.52 °C. The correlation between μ and σ was -0.55 indicating that the day-to-day temperatures were generally less variable when the overall mean temperature was higher than that of a typical year.

For the Pacific Northwest there were no near-shore NBDC buoys recording water temperatures that could be considered representative of oyster growing areas. Consequently, for this region, seasonal and year-to-year variations in water temperature distributions were developed based on compiled data from the Washington State Department of Health from 1988 through 1999. These water temperature data were recorded in association with collection of samples for monitoring of Vibrio species and fecal coliforms and are therefore directly representative of temperatures for oyster growing areas. Averages of water temperature were substituted when multiple measurements were recorded for any given day. Year-to-year variations in the water temperature distributions for the Pacific Northwest were developed in the same manner as that for the other regions.

Differences from one year to the next were evident for all regions and seasons. Therefore, the potential effect of year-to-year variation in the water temperature distributions was included in the model. First, the mean and the standard deviation of the parameters of the fitted normal distributions for each region/season combination were determined across all available years of data (see Table IV-2 and Appendix 5 for more details). The mean and standard deviation where then used to sample, assuming a normal distribution, a simulated set of 1,000 parameter values for each region/season combination. These sampled values were used to characterize the year-to-year variation of water temperature distributions in model uncertainty simulations. The simulated normal distributions used in model simulations were truncated at the observed upper and lower temperatures for each region/season combination.

Relationship Between Water Temperature and Total Vibrio parahaemolyticus Levels in Oysters

The relationship between total V. parahaemolyticus densities in oysters and water temperature was quantified using three comprehensive survey data sets: DePaola et al. (1990); FDA/ISSC (2001); and Washington State Department of Health (2000, 2001). These data sets were selected for quantitative modeling based on the criteria listed above (Table IV-1b).

Because different methodologies were used for enumeration in these three surveys (Table IV-3), the data sets were not pooled together. Instead, regression models were fit separately to each data set. A relatively large proportion of samples within the data sets had non-detectable levels of V. parahaemolyticus. In the DePaola et al. (1990) study, 26 of 61 oyster samples (43%) did not have detectable V. parahaemolyticus (the lower limit of detection is approximately 10 cfu/g). In the 2001 FDA/ISSC study (later published as Cook et al., 2002b), 232 of 624 (37%) samples analyzed for total V. parahaemolyticus were found to have less than the limit of detection (10 cfu/g) and 93 of 262 (36%) oyster samples were less than the limit of detection (0.3 cfu/g) in the Washington State monitoring data (Washington State Department of Health, 2000; 2001). For regression analysis, it was assumed that V. parahaemolyticus was present in these non-detect samples at levels less than the detection limit (i.e., the true density was below the limit of detection) but never zero (see discussion of Tobit regression below).

Regression Analysis. Tobit regression is a maximum likelihood procedure for which the likelihood of the data reflects both the probability of obtaining non-detectable and detectable density levels. The influence of non-detectable outcomes is determined by the probability of the density in a sample falling below a fixed limit of detection. The Tobit regression method was used to avoid bias and underestimation of variance of the total predicted V. parahaemolyticus densities. For example, if the non-detectable values are replaced with zeros or with half the limit of detection and a regression line is fit to the data then the estimated relationship of total V. parahaemolyticus densities versus water temperature could be substantially biased towards higher or lower levels. Imputing the non-detectable values (such that the value is between zero and the non-detectable limit) rather than assume they are zero or half the limit of detection reduces the bias of the estimate. See Appendix 5 for details about the Tobit regression analysis procedures and results.

Plots of the best fitting regression line versus temperature and the associated 5^th and 95^th confidence intervals are shown in Figures IV-3 through IV-5 for each of the three data sets. In these figures, non-detectable V. parahaemolyticus levels were replaced with randomly imputed values (open circles) based on the maximum likelihood estimate (MLE) of the regression relationship. Regression analysis of the three data sets indicated that the effect of temperature on the mean log₁₀ total V. parahaemolyticus densities was approximately linear in the range of water temperatures sampled.

Figure IV-3. Tobit Regression Fit of Vibrio parahaemolyticus Densities in Oysters Versus Water Temperature Using the DePaola et al. (1990) Data Set
[Solid line is the best estimate of the median V. parahaemolyticus/g. Dashed lines show the 5th and 95^th % confidence limits. Closed circles are V. parahaemolyticus detectable values from DePaola et al., 1990. Open circles are randomly imputed values for samples with densities less than the limit of detection (10 cfu/g).]

Figure IV-4. Tobit Regression Fit of the Vibrio parahaemolyticus Densities in Oysters Versus Water Temperature Using the FDA/ISSC (2001) Data Set
[Solid line is the best estimate of the median V. parahaemolyticus/g. Dashed lines show the 5th and 95^th % confidence limits. Closed circles are V. parahaemolyticus detectable values from FDA/ISSC, 2001. Open circles are randomly imputed values for samples with densities less than the limit of detection (10 cfu/g).]

Figure IV-5. Tobit Regression Fit of the Vibrio parahaemolyticus Densities in Oysters Versus Water Temperature Using the State Department of Health (2000; 2001) Data Sets
[Solid line is the best estimate of the median V. parahaemolyticus/g. Dashed lines show the 5th and 95^th % confidence limits. Closed circles are V. parahaemolyticus detectable values from Washington State Department of Health (2000; 2001). Open circles are randomly imputed values for samples with densities less than the limit of detection (0.3 cfu/g).]

In order to develop a more accurate predictive distribution for total V. parahaemolyticus density (cfu/g oyster) in harvest waters, the method error for the data described in Table IV-3 was estimated and then subtracted from the estimated variance about the regression fit to obtain an estimate of population variation. This correction is important to prevent an inappropriate over estimation of the variance of V. parahaemolyticus densities. See Appendix 5 for the determination of independent estimates of method error to correct the variances.

Uncertainty. The results of the Tobit regression analysis of the three data sets were used to generate 1,000 sets of parameters for the relationship of water temperature to total V. parahaemolyticus densities in oysters. These sets of regression parameters were used to represent uncertainty of the water temperature relationship and variance of total V. parahaemolyticus densities in the Monte Carlo simulations. For the Gulf Coast, Mid-Atlantic and Northeast Atlantic regions, the uncertainty from the regression analyses shown in Figures IV-3 and IV-4 were used. Approximately 500 sets of parameters from distributions of the model fits to these data sets were obtained and combined. The resulting 1,000 sets of parameters were used once for each of the 1,000 model simulations for these three regions. For the Pacific Northwest region the 1,000 parameters were obtained from the distribution shown in Figure IV-5.

The effect of regression parameter uncertainty was implemented in the risk assessment by using a multivariate normal approximation for parameter uncertainty for each of the three data sets. Accounting for the effect of the uncertainty in the data sets was implemented in Monte Carlo simulations by generating a sample of 1,000 sets of parameters from the uncertainty distributions. Independent estimates of method error for each of the three data sets were then used to correct this additional variance in the observed data. See Appendix 5 for detailed discussion of how the regression parameter uncertainty was assessed based on a multivariate normal approximation.

