Challenges Associated with the Use of Biologically-Based Models in Risk Assessment. Kenny S. Crump, ICF Kaiser International, 602 East Georgia, Ruston, LA 71270
Mechanistic models of cancer that are currently available generally express the cancer risk as a function of underlying physiological parameters such as mutation rates and cell division rates. These models thus provide a means for predicting the effect of exposure upon cancer from data on the effect of exposure upon underlying physiologic parameters, and consequently provide the opportunity to incorporate a much richer body of data than statistical models that simply express the probability of cancer as a function of exposure. However, these models generally do not specify parametric expressions for the dose-responses for changes in these physiologic parameters. Different parametric expressions can generally be found that will fit all of the underlying data equally well but predict very different risks at doses lower than those at which the data reliably specify the dose-response. Thus, when using these models to predict low dose risks it is crucial to obtain data that quantify the dose-response for parameters in mechanistic models as accurately as possible at the lowest feasible doses. Another point that may be important in evaluating the accuracy of detailed mechanistic models of carcinogenesis is that they introduce "uncertainty of relevance" into the estimation process. For example, ancillary data on cell proliferation rates may pertain to a different group of cells than those that are actually at risk of progressing to malignancy. Although such uncertainty is difficult to quantify, it is potentially very important. These points will be illustrated using the results from the USEPA reassessment of the carcinogenic risk from exposure to dioxin.