Lessons from Hiroshima and Nagasaki: The most exposed and most vulnerable

Abstract

Because of its scope, the Life Span Study of Japanese atomic-bomb survivors has become highly influential and widely accepted by regulators, policy makers, and courts of law. The usefulness of the study in regard to calculating excess cancer risk, however, should not disguise its limitations. The Life Span Study examines survivors, but not those who died in the years immediately following the Hiroshima and Nagasaki bombings. The study design therefore intentionally omitted frail people—including the very young and the very old—who may have been especially vulnerable to radiation. Regulators and risk assessors need to be aware of this shortcoming and to supplement Life Span Study results with information from studies that focus on those most vulnerable to radiation. A failure to do so may result in an underestimate of the harm caused by radiation, particularly at low doses.

In August 1945, US aircraft dropped atomic bombs on Hiroshima and Nagasaki, Japan, with profound effect. Reported numbers vary, but estimates show that 90,000 to 120,000 civilians in Hiroshima (out of a population of about 330,000) and 60,000 to 80,000 in Nagasaki (out of 280,000) died by the end of that year, after succumbing to burns, marrow depletion, and other consequences of the heat, blast, and radiation emitted by the bombs (Committee for the Compilation of Materials on Damage Caused by the Atomic Bombs in Hiroshima and Nagasaki, 1981; Douple et al., 2011).

A decade after the attacks, a critical review of the Atomic Bomb Casualty Commission’s investigations into the health effects of the bombings by an ad hoc committee of the National Research Council of the National Academy of Sciences led to the recommendation that investigators establish fixed, well-defined groups of exposed and non-exposed people and follow them over time for a study of mortality and disease risk as they relate to survivors’ distance from the hypocenters of the bombings (Francis et al., 1955). The resultant study, known as the Life Span Study of Japanese atomic-bomb survivors, has become highly influential and widely accepted by regulators, policy makers, and courts of law.

Japan’s October 1950 national census provided a list of people who reported being in Hiroshima or Nagasaki at the times of bombings and who were still living five years later. From this list, the investigators identified a large proportion of atomic-bomb survivors who were within 2.5 kilometers of ground zero when the bombs fell, together with a similar-sized sample of age- and sex-matched people who were between 3 and 10 kilometers from the hypocenters. Information on vital status and cause of death was initially gathered retrospectively—that is, back to the date of the October 1950 census—but since the 1960s, the data have been collected through triennial searches of death records. A tremendous effort was undertaken to estimate bomb survivors’ doses from external exposure to gamma radiation and neutrons. (Doses from fallout and other sources of residual or induced radiation have not been quantified.) For survivors within 2 kilometers of the hypocenters, estimates of penetrating radiation emitted by the bombs were derived using self-reported information on location and shielding that was collected by interviews during the late 1950s and early 1960s. Dose estimates for other survivors are derived without the detailed shielding history information obtained for proximal survivors.

The Life Span Study is appealing to regulators and risk assessors in part because of its scope: It is a large study that includes data on men, women, and children exposed to ionizing radiation, including some with very high estimated radiation doses. Among the study’s notable findings are a peak and subsequent decline in the relative risk of leukemia mortality, a relatively rare disease in the general population that appeared in excess promptly after the bombings, particularly among those who were children. The solid cancers, less rare than leukemia, also occurred at elevated rates and over time have contributed much more to the total number of excess cancer deaths. The study shows that the relative rate of solid cancer appears to increase proportionally with the dose of radiation, in a quite linear fashion. And the striking finding is that the increased risk of cancer among atomic-bomb survivors has persisted for six decades.

But the Life Span Study looks at the atomic-bomb survivors’ health effects starting in 1950, five years after the attacks occurred, with exposure information derived from field interviews conducted under the direction of an occupying military authority. The usefulness of the study in regard to calculating excess cancer risk should not disguise its limitations.

In addition to several factors that affect the internal validity of the Life Span Study, it is, by design, incomplete. It did not investigate the Hiroshima and Nagasaki residents who died in the years immediately after the atomic bombings. The design, therefore, intentionally omitted frail people—including the very young and very old—who may have been especially vulnerable to radiation. To improve current-day estimates of cancer rates resulting from radiation, regulators and risk assessors need to be aware of this shortcoming in the study and to supplement its results with information from studies that include—or even focus on—those most vulnerable to radiation. A failure to do so may well result in an underestimate of the harm caused by radiation, particularly at low doses.