Growth of Vibrio parahaemolyticus During Intertidal Exposure

A significant portion of the oysters in the Pacific Northwest are harvested when oyster reefs are exposed during the course of the tide cycle. Exposure to the air and radiative heating of oysters in bright sunlight can elevate oyster temperatures substantially above that of the water (and air) temperature. To model the effect of intertidal harvesting on V. parahaemolyticus densities in the Pacific Northwest, the effect of elevated oyster temperatures and duration of exposure during the collection process was modeled as a separate growth step occurring prior to that associated with transport of the harvest to processing facilities at ambient air temperature. The loglinear growth rate model described in the Post-Harvest module below was used.

To predict the growth of V. parahaemolyticus in intertidal harvested oysters prior to refrigeration, the growth rate model was applied twice. It was first applied to determine the extent of growth that corresponds to 4 to 8 hours of intertidal exposure and secondly to determine the extent of growth that occurs during subsequent transportation (1 hour).

The proportions of days that are cloudy, partly cloudy and sunny during the summer in the Pacific Northwest are about 33% each, respectively (National Weather Service, 2002). Given that the most significant elevation of oyster temperature is likely to occur during exposure under sunny conditions the recent studies of intertidal exposure in the Pacific Northwest (DePaola et al., 2002; Herwig and Cheney, 2001), conducted over multiple sampling occasions, likely reflect the varying effects of sunny versus cloudy conditions. The range of oyster versus air temperature differences observed in these studies was 0 to 10°C. More definitive information is lacking and, based on the range of observations alone, a uniform distribution with a range of 0°C to 10°C was considered a reasonable representation of both the variability and uncertainty of the average difference in oyster versus ambient air temperature during periods of intertidal exposure. With respect to duration of exposure, oysters are typically collected by barge at the time of the incoming tide at the collection site. Consequently, the duration of exposure can be expected to vary as a consequence of the varying depth of the oyster reefs relative to the maximum tide height. Considering the likely range of depths of commercial reefs, a range of exposures of between 4 to 8 hours was assumed with all values within this range considered equally likely. The uniform distribution chosen represents uncertainty as well as the variability in the duration of exposure likely to occur.

Not all of the Pacific Northwest harvest is collected after intertidal exposure. A smaller, but still significant portion of the overall harvest is collected by dredging submerged oyster reefs and, consequently, for this portion of the harvest the densities at time of collection were modeled based on water temperature (i.e., without an intertidal growth step), as was done for the other regions of the country where there is no intertidal harvesting. The estimate of the proportion of the Pacific Northwest harvest that is collected during intertidal cycles was obtained based on data for average shellstock harvest volume in four major harvest areas of Washington State from 1990 to 2000 (Kaysner, 2002) and expert opinion on the percentage of harvest that is collected intertidally in these selected areas. This combination of harvest data and expert opinion indicated that the overall statewide percentage of shellstock harvested after intertidal exposure is approximately 75% of the total harvest for all seasons. Since Washington State is the largest harvest area in the Pacific Northwest this statistic was considered representative of the region as a whole. Thus, the intertidal growth calculation described here was assumed to apply to 75% of the Pacific Northwest harvest.

Ratio of Pathogenic to Total Vibrio parahaemolyticus Levels in Oysters

Seven studies were identified which provide data on the relationship between total and pathogenic V. parahaemolyticus in oysters (Table IV-4). In these studies, samples were analyzed for pathogenic V. parahaemolyticus (TDH⁺). The microorganisms isolated from the TDH⁺ samples were further analyzed to determine the percentage of the total V. parahaemolyticus microorganisms in the oysters that are pathogenic. Differences were observed in the various United States regions with higher percent pathogenic values observed in the Pacific Northwest compared to the Gulf Coast and Atlantic regions.

Two studies, DePaola et al. (2002) and Kaufman et al. (2003) were selected as the most appropriate for estimating the distribution of pathogenic to total V. parahaemolyticus in oysters, based on the criteria described in Table IV-1c. The data from these two studies indicated that the number of pathogenic V. parahaemolyticus in sample portions was frequently non-detectable. In addition, high numbers of pathogenic microorganisms were sometimes observed in samples that had low counts of total V. parahaemolyticus in replicate samples. Some degree of variation is expected due to the natural processes of growth and competition between different strains of V. parahaemolyticus in the presence of other micro flora in the oysters. Additionally, the study by DePaola et al. (2003a) suggests that there may be some seasonal variation in the percentage of V. parahaemolyticus that are pathogenic. However, this finding has not been replicated in other studies. Accordingly, for the purpose of this risk assessment, the ratio between pathogenic and total V. parahaemolyticus densities was assumed to be temperature independent.

The studies representing different regions in the United States were analyzed separately. The study by DePaola et al., (2002) was conducted in the Hood Canal area and represented the Pacific Northwest region. The study by Kaufman et al. (2003) was conducted in the Gulf Coast. It was assumed that the percentage pathogenic data from the Gulf Coast region can also be used to represent the Mid-Atlantic and Northeast Atlantic regions. This assumption was based on the data by Cook et al. (2002b) which showed that there was no apparent difference in the percentage of TDH⁺ V. parahaemolyticus in oyster samples among the Gulf Coast, Mid-Atlantic, and Northeast Atlantic regions.

Given the low densities of pathogenic V. parahaemolyticus in oysters and the resulting high frequency of non-detectable amounts in samples, the distributions of percentage pathogenic were estimated based on the assumption that pathogenic counts in sample portions were distributed according to a Beta-Binomial distribution. The Beta-Binomial distribution is a flexible two-parameter distribution commonly used to model variability of proportions (see Appendix 5 for additional information). In applying the Beta-Binomial distributional model to the Gulf Coast and Pacific Northwest data, the amount of pathogenic V. parahaemolyticus observed in a given sample portion is assumed to be binomially distributed with size parameter equal to the number of total V. parahaemolyticus expected in that sample volume. This is based on the number of total V. parahaemolyticus actually observed in the corresponding sample portion assayed for total V. parahaemolyticus. The probability parameter of the binomial distribution for pathogenic counts per sample is assumed to be randomly distributed according to a Beta distribution with unknown parameters a and b. The a and b parameters defining the distribution of percentage pathogenic were estimated based on the observed counts of total and pathogenic V. parahaemolyticus and sample volumes by maximizing the Beta-Binomial likelihood of the observed data. The resulting estimates of the mean of the distribution of percentage pathogenic (P) for the various harvest regions are given in Table IV-5. See Appendix 5 for details.

