A comparison of dose-response models for death from hematological depression in different species.

Many radiation-induced lethality experiments have been published for various mammalian species. From those studies a subset of studies reflecting useful biological and physical variables has been compiled into a database suitable to study interspecific variability of radiosensitivity, dose-rate dependence of sensitivity, dose-response behavior within each experiment, etc. The data compiled were restricted to continuous and nearly continuous exposures to photon radiations having source energies above 100 keV. Photon source energy, exposure geometry, and body weight considerations were used to select studies where the dose to hematopoietic tissue was approximately uniform. The database reflects 13 mammalian species ranging in size from mouse to cattle. Some 211 studies were compiled, but only 105 were documented in adequate detail to be useful in development and evaluation of dose-response models of interest to practical human exposures. Of the 105 studies, 70 were for various rodent species, and 35 were for non-rodent groups ranging from standard laboratory primates (body weight approximately 5 kg) to cattle (body weight approximately 375 kg). This paper considers seven different dose-response models which are tested for validity against those 105 studies. The dose-response models include: right-skewed extreme value, left-skewed extreme value, log-logistic, log-probit, logistic, probit, and Weibull models. In general, the log transformation models did not improve model performance and the extreme value models did not seem consistent with the preponderance of the data. Overall, the probit and the logistic models seemed preferable over the Weibull model.