Tobacco Documents Online

10 Harris: It turns out that a great many chemicals that can cause cancer in one species don't seem to do anything at all in another species. Here's an analogy. (Excerpt from music from a CD) Harris: The difference between rodents and people can be as dramatic as the difference between this CD and an LP. You could drop this CD, get it dusty, even scratch it, and you wouldn't necessarily hurt it. (Sound of record being scratched by a stereo needle and music played from an LP) Harris: By try the same thing with a record, and you can just hear the damage. To be sure, some things will damage either a CD or a record album--say, a hot windowsill. Likewise, John Doull from the University of Kansas says some chemicals do cause cancer in all sorts of animals ... Now nobody's suggesting that these chemicals are harmless, but in some cases scientists believe that the standards may be vastly overstating the health risks. Again, this comes down to a necessary but flawed shortcut the EPA uses to size up a chemical. Scientists give a huge dose of chemicals to rats and then estimate the effects of that chemical at lower doses. By

11- way of analogy, if you drop a bottle from 10 feet off the ground, it's pretty obvious what's going to happen. (Sound of glass shattering) Harris: This large drop is equivalent to a large dose of a chemical, and it can be deadly. But what if, instead of taking one bottle and dropping it from 10 feet, you take 10 bottles and drop them from one foot? It's like giving many people a smaller dose of that toxic chemical. Here's what the EPA assumes will happen. (Sound of several bottles hitting the ground and one of them shattering) Harris: They figure one of the 10 bottles will break. The reasoning is that one-tenth the dose, or one-tenth the drop . distance, will do one-tenth the damage. In reality, though, this is what happens. (Sound of several bottles hitting the ground) Harris: There is, in fact, a safe height you can drop a bottle from without breaking it, and John Doull from the University of Kansas says the same idea holds for toxic chemicals.

12 Doull: It is the dose, not the compound, that determines its adverse effects . . . . Harris: So, recently, researchers like Swenberg have started to dig deeper and ask why some chemicals trigger cancer in some animals. It's as though they're trying to understand the difference between turntables and CD players. And Swenberg says one especially interesting example is unleaded gasoline. You may have been the sticker at the pump warning that gasoline causes cancer in laboratory animals. Well, here's the story with gasoline. Swenberg: It causes kidney cancer in male rats only, not in female rates and not in mice. Harris: So what's going on? Swenberg decided to find out by studying those animals, and he discovered that a chemical in gasoline binds to a naturally-occuring protein that's only found in the kidneys of male rats. Swenberg: And this results in a build-up of the protein and ultimately leads to the development of cancer. And since humans • do not synthesize this protein, this is not likely to be a -- a mechanism important to humans. Harris: Swenberg says dozens of other chemicals besides gasoline cause this specific kidney cancer in male rats,

1 3 including copy machine toner, a bathroom deodorizer and even a natural chemical called D-limonene. Swenberg: It turns out that about two glasses of orange juice contains a carcinogenic amount of D-limonene for the male rat, but it has absolutely no effect on mice or on female rats, and I'm sure it has no effect on humans. Harris: As a result of this research, the Environmental Protection Agency recently decided that if a chemical like gasoline only triggers this kind of kidney tumor in male rats and it doesn't do anything else bad, it's probably not going to cause cancer in people. So far there are just a handful of stories like this where scientists have actually figured out why a compound is causing tumors in certain animals. But there are a lot more studies in the works, including reassessments of dioxin, formaldehyde and certain PCBs.21 Knowledge of mechanisms yields far greater discriminatory power. With such knowledge scientists can determine whether there is a threshold below which there is no damage or whether harm occurs proportionate to the dose, however low that dose is. No mechanism, no dose-response relationship. The War over the Dose-Response Threshold .

14 There is disagreement over whether there is a dose-response threshold, so that below a certain level no harm occurs, or whether the damage is linear, such that harm from a chemical increases or decreases as a proportion of the dose. It is important, to start with, to ask why such an apparently technical matter has occasioned so much dispute. Because the field of toxicology is built on the principle that the poison is in the dose, the opposing linear (or proportional) principle-- there is no threshold dose below which damage cannot occur--is a challenge to toxicological science. A common statement about dose response levels is that no one really understands what happens when people are exposed to very low levels of chemicals.22 There is no difficulty in finding substances, such as the heart medicine digitalis, that are helpful at low doses but can be fatal at large doses. But that does not answer the question of whether there are substances for which no threshold exists.23 Given there is a considerable range of sensitivity among human beings, it can always be 'said that some hypersensitive people might be adversely affected. The traditional response has been to use a margin of safety to take care of the supersusceptible. Given also that chemicals may interact with each other to create cancers that neither substance would alone, it cannot be said definitively that either is safe. By the same token, however, one chemical may render another harmless or less harmful.24 The regulatory response is that the dose-response relationship is linear. The rationale is that this provides a margin of safety for the public. The question is whether this assumption is true.

