Panel Discussion

ELIZABETH L. ANDERSON, Ph.D., Director, EPA Office of Health and Environment Assessment, Office of Research and Development, Washington, D.C.

NORTON NELSON, Ph.D., New York University Medical Center, Institute of Environmental Medicine, New York, NY

ALAN W. ECKERT, Esq., Moderator, Member Standing Committee on Environmental Law; U.S.EPA, Office of General Counsel, Washington, D.C.

DR. ELIZABETH L. ANDERSON: Dr. Whipple has emphasized that risk assessment is on the rise because of policy needs, and I think this is certainly true. Two other points come to mind, and I think they are worth mentioning. First, as the number of our laws increased, scientists searched for their role in the public policy arena. Over time, scientists have become more cautious about making decisions to define a safe level of exposure in the face of scientific uncertainty. Policy makers and lawyers often press science for the answer to what is safe. When faced with considerable uncertainty, [16 ELR 10196] scientists are becoming less willing to make these judgments. Final decisions to define safety frequently cannot be based solely on scientific answers but rather must include social and economic considerations in deciding "safety" in terms of how much risk to accept.

In many cases, scientists have started to include threshold as well as non-threshold effects in risk assessment, thus tossing the decision about acceptable risk back to the policy arena for an increasing number of chemicals.

The second important factor contributing to the increase in risk assessment, particularly in the United States, is the public's perception of risk and the public's desire to know the nature of the residual risk. Whether justified or not, there is an enormous public fear of toxic substances, and this fear drives us to an explicit statement of risk. The "doctor-in-the-white-coat approach" no longer satisfies an inquisitive public. Yet it is very difficult to explain risk to the public. People often feel extremely threatened by something that we may regard as a very trivial risk. Unfortunately, there appear to be no easy answers as to how we might improve risk communication with the public.

I would like to comment briefly on the role of uncertainty in risk assessment. Indeed, risk assessment is an assessment of uncertainty. Some time ago, Dr. Richard Wilson, addressing a conference at the Banbury Center, said that if one knows something for sure, one does not need to perform a risk assessment. So it makes no sense to discard risk assessment because it is too uncertain; there is no alternative. I would like to see the scientific uncertainties in risk assessment find better translation in the subsequent policy/legal process.

Risk assessment is basically a process that addresses two questions: how likely is a risk to occur? and, if it does occur, how bad will it be? The first step is to identify the likely hazard. In identifying carcinogens, it is important not to focus only on the numbers and fail to pay attention to the strength of the evidence; the two do not necessarily go hand in hand. For example, when comparing arsenic, benzene, and vinyl chloride, we see a 1,000-fold higher potency estimate for arsenic than for the other two, and all three chemicals have convincing human data. Sometimes the weight of the evidence for a particular compound will appear much stronger simply because the compound has been tested more thoroughly. This does not mean that the compound is more potent than other compounds. The new Environmental Protection Agency (EPA) guidelines incorporate hazard identification explicitly through a coding of letters A through F, which will be expressed side by side with the quantitative work.

In assessing risk, it is important to separate the science from the risk management. Precisely because policy judgments are involved in risk assessment, assumptions must be made. To make sure these assumptions are based on the best science available rather than on some ultimately desired policy outcome, the risk assessment must be performed independently. The risk manager should use the risk assessment information as an impetus to the choice of action; only then should social and economic considerations be factored into the decision.

Finally, I would like to speak about the use of models in risk assessment. There are basically two classes of models: mechanistic models and tolerance models. Mechanistic models attempt to model what we understand about the cancer process. We do not understand enough, but there are some plausible explanations of how cancer progresses. The notion that a single cell can be disrupted in such a way that the cell is damaged — probably the DNA — and then goes through multiple stages to produce cancer underlies the linear, non-threshold concept of cancer induction. Tolerance models, which attempt to model the sensitivity of populations to disease, do not rely on this kind of basis.

The upper bound estimate of risk process used by EPA relies on a linear model at low doses. This simply means that the risks are unlikely to be higher than the dose-response model, although they could be considerably lower. Both types of models most often fit the data equally well in the observed dose range, but come apart very quickly as we go to lower doses, where some of the models give answers of zero.

All of this is to say that one can run all kinds of models and obtain a host of answers. The answers are not very instructive to policy makers because the scientists cannot confirm that any one model provides the right answer. For the majority of chemicals, the plausible upper bound on the risk is about the most accurate answer that scientists can provide.

DR. NORTON NELSON: In 1980 the Doll-Peto report on cancer concluded again what was already widely understood by those in the field: that cigarette smoking is a major source of cancer.

Likewise, the implication that dietary factors may be related to cancer goes back several decades, when a series of migrant studies called attention to large disparities in cancer rates according to organ site in migrant populations. As the issue was studied more intensively, it became apparent that the findings could only be attributable to factors of cultural or lifestyle factors, and not to genetic differences. On the basis of these disparities, foods were (and still are) considered likely factors in the development of cancer.

In using the term, "environmental cancer," many scientists, including myself, may have made a poor choice of words. This has permitted a widely held misconception to persist. By "environmental," we did not mean man-made chemicals. We meant merely that the risk factors were exogenous, that is external. By that definition, diet and smoking are clearly external factors.

