How Do Scientists Determine the Long-Term Risks from Radiation?

How Do Scientists Determine the Long-Term Risks from Radiation?

Where did the risk estimates in this report come from?

How are risk estimates expressed?

It is well established that cancer rates begin to rise above the normal background rate only some time after exposure, the latent period, which varies with the type of cancer and other factors such as age. Even after the latent period has passed and radiation effects begin to appear, not all effects are due to radiation. The excess rate may still vary by age, latency, or other factors, but for many cancers it tends to be roughly proportional to the rate in the general population. This is known as the constant relative risk model, and the ratio of rates at any given age between exposed and unexposed groups is called the relative risk. Many advisory committees have based their risk estimates on models for the relative risk as a function of dose and perhaps other factors. Other committees, however, have based their estimates on the difference in rates between exposed and unexposed groups, a quantity known as the absolute risk. This quantity also varies with dose and other factors, but when this variation is appropriately accounted for, either approach can be used to estimate lifetime risk.

What are the types of data on which such estimates are based?

Why are human data preferable to data on animals or tissue cultures for most purposes?

How have human data been obtained?

Why is it necessary to compare exposed populations with unexposed populations?

Why is the analysis of epidemiologic data so complicated?

1. Random variation due to sample size.

2. Multiple variables.

3. Limited time span of most studies.

4. Problems of extrapolation.

In addition, individual studies may also be biased in their design or implementation.

What is random variation?

A simple-minded example of this is the classic puzzle of determining, in a drawer of 100 socks, how many are white and how many are black, by pulling out one sock at a time. Obviously, if we pull out all the socks, we know for certain. In most areas of study, though, "pulling out all the socks" is far too expensive and time-consuming. But if we pull only 10, with what degree of confidence can we predict the color of the others? If we pull 20, we will have more confidence. In other words, the larger the sample, the greater our confidence. Using statistical techniques, our degree of confidence can be calculated from the size of the entire population (in this case 100 socks) and the size of the actual sample. The result is popularly called the margin of error.

The most common examples of this in everyday life are the public opinion polls continually quoted in the news media. As can be seen in the simple example of the drawer of socks, the highest degree of confidence can be achieved simply by pulling all the socks out of the drawer. For public opinion polls, this would be far too expensive; instead, a small sample is selected at random from the population. Nowadays it is common to report not only the actual results, but also the sample size and the margin of error. The margin of error depends not only on the sample size, but also on how high a degree of confidence we desire. The degree of confidence is the probability that our sample has provided a true picture of the entire population. For example, the margin of error will be smaller for 80 percent degree of confidence than for 95 percent. Even where a study covers an entire exposed population, such as the atomic bomb survivors, the issue of random variation remains when we wish to generalize the findings to other populations.

What are multiple variables?

A more subtle problem is misspecification of the model finally chosen to calculate risks. The model may weigh selected factors in a manner that best fits the data from a statistical viewpoint. This model, while fitting the data, may not actually be a "correct" view of nature; another model that does not fit the data quite as well may actually better describe the as-yet-unknown underlying mechanisms.

Why does a limited time span reduce the value of a study?

However, another concern is understanding the effects of much lower levels of radiation. Unlike the more acute effects of large exposures, these may not appear for some time. Some cancers, for example, do not appear until many years after the initial exposure. These latent effects may continue to appear in a population throughout their entire lifetimes. Calculating the lifetime risk of an exposure requires following the entire sample until all its members have died. Thus far, none of the exposed populations have yet been followed to the ends of their lives, although the radium dial painter study for the group painting before 1930 essentially has been completed, and the follow-up has been closed out.[110]

Why does extrapolation among human populations pose problems?

Even though humans are all members of the same species, there are similar problems when extrapolating results from one group of humans to another group. Within the human species, different groups can have different rates of disease. For example, stomach cancer is much more common and breast cancer much rarer among Japanese than among U.S. residents.

