Uncertainty and the "Flavors" of Risk

Let's give risk assessment a reality check
by Robert J. Scheuplein
(Director of the Office of Special Research Skills at the Food and Drug Administration.)

EPA JOURNAL JAN UARY/FEBRUARY/MARCH 1993 19(1):16-17

If you are an average person (or even above average), you're exposed to a large amount of risk. Like ice cream, risk comes in a variety of flavors, and some people like to add nuts and berries to suit their tastes. So it is with risks. But the basic ones-the vanilla, chocolate, and strawberry of risks, if you will-are the following three: personal activities, natural disasters, and chemical exposures.

Consider first certain dangerous personal activities like firefighting, coal mining, skiing, motorcycling, driving automobiles, etc.-things you do for a living or for fun. Here the danger in the activity may be self evident, and the risk is ordinarily self imposed. As a class, these risks are the highest, along with risks from ordinary diseases, which personal behavior can also affect. For example, the annual risk associated with motorcycling is about 2 percent, or about 2,000 deaths per 100,000 persons at risk. Firefighting is much safer, only about 80 annual deaths per 100,000, or 0.08 percent. The annual death rate from motor vehicles is around 24 per 100,000, or 0.024 percent. This is just the death rate; of course, the accident rate and injury rate are much higher.

The risks above might be described as part of the price we pay for living in a civilized world, but the next category of risks, natural disasters, are not wholly our fault. The risks from floods, hurricanes, earthquakes, lightning, meteorite hits, etc., are the price we pay for our 70-year or so lease on the planet. Of course, if you want to live on an earthquake fault, you share some responsibility. These risks in the aggregate are quite small. But try telling that to the folks in South Florida who experienced Hurricane Andrew. Lightning kills 0.05 people per year per 100,000, or about 0.00005 percent. The risks from meteorite hits are about 0.000006 per 100,000, or 0.000000006 percent.

Now the last major category of risks, the "strawberry of risks," derive from chemical exposures. Before we discuss them, let me emphasize an important point. The risks above are real, obtained by counting victims. They are actuarial risks. They depend only on how accurately the deaths and the populations at risk were attributed and recorded. They are not based on inferences from animal data, nor on prudent extrapolations of adverse effects in animals. This distinction is essential to make because the risks from chemical exposures are for the most part based on such inferences and extrapolations

Currently the regulatory objective is often fulfilled

at the expense of the scientific one.

Everyone knows about poisons, drugs, and acute occupational exposures to industrial chemicals. For many of these exposures there is human data. But for chronic low-level risk from chemicals in the environment we live in, in the air we breathe, in the water we drink, or in the food we ingest, we need to depend on animal data. These risks are ordinarily small. For example: The cancer risk from chlorinated drinking water has been estimated as 0.8 per year per 100,000 persons exposed or 0.0008 percent. While this number looks the same as the others and can be expressed in the same units, it is not the same and can be compared with the actuarial risks only if the differences are kept in mind.

What are these differences?

First, as stated above, chemical risk is based on the finding of an adverse effect in an animal study. In the case of chlorinated drinking water, it is based on several carcinogen bioassays conducted in mice using various chlorinated compounds. It is inferred that humans will be similarly susceptible to these same compounds. But this is not necessarily true.

Second, the quantitative result is a worst- case estimate sometimes called an upper- bound estimate. It is based on a mathematical extrapolation of adverse effects in animals, exposed at high dose levels, to the much lower levels anticipated for humans. Why is this done? Why not just expose the animal to the appropriate lower doses? The reason is there would be no effect at low doses, unless the number of animals in the experiment were increased dramatically-say to several thousand. The problem lies in trying to detect in a population of 100 a disease incidence that you might believe to be one in 1,000. So toxicologists need to exaggerate the animal doses and extrapolate downwards-hopefully.