Uncertainty. The studies by Kaufman et al. (2003) and DePaola et al. (2002) provide information which is sufficient for estimation of the parameters for the Beta distribution of the percentage pathogenic V. parahaemolyticus. However, there is uncertainty associated with the estimates due to the limited sample sizes of the studies, particularly in regard to the volume of sample examined for pathogenic V. parahaemolyticus. There is also the possibility that the distribution of percentage pathogenic V. parahaemolyticus changes from one year to the next in response to changing environment conditions. In this regard, conditions in the Gulf Coast and Pacific Northwest during the summer of 2001 (when the two studies were conducted) appear to have been close to the norm. That is, the estimates of the mean percent pathogenic V. parahaemolyticus obtained on the basis of these studies are comparable to the estimates reported in Table IV-4 based on studies conducted in previous years. It is unknown at present the extent to which the distribution of percentage pathogenic may vary or how extreme (high or low) the mean and variance of the percent pathogenic V. parahaemolyticus distribution might fluctuate from one year to the next. In order to evaluate the effect of these uncertainties on the predicted illness rates, the uncertainty associated with the a and b parameter estimates was determined by using a parametric bootstrap procedure. See Appendix 5 for details.

For each region/season combination, the density of pathogenic V. parahaemolyticus at harvest was obtained by multiplying the density of total V. parahaemolyticus at harvest, as influenced by water temperature, with a value for the percentage of total V. parahaemolyticus that are pathogenic that was generated by a beta distribution with specific parameters. These parameters were derived to account for the uncertainty of what the actual percent pathogenic truly is by a multivariate analysis of the harvest data. Based on an analysis of the data, 1,000 plausible beta distribution parameters with an overall mean of 2.3 % was generated for the Pacific Northwest and 0.18% was generated for all other regions except the Pacific Northwest. These 1,000 plausible beta parameters were used once in the 1,000 simulations, but each set of parameters was used to generate 10,000 individual estimates of percent pathogenic during the model iterations.

Output of the Harvest Module

The output of the Harvest Module is the level of total and pathogenic V. parahaemolyticus in oysters at the time of harvest. For each region/season combination, the distribution of pathogenic V. parahaemolyticus at harvest was obtained by combining the distribution of total V. parahaemolyticus at harvest, as influenced by water temperature, with the appropriate distribution for the percentage of total V. parahaemolyticus that are pathogenic. Specific details of these calculations, the Monte Carlo methods used, and their implementation in @Risk (Palisade) based on the distributions and relationships as described above, can be found in Appendix 3.

Table IV-6 shows the mean and confidence intervals of the uncertainty distributions of the mean levels (i.e., the averages with respect to variability) of total and pathogenic V. parahaemolyticus at harvest for each of the 24 region/season combinations. The uncertainty in the mean estimates is also represented in Table IV-6 as the upper and lower bounds of the confidence limits (see discussion below). A comparison of mean total and pathogenic V. parahaemolyticus levels across these 24 region/season combinations indicates that, as expected, the Gulf Coast values are considerable higher than the other regions due to the warmer water temperatures in the Gulf. The levels of V. parahaemolyticus in the mid-Atlantic and Northeast Atlantic Summer are higher than those of the Pacific Northwest (when harvest occurs by dredging). Even during the summer, water temperatures in the Pacific Northwest are cooler (~11 °C), on average, than in the other Gulf and Atlantic regions. However, exposure to ambient temperatures for longer time periods, such as occurs during intertidal harvest in some Pacific Northwest areas, allows for additional growth of the microorganism, resulting in an increase in those levels to levels higher than for the mid- and Northeast Atlantic.

Uncertainty. The output of the model simulations is a two-dimensional variability and uncertainty distribution for each region/season combination. At fixed values of the uncertainty parameters, the resulting one-dimensional distributions represent model predictions of the intrinsic variation of V. parahaemolyticus densities at time of harvest (i.e., variation from one collection of oysters to the next), conditional on the values of the uncertainty parameters. These variability distributions were found to be positively skewed (i.e., close to lognormal) suggesting that the variability of total V. parahaemolyticus/g at fixed temperature dominates the effects of variations of temperature (within each region/season).

It should be noted that, while the ratio of pathogenic to total V. parahaemolyticus values are close to the mean of the percent pathogenic distribution (as estimated and discussed above) the values do not match precisely because of the random approximation inherent to the Monte Carlo simulation (Appendix 3). The width of the confidence intervals gives an indication of the uncertainty of the predictions with an approximate 10-fold to 20-fold range, depending upon the region/season and the output variable.

It is also worth noting that the variability of pathogenic V. parahaemolyticus/g is greater than that of total V. parahaemolyticus/g. This is a consequence of the fact that, for pathogenic V. parahaemolyticus/g, there is the added effect of the variability of the percent pathogenic from one collection of oysters to the next. An appropriate summary of these two-dimensional distributions of the output variables is the one-dimensional uncertainty distribution of the mean of the variability distribution(s). Although other statistics and percentiles of the variability distributions have relevance with respect to the extremes of exposure that may occur on the individual level, it is the mean of the variability distributions that is the single most relevant measure of population exposure and hence the most pertinent for comparisons across different region and season categories.

Post-Harvest Module

The Post-Harvest Module predicts the effects of typical industry practices on V. parahaemolyticus densities in oysters during transportation, distribution and storage from harvest through retail. Factors that influence the levels of pathogenic V. parahaemolyticus in oysters (i.e., growth or die-off) include: ambient air temperatures at time of harvest; time from harvest until the oysters are placed under refrigeration; time it takes the oysters to cool once under refrigeration, and length of refrigeration time until consumption.

Growth and Survival. The growth and survival of V. parahaemolyticus in shellstock oysters has been studied. Cook and Ruple (1989) reported that levels of V. parahaemolyticus increase at temperatures above 10 °C, but in most cases did not detect an increase during storage at 10 °C. After one day of storage at either 22 °C or 30 °C the levels of V. parahaemolyticus were 2 to 3 orders of magnitude higher than those at harvest. Gooch et al. (2002) reported a 50-fold increase in V. parahaemolyticus levels after storage at 26 °C for 10 hours and a 790-fold increase after 24 hours. After refrigeration at 3 °C for approximately 14 days a 6-fold decrease in the levels was observed. The results from these studies indicate that V. parahaemolyticus can grow rapidly in unrefrigerated oysters.

Data Selection and Criteria for the Post-Harvest Module

The selection of data for use in the Post-Harvest Module considered the availability of data and limitations of the data sources. Model inputs (i.e., data or assumptions) included the following.

1. To calculate the growth of V. parahaemolyticus in oysters from harvest to initial refrigeration, model inputs were needed for the duration of harvest, time-to-refrigeration, oyster temperature, and growth rate. Air temperature was used as a surrogate to estimate oyster temperature.
2. To calculate the growth of V. parahaemolyticus in oysters from initial refrigeration until cooled to a no-growth temperature, model inputs were needed for the cooldown time and growth rate during cooling.
3. To calculate the levels of V. parahaemolyticus in oysters from refrigeration to retail, model inputs were needed for the die-off rate and duration of cold storage.

