Interspecies extrapolation: a reexamination of acute toxicity data

We reanalyze the acute toxicity data on cancer chemotherapeutic agents compiled by Freireich et al.(1) and Schein et al.(2) to derive coefficients of the allometric equation for scaling toxic doses across species (toxic dose = a.[body weight]b). In doing so, we extend the analysis of Travis and White (Risk Analysis, 1988, 8, 119-125) by addressing uncertainties inherent in the analysis and by including the hamster data, previously not used. Through Monte Carlo sampling, we specifically account for measurement errors when deriving confidence intervals and testing hypotheses. Two hypotheses are considered: first, that the allometric scaling power (b) varies for chemicals of the type studied; second, that the same scaling power, or "scaling law," holds for all chemicals in the data set. Following the first hypothesis, in 95% of the cases the allometric power of body weight falls in the range from 0.42-0.97, with a population mean of 0.74. Assuming the second hypothesis to be true-that the same scaling law is followed for all chemicals-the maximum likelihood estimate of the scaling power is 0.74; confidence bounds on the mean depend on the size of measurement error assumed. Under a "best case" analysis, 95% confidence bounds on the mean are 0.71 and 0.77, similar to the results reported by Travis and White. For alternative assumptions regarding measurement error, the confidence intervals are larger and include 0.67, but not 1.00. Although a scaling power of about 0.75 provides the best fit to the data as a whole, a scaling power of 0.67, corresponding to scaling per unit surface area, is not rejected when the nonhomogeneity of variances is taken into account. Hence, both surface area and 0.75 power scaling are consistent with the Freireich et al. and Schein et al. data sets. To illustrate the potential impact of overestimating the scaling power, we compare reported human MTDs to values extrapolated from mouse LD10s.