|
July 19, 2005
Surveillance data were compared to the model predictions as one of the approaches to validate the risk characterization portion of the model (i.e., the predicted illnesses attributed to oysters harvested from each region and season). Surveillance for Vibrio illness in the United States is conducted by the Centers for Disease Control and Prevention (CDC). State health departments submit reports of Vibrio illness to CDC's Cholera and Other Vibrio Illness Surveillance System (COVISS) (http://www.cdc.gov/foodborneoutbreaks/report_pub.htm).
Understanding the uncertainties associated with this approach to validating the risk assessment requires an understanding of how the data are acquired and interpreted. The difference can become important when a substantial portion of the oysters consumed in a region is not harvested from that region. The risk assessment model predicts illnesses associated with oysters harvested from a given region. Conversely, surveillance data are used to estimate the total number of cases based on the illnesses reported within a region. An illness caused by V. parahaemolyticus is reported to COVISS when the following occurs:
Completeness of reporting varies among state health departments. Reporting clinical isolation of V. parahaemolyticus is mandatory in some states but not in others. Reporting to CDC is voluntary. FDA and state shellfish authorities attempt to gather traceback information on illnesses associated with bivalve molluscan shellfish. However, information on the source of illness may be incomplete. Consequently, there are limitations to be considered in comparing the results of model predictions to observational surveillance data. These limitations are discussed in detail below.
As indicated in Chapter III: Hazard Characterization, the dose-response model is "anchored" using CDC's estimated average annual incidence of cases associated with raw oyster consumption (i.e., 2789 V. parahaemolyticus illnesses (Painter, 2003). This estimate is based on an analysis of V. parahaemolyticus illnesses reported in the National Notifiable Diseases Surveillance System (NNDSS) and the Cholera and Other Vibrio Illness Surveillance System (COVISS) from 1998 to 2002. Because some cases may be reported in both systems, a "capture-recapture" method was used to obtain an estimate of the number of V. parahaemolyticus cases for the five-year period. The reported cases were adjusted to account for CDC's estimate of underreporting (a factor of 1:20) and the estimate that 62% of cases are associated with oyster consumption. A complete description of the data and information that CDC used to estimate the annual illness burden in a manner appropriate to be considered in this risk assessment is provided in Chapter II. Hazard Identification, section titled "Annual Incidence." For the purposes of this specific comparison of predicted cases versus those estimated from surveillance data, COVISS surveillance data from 1998 to 2003 were used (Painter, 2004a and 2004b).
Table A9-1 provides a comparison of the seasonal distribution of V. parahaemolyticus illnesses within the United States predicted by the risk assessment model and the number of cases estimated by the CDC using reported illnesses. Between 1998 and 2003, COVISS received 1018 reports of V. parahaemolyticus illnesses in the United States (excluding Guam). Of those, 104 were associated with wounds 914 were food borne, of the foodborne cases 78% (713) are estimated to be oyster-associated. The observed seasonal frequency of illness occurrence for those 713 illnesses was then applied to the estimated total number of oyster associated cases per year (i.e., 2,789) and compared with number of illnesses predicted by the risk assessment model.
| Season | Illness Estimated from | |||
|---|---|---|---|---|
V. parahaemolyticus
Risk Assessmenta |
Reported Illnessesb | |||
| Number | % Annual | Number | % Annual | |
| Winter (January-March) |
10 | 0.3% | 156 | 5.6% |
| Spring (April-June) |
723 | 25.6% | 841 | 30.1% |
| Summer (July-September) |
1,903 | 67.3% | 1,474 | 52.9% |
| Fall (October-December) |
190 | 6.7% | 318 | 11.4% |
| Total | 2,826 | 100% | 2,789 | 100% |
a Model-predicted
illnesses associated with consumption
of oysters harvested from all regions.
b Values in the column "% of Annual" were calculated
from illnesses reported to COVISS from 1998-2003, excluding patients with
isolates from wound. Values in the column "Number" were calculated by
multiplying the percent of annual for each season by the estimated total (2,789. Source: Painter, 2005.
As shown in Table A9-1, the risk assessment model and the surveillance data indicate similar trends in the seasonal distribution of V. parahaemolyticus illnesses. For spring, summer, and fall, estimated illnesses based on reported illness were similar to that predicted by the risk assessment model (Table A9-1). The percentage of illness reported during winter months was substantially higher than the percentage of illnesses predicted by the risk assessment model. However, this difference accounts for a relatively small percentage (5%) of the total illnesses.
Preliminary data and observations provided by Canada (Buenaventura et al., 2002; Banerjee and Farber, 2005) suggest a significantly lower incidence of cases in the winter months in the British Columbia region. This observation is consistent with the model predictions. It is possible that the divergence between the CDC surveillance data and the predicted values reflect the existence of additional factors related to post-retail handling or consumption patterns of raw oysters during the winter months that have not been previously recognized and thus not incorporated in the model. Any consideration of such factors would require more sophisticated epidemiological investigations than those that are currently being performed. Alternatively, the differential could reflect the substantial uncertainty associated with the model and surveillance estimates.