Having, I hope, corrected that misconception, I would like to move on, and compare the fields of toxicology and epidemiology as sources of risk assessment. Unfortunately, epidemiology is rarely adequate or even available, but it is useful to remember some of the trade-offs involved when one moves from animal studies in the laboratory to epidemiological studies out in the field. Obviously, epidemiology is the more relevant, since the studies involve human beings. We are interested primarily in human disease, not rat or mouse disease. All animal studies are to some degree suspect as to relevancy.

Control of variables — and these are numerous — such as exposure, environmental confounding factors, and so forth is excellent in the toxicology laboratory but poor in epidemiology.

Toxicological laboratory studies are also excellent in terms of identifying causal factors. Identification of causal factors can be very poor if one depends exclusively on human studies.

In terms of size of population, laboratory studies are very difficult because of the cost. One of the largest scale studies ever done in chemical carcinogenicity was performed at the National Center for Toxicology Research. The study used a population of such size that the researchers could detect an effect on the order of one percent; that study was very expensive and is not likely to be done many times again. Size [16 ELR 10197] of population is unlimited as far as humans are concerned, although it costs money.

Whether one uses laboratory or epidemiological studies, sensitivity is poor. In laboratory studies, one has the advantage of being able to increase the dosage in order to increase the apparent sensitivity. It is done literally to overcome the insensitivity of the small groups that are under study. This practice is one source of the criticism of unrealism that has been levelled against the use of laboratory studies.

The contrast in genetic diversity between epidemiological and toxicological studies is enormous. The genetic diversity of the human population is wide, and brings with it a wide range in cancer sensitivity that we cannot always define. For better or for worse, there has been a deliberate attempt to narrow genetic diversity in laboratory studies. To obtain reproducible results, one needs to use narrowly-defined genetic strains, and this may be a mistake. To use genetically diverse studies in animals means that the studies would have to be expanded enormously in size to achieve the same degree of reproducibility. There is some sense, however, in choosing a deliberately narrow genetic strain, if one is sure that one has chosen the most sensitive strain.

Intercurrent disease is theoretically controllable in the laboratory. It is not controllable in human studies, where outside factors can alter responses; however, epidemiologists have devised ingenious methods to partially deal with such interferences.

Finally, study of mechanism is easily accessible in the laboratory. Ethical considerations prevent one from going very far in terms of human studies. One can kill one's animals halfway through an experiment and look at their livers; this approach can hardly be regarded as practical with humans!

In closing, let me address the issue of the nonthreshold view of risk. When the conclusion is reached that, in some instances, there is probably no threshold, one has to acknowledge that some degree of danger exists. If there is, in fact, a finite risk at every conceivable dosage level of a chemical carcinogen, the traditional safety-factor approach simply cannot work. This means that one will have to live with a judgment as to risk, and that judgment may involve much guesswork.

Regulators may have to tell the public "there is a risk, but the things you're eating are safe." That, in essence, is the dilemma. Even though a risk exists, one must manage to convey the idea that the danger is an acceptable one.

I can never think of this dilemma without thinking of the Delaney Amendment. I was one of those who strongly opposed the Delaney Amendment when it first came up, because I said that science should be sufficiently advanced to estimate the risk and thereby provide some guidance as to the legislative control of food additives. Today, having thought more about the difficulties, the underlying principle seems one of social economy. For instance, is finding another way to make ice cream pink so important that we are going to perform elaborate and expensive tests in order to judge the additive safe?

Accordingly, I question whether we should even bother with risk assessment in cases where the social benefit is so insignificant that the benefit justifies only a qualitative judgment on a chemical. I find, thus, that the Delaney Amendment constitutes a legitimate approach in those few instances where the social benefit is so minor that there is no reason to get into a big sweat about abandoning use altogether.

PARTICIPANT: I am troubled by the implication, perhaps inadvertent, that if a risk assessment involves a large factor of uncertainty, it is somehow a bad risk assessment. To the contrary, is it not really incumbent upon scientists to preserve the uncertainty? By preserving this information — which we, as scientists, know is very good scientific information — the risk assessor has preserved that structure and has allowed the risk manager to become aware of the uncertainty.

The Science Advisory Board is currently drafting guidelines that would require every risk assessment to state explicitly the range of uncertainty.

ANDERSON: A good risk assessment ought to express what you know, as well as what you don't know. The reason for the guidelines proposed by the Science Advisory Board is to organize the generic uncertainties that come up repeatedly. If we have more information in a specific case, we can deal with the risk more precisely.

I do not know exactly what the Science Advisory Board has in mind, but do not be misled into believing that we have methods at hand for the majority of chemicals to give you on the dose-response curve some range around the upper bound. We can establish ranges for exposure, but we do not have any good ideas about how to express a range around the upper bound. We say the upper bound puts an upper bound on the risk, and that risk's lower bound could be zero. The Science Advisory Board may have something else in mind.

DR. CHRIS G. WHIPPLE: I agree that risk assessments in which few sources of uncertainty are considered are not, as a general rule, preferable to those with a more detailed treatment of uncertainty. I have, however, seen that notion gone berserk. In some cases the concern that all sources of uncertainty be recognized is so great that a range is given without comments regarding the plausibility of the numbers in the range.

It is incumbent upon risk assessors to find a mechanism to communicate what they know — to tell what they have estimated, based on assumptions that they think are reasonable, and to describe the uncertainty. I do not think that a range is adequate. Frequently, estimates range from zero or near-zero risk to estimates which are zero, and they will be so high as to be implausible. We have to learn how to define range estimates in a more sophisticated manner.