Data were generally not available for the temperature of oysters after harvest. It was assumed that the temperature of oysters would equilibrate with the air temperature. Therefore, the air temperature data from the comprehensive NBDC database were used for each region/season combination. All identified studies were used in the model to provide information for time from harvest to refrigeration, growth/decline rate of V. parahaemolyticus in oysters during storage, and storage time between refrigeration and consumption.

Assumptions for the Post-Harvest Module

1. The growth and survival of pathogenic V. parahaemolyticus in harvested oysters is the same as total V. parahaemolyticus.
2. The relative growth rate of total V. parahaemolyticus in oysters versus broth culture conditions is temperature independent.
3. Oysters equilibrate rapidly with that of ambient temperature after harvest and prior to refrigeration; ambient air temperature is a surrogate for oyster meat temperature. For Pacific Northwest (Intertidal) region, oyster temperature is greater than air temperature because of the effect of direct sunlight.
4. Air temperature at noon is representative of the environmental temperature that oysters are subject to after harvest and prior to refrigeration. (This assumption does not apply to the Pacific Northwest (Intertidal) region.)
5. Water activity of oysters does not vary substantially.
6. NSSP guidelines for the maximum time that oysters can remain unrefrigerated after harvest are never exceeded.
7. The extent of growth occurring over time at a given average temperature and predicted maximal growth rate is assumed to follow a simple three-phase loglinear model with no lag phase (Buchanan et al., 1997).
8. Value for the maximal density at all temperatures approaches a plateau of approximately 10⁶ total V. parahaemolyticus per gram after 24 hours (Gooch et al., 1999; 2002). [Note: To ensure that levels of pathogenic V. parahaemolyticus do not exceed the value equivalent to 10⁶ total V. parahaemolyticus, the simulation model was run separately, but in parallel for total and pathogenic V. parahaemolyticus (see Appendix 3).]
9. Oysters are harvested uniformly from the start of the harvest up to one hour prior to conclusion of the harvesting operation. (This assumption does not apply to the Pacific Northwest (Intertidal) region.)
10. The duration of time until oysters reach "no-growth" temperature after being placed under refrigeration varies uniformly between 1 and 10 hours.
11. Once "no-growth" temperature is attained no further growth occurs during storage and transport through the retail market.
12. No temperature abuse or mishandling occurs at retail, eating establishments, or as a result of consumer behavior.

Modeling the Post-Harvest Module

The various model inputs and output for the Post-Harvest Module are illustrated in Figure IV-6 and discussed in detail below.

Figure IV-6. Schematic Depiction of the Post-Harvest Module of the Vibrio parahaemolyticus Exposure Assessment Model
[Vp/g is Vibrio parahaemolyticus per gram oyster. Levels of total and pathogenic V. parahaemolyticus were simulated by the model separately and in parallel.]

Studies and Data Sources Used for the Post-Harvest Module

1. Growth rate of V. parahaemolyticus: The growth rate was based on estimates obtained from Miles et al., 1997 and Gooch et al., 2002.
2. Time from harvest to refrigeration: Information from a 1997 GCSL survey was used to estimate the duration of harvesting operations under current industry practices (Gulf Coast Seafood Laboratory, 1997) for the Gulf Coast States. The Gulf Coast practices were assumed to be representative of the Pacific Northwest, Mid-Atlantic, and Northeast Atlantic regions.
3. Oyster Temperature Distributions: Air temperature data from the National Buoy Data Center (NBDC) were used as a surrogate for oyster temperature for all regions with the exception of the Pacific Northwest intertidal. For intertidal harvesting, oyster temperature was based on NBDC air temperature, oyster versus air temperature differences (DePaola et al., 2002; Herwig and Cheney, 2001), and the National Weather Service (NWS, 1999) data on the proportion of days that are cloudy, partly cloudy and sunny.
4. Die-off rate during cold storage: Data (a point estimate) from Gooch et al. (2002) were used for all regions and seasons.
5. Cold storage time: Data from Cook et al. (2002a) (originally reported as FDA/ISSC, 2000) were used for all regions and seasons.

Growth of Vibrio parahaemolyticus from Harvest to First Refrigeration

The extent of growth that occurs during the period of time from harvest until the time that oysters are first placed under refrigeration is determined by four factors:

1. the duration of harvest,
2. the growth rate of V. parahaemolyticus as a function of air temperature,
3. the temperature of oyster meat following harvest, and
4. the length of time held unrefrigerated.

Additionally, for the Pacific Northwest, V. parahaemolyticus densities at time of harvest are influenced by whether or not oysters are collected intertidally.

Growth Rate Model

Gooch et al. (2002) is the only study identified which observed the post-harvest growth in oysters and it was limited to only one temperature (26 °C). Therefore, a model of V. parahaemolyticus growth in microbiological broth medium was used (Miles et al., 1997) to predict growth of V. parahaemolyticus in oysters at a range of temperatures. The predictions of this model were adjusted to predict the growth rate of total V. parahaemolyticus in oysters. An upper limit of 10⁶ was set for the maximum density of total V. parahaemolyticus in oysters. Based on a study by Cook (2002a), the growth and survival of pathogenic and total V. parahaemolyticus in oysters after harvest were considered to be the same. Cook (2002a) reported that the presence of the tdh gene that codes for pathogenicity does not alter the growth rate of V. parahaemolyticus under typical temperature conditions.
Miles et al. (1997) studied the growth rate of four strains of V. parahaemolyticus in broth cultures at different temperatures and water activities. For each combination of temperature and water activity, the extent of bacterial growth observed was modeled using the Gompertz function. This is a sigmoid growth curve with a growth rate (slope) that increases up to a maximum rate (μ_m ) and then falls to zero as the bacterial population reaches a steady state. A plot of the resulting model prediction for μ_m as a function of temperature is a unimodal function with a maximum value and no growth rate outside of the predicted range of temperatures favorable for growth.
It was assumed that water activity of oysters does not vary substantially with a nominal value equal to the optimal value of 0.985 predicted to occur under broth culture conditions. At this water activity, the predicted growth rate in broth at 26 °C (78.8 °F) is 0.84-log₁₀ per hour, which is approximately a 7-fold increase in density per hour. This is approximately four times greater than the rate of growth observed for V. parahaemolyticus in oysters held at 26 °C (78.8 °F) (Gooch et al., 2002).

Therefore, for the risk assessment model, the predictions of the growth rate in broth cultures were divided by a growth rate factor. This factor was estimated based on Gooch et al., (2002) experimental data, but to account for uncertainty, a triangle distribution with a range of 3 to 5 and mean of 4 was used in the model.

