14003-96-4 br b w human b
(b.w.human/b.w. animal)1/3. Furthermore, a difference is that in the additive model the “dose per day” is used, compared to the relative
model that uses the “lifetime dose”. Thus, the factors that are considered in our relative risk model and not in an additive model are:
1) the background tumor incidence; 2) the relation between administered and internal dose (for consideration of pharmacokinetics in the extrapolation between exposure doses and between species); and, 3) the estimated lifetime dose; (c.f. Törnqvist and Ehrenberg., 1992; Paulsson et al., 2001). The additive model is for instance used by the U.S. Environmental Protection Agency for estimations of cancer risk in humans and for regulations of carcinogens (U.S. EPA, 2005a). For a comparison of the risk coefficients estimated with the risk models, there are only a few earlier evaluations performed with the relative risk model. The evaluations with the relative risk model indicated in the order of 10 times higher estimate for 14003-96-4 oxide (Granath et al., 1999), and about 3 times higher estimate for acrylamide (Törnqvist et al., 2008), respectively, than corresponding estimates by U.S. EPA.
The suggestion to project to human cancer risk using background risk relies on the observation of an approximately equal excess relative risk coefficient between tumour sites seen among A-bomb survivors (Pierce et al., 1996; UNSCEAR, 2000). Furthermore, evaluations of carcinogenicity test data for ionizing radiation suggest a common re-lative risk coefficient which is approximately the same in responding sites in several strains of mice (Storer et al., 1988) and is approximately the same for mice, dogs and humans (Granath et al., 1999). The background risk for cancer is assumed to depend on background mutations interacting with “background promotion” (conditions that stimulate growth and clonal expansion). Background mutations could be inherited and/or caused by mutagenic factors from endogenous and exogenous sources, and could also originate from spontaneous DNA replication errors, referred to as “random mutations” by Tomasetti and co-workers (Tomasetti et al., 2017; Tomasetti and Vogelstein, 2015). In the relative risk model, the background cancer incidence is assumed to give a rough estimate of background conditions interacting with the exogenous genotoxic factor, like glycidol in the present study, to de-velop cancer (cf. Granath et al., 1999).
As the strains of mice and rats, which are commonly used in car-cinogenicity studies, have been developed to be sensitive for tumor development, the background incidence for certain types of tumors are high in these strains (e.g. in the liver). Using an additive risk model can result in overestimations in the species-extrapolation of a risk coeffi-cient to humans, in general for populations with lower background incidence than the test species. This was observed by Kuo et al. (2002) who compared cancer risk predictions by either an additive risk model or a relative risk model (without consideration of internal doses) for about 100 tumor types in mice and rats. In general, extrapolating risks Food and Chemical Toxicology 128 (2019) 54–60
from species with high background rates to species with low back-ground rates using the additive risk model resulted in overpredictions, whereas the relative risk model matched the observed tumor incidence better (Kuo et al., 2002). This supports that a relative risk model is preferable because it considers the background incidence of the studied species, which the model is further improved if internal doses are considered.
Background tumor incidence of different cancer types vary between human populations, likely related to lifestyle. Another support for the relative cancer risk models is a study by Korobitsyn (2011). He devel-oped a relative (multiplicative) cancer risk model for cross-population extrapolations, considering population specific health and demographic parameters in the studied Russian population. Compared to an additive risk model, the relative risk model enabled calculations of risks for cancer development at different ages for males and females in different subpopulations. This approach is similar as for cancer risk projection of ionizing radiation using a relative risk model, where population specific background cancer rates and dose rates at different ages are considered in the calculations of the lifetime attributable risk (LAR) from the excess relative risk coefficient (ERR) for ionizing radiation. This is required to adjust for the high exposure doses of ionizing radiation underlying the ERR, obtained from survivors of the Hiroshima/Nagasaki A-bombings (ICRP, 2007; BEIR, 2006).
The experience from ionizing radiation should be further explored for improvement of species-extrapolation to humans of risk estimates for chemicals, as e.g. have been discussed by the U.S. EPA for age-de-pendent susceptibility (U.S. EPA, 2005b; Barton et al., 2005). Appli-cation of a relative risk model to chemical genotoxic carcinogens would give possibilities to further refine the transfer of the cancer risk coef-ficient to different human populations.