The Delaney Clause
The influence of dose response modeling on food safety cannot
be discussed without reference to the Delaney clause, which was included in the
1958 Food Additive Amendment:
the Secretary of the Food and Drug
Administration shall not approve for use in food any chemical additive found to
induce cancer in man, or, after tests, found to induce cancer in animals
In 1958, it was thought that the Delaney clause would apply
to very few chemicals, and would therefore be invoked only on rare
occasions . Furthermore, it was supposed that cancer, or at least some cancers, are caused solely as a result of exposure to cancer causing chemicals. But, the science rather quickly rendered the supposition the law was based on obsolete (Blank, 1974). First,
chemists became much better at detecting very small amounts of chemicals,
including those that were present as minor impurities in food additives and
chemicals that migrate in very small amounts from packaging to food. Second, it was discovered that pumping
nearly lethal amounts of a chemical into rats and mice over a long period of
time could implicate many chemicals in food as being carcinogenic. It also came to be appreciated that rather than causing cancer to occur all by themselves, at least some chemicals acted by accelerating the cancer development process.
To limit the
impact of the Delaney clause, the FDA developed two strategies:
- Divide and Conquer. First, because it was known that cancer develops in multiple stages, chemicals were differentiated into two classes; genotoxic chemicals “initiators” that act by causing mutations, and non-genotoxic chemicals “promoters” that act at later stages by facilitating the growth of cancerous cells. The ability of a compound to cause a mutation either in vivo or in vitro is often used to differentiate genotoxic compounds from those that are not. For Delaney purposes, the label of “carcinogen” was reserved for genotoxic compounds. This diversion worked, probably because not very many people, including congress, really wants to enforce the Delaney clause by the letter of the law (Blank, 1974). Even though the disease is the same either way, this genotoxic/nongenotoxic distinction has also come to be used for contaminants; since though the Delaney Clause doesn’t apply anyway, that can be rather annoying.
- De Minimis. The second strategy was to use quantitative dose-response modeling to argue that the risk from chemicals occurring in extremely small amounts is trivial or negligible. This involves the use of a mathematical model to extrapolate high dose-results to estimate immeasurable effects, usually from experiments with animals. Although both empirical and biophysical models have been used for this purpose, it is the latter category of models that have dominated scientific discussion, largely because they promised to deliver more certainty to a process that all involved agree is fraught with theoretical speculation. Even though those promises largely failed to materialize, the language associated with the use of those models persists. In any case, estimates were produced and legal arguments were advanced upon them. In particular, a standard of one-in-a-million was invoked to delineate the point at which a risk became so small that it was beneath the purview of the federal government, at least in some cases.
A Flawed Paradigm Foundation
Like the Safety Assessment Paradigm, quantitative cancer risk
assessment evolved and came to be used by other agencies and for purposes other
than the regulation of food additives. In
those circumstances, the risk estimate came to be appreciated as source of
information, rather than as a standardized policy-making procedure. Perhaps the attractive feature, at least for
some, was that standard of risk that would be considered to be acceptable could
be chosen by a non-technical regulator. And
thus, the Redbook Paradigm was born. Yet, even at the beginning, quantitative cancer risk
assessment never really conformed to the ideals of the Redbook. There were two problems:
- For the purpose of arguing de minimus, early cancer models were largely intended to estimate how big the risk might be by providing a worst-case estimate. As a result, rather than attempting to characterize the uncertainties associated with the estimating the risk, they were produced with the intent of erring on the side of safety. While this is fine when the goal is to establish negligible risk, the fact that it also produces estimates that are likely to be wrong makes it less suited for other purposes like cost-benefit analysis or informing the agency managers or the public about how likely the risks really are.
- Early cancer models were based on biophysical models of radiation effects. This brought the Copenhagen interpretation into risk assessment. Like Schrödinger’s cat, cancer risk estimates were thought of as properties of an individual rather than a population, so that any given person has a cancer risk state. Models of cancer occurrence in a population of rodents were used to characterize the individual risk state that would be expected in a human. In a population of humans, that risk state may be both variable and uncertain (e.g. see NRC, 1996). The Ensemble interpretation avoids all that silliness by treating cancer risk models as population models: Models of populations of rodents may be used to predict what will happen in populations of humans; they simply do not predict individual risk. Unfortunately, the Copenhagen habit has been hard to break, which has meant that risk estimates are often referred to as being “unitless”, which pretty much makes them vague fear factors.
These flaws existed before the Redbook was ever written, and
to some extent they were adopted by the Redbook itself. The paradigm battle has raged ever since.
References
Blank, Charles H. (1974).
The Delaney Clause: Technical Naivete and Scientific Advocacy in the
Formulation of Public Health Policies. California Law Review 62:1084-1120.
National Research Council (1994). Understanding
Risk. National Academy Press,
Washington, DC.
No comments:
Post a Comment