on determination of a NOAEL from a controlled study in animals. In this context, the NOAEL is defined as the highest experimental dose that does not produce a statistically or biologically significant increase in adverse effects over those of controls. An “acceptably safe” daily dose for humans is then derived by dividing the NOAEL by a safety factor, usually 10 to 1,000, to account for sensitive subgroups of the population, data insufficiency, and extrapolation from animals to humans. The U.S. Environmental Protection Agency (EPA) refers to the resulting quantity as the reference dose (RfD), the Food and Drug Administration (FDA) uses the term allowable daily intake (ADI) and the Agency for Toxic Substances and Disease Registry (ATSDR) uses minimum risk level (MRL). The concept is also similar to the upper limits (ULs) recently introduced by the National Academy of Sciences for nutrient recommendations. In the event that the lowest experimental dose shows a significant difference from the control, it is termed a LOAEL (lowest-observed-adverse-effect level), and an extra factor of 10 is used in the determination of the RfD, ADI, or MRL (see, for example, EPA 1998). Various reports have provided RfDs for MeHg that are derived from animal studies (Rice 1992; Gilbert et al. 1993; Zelikoff et al. 1995; Rice 1996). Typically, these calculations have used the results from a series of nonhuman primate studies, which indicate that adverse developmental effects in several outcomes occur at 50 µg/kg per day maternal dose. Uncertainty factors of 10 were used for LOAEL to NOAEL, species differences, and individual variation in response for an RfD of 0.05 µg/kg per day.
In recent years, use of the NOAEL has become controversial among risk assessors and regulators because of several serious statistical drawbacks (Gaylor 1983; Crump 1984; Kimmel and Gaylor 1988; Kimmel 1990; Leisenring and Ryan 1992). For instance, because the NOAEL must, by definition, correspond to one of the experimental doses, it can vary considerably across different experiments, yet this statistical variation is usually ignored when computing RfD values. Furthermore, estimation of the NOAEL is sensitive to sample size: because the NOAEL is based on statistical comparisons between exposed and unexposed dose groups, larger studies have higher power to detect small changes and therefore tend to produce lower NOAELs. In contrast, smaller studies tend to yield higher NOAELs due to their lower power to detect real effects. Because NOAEL calculations are traditionally based on pairwise comparisons of exposed groups and controls, there is