their HERP (Human Exposure/Rodent Potency) index as the fraction of the TD50 accounted for by human exposure, Ames et al. (1987) implicitly assume a linear dose-response relationship below the TD50. Wartenburg and Gallo (1990) have objected to the latter application of the TD 50 on the grounds that many dose-response relationships are highly nonlinear. In practice, however, linear extrapolation from the TD 50 will often approximate q1* in experiments employing only two or three doses because of the limited opportunity to observe curvature (Krewski, 1990). Although the HERP index appears to be based on the tacit assumption of a linear dose-response, Gold & Ames (1990) and Gold et al. (1992) emphasize that the index is intended for priority ranking rather than quantitative risk assessment.
5.3. Preliminary Estimate of Risk
The fact that q1* is highly correlated with the MDT suggests that preliminary estimates of cancer risk may be based on the MTD. Gaylor (1989) exploited this correlation to estimate the 10-6 RSD. Estimates of the RSD were obtained by the linear interpolation procedure given by Gaylor & Kodell (1980), as modified by Farmer et al. (1982), for 38 chemical carcinogens tested by oral administration in the U.S. National Toxicology Program. Estimates of the RSD were compared with the MTD for up to 69 tumor sites in both rats and mice, for a total of 138 cases. The ratio of the MTD to the RSD varied over a 184-fold range, which is considerably larger than the 32-fold range suggested by Bernstein et al. (1985) for the range of TD50 values relative to the MTD. The overall geometric mean of the ratio MDT/RSD was 3.8 × 105; only 3 of the 138 ratios were more than a factor of 10 from the mean. This suggests that a preliminary estimate of the RSD may be obtained by dividing the MTD by 380,000.
As in predicting the TD50 from the MDT (section 4.1), linear regression analysis may also be used to predict low dose slopes from the MDT. This may be illustrated using the 191 rodent carcinogens considered previously. The results of regressing the logarithms of the linearized upper bounds on low dose slope based on either the LMS model or MFX are given in Table 4. The estimated slope of the linear regression model is approximately -1 for both the LMS and MFX methods. The