ASP Risk Assessment
ENVIRON International has created and updated a database of information pertaining to domoic acid (DA) concentrations in king scallops in Scottish offshore waters. In addition to geospatial, bathymetric, and some phytoplankton data, the database contains DA concentration measurements taken between July, 1998 and March, 2004.
That database has been used to investigate the time-course of DA concentrations in boxes that have been “closed” due to concentrations exceeding a risk-based threshold of 20 μg DA per g of tissue. Both gonad and whole tissues were examined; only samples collected up to the minimum of 335 days after closure or the date of reopening were included. The probabilistic analysis that is reported here used the statistical modeling approach of logistic regression to predict, based on the observations in the database, the likelihood that samples taken from closed boxes at various times after closure will have DA concentrations less than 20 μg/g (referred to as a “low concentration”).
One of the major efforts of the analysis was to determine what factors might affect the time-course of interest. It was determined that the following factors played an important and significant role, in the sense of modifying the probability of obtaining a low-concentration sample from a closed box:
- the area (as taken from the FSAS designation, being E, M, NM, etc.) in which the closed box is located;
- the month in which closure occurred; and
- the DA concentration causing closure.
These factors were determined to modify the probability of obtaining a low-concentration sample at any time after closure. The magnitude of their effects, and the specific effect on the time-course probability predictions was determined by finding the best-fitting logistic regression models that included some function of time after closure. Models that provided a satisfactory fit to the data were found; the predictions (in both graphical and tabular form) of those best fitting models have been made available in an accompanying Excel spreadsheet named LOGISTIC_PREDICTIONS.xls.
Implementation of the modeling results can be based on the LOGISTIC_PREDICTIONS.xls spreadsheet. If one wants to base the timing of resampling of a closed box on the likelihood of reopening it, then the predicted probabilities in that spreadsheet can be used. To successfully reopen a box, one needs 4 consecutive samples (2 per day on 2 separate days at least seven days apart) all with low DA concentration. If, for example, one wanted the probability of successfully sampling for reopening to be at least 50% (P = 0.5), then that means that there would need to be a high probability (in this example, about 0.84 = (0.5)¼) that individual samples will have low concentrations. The spreadsheet tabular read-outs of the individual-sample probabilities can be used to approximate when that will be the case.
The time after closure when such a high probability will occur depends on the factors of area, month of closure, and level causing closure, as indicated above. In general, however, it appears that at short times after closure, e.g., less than 60 days and in some instances well beyond 60 days after closure, the probability of getting low-concentration samples may not be sufficiently high to offer much likelihood of successful reopening. Individual instances (based on area, month of closure, and level causing closure) should be evaluated on a case-by-case basis using the spreadsheet predictive tools, but when the likelihood of successful reopening is estimated to be relatively low, then allocation of resources to other sampling efforts (e.g., monitoring of open boxes that would still be yielding scallops for consumption) may be more cost-effective and health-protective.
The logistic regression analysis was successful in identifying models that appeared to be consistent with the data included in the database. There are uncertainties associated with those models, as there are with any modeling effort. We have identified several key follow-on tasks that would help characterize or reduce those uncertainties, such as refinement of the definition of variables in the models, consideration of additional covariables, calculation of bounds on predicted probabilities, and elaboration of the modeling approach. Of greatest importance, however, is the continued augmentation of the database used for this analysis. Additional data are already available (from April to October 2004); those data offer the rare opportunity to test the model predictions against data that were not used to fit the models. Such a model validation effort would identify if and how the models need to be improved. In general, an on-going effort to maintain the database and integrate the new data with the old (ideally, in terms of using those data to update the model predictions) has been identified as a key effort for improving the ability of FSAS to allocate sampling resources.
Project Code: S02019