The scientific jigsaw puzzle: Fitting the pieces of the low-level radiation debate

AbstractQuantitative risk estimates from exposure to ionizing radiation are dominated by analysis of the one-time exposures received by the Japanese survivors at Hiroshima and Nagasaki. Three recent epidemiologic studies suggest that the risk from protracted exposure is no lower, and in fact may be higher, than from single exposures. There is near-universal acceptance that epidemiologic data demonstrates an excess risk of delayed cancer incidence above a dose of 0.1 sievert (Sv), which, for the average American, is equivalent to 40 years of unavoidable exposure from natural background radiation. Model fits, both parametric and nonparametric, to the atomic-bomb data support a linear no-threshold model, below 0.1 Sv. On the basis of biologic arguments, the scientific establishment in the United States and many other countries accepts this dose-model down to zero-dose, but there is spirited dissent. The dissent may be irrelevant for developed countries, given the increase in medical diagnostic radiation that has occurred in recent decades; a sizeable percentage of this population will receive cumulative doses from the medical profession in excess of 0.1 Sv, making talk of a threshold or other sublinear response below that dose moot for future releases from nuclear facilities or a dirty bomb. The risks from both medical diagnostic doses and nuclear accident doses can be computed using the linear dose-response model, with uncertainties assigned below 0.1 Sv in a way that captures alternative scientific hypotheses. Then, the important debate over low-level radiation exposures, namely planning for accident response and weighing benefits and risks of technologies, can proceed with less distraction. One of the biggest paradoxes in the low-level radiation debate is that an individual risk can be a minor concern, while the societal risk—the total delayed cancers in an exposed population—can be of major concern.Keywordsatomic bombdebatedoselinear no threshold modellow level radiationone time exposureprotracted exposureradiationrisk FundingThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.NotesNotes1 There had been hints in some earlier, smaller, occupational studies, e.g., CitationWing et al. (1991).2 When exposures are only to gamma and beta rays, they are usually measured in units of grays (Gy). But when other types of radiation are involved in a dose calculation—for example, neutrons and alpha particles—the comparable unit is either a "weighted gray" or a sievert. These latter units weigh different types of radiation by their estimated effectiveness in increasing delayed cancer rates; therefore, doses and health risks can be compared among epidemiologic studies that involve different mixtures of radiation types.3 Some analysts add a quadratic term to account for any curvature in the dose-response curve. However, at low doses, this linear-quadratic model becomes a linear model.4 Weighted colon dose multiplies the neutron dose component by an effectiveness factor of 10 (CitationChomentowski et al., 2000).5 No model is assumed, which means the fit is nonparametric. Instead, the data are smoothed and used to define straight-line segments. The circles indicate the excess relative risk at the mean dose in each of the 22 specific dose categories.6 An ERR of 0.1 means that the cancer rate has increased by 10 percent. To get an absolute risk, one multiplies the ERR by the baseline cancer rate.7 Some critics also argue, incorrectly, that the atomic-bomb data have no control groups. In fact, it has the best types of control groups epidemiologic methods can offer. When a study can present excess health effects as a function of dose, it has internal controls for comparison. Subjects in each dose category serve as controls for each other, all the way down to zero dose.8 The 1980 Biologic Effects of Ionizing Radiation (BEIR) report found that the slope of the response was 10 times lower than the value in the 2006 BEIR report; the 1995 BEIR report found it was twice as low (CitationMarshall, 1990; CitationNational Research Council, 2006; CitationReissland, 1981). Interestingly, as the slope of the birth-defect response decreased considerably over time, so, too, did public interest in the matter; perhaps the direction of change in these slopes defines, in part, public concern.9 Some caveats: First of all, unexpected results are always a possibility and could very well appear in the next study. Second, molecular analysis of future tumors may one day lead to identification of subgroups with dose-responses that are much higher than the population average. Also, new cancer cases will emerge in the future for those exposed at young ages; possibly, these rates will be dramatically different in old age than current projections expect. As a result, lifetime risks for those exposed at young ages could conceivably change significantly.10 At the time, this theory was supported by the interpretation of experiments on cells showing that, for the same total dose, cellular damage increased the faster the dose was delivered. Consequently, there was strong support for scaling down the risks determined from the bomb data, when it came to predicting cancer from chronic, protracted exposures. These experiments did not involve human cancer, in which development is a more complex process than can be inferred from experiments on isolated cells. Furthermore, the dose response measured in these experiments was for the total cellular damage, not the damage restricted to those components that affect cancer development, which could have their own, distinct dose response.11 Cardis's critics do not accept that the Canada results simply reflect the highest data point among a set with wide variance; they have spent extensive energy focusing on the weaknesses of only this study, while overlooking or ignoring weaknesses in other studies with lower ERR/Sv results (CitationAshmore et al., 2010; CitationBoice, 2010). If Canada, the study with the highest ERR/Sv, is removed from the study, as the critics wish, the overall results will no longer be statistically significant. Blind reliance on 95 percent confidence limits is no longer the practice in epidemiology, but it remains a widespread practice among stakeholders—when they will benefit from discounting data. Studies are assessed as a group, for instance, in the context of the other studies discussed in this article.12 The findings were not greatly influenced by data from any one country: Formal statistical tests provided no evidence for differences in risk between countries. Thus, there was no statistical basis for removing the highest data point. Analyses excluding one country at a time produced excess relative risks per Sv ranging from 0.58 (excluding Canada) to 1.25 (excluding the United Kingdom). Only by excluding Canada