An Integrated Dose-Response Method for Estimating Noncancer Risks. S. J. S. Baird and G. M. Gray, Harvard Center for Risk Analysis, Harvard School of Public Health, 718 Huntington Ave., Boston, MA 02115; J. T. Cohen, Gradient Corp., 44 Brattle Street, Cambridge, MA 02138; and L. Rhomberg, Harvard Center for Risk Analysis, Harvard School of Public Health, 718 Huntington Ave., Boston, MA 02115
Probabilistic methods proposed to quantitatively characterize the uncertainty in noncancer risk assessment (NCRA) have been hampered by their reliance upon the current NCRA paradigm which separates the extrapolation into four areas — animal to human, human sensitivity, duration of exposure and accounting for lack of a NOAEL [Baird et al. (1996), HERA 2(1):79-102; Swartout et al. (1997), FAT 36(1, Part 2):208]. In the current paradigm uncertainty and variability are double counted across the numerator and the uncertainty factors, and the empirical data does not sort cleanly across the animal to human and human sensitivity extrapolations. Here we present a framework that disentangles the NCRA process from these limitations by reformulating the problem as an extrapolation of the dose-response function from animals to humans, instead of an extrapolation from a single dose. The Integrated Dose-Response Method begins by adjusting all available data for the chemical of interest to a common dose metric, representing a human-equivalent dose, through scaling while accounting for uncertainty in the scaling power. All adjusted bioassay data for each sex, strain and species (SSS) tested are then modeled separately using a common dose-response (D-R) function. The minimum ED50 across endpoints for each SSS combination is incorporated into a distribution representing the variability in sensitivity across SSSs accounting for our lack of knowledge, or uncertainty, about where human sensitivity falls in comparison to the tested animal sensitivity. The slopes of the D-R curves for each SSS endpoint are incorporated into a distribution representing the variability in sensitivity across animal bioassays. After the animal slopes are modified to account for the expected extra-heterogeneity in human population responses, a human D-R curve is created using the distribution of ED50s across SSSs and the slopes. The EDX for X response level is calculated iteratively using the ED50 and slope distributions to generate a distribution that characterizes the uncertainty in the dose in humans that will cause X % response. The response rate, X, is selected as part of a risk management or policy process and may differ for different types of decisions. The advantages of this method are that all of the available data are used to estimate the expected human response; the response rate of interest can be explicitly defined at any rate, X; the uncertainty in the dose for any response rate can be characterized, and there is decreased bias in the characterization of human sensitivity through (i) assuming that humans are just one draw from the distribution of animal responses, (ii) using the ED50 as a response metric for comparison, and (iii) using a distribution of human slopes for calculating the EDX from the ED50. Limitations of the method include the extensive data requirements, and the limited empirical information in the difference between human and animal slopes across endpoints. Finally, this method is not dependent on the assumption of a threshold, which alters the interpretation of these results from those of the current NCRA paradigm. (Work supported in part by Health Canada.)
.