Dose-response assessment plays a more prominent role where the adverse effect is cancer, for which usually no threshold is assumed. Because human doses are typically lower than the experimental LOEL (lowest observed effect level), there is a great deal of uncertainty about how a dose behaves biologically in the region below the experimental LOEL, especially when a no threshold effect is assumed. The current practice is to extrapolate from responses at the experimental doses down to the region below the LOEL, using a theorized mechanistic model such as the linearized multistage model (see Lecture 3). This type of extrapolation is part of the dose-response assessment.

In addition, currently a great deal of attention is being given to using the dose-response data to derive a benchmark dose (BMD), which is supposed to yield a more accurate account of the NOEL or LOEL. Unlike the NOEL approach, BMD modeling considers the entire dose-response curves across studies. As Rees and Hattis (1994) put it, the magnitude of the confidence limits for BMD is determined by more aspects of the experimental design including sample size, dose levels, and dose spacing. Technical guidance on BMD modeling, including free software from the Internet, is provided by USEPA (2000c).

There appear to be some uncertainties and skepticism about these types of dose extrapolation. These and the uncertainties associated with human exposure assessment are discussed extensively in the last part of this lecture, after a due consideration of the numerous ways in which an estimated exposure level can be assessed against a level pre-established as safe.