did the results lose statistical significance.13 To date, none of the epidemiology studies dealing with protracted exposures have explicitly accounted for the effect of uncertainty in dose estimation. Paying attention to uncertainty in dose estimation is perhaps most important with the Techa River data and some of the studies of background radiation discussed later in the text, because mathematical models are used in part to reconstruct individual exposures, filling in gaps in direct measurements. Accounting for measurement uncertainty strengthened the findings of an association with radiation in studies of thyroid disease following nuclear weapons tests (CitationLyon et al., 2006).14 Two epidemiologic studies, one in India (CitationNair et al., 2009) and one in China (CitationTao et al., 2012), are notable because they include individualized doses, which typically would mean they rely on internal controls by dose category, like the analysis of the atomic-bomb survivors. However, the individualized dose method is weakened in these studies, because the authors mixed together distinct geographic regions, some with low doses and some with high. No separate analyses were given in the studies to account for different baseline cancer rates in each region; nor was there stratification by region. Had there been, it is likely that the confidence intervals would have increased. As it was, the confidence intervals around their slightly negative slopes were quite wide, limiting their usefulness in potentially contradicting other studies, as EPRI maintains they do. Early markers of cancer risk, namely dose-related chromosomal aberrations (China) and mitochondrial DNA mutations (India) have been found in these high-background regions, casting some doubt on the null epidemiologic findings in these studies. Both studies exclude persons under 30, the age period when radiation-induced leukemia is extremely high according to atomic-bomb results. In the Chinese mortality study the authors note the difficulty in diagnosing liver cancer mortality. When liver cancer is removed from the analysis, the ERR/Sv is positive and the upper confidence interval for the ERR/Sv is much greater than that for the other studies mentioned in this article, including the 15-country study and the Techa River cohort study. Thus, from this perspective, the study of high background radiation in China is not at all inconsistent with studies showing risks of protracted exposure greater than predicted by analysis of the atomic-bomb data. The study in India did not provide an analysis with liver cancer removed, so a comparison cannot be made. For a more positive view of these types of studies, see CitationBoice et al. (2010). For a discussion of other studies in high-background-radiation regions, see CitationHendry et al. (2009).15 The weights here correspond to different times spent near the riverbank (CitationStandring et al., 2009).16 To obtain this curve, no model was assumed, which means the fit is called nonparametric: Instead of fitting to a model (e.g., a linear model), a running average of the data was used to define the curve.17 The next BEIR report is not yet in the planning.18 Because many more cancers are expected in the atomic-bomb cohort in the next 20 years, more information can be expected to emerge on the low-dose range in the future. Given the 50 years it has taken for the atomic-bomb analyses to produce powerful results, we cannot expect Fukushima data to contribute much for a great many decades, particularly because the average dose is likely to be lower. However, the ability to perform genetic sequencing of removed tumors offers a new opportunity to expand the power of epidemiologic studies.19 Presentations to the BEIR committee are open to the public. The biographies of the committee members appear in the report. So, too, are the names of the numerous outside reviewers listed in the final report. Both of these procedures allow outsiders to assess whether or not the committee is balanced.20 It is perhaps out of the same concern to avoid social amplification of risk that the UNSCEAR authors have declined to use the LNT to predict the total number of excess cancers that will result over time as a result of the Chernobyl accident. Such a number would be very small compared with the number of cancers that would have appeared without the accident. One of the biggest paradoxes in the low-level radiation debate is that an individual risk can be a minor concern, while the societal risk (total delayed cancers in an exposed population) can be of major concern.21 Some argue that bystander effects could also be protective, signaling the body to increase repair efforts.22 The use of a pure threshold in the fit was a mathematical convenience; a quasi-threshold would have fit just as well.23 According to the criticism, too many members had strong prior views tied to their relationships with the French nuclear industry or their medical practice in radiation medicine. In contrast to the policy followed in preparation of the BEIR reports, no biographies of the authors were provided in the report of the French Academies, deepening suspicion that something was being hidden.24 Furthermore, the French study did not undergo the intensive peer-review process used with the BEIR committee reports. Of course, the BEIR report process is not perfect, either, as history shows. In its 1980 report, for example, the committee muffed the slope of the LNT (CitationMarshall, 1990). However, the BEIR committees do follow processes established by the US National Academies of Sciences and the US Institute of Medicine, which require assembling a group of scientists with a broad range of perspectives. Attempts are made to balance biased views, which are inevitable in any sizeable committee of active scientists or medical professionals.Additional informationAuthor biographyGuest editor of the Bulletin's special issue on low-dose radiation risks, Jan Beyea is a nuclear physicist, who, for 40 years, has listened to, and sometimes participated in, the debate over risks from low-level radiation exposure. He is a co-author of papers on environmental epidemiology, including a study of the Three Mile Island accident. Beyea has been a member and reviewer of numerous studies for the National Research Council, most recently serving as a committee member for the completed study America's Energy Future, and the ongoing study, The Feasibility of Inertial Fusion Energy. He currently is a member of the World Trade Center Health Registry Scientific Advisory Committee. He advises plaintiff law firms on litigation strategy in large, toxic tort cases, where he has watched some of the most prominent researchers in the world go head to head, debating the risks of low-level radiation. He is chief scientist at Consulting in the Public Interest.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.