How then should estimates of the radiation-induced excess of cancer among the atomic bomb survivors be applied to the U.S. population? Assumptions are needed to "transport" risk estimates from one human population to another human population that may have very different "normal" risks.

Why does extrapolation from high to low doses pose problems?

The basic assumption concerns the dose effect. Is the effect of a dose linear? This would mean that half the dose would produce half the effect; one-tenth of the dose would produce one-tenth of the effect, and so forth. Nature is not always so reasonable, however. There are many instances in nature of nonlinear relationships. A nonlinear dose effect, for example, could mean that half the dose would produce 75 percent of the effects. Or, going in the other direction, a nonlinear dose effect could mean that half the dose would produce only 10 percent of the effect. Reliable data are too sparse to settle the issue empirically. Much of the ongoing controversy over low-dose effects concerns which dose effect relationship to assume. Concerning dose response, most radiation advisory committees assume that radiation risks are linear in doses at low levels, although these risks may involve nonlinear terms at higher doses.

Another assumption concerns the effect of dose rate. It is generally agreed that the effect of high-dose x rays is reduced if the radiation is received over a period of time instead of all at once. (This reduction in acute effects, due to the cell's ability to repair itself in between exposures, is one of the reasons that modern protocols for radiotherapy use several fractionated doses.) The degree to which this also happens at low doses is less clear. There are few human data on the effect of dose rate on cancer induction. Most estimates of the effect come from animal or cell culture experiments. There is also evidence of quite different dose-rate effects for alpha radiation and neutrons.

How can a specific study be biased?

What is called a confounding bias may result if factors other than radiation have affected disease rates. Such factors, as mentioned earlier, might be a rate of smoking higher than the general population.

A selection bias may result if the sample was not truly a random selection from the population under study. For example, the results of a study that includes only employed subjects might not be applicable to the general population, since employed people as a group are healthier than the entire population.

An information bias may result from unreliability in a source of basic data. For example, basing the amount of exposure on the memory of the subjects may bias the study, since sick people may recall differently than healthy people. Dose, in particular, can be difficult to determine when studies are conducted on populations exposed prior to the study, since there usually was no accurate measurement at the time of exposure. Sometimes when dose measurements were taken, as in the case of the atomic veterans, the data are not adequate by today's standards.[111]

Finally, any study is subject to the random variation discussed earlier, which depends on how large the sample is. This is more important for low-dose than for high-dose studies, since the low-dose effects themselves are small enough to be lost amid random variations if the sample is too small.

To summarize, multiple studies may produce somewhat different results because there is an actual difference in the response between populations or because studies contain spurious results due to their own inadequacies. In addition, it must be recognized that the entire body of scientific literature is itself subject to a form of bias known as publication bias, meaning an overreporting of findings of excess risk. This is because studies that demonstrate an excess risk may be more likely to be published than those that do not.

In view of all these uncertainties, what risk estimates did the Committee choose?

The BEIR V Committee of the National Academy of Sciences estimated in 1990 that the lifetime risk from a single exposure to 10 rem of whole-body external radiation was about 8 excess cancers (of any type) per 1,000 people. (This number is actually an average over all possible ages at which an individual might be exposed, weighted by population and age distribution.) For continuous exposure to 0.1 rem per year throughout a lifetime, the corresponding estimate was 5.6 excess cancers (that is, over and above the rate expected in a similar, but nonexposed population) per 1,000 people. It is widely agreed that for x rays and gamma rays, this latter figure should be reduced by some factor to allow for a cell's ability to repair DNA, but there is considerable uncertainty as to what figure to use; a figure of about 2 or 3 is often suggested.[112]

The estimates of lifetime risk from the BEIR V report have a range of uncertainty due to random variation of about 1.4- fold. The additional uncertainties, due to the factors discussed earlier, are likely to be larger than the random variation.

In comparison, for most chemical carcinogens, the uncertainties are often a factor of 10 or more. This agreement among studies of radiation effects is quite remarkable and reflects the enormous amount of scientific research that has been devoted to the subject, as well as the large number of people who have been exposed to doses large enough to show effects.