Extrapolation from large-scale radiation exposures: cancer.

Even though much is known about cancer risk associated with exposure to ionizing radiation, societal demands for detailed risk assessment go far beyond our ability to satisfy them according to customary standards of scientific accuracy. Society's requirement, however, is for the best information available, however good it may be, and not necessarily for the definitive solution to the problem posed. Bayesian methods may be useful for presenting incomplete and varied data and informed scientific opinion in a form suitable for use in societal decision making, while at the same time providing a disciplinary framework for incorporating opinion into scientific recommendations. Inferences about individual cases and their relationship to risk of radiation carcinogenesis provide an especially severe test of the completeness of our understanding of the relationship between exposure and risk. This is particularly true with respect to the distribution of excess risk over time following exposure. Recent work suggests that a standard model often used for projection of risk forward in time, the constant relative excess model, may give a surprisingly accurate picture of time to tumor for a number of cancer sites.