Assessing Risk at Low Doses

Abstract

Although a risk assessment based on extrapolation of risk from high-dose studies to the low-dose region gives the appearance of a relatively precise estimate of low-dose risk, in fact knowledge is required about the biological mechanisms involved before one can derive an appropriate extrapolation model.

To avoid this problem, one strategy is to attempt to assess risk from studies of populations with relatively low radiation exposures. This superficially appealing approach has distinct weaknesses which must be considered. First, the precision of the estimates (i.e., the narrowness of the confidence interval) is very limited with most low-dose data. Two corollaries to the lack of precision are that the sample size needed to detect an effect will be extremely large, and that the statistical power to detect an effect or a non-zero dose-response regression slope will often be limited. The lack of statistical power also means that if one should detect a “statistically significant” effect, the estimated magnitude of that effect is likely to be seriously biased on the high side.

Another problem with low-dose studies is that the magnitude of subtle biases or confounding in the data set may well be greater than the magnitude of the expected effect. This confounding can either mask a true effect or yield false-positive findings, so there is little certainty about the results.

In general, one can hope to obtain meaningful measures of risk from low-dose studies only if the number of persons studied is very large, as may be achieved by a pooling of studies. It is recommended that studies be developed studies that address gaps in our knowledge about the effects of ionizing radiation exposure, rather than focusing primarily on low-dose studies.