After transfer of an inoculum to different medium or environmental conditions there is typically a demonstrable lag phase during which time the bacterial population adapts to different environmental conditions and growth is sub optimal. This lag phase is commonly modeled by a sigmoid growth function such as the logistic or Gompertz. However, a sigmoid growth function (e.g., Gompertz) is not an appropriate model for growth of V. parahaemolyticus in oysters after harvesting, as changes in environment are typically gradual and do not arrest the growth rate and induce a lag phase. Consequently, the extent of growth occurring over time at a given average temperature was assumed to follow a simple three-phase loglinear model with no lag phase (Buchanan et al., 1997). This model is of the form:

log₁₀(N(t)) = min{log₁₀(N(0)) + μ_m * t,A}

where N(0) refers to bacterial density at harvest, N(t) refers to the bacterial density at a given time (t) post-harvest, A is the logarithm of the maximum attainable density of V. parahaemolyticus in oysters, and the parameter μ_m (the maximal growth rate) is a function of ambient temperature. At 26 °C, the density of V. parahaemolyticus in oysters was observed to approach a plateau of approximately 6.0-log₁₀ per gram after 24 hours (Gooch et al., 1999; 2002). This value was assumed for the maximal density (A) at all temperatures. Figure IV-7 shows the predictions (mean) of the log₁₀ increase in V. parahaemolyticus density from an initial level of 1,000/g as a function of time for three ambient temperatures, 20, 26 and 32 °C (68, 78.8, and 89.6 °F).

Oyster Temperatures

Ideally, the average temperature of oyster meat would be used to determine the growth rate parameter (μ_m) in the above equation. This temperature varies due to the temperature of both the air and water at the time of oyster harvest. The temperature of the oyster meat after harvest can be reasonably expected to equilibrate to that of the air although this may be modified somewhat by evaporative cooling and the extent to which oysters are properly shaded from direct sunlight aboard ship. This expectation was confirmed by warming/cooling experiments using a temperature probe, which indicated that individual oysters equilibrate rapidly to air temperature (i.e., <30 minutes) from initially wide temperature differences. When oysters were placed in a sack the rate of equilibration was observed to be slower (i.e., ~2 hours) and complete equilibration did not occur due to the effect of evaporative cooling (Cook, 2001). However, it was assumed that the temperature of oyster meat equilibrates rapidly with that of the ambient air. Therefore air temperature was used as a surrogate for oyster meat temperature for oysters harvested by dredging. For oysters harvested in intertidal areas, additional growth of V. parahaemolyticus was considered (see section titled, "Growth of Vibrio parahaemolyticus During Intertidal Exposure" in the Harvest Module section).

Figure IV-7. Predicted Mean Loglinear Growth of Vibrio parahaemolyticus in Oysters from an Initial Density of 1,000 (3-log₁₀) Vibrio parahaemolyticus per gram as a Function of Ambient Air Temperature

Air Temperature Distributions

Air temperature data were used as a surrogate for oyster temperature data because of limited data of the temperatures in oysters under different environmental conditions. For all regions except the Pacific Northwest (Intertidal), ambient air temperature data recorded at midday from the near-shore NBDC (National Buoy Data Center; http://www.ndbc.noaa.gov/index.shtml) buoys were used for this purpose. Examination of water and air temperatures obtained from the NBDC database show a strong correlation between water and air temperature. This correlation has been incorporated into the model by using the distribution of the difference in water temperature versus air temperature. The temperature difference distributions along with the water temperature distributions (from the Harvest Module) are used in the Post-Harvest Module simulations to predict air temperature. The difference in air and water temperature was found to be well characterized by a normal distribution. The parameters for the normal distribution were different for each region/season combination (see Appendix 3 for link to spreadsheets for this information). The distributions of difference in air temperature versus water temperature were obtained by pooling the data available for each near-shore buoy across all available years. The mean and standard deviation of these distributions are shown in Table IV-7.

Distribution of Time Oysters are Unrefrigerated

For oysters harvested by dredging, the distribution of the length of time that oysters are held unrefrigerated was inferred based on the distribution of duration of daily oyster harvesting operations (i.e., the combination of harvesting and transportation time). The distribution of time that oysters are unrefrigerated was obtained by assuming that oysters are collected uniformly from the start of the harvest up to one hour prior to conclusion of the harvesting operation when oysters are landed and placed in cold storage. An additional hour was assumed to be representative of the duration of transportation time to the processing facility, although this may vary somewhat for different harvesting regions. The derived distribution for time unrefrigerated reflects the fact that oysters collected at the start of the harvesting operation are exposed to ambient air temperatures for a longer period of time than those collected towards the end of harvesting operations. Consequently the mean time that oysters remain unrefrigerated is much less than the maximum duration of harvesting might suggest.

Information from a 1997 GCSL survey was used to estimate the duration of harvesting operations under current industry practices (GCSL, 1997). The survey was conducted in several Gulf Coast states during the fall of two successive years; one season prior to initiation of the NSSP time-to-refrigeration requirements (for states whose product has been confirmed as the source of two or more V. vulnificus illnesses), and then the following year after implementation. Duration of harvest was reported to be longer in Louisiana than in Florida and Texas, during both years. This probably reflects more remote oyster harvesting areas in Louisiana relative to other states on the Gulf Coast. Also, the duration of harvesting operations was reported to be shorter after the implementation of the NSSP guidelines due to compliance of the harvesters with the new requirements that took effect in 1996.

Data on the duration of harvesting during seasons other than the fall were not obtained during the 1997 GCSL survey. However, given the water temperature thresholds at which the NSSP time-to-refrigeration requirements are specified to be in effect, duration of harvesting during the spring and summer can be reasonably inferred to be similar to that reported during the fall. Therefore, the current duration of harvesting in the Gulf Coast during the spring, summer and fall was assumed to be equal to that reported in the 1997 GCSL survey during the fall of 1996, when the NSSP time-to-refrigeration requirements were in effect. The current duration of harvesting during the winter was assumed to be equal to the duration of harvesting that was reported prior to the implementation of the NSSP guidelines (fall of 1995) because, when cooler water conditions prevail, the NSSP requirements are not as stringent. A distinction between Louisiana and the rest of the Gulf Coast states was made based on the apparent differences in the reported durations of harvesting in the 1997 GCSL survey. Louisiana represents roughly half of the Gulf Coast harvest.

No data were identified for the duration of harvesting operations in regions other than the Gulf Coast. Consequently, estimates for other regions were inferred based on selected states included in the 1997 GCSL survey. The practices of Florida and Texas were assumed to be representative of the Pacific Northwest, Mid-Atlantic, and Northeast Atlantic regions. In the absence of conflicting information, the longer (pre-1996) reported harvesting durations were taken to be appropriate for all seasons, since temperature thresholds at which more stringent time-to-refrigeration requirements would take effect would not commonly be exceeded outside of the Gulf Coast.