V. parahaemolyticus illnesses were most frequently reported to CDC's Cholera and Other Vibrio Illness Surveillance (COVISSS) system from Pacific Coast states (Table A9-2). However, the reporting state typically indicates the state of residence of the patient, not the oyster harvest state.
| Region | Percentage llnessesa |
|---|---|
| Atlanticb | 21.3% |
| Gulf Coastc | 26.4% |
| Pacific Coastd | 45.9% |
| Non-coastal States | 6.4% |
| Total | 100% |
a Percentages were calculated from the number of
illnesses reported to COVISS from 1998-2003,
excluding patients with isolates from wound.
Source: Painter, 2005
b Includes mid-Atlantic and Northeast Atlantic coast states.
c Florida is included in the Gulf Coast regions.
d The Pacific Coast includes Hawaii
In general, most oysters consumed in the Gulf Coast are harvested from that region. For other regions in the United States, the source of the oysters consumed is a mix of multiple harvest regions. As a means of comparing the model predictions with comparable surveillance data, illness cases reported to COVISS between 1998 and 2003 were sorted by reporting region and the source of the oysters, if known (Table A9-3). Of the 713 oyster-associated V. parahaemolyticus reported illnesses only 18.4% (131) were traced to a specific harvest site. Of those 131 illnesses, the percent of illnesses from each reporting region that were traced to harvest regions are indicated in Table A9.3. This table illustrates the differences across regions. Of the illnesses reported in the Atlantic only 31% were traced to oysters harvested from that region. However, in the Gulf Coast, the vast majority of the illnesses were traced to that region (93%). In addition, the majority (57%) of the illnesses reported in the Pacific Northwest are associated with oysters from that same region.
| Patient Residence | Oyster Harvest Regiona | |||
|---|---|---|---|---|
| Atlanticb | Gulf Coast | Pacific Northwestc | Other Pacific States | |
| Atlanticb | 31% | 54% | 15% | 0% |
|
Gulf Coastd |
7% | 93% | 0% | 0% |
|
Pacific Coaste |
10% | 12% | 57% | 21% |
|
Non-coastal States |
40% | 40% | 20% | 0% |
a Source: Painter, 2005.
b Includes mid-Atlantic and Northeast Atlantic coast states.
c Includes the states of Oregon and Washington.
d Florida is included in the Gulf Coast region.
e The Pacific Coast includes Hawaii.
The percentage of illnesses attributable to each harvest region was estimated by combining the data from Tables A9-2 and A9-3. The total attributable illness for each region was calculated as a weighted average of the percent of cases attributed to each harvest region, weighted by the percentage of cases reported from each region (Table A9-4). For example, the following calculations were performed to determine the percentage of illnesses attributable to Atlantic region oysters:
Thus, a total of 15.6% (6.6% + 1.8% + 4.6% +2.6%) of all oyster-associated V. parahaemolyticus cases were attributed to oysters harvested from the Atlantic region.
| Patient Residence | Oyster Harvest Regiona | |||
|---|---|---|---|---|
| Atlanticb | Gulf Coast | Pacific Northwestc | Other Pacific | |
| Atlanticb | 6.6% | 11.5% | 3.2% | 0% |
| Gulf Coastd | 1.8% | 24.6% | 0% | 0% |
| Pacific Coaste | 4.6% | 5.5% | 26.2% | 9.6% |
| Non-coastal States | 2.6% | 2.5% | 1.3% | 0% |
| Total Attributed Illnesses | 15.6% | 44.1% | 30.7% | 9.6% |
a Source: Painter, 2005.
b Includes mid-Atlantic and Northeast Atlantic coast states.
c Includes states of Oregon and Washington.
d Florida is included in the Gulf Coast region.
e Hawaii is included in the Pacific Coast region.
Differences between the illnesses estimated based on COVISS data and the number of illnesses predicted by the risk assessment is evidence that there are as yet unaccounted for factor(s) in the either the model or the surveillance data, or both. Surveillance data are limited by variation in reporting rates between states, incomplete food history, and incomplete traceback information. Risk assessment models may be limited by unrecognized factors in post-retail handling or in consumption patterns of raw oysters during the winter months. Nonetheless, the above information provides the best available description of the data patterns that are observed.
Although the magnitude of the numbers are different, information from reported illness and the risk assessment model predictions indicate that most oyster associated V. parahaemolyticus illnesses are associated with the Gulf Coast oysters, followed by Pacific Northwest oysters. Thus, the predictions of the risk assessment model is consistent, both in terms of seasonal and regional differences, are consistent with the surveillance data. Because of the intrinsic difference in what the two systems measure (location of illness occurrence vs. harvest region of oysters that cause illness), full validation of the regional model predictions of illness based on regional surveillance data would benefit from additional research and targeted surveillance initiatives to acquire more thorough traceback data.
