home edit2 whole TSD November 2002 PDF format - OEHHA

Several models proposed for extrapolating low-dose human cancer risk from animal carcinogenicity
data are described in the Guidelines; these models include the Mantel-Bryan method (log-probit model),
the one-hit model, the linearized multistage procedure, the gamma multihit model, and a number of timeto-tumor
models. The Guidelines state that time-to-tumor models (i.e., a Weibull model) should be
used for low-dose extrapolation in all cases where supporting data are available and particularly when
survival is poor due to competing toxicity. However, the Guidelines also note the difficulty of
determining the actual response times in an experiment; internal tumors are generally difficult to detect in
live animals and their presence is usually detected only at necropsy. Additionally, use of these models
often requires making the determination of whether a tumor was the cause of death, or was found only
coincidentally at necropsy when death was due to other causes. Further, competing causes of death,
such as chemical toxicity, may decrease the observed time-to-tumor for nonlethal cancers by allowing
earlier necropsy of animals in higher dose groups. The linearized multistage procedure is noted as being
an appropriate method for dose extrapolation in most cases, with the primary exception being a situation
in which sufficient empirical data are available to indicate a dose-response curve of a “quasi-threshold”
type (e.g., flat for two or three dose levels, then curving sharply upwards). In this case, the linearized
multistage procedure may underestimate the number of stages and overestimate the low-dose risks. In
this case, the gamma multihit model is suggested as being a potential alternative. The Mantel-Bryan
model is described as having little biological basis as applied to carcinogenesis, and being likely to
underestimate risks at low doses. The Guidelines state that this model should not be used for low dose
extrapolation.
The Guidelines state that both animal and human data, when available, should be part of the doseresponse
assessment. Low-dose extrapolation of human cancer risk from animal carcinogenicity data is
generally based on the most sensitive site, species and study demonstrating carcinogenicity of a
particular chemical, unless other evidence indicates that the data set in question is not appropriate for
use. Where both benign and malignant tumors are induced at the same site and the malignant tumors are
significantly increased, the data on both types of tumors may be combined to form the basis for risk
assessment. Pharmacokinetic data on chemical metabolism, effective dose at target site, or species
differences between laboratory test animals and humans are considered in dose-response assessments
when they are available. In performing exposure scaling from animals to humans, the “surface area”
correction (correcting by the 2/3 power of body weight) is used unless specific data indicates that this
should not be done. The Guidelines assume that in the absence of evidence to the contrary, chemicals
that cause cancer after exposure by ingestion will also cause cancer after exposure by inhalation, and
vice versa. Cancer unit risk factors (µg/m 3 ) -1 are calculated from cancer potency factors (mg/kg-day) -1
using the following relationship:
UR =
CPF * 20 m
3
70 kg * CV
where UR is the cancer unit risk, CPF is the cancer potency factor, 70 kg is the reference human body
weight, 20 m 3 is the reference human inspiration rate/day, and CV is the conversion factor from mg to
µg (= 1000). The cancer unit risk describes the excess cancer risk associated with an inhalation
13