To continue with our particular example, the actual amount of chlorinated hydrocarbon chemicals in drinking water, the chemical byproducts of the chlorination process, is very small, typically a few parts per billion. The doses to which the animals were exposed are very high, many thousands of times higher than human exposure levels. The high-to-low dose extrapolation is used to estimate the effect of the lower dose using various conservative assumptions. The most important of these assumptions is that there will inevitably be some cancer risk no matter how small the dose.

Third, the risk is an average attributed risk; it applies to no one in particular and to everyone on the average. 'if you're an average person and you drink the expected amount of water with the expected concentration of chlorinated compounds, your possible risk is no greater than the given risk number. But in no way is this intended to be a predicted risk for you individually; your particular pattern of exposure, your exposures to other carcinogens, your genes, your diet, and other factors determine your particular susceptibility. This is the way carcinogenic risks are determined for most regulated chemicals: foods and cosmetics, pesticides, household chemicals, most industrial and workplace chemicals, air and water pollutants, and toxics and waste site contaminants. (Drugs and biologics are usually regulated with human data.)

Of course, not all chemicals present a cancer risk but they can pose other risks. There are chemical substances that affect developmental, reproductive, neurobehaviorial, and other body functions. Typically, such substances are regulated by determining "'no-effect levels" in animals and applying safety factors. Numerical risk estimates are not made because thresholds are assumed. In other words, unlike for carcinogens, risk is not assumed to be present at all doses.

Cancer risks of less than 10-6-one in a million per lifetime or one in 14,000 per year or 7 per 100,000 per year or 0.007 percent-are usually not considered worth regulating. (Lifetime risks are approximately 70 times higher than annual risks if the risks are similar from year to year for a lifetime.)

The inherent conservatism in estimates of the cancer risk may be illustrated the following way. Suppose you work for a regulatory agency and you are asked for the agency's official estimate of the average height of a person. You remember that the average height of an American male is, say, 5 feet, 10 inches. But that applies only to American men, and probably doesn't include modern American basketball players, some of whom are over 7 feet.

Getting the data on all the people in the world is impractical, but unless you do, you can't really give a figure without including some certain error. And the size of the error is also impossible to obtain. So in order to be absolutely clear and correct in your response you decide to give a worst-case estimate. You will cast your response in the form: "The average height of a man will in no case exceed .... "

This is a very strong statement, so to hedge your bet and to be sure you're right, you will have to make conservative assumptions. One you might make is that the average height of a person will in no case exceed the tallest person in the world. This contains the inherent reliability one likes to have when called upon to defend the regulatory decision against tall activists. Now, the tallest people you know about from your research are all less than 8 feet. But there may be giants somewhere, and there is some anecdotal evidence. (Remember the stories about "Bigfoot.,") Let's assume you find a record of a 12-foot giant now deceased. On the possibility that he might have left living relatives, you assume a maximum height of 15 feet because there is plenty of data indicating that better nutrition over the last 50 years has increased the average body size by about 20 percent. So your official response, supported by several pages of data, reads:

"The average height of a person will in no case exceed 15 feet."

This statement has all the required regulatory qualities needed for the Federal Register. It is impeccably correct. It will withstand any legal challenge. It is prudent and does not underestimate the height. It also has at least two undesirable qualities: It is not very helpful. And it discriminates against short people. (Translation: The recasting of the regulatory problem away from probable risk (average height) to worst- case risk results in the under appreciation of risk-lowering factors.)

The linear extrapolation of rodent bioassay data embodies the regulator's credo ("It's better to be safe than sorry") far more than it does the scientist's, ("It's better to be right than wrong"). Currently the regulatory objective is often fulfilled at the expense of the scientific one.

When carcinogens were few, biological understanding of mechanisms more primitive, and analytical sensitivity in the parts per million range, the differences in these two points of view were not large and didn't really matter much. Today, for many substances, the situation has changed in each of these areas, and we face an ever- growing separation between the application of good science and credible, efficient regulation.