Table IV-8 shows the minimum, maximum and the most likely durations of oyster harvesting that have been inferred to apply for each of the different regions and seasons based on the 1997 GCSL survey data. Beta-PERT distributions were fit to these data to obtain smooth and continuous estimates of the distributions of the harvest durations. A Beta-PERT distribution is commonly used to infer a continuous distribution when the available data or expert opinion identifies only the range and most likely value of the parameter to be modeled. Figure IV-8 shows an example Beta-PERT distribution with minimum of 2, maximum of 11 and mode of 8 hours.

Figure IV-8. Example Beta-PERT Probability Density Distribution for Duration of Oyster Harvesting

Growth of Vibrio parahaemolyticus During Cooldown

Vibrio parahaemolyticus will continue to grow in oysters after they are placed under refrigeration until the temperature of the oyster tissue falls below a certain threshold (e.g. 8 °C) (46.4 °F) (Cook and Ruple, 1989). The time it takes for oysters to cool once under refrigeration is presumably quite variable depending on efficiency of the cooler, quantity of oysters to be cooled and their arrangement in the cooler. Data on cooling rates of commercial oyster shellstock could not be located. Preliminary GCSL experiments with a single in-shell oyster at 30 °C (86 °F) in which a temperature probe was inserted into its tissue indicated a cooling rate of approximately 0.5 °C (0.9 °F)/min when placed into a 3 °C (37.4 °F) cooler (DePaola, 1999). However, 24 oysters in an uninsulated plastic container required approximately 7 hours to drop from 26 °C (78.8 °F) to 3 °C (37.4 °F). In another GCSL study, one bushel of commercial size oysters (>3" hinge to bill) contained in a burlap sack was tempered to 25 °C. Using thermocouples inserted in oysters at different depths of the bushel, the investigator found that the oyster on the bottom of the sack cooled to 10 °C in 1.9 hr. (Contact with the cold floor of the cooler probably hastened its cooling.) The oysters in the center of the sack required 2.1 and 2.6 hr. to cool to 10 °C. The oyster in the top of the sack cooled in 2.2 hr. The single oyster outside the sack cooled to 10 °C in 0.3 hr (Cook, 2002b).

These data suggest considerable variability in the cooling rate depending upon the load and/or configuration of the oysters to be cooled. The cooling rate would also depend on the temperature of the cooler, which is likely to vary (FDA/ISSC, 2000). The distribution of cooler temperatures/efficiencies in the industry (e.g., both wholesale and retail establishments) is an uncertainty impacting the estimation of an appropriate distribution for the cooldown time. Based on this observation, a rectangular distribution between 1 and 10 hours was used for the cooldown time to represent both the variability (e.g., due to load and/or configuration of oysters in a cooler) and the uncertainty inherent due to lack of knowledge concerning cooler temperatures and typical loading conditions.

As oysters cool down to storage temperatures the growth rate of V. parahaemolyticus slows with the declining temperature of the oyster tissue. At the start of the cooldown period, when oysters are first placed under refrigeration, the growth rate is still equal to the initial rate as determined by ambient air temperature. Assuming that no appreciable temperature abuse occurs after oysters have been placed in cold storage, further growth stops at the end of the cooldown period when oysters have reached a no-growth storage temperature. Beyond these reasonable assumptions little data are available as to the shape of the cooling curve, which is likely to depend on the loading and/or configuration of oysters in the cooler and the cooler temperature. Both of these factors are likely to vary under actual industry practice. Given this identified uncertainty, it was assumed that during the period of cooldown, the growth rate of V. parahaemolyticus drops linearly down to zero. This assumption may overestimate the growth that occurs if the temperature equilibration follows an exponential law (i.e., Newton's Law of Cooling). However, typical loading and configuration of oysters in sacks stacked on pallets can be reasonably expected to reduce convective flow of chilled air and thereby slow equilibration of oysters to the cooler temperature (Schwarz, 2003b). Thus an exponential cooling rate was considered unlikely with respect to most of the harvest.
Given the assumption of a linear cooling curve, a discrete approximation was used to model the amount of growth occurring during cooldown. Conditional on the duration of the cooldown period, the extent of growth during each hour of the cooldown period was approximated as an average growth rate during that hour times a duration of one hour. These average growth rates were determined by the duration of the cooldown period, the growth rate prior to refrigeration (i.e., as determined by the ambient air temperature for a given oyster lot), and the assumed linearity of the cooling curve. These calculations of average growth rate per hour consistent with the linear cooldown rate assumption are illustrated in the Table IV-9, where, for example, it takes T hours for a particular oyster lot to reach cooler temperature.

The total additional growth was then obtained as the sum of these values over the cooldown period subject to the restriction that the maximum density of 6.0-log₁₀ per gram could not be exceeded. Specifically, the potential amount of additional growth is the sum of the growth over the T hours:

and this amount of additional growth is truncated by the assumption of a maximum density according to the following formula:

where N represents the density of V. parahaemolyticus at the time of first refrigeration and A is the maximum attainable density (6-log₁₀ per gram). Since the cooldown time T is a random variable with a mean of 5.5 hours, the average extent of growth is 3.25*μ_m in the absence of the truncation effect, where μ_m is the maximal growth rate determined by ambient air temperature at time of harvest. Thus, for an initial growth rate of 0.19-log₁₀ per hour (i.e., at 26 °C), the average growth occurring during cooldown is approximately 0.6-log₁₀ when densities at time of first refrigeration are generally below the maximum density, as is typically the case.

Change in Levels of Vibrio parahaemolyticus During Cold Storage

Gooch et al. (2002) showed that in oysters, V. parahaemolyticus levels declined 6-fold (0.8-log₁₀ cfu/g) when stored 14 to 17 days at 3 °C. This average rate of change was used as a point estimate of the rate of decline considered typical of refrigerated oysters in the marketplace, although some error may be introduced because commercial oysters are typically stored at higher temperatures (5-10 °C). This observation is supported by analysis of V. parahaemolyticus levels in retail oysters sampled from commercial establishments which suggests a decline of 0.04-log₁₀ cfu/g per day (FDA/ISSC, 2000; Cook et al., 2002a). Both estimates are potentially biased to over predicting the extent of decline due to the fact that chill-stressed V. parahaemolyticus may not have been recovered by the methods used in these studies. However, in the Gooch et al. study, one of the enumeration methods used employed a repair step in a medium containing magnesium, which has been shown to increase recovery of chill-stressed cells. This method did not result in higher V. parahaemolyticus counts after refrigeration than the other measurement methods that were used. Therefore, the potential bias due to the effect of chill-stress was considered negligible. The estimate of the storage effect based on the Gooch et al. study was considered the more reliable estimate because the study was conducted under controlled conditions. The estimate based on the ISSC/FDA retail study is potentially confounded and/or biased by factors other than storage time.

