PRINCIPLES OF TOXICOLOGY

456 RISK ASSESSMENT
Because low-dose responses cannot be measured, they must be modeled. There are three types of
models:
1. The first category of models consists of the “mechanistic” models. These are dose–response
models that attempt to base risk on a general theory of the biological steps that might be involved in
the development of carcinogenesis. Examples of mechanistic models include the early “one-hit” and
the subsequent “multihit” models for carcinogenesis. These models were based on assumptions
concerning the number of “hits” or events of significant genetic damage that were necessary to induce
cancer. A related model, the “linearized multistage” (LMS) model of carcinogenesis, is based on the
theory that cancer cells develop through a series of different stages, evolving from normal cells to
cancer cells that then multiply.
2. The second category of cancer extrapolation models includes the “threshold distribution”
models. Rather than attempting to mimic a particular theory of carcinogenesis, these models are based
upon the assumption that different individuals within a population of exposed persons will have
different risk tolerances. This variation in tolerance in the exposed population is described with
different probability distributions of the risk per unit of dose. Models that fall within this category
include the probit, the logit, and the Weibull.
3. The third category of model is the “time-to-tumor” model. This type of model bases the risk or
probability of getting cancer on the relationship between dose and latency. With this model, the risk
of cancer is expressed temporally (in units of time), and a safe dose is selected as one where the interval
between exposure and cancer is so long that the risk of other diseases becomes of greater concern.
Each of these models can accommodate the assumption that any finite dose poses a risk of cancer, the
essential tenet of a nonthreshold model. However, the shape of the dose–response curve in the low-dose
region can vary substantially among models (see Figure 18.6). Because the shape of the dose–response
curve in the low-dose region cannot be verified by measurement, there is no means to determine which
shape is correct. A simple example of the impact of choosing one cancer extrapolation model over
another is given in Table 18.1, which compares the results of dose–response modeling using three
different models where it was assumed in each model that a relative dose of 1.0 produced a 50% cancer
incidence. The results generated by all three models are essentially indistinguishable at high doses
where the animal cancer incidence might be observable, and so one would conclude they all “fit” the
experimental data equally well. However, when modeling the risks associated with lower doses, the
dose/risk range in which regulatory agencies and risk assessors are most frequently interested, there
is a wide divergence in the risk projected by each model for a given low dose. In fact, at 1 / 10,000th
of the dose causing a 50% cancer incidence in animals, the risks predicted by these three models
produce a 70,000-fold variation in the predicted response.
Regulatory agencies utilize cancer risk estimates in regulating carcinogens, but they are faced with
many models that yield a wide range of risk estimates. In the absence of any scientific basis to determine
which is most correct, they must make a science policy decision in selecting the model to use. Generally,
in the face of this uncertainty, they have selected models that tend to provide higher estimates of risk,
particularly when combined with conservative exposure assumptions (see Table 18.2). This is consistent
with their mission to protect public health, and consequently the need to avoid underestimating
risks. For example, the USEPA has historically used conservative models such as the one-hit or LMS
model in calculating cancer risks from exposure to all carcinogens. These models assume linearity in
the low-dose range, and as shown in Table 18.1, tend to require a larger reduction in dose to attain a
certain low level of risk relative to other models.
Extensive research in the area of chemical carcinogenesis indicates that many chemical carcinogens
act via epigenetic or promotional mechanisms that, like noncancer toxicities, do not involve or require
genetic damage. It has been proposed that these mechanisms and carcinogenic responses should have
thresholds. Similarly, numerous enzyme systems have been identified as responsible for maintaining
the integrity of the genetic code. These repair enzymes and pathways could provide an effective dose