Dose-Scaling and Distributional Approaches to Noncancer Risk Assessment. Lorenz Rhomberg and Sandra J. S. Baird, Harvard School of Public Health, 718 Huntington Ave., Boston, MA 02159; Joshua Cohen, Gradient Corporation, 44 Brattle St., Cambridge, MA 02138; and George Gray, Harvard Center for Risk Analysis, Harvard School of Public Health, 718 Huntington Ave., Boston, MA 02159
The problem of scaling doses for toxicological equivalence across species has been relatively little studied for noncancer endpoints. In noncancer risk assessment, it has been traditional to express doses on a mg/kg/day basis and to apply a 10-fold uncertainty factor for extrapolating critical-effect NOAELs from animals to humans. It might seem that distributional approaches to risk assessment—in which fixed extrapolation factors are replaced with statistical distributions derived from observations of the varying relative size of the toxicologically equivalent values over chemicals—would obviate concerns over dose-scaling rules by relying instead on empirically derived patterns. We argue, however, that proper dose scaling is still critical to well conceived distributional methods. A poorly chosen dose-scaling assumption has the effect of artificially inflating the distribution among species in equally toxic dose. It skews the identification of critical NOAELs among species. Moreover, when using a distribution of species-to-species ratios of equitoxic doses to extrapolate probabilistically from small experimental animals to humans, a poorly chosen scaling method introduces a considerable systematic bias, since the human value can no longer be considered to be randomly drawn from the distribution. The rationale for the traditional 10-fold animal-to-human factor is vaguely defined, combining the notion of allowance for agent-by-agent variability with that of correction for systematic differences in sensitivity in animals and humans. We explore the impact of firmer definitions on the interpretation of empirical data for defining probabilistic cross-species extrapolation, and we argue that best-fitting allometric relationships across species in equitoxic doses lead to "proper" scaling, i.e., scaling that avoids introducing systematic bias, minimizes the uncertainty in probabilistic extrapolation of equivalent doses across species, and gives all datasets appropriate consideration when choosing critical effects.
This work was supported in part by Health Canada under Contract No. H1201-6-9888.