Cold Storage Time

Data from the ISSC/FDA retail study for the time between harvest and sample collection were assumed to be a reliable estimate for the length of refrigeration time (Cook et al., 2002a). Summary statistics on the storage time for samples obtained during the study are shown in Table IV-10. A small degree of error may be introduced by assuming that these data are representative of storage time in so far as samples were generally collected on Monday or Tuesday and most servings are consumed in restaurants on weekends. Since this was a year long nationwide survey, the mean of 7.7 days and range of 1 to 21 days was assumed to be representative of all seasons and regions. A Beta-PERT distribution was utilized based on these statistics to infer the range and magnitude of variation expected to occur in the duration of storage time.

The effect of storage was modeled by combining the distribution of storage times with the point estimate of the rate of change in V. parahaemolyticus levels per day. Thus, it is assumed that storage temperatures are always below the "no-growth" temperature for V. parahaemolyticus. The effect of this assumption is to likely underestimate the variance of the change in V. parahaemolyticus densities. During the FDA/ISSC retail study 25% of coolers were found to be >5.5 °C (42° F) and 2.5 % were >10°C (50 °F) at the time of sample collection (FDA/ISSC, 2000; Cook et al., 2002a). A report by the FDA Retail Food Program Steering Committee suggests that 34% of "seafood retailers" practice improper storage conditions, i.e., temperatures >5.5 °C (FDA Retail Food Program Steering Committee, 2000). These estimates of deviation from compliance are relatively consistent and suggest that it is possible that V. parahaemolyticus levels increase in stored oysters However, the ISSC/FDA retail study data indicate an overall average decrease in V. parahaemolyticus levels during storage. The rate of decrease would be anticipated to be higher and the effect less variable if the 5.5 °C standard was consistently maintained.

Output of the Post-Harvest Module

The output of the Post-Harvest module, like that of the Harvest Module, is a two-dimensional variability and uncertainty distribution for each of a set of selected output variables and for each region/season combination. The output variables of interest for the Post-Harvest Module include the levels (i.e., densities) of total and pathogenic V. parahaemolyticus in oysters at the time of consumption. As discussed previously with respect to output of the Harvest module, the most pertinent summary of the two-dimensional variability and uncertainty distributions is the one-dimensional uncertainty distribution of the average levels (i.e., the averages over variability).

Table IV-11 shows the predicted levels of total and pathogenic V. parahaemolyticus in oysters post-harvest. The post-harvest results, in comparison to those shown in Table IV-6 for at-harvest, are indicative of the nominal effects of current post-harvest handling and processing practices on the potential for growth of V. parahaemolyticus in oysters. Vibrio parahaemolyticus levels post harvest are highest in the Louisiana and non-Louisiana Gulf Coast regions as expected, because the levels at harvest were the highest and ambient temperature is much higher in this region than in the other regions, allowing for more growth. The levels in the Louisiana Gulf Coast region are much higher than those in the non-Louisiana Gulf Coast region reflecting the longer time-to-refrigeration data used in the model for the Louisiana oyster harvest.

Consumption Module

The Consumption Module estimates the levels of pathogenic V. parahaemolyticus in a single serving of an oyster meal. The quantity and weight of oysters consumed per serving and the density of pathogenic V. parahaemolyticus/g shellfish at consumption are included in the modeling of this module. The determination of the number of raw oyster servings per annum is also discussed in this chapter and is used in the risk characterization portion of the model to calculate the illnesses per annum from the model-predicted illnesses per serving. Because raw oysters are infrequently consumed in the United States, the number of raw oyster servings was derived using the amount of oyster landings reported by the National Marine Fisheries Service (NMFS) for each region season, the mean weight of oysters per serving, and the likely amount of the harvest that is consumed raw.

Consumption was restricted in scope to domestically harvested product because most United States raw consumption is associated with domestically harvested oysters. Total United States imports of live oysters (which may then be consumed raw) have averaged approximately 3.5 million pounds (meat weight) per year from 1991 to 1998 (Hardesty, 2001). This corresponds to approximately 10% of the average yearly United States domestic harvest volume as reported by the National Marine Fisheries Service (NMFS) from 1990 to 1998. Most of these imported live oysters are from Canada (British Columbia and Prince Edward Island) and are of relatively low risk in consideration of generally cooler water temperatures of northern harvest areas. Although some confirmed United States illnesses have been traced back to imported oysters from Canadian harvest areas (i.e., in the Pacific Northwest), the relative number is very small and hence there is little bias associated with excluding imported oysters from the assessment.

United States exports of domestically harvested oysters generally account for less than 10% of the total United States harvest volume in any given year (Muth et al., 2000; Hardesty, 2001). While oyster landing statistics reported to the NMFS include that intended for both domestic and export markets, the reported landings themselves are likely to be somewhat lower than actual landings (Muth et al., 2000) and therefore there is little bias in assuming that reported landings of oysters to the NMFS provide a reasonable estimate of total domestically produced oyster harvest available for domestic consumption.

Data Selection and Criteria for the Consumption Module

The selection of data for use in the Consumption Module considered the availability of data and limitations of the data sources. Data used in the model included the following:

1. the number of oysters consumed per serving, and
2. the weight of oyster meats.

Number of Oysters Consumed per Serving. The criteria used to select the data used to estimate the distribution of the number of raw oysters consumed per serving is provided in Table IV-12. A nationally representative survey with a large number of raw oyster consumers would be preferable. However, because the best available national survey included a small number of oyster consumers, a regional survey was selected.

Weight of Oyster Meats. Only one large, nationally representative study was identified.

Assumptions for the Consumption Module

1. The consumption patterns by immunocompromised and healthy populations are the same.
2. The percentage of raw oyster consumption does not vary by region or season.
3. All V. parahaemolyticus illnesses are associated with consumption of domestic oysters (i.e., the impact of imported oysters on total illnesses was not evaluated).
4. Raw oyster consumption patterns in Florida are representative for the United States

Modeling the Consumption Module

Distributions of doses of pathogenic V. parahaemolyticus ingested with oyster servings were obtained by combining predicted distributions of pathogenic V. parahaemolyticus per gram with estimated distributions for the number of oysters per serving and the mean weight of individual oysters as shown in Figure IV-9.

Figure IV-9. Schematic Depiction of the Consumption Module of the Vibrio parahaemolyticus Exposure Assessment Model

Studies and Data Sources Used for the Consumption Module

1. Number of raw oysters consumed per serving: Data from a regional telephone survey, conducted by the Florida Agricultural Market Research Center, University of Florida (Degner and Petrone, 1994) was used to estimate the distribution of the number of oysters/serving. This estimated distribution was used for all regions and seasons.
2. Oyster meat weight: Data from the ISSC/FDA retail study (FDA/ISSC, 2000; DePaola, 2002) were used to estimate the distribution of the average gram weight of oysters in a serving at the time of consumption. This estimated distribution was used for all regions and seasons. Data from Kaufman et al. (2003) were used to adjust the reported oyster weights from the ISSC/FDA study for the weight of the mantle fluid.