1 5 Going further into the "furious battle [that] rages around the 'threshold controversy,"'25 it will be instructive to read a semiofficial account by high-ranking Environmental Protection Agency officials published in a major journal, Risk Analysis. The models EPA uses attempt to establish an upper-bound, nearly the worst that could happen, on the basis of a no-threshold linear response. Anderson et al. are quite open in saying that "This recognition that the lower bound may . . . be indistinguishable from zero stems from the uncertainties associated with mechanisms of carcinogenesis including the possibility of detoxification and repair mechanisms, metabolic pathways, and the role of the agentt in the cancer process."26 In short, for all EPA knows, there may be no damage at low doses. Furthermore, Anderson et al. continue, "Most often there is no biological justification to support the choice of any one model to describe actual risk."27 While this task would be easy if there were data on actual environmental exposures to human beings, in which case an appropriate model could be fitted to the data, the EPA article continues, "In the absence of such data a variety of models can be used to fit the data in the observed range, but these models differ sharply [in the danger estimates they produce] at low doses."28 If the choice of model determines the results, because they "differ sharply at low doses," why bother with the experiment? Exactly. Employing the justification that nevertheless these models are the best available, Anderson et al. state that "It should be clear from the preceding discussion that the linear non-threshold model has been used by the EPA to place plausible upper bounds on risk, not to establish actual risk."29 This is a significant admission.

16 My question is "best for what"? Use of the upper bound misleads people into thinking it is an actual estimate of hazard by an authoritative government agency when it is not. Use of "worst case" scenarios makes no sense, moreover, when there is reason why the outcome may be zero and there is no biological sense in anticipatory epidemic consequences. Now we know that everything depends on which of the available statistical models are used and whether whichever one is chosen in the absence of biological indications, tells us what we need to know. Does it? EPA claims that the linear, no-dose-response model best fits knowledge about cancer causation. B ut its officials could not know this without knowledge of the mechanisms at work, in which case they would be able to choose one they knew fit the causal relationship. At other times, they acknowledge the real basis for their choice of model, the desire to choose the most conservative estimate so as to, as the saying goes, act on the side of safety. But, are they so acting? If it is true, as Anderson et al. say in their appendix, that "There is no really solid scientific basis for any mathematical extrapolation model relating carcinogen exposure to cancer risks at the extremely low levels of concentration that must be dealt, with in evaluating environmental hazards,"30 then why make one? The answer must be that with going from rodents to people most regulation of chemicals would lack a rationale that could be called scientific. No science, no regulation; no man-mouse extrapolation, no N Ln r science, no regulation. The models that make this extrapolation E., ~ ~ ~

1 7 plausible are the important thing. Extrapolations from animals dosed at very high levels to people exposed to far smaller levels make sense only in the context of the models of cancer causation into which they are meant to fit. Multistage Models Is this chemical guilty and to what degree at which dose and to whom? Interpretation of cancer causation depends on models, which we can think of as symbolic representations of theories, with numbers attached, that give meaning to the data. No model, no interpretation, no meaning, no result. Actually, there have to be two models in one; first, a model of the biology underlying cancer causation and, second, the statistical approximation of that model. Getting accurate results depends both on the predictive power of the biological model of cancer causation and on whether the statistical approximation captures the causal structure of the model. If the . model does not well describe cancer causation in human beings, and if, on top of that, the statistical approximation does not well describe the model, the errors in both models multiply to give unsatisfactory results. The task is a daunting one. We can take the Armitage-Doll model as representative of those used by governmental agencies in regulation. It seeks to describe the relationship between exposures to chemicals and the incidence of cancer at various ages for men and women.3 t The biological version portrays human cells as going~ through a number of stages that ultimately result in cancer. The hypothesis is that one or NJ c.n ~ ~ ~ ~ ~ ~ w ~

18 more cells receives an insult and then goes through several changes that turn them into malignant cells after which they proliferate. The times and the different stages are not specified. All these stages are probabilistic in that some cells under the same exposure will become cancerous and others will not. "Put another way, with multistage models," Richard Peto tells us, "when all the predisposing factors have been allowed for, luck has an essential role in determining who gets cancer and who does not."32 Thus the stages in the models are essentially probabilities without the users knowing whether human cancer proceeds in those stages or according to those odds.33 In the field of economics, these would be called Markov chain models, which means essentially that every present stage depends on results of previous stages. The time spent in the various stages is assumed to be proportionate to the exposure of the affected individual. The basic difficulty with multistage models, as the reader might imagine, is that there is little reason to believe they actually capture the biological process of cancer formation. At the same time, the statistical manipulations are very far from the causal requirements of the model, so that one has no idea what one has got when the result is cranked out.34 Which brings us to the statistical interpretation of animal cancer studies, the most critical -and least understood part of modeling cancer causation. The existence of twelve different test groups tells us that statistical inference is the essence of the matter: after all, conclusions are to be drawn observing differences between the 0 control group and other animals and between sexual and dose

19 groups. This is not something that can be done by counting on the fingers of one hand. It requires methods based on statistical theory. Biological Interpretation of Animal Cancer_ Tests Fears at al. are wise in concluding that "There is danger in relying solely on the finding of statistical significance without incorporating biological knowledge and corraborative evidence such as the presence of a dose-response relationship for experimentally consistent results in different species or sexes."3S But what if there is little or no biological knowledge? In order to get accurate estimates of the probability that chemicals that cause cancer in animals also cause cancer in human beings, Salzburg recommends applying "the bioassay to a number of innocuous substances. There have to be some compounds that are not human carcinogens, or the whole exercise of looking for carcinogens makes no sense." Yet, after examining the literature,, he finds that "this was never done for the jrodent) lifetime feeding study ...."3b His argument needs to be heard in full: Thus, it would appear that no attempt has ever been made to determine how well society can identify human carcinogens by feeding groups of 50 rats and mice, each, the suspect substance at maximum tolerated doses for their entire lives. Common scientific prudence would suggest that this assay be tried on a group of known human carcinogens and on a group of supposedly innocuous substances (such as sucrose or amino