Letter writers Jeffry Siegel, Charles Pennington, and Bill Sacks state that modern-day concepts of a linear dose response to ionizing radiation are based on fruit-fly data collected 70 years ago. Actually, fruit-fly data haven’t been important since William Russell’s Oak Ridge mice data became available in the 1960s. Today, linearity is based on fits to data for cancer incidence or mortality as a function of dose received by individuals in large exposed populations, such as the atomic-bomb survivors (125 000), the Techa River cohort in Russia (17 000), and radiation workers (300 000–600 000). A linear fit is taken as the conservative starting point for dose response, with quadratic terms turning out to be modest.1 Widespread consensus exists that linearity holds at least down to 100 millisieverts, and there is a broad but not unanimous view that it is likely to continue to apply at lower doses—that risk will continue to decrease in proportion to dose.
Arguments about repair and evolutionary protection are not sufficient. On occasion, repair systems can fail—for example, mismatched repairs of breaks in double-stranded DNA. Protective systems, such as tumor-suppression genes, can be damaged or turned off by ionizing radiation. Furthermore, ionizing radiation is a promoter, not just an initiator; it can affect cells already genetically damaged by other causes.
Siegel and coauthors cite a claim Siegel makes2 that a graph of atomic-bomb-survivor data for cancer incidence suggests a threshold in the epidemiologic noise region of the dose response (below 100 mSv) where uncertainty of data points is great, but the comparable graph for mortality,3 suggesting a supralinear response, is not shown. They also cite a 58-page review by the French Academy of Sciences and National Academy of Medicine,4 but they do not mention the more comprehensive 500-page BEIR VII report by 17 experts and 16 reviewers assembled by the US National Academies.1 That review concluded “that the risk would continue in a linear fashion at lower doses without a threshold and that the smallest dose has the potential to cause a small increase in risk to humans” (page 7).
In the decade since the French academies review, a wave of studies of protracted human exposure, as discussed in a 2009 meta-analysis,5 has suggested that protracted exposures have dose responses similar to or greater than single exposures. It is hard to justify a threshold if a dose accumulated from a large number of small exposures has an impact the same as or larger than the same dose delivered in a single exposure.
The public-health and risk-assessment communities should ignore partisan views and assume the linear no-threshold (LNT) dose-response relationship at low doses, with uncertainty bands above and below the LNT that cover alternative hypotheses. Disputes over radiation dose-response models distract attention from the fact that the individual risks at the low-dose levels that are being debated are small, whether assessed using a linear, supralinear, or threshold model. However, for situations in which hundreds of thousands of people are irradiated, risk is spread over a huge population in a kind of reverse lottery. To estimate such social risks is considered inappropriate by some, but it is necessary for cost–benefit calculations in retrofit analysis of nuclear power plants or for seeing whether medical diagnostic procedures carry a net population benefit. Accounting for uncertainty should make such calculations more palatable. The concern that the public can’t handle bad news about risk is misplaced. What destroys public trust is the idea of a cover-up, which is implied by an unwillingness to calculate possible risks.