Number of Raw Oysters per Serving

Data from a regional telephone survey, conducted by the Florida Agricultural Market Research Center, University of Florida (Degner and Petrone, 1994) was used to determine the number of oysters consumed per serving. The survey was conducted during April and May of 1994. It included 1,012 adults in seven metropolitan areas in north and central Florida. Three hundred and six of the respondents reporting raw oyster consumption at least once in the previous year provided self-reported or recall information as to the number of oysters that they typically consumed per serving. These data were used as an estimate of the distribution of number of oysters per serving. The empirical distribution of the survey data is shown in Figure IV-10. The most typical serving sizes reported by the respondents were 6, 12 and 24 oysters, with 12 being the most frequent.

The Florida survey data was assumed to apply nationwide. Potentially, this may be biased somewhat with respect to the number of oysters per serving on the national level since the consumption survey was conducted in a region which is not necessarily representative of the entire country. Also, the survey was conducted in 1994 and even though consumption behavior may be changing from year to year, the estimated distribution of oysters per serving was assumed to apply to current consumption behavior. The magnitude of these potential biases is expected to be small relative to other identified uncertainties.

Figure IV-10. Self-reported Frequency of Number of Oysters Consumed per Serving (University of Florida Consumption Survey) (Degner and Petrone, 1994).

Oyster Meat Weight

The ISSC/FDA retail data (FDA/ISSC, 2000; DePaola, 2002) was used to estimate the gram weight of oysters consumed per serving. In this study, oyster weights were taken for 339 of the 370 samples collected from wholesale and retail locations. Samples generally consisted of 12 oysters (range, 4 to 15) and this included both the oyster meat and the mantle fluid. The average oyster weight per sample (meat and mantle fluid) was calculated by dividing the total gram weight by the number of oysters in the sample. The resulting distribution of average oyster weight per sample was found to be positively skewed (Appendix 5, Figure A5-11). This is likely because the oyster samples collected from retail establishments were harvested from many different growing areas; the Gulf Coast, Mid-Atlantic, Northeast Atlantic and Pacific Northwest regions were all equally represented.

Although there were some apparent differences in the mean oyster weight distribution by region and season of harvest, the differences were not large. A single estimate of the distribution of average gram weight per oyster based on pooling all of the data was considered appropriate and this estimate was assumed to apply to oysters harvested from all regions and seasons. A lognormal distribution was fit to the observed average oyster weight data in order to obtain a smooth estimate of the average oyster weight, rather than using the empirical distribution of the data. The maximum likelihood estimates obtained corresponded to a geometric mean average oyster weight of 15.2 grams and a geometric standard deviation of 1.4 grams.

Since the samples in the retail study were a combination of both oyster meat and mantle fluid a correction is needed to infer the average meat weight per oyster. Mantle fluid is typically not consumed. Based on mantle fluid versus meat weight measurements of individual Gulf Coast oysters collected during the Kaufman et al. (2003) study and the weight of oysters at retail (DePaola, 2002), approximately 90% of the total oyster weight is the meat weight. Therefore, the average oyster weight distribution was multiplied by this average percentage to obtain a distribution of the average meat weight per oyster.

Oyster Meat Weight per Serving

The total gram weight of oyster meat consumed per serving was obtained as the combination of the distribution of the number of oysters consumed and the distribution of the average meat weight per oyster at retail. The distribution of total consumption per serving was truncated at less than 10 grams or more than 2,000 grams because consumption outside these levels is unlikely. The best estimate of the mean meat weight per serving was approximately 200 grams.

Number of Raw Oyster Servings per Annum

The total annual number of servings consumed was estimated using data on the total landings of oysters, the mean weight of oysters per serving, and the likely amount of the total harvest that is consumed raw. Industry estimates suggest that approximately 50% of the Gulf Coast harvest is consumed raw (Muth et al., 2000). This estimate was assumed to apply for each region/ season combination. The total amount (weight) of oysters harvested from different regions and seasons in the United States was obtained from the National Marine Fisheries Service (NMFS). For this risk assessment, the average NMFS landings data from 1990 to 1998 were used as shown in Table IV-13.

Total landings across different regions and seasons vary from year-to-year, presumably due to the influence of numerous factors (e.g. closures due to water quality, market forces). Although some year-to-year trends and fluctuations are evident in the oyster landings data, these year-to-year differences are generally less than 25% of the overall average oyster landing for the identified period from 1990 to 1998. This is a relatively small variation relative to other identified modeling uncertainties impacting risk characterization.

The total amount of oyster meat consumed equals the sum of the amounts in each serving consumed. Thus, the total number of servings can be estimated using the following equation:

where N denotes the total number of servings, S_k denotes amount of meat weight consumed in each of the N servings, E[S] denotes the average of the S_k, f denotes the percentage of the total landed oyster meat weight that is consumed raw, and L denotes the total weight of oyster meat landed (i.e., for a given region and season combination). This equation was used to solve for N, the total number of servings, for each region/season combination.

Table IV-14 provides the calculated number of raw oyster servings for each region/season combination. The total annual number of raw oyster servings is approximately 40 million (i.e., N = [(0.5 x 16,400,000 kg)/0.2 kg]. In this calculation, the total landings (L), from Table IV-14, is approximately 36 million pounds (16 million kg). The mean meat weight per serving (E[S]) is estimated as 200 grams (based on the ISSC/FDA retail study) and the percentage of total landed oyster meat weight consumed raw (f) is assumed to be 50%.

Assuming that children do not eat raw oysters and the adult U.S. population is approximately 200 million, the annual consumption rate is approximately 0.2 servings per adult per year (40/200 = 0.2). This consumption rate was calculated. This consumption rate is consistent with the estimate of 0.0005 servings per day or 0.18 servings per person per year based on the 1989-1992 CFSII survey data. It should be noted that regional consumption rates are likely. For example, the consumption rate reported in the Florida consumer survey (Degner and Petrone, 1994) is considerably higher (5.2 servings per year) than the national estimates described above (approximately 0.2 servings per year).

Output of the Consumption Module

The output of the Consumption Module is the level of pathogenic V. parahaemolyticus associated with typical serving sizes. The output of the simulation consists of a two-dimensional variability and uncertainty distribution or, alternatively, a sequence of variability distributions indexed by selected sets of uncertainty parameters. An appropriate summary of this two-dimensional variability and uncertainty distributions is the one-dimensional uncertainty distribution of the mean of the variability distribution(s).

Table IV-15 shows the predicted mean levels of pathogenic V. parahaemolyticus at consumption. As would be expected, the relative level of exposure for the different region/season combinations at consumption should be no different from the levels at post-harvest; consumption levels are derived from the post-harvest levels and the serving size and it is the same average (200 g) for all region/season combinations. The mean levels of pathogenic V. parahaemolyticus per serving are higher at time of consumption for the Gulf Coast (Louisiana and non-Louisiana) compared to the other regions. The highest levels are attributed to the Gulf Coast (Louisiana) region.

---

Table of Contents