carcinogenicity testing would be most effective in providing information to assist risk managers, given the incomplete scientific understanding of chemical carcinogenesis in rodents and humans.
The Two-Stage Model of Carcinogenesis
Efforts to improve cancer risk assessment have resulted in the development of a mathematical dose-response model, called the two-stage model, that is based on a two-stage paradigm for the biologic phenomena thought to be associated with carcinogenesis. This paradigm is based on the relationship between tumor incidence and age, which suggests that at least two critical cellular changes are necessary for the development of many nonhereditary tumors. Current evidence suggests that some tumors might require more than two critical events to be expressed as human cancer. More complex models might be needed to describe multistage carcinogenesis accurately; however, it is hoped that the two-stage model will provide more accurate estimates of the cancer potency of chemicals that the multistage models currently in use by regulatory agencies.
Applying the two-stage model requires more extensive biologic data than current procedures; and because its feasibility as a tool for routine regulatory use has been questioned CRAM chose as its second task to evaluate the data needs and regulatory applicability of two-stage models of carcinogenesis. The committee considered several applications of the two-stage model to rodent carcinogens with different mechanisms of action and different quantities of available data. The committee noted that numerous assumptions were required to apply the model in each case. Assumptions must be made about mechanisms of action, appropriate target cells, time dependence, and the shape of the dose-response relationship. Extensive data would have to be obtained to reduce the current uncertainty in these assumptions. In fact, for very few chemicals are data sufficient to support the use of this model.
By studying specific application of the two-stage model, the committee determined that when different forms of the model are consistent with a particular data set, risk estimates can differ by several orders of magnitude. Therefore, the committee concluded that even if an agent's