The epidemiological challenge

There is a great deal of interest in the health effects of low-level exposures to ionizing radiation, as occurs in some environmental and occupational settings. Yet epidemiologists face a tremendous challenge when they are asked about the number of excess cancers expected in a population as a result of an increase in radiation exposure (for example, the release of radioactive elements during the recent disaster at the Fukushima Daiichi Nuclear Power Station). Since the long-term effects of the radiological releases have yet to occur, epidemiologists can’t empirically investigate the question. Rather, they only can make an estimate of the possible consequences of the exposures; they do this by using a statistical model of radiation-related excess cancer risks derived from a study of a different group of people who were exposed at an earlier time (known as the reference population). The epidemiologist relies upon analogy, in the form of a statistical model. In some situations, the analogy works well. But there are at least two ways that this approach may lead to erroneous conclusions.

First, the insights drawn from a reference population may provide a distorted view of the true association in the reference population. In other words, the statistical model for the radiation dose-to-cancer association in the reference population may be biased. Applying a biased estimate derived from one population to another population offers little prospect for obtaining a correct answer. Such bias might result from, for instance, mismeasurement of radiation exposures or from imbalances between the exposed and unexposed in the distribution of other factors that cause cancer (an obvious example being cigarette smoking). ¹

Second, the analogy may simply be inappropriate. The population that we wish to draw inferences about may not be comparable to the reference population. Differences of place and time, of constitution, difference in exposures to other hazards, or in the prevalence of chronic diseases may work against a valid transporting of radiation risk estimates from the reference population to the population of interest.

To properly estimate the excess cancer that a radiation-exposed population may experience requires confidence both in the internal validity of the risk coefficients derived from a reference population, and in the ability to generalize from a study of the reference population. A small difference between the radiation risk coefficients derived for a reference population and the “true” coefficient for the population of interest may lead to relatively large distortions in estimates of the number of excess cancers.

The internal validity of the Life Span Study

Early analyses of the Life Span Study acknowledged the uncertainty of determining the location and shielding of the survivors. The study relied on crude proxies for radiation doses received: self-reported distance from hypocenter; major categories of shielding (such as location inside a building versus outdoors); and the presence of acute radiation effects (such as hair loss and skin reddening). Over time, however, more sophisticated dosimetry systems have been developed to derive estimates for many—but not all—survivors of the bombings. Investigators have also used statistical methods to correct for some apparent errors in exposure measurement.

Beyond problems in dose estimation, early investigators also realized that an observational study of Japanese atomic-bomb survivors was susceptible to bias that arose from the confounding of the radiation effect by other factors—such as occupation—that were related to both location at the time of the bombing and subsequent cancer risk. Occupation and other socioeconomic factors have not been accounted for in most major analyses of mortality among the Life Span Study cohort members.

Although early investigators noted the above problems, another issue is less often discussed: a factor such as cigarette smoking, which was not associated with estimated radiation dose at the time of bombing, could still lead to bias in estimates of the association between estimated dose and mortality in an analysis restricted to five-year survivors of the bombing. ²

Distinct from such problems of bias, are problems of inference regarding radiation effects. The experience of being close to ground zero may have been so profound as to have effects that influenced later life experiences—e.g., marriage, health-related behaviors, use of medical services—and in this way also influenced mortality. The Life Span Study is often taken as a study of the effects of ionizing radiation on mortality, and yet the consequences of the atomic bombings were not purely radiological. The radiological and non-radiological effects of proximity to ground zero have yet to be untangled. In other words, the effects of proximity to ground zero on health may extend beyond the direct carcinogenic effects of the ionizing radiation from the atomic bombings. Such effects are plausible, and they pose another challenge to those who attempt to apply the dose-response findings on bomb survivors to populations exposed to radiation in other settings.

Can the Life Span Study be generalized?

One-third of the population of Hiroshima and one-quarter of the population of Nagasaki died relatively promptly after the atomic bombings. These people are not part of the Life Span Study, and this omission may introduce problems in generalizing the study’s findings to other groups of people exposed to radiation. An example of possible bias: Suppose that the population of those exposed to the Hiroshima and Nagasaki bombings included some people who had a strong physical constitution and some people who were relatively frail. If frailty increased one’s susceptibility to both the early and late effects of the atomic bomb, and the study included the entire population, then an overall estimate of the radiation–mortality association would represent a weighted average of radiation risks for those who were frail and those who were not. But this is not what occurred in the study, which excludes those who succumbed to the early effects of the bomb (meaning they died between August 1945 and October 1950). By so doing, the study includes a smaller proportion of the frail people than likely existed in the source population, since those who were frail disproportionately died in the early years. This omission could lead to artificially low estimates of radiation cancer risk if generalized to a population that was exposed to radiation but was not subject to the same selective survival criterion.

University of Birmingham Medical School Professor Alice Stewart has cautioned that the study only focused on exceptional people who survived the bombings; if constitutional factors that led to their survival during the early period, from 1945 to 1950, also implied lower susceptibility to later carcinogenic effects of ionizing radiation, then inferences regarding cancer risks in other populations might be underestimated (Stewart and Kneale, 1990). She further noted that selective survival likely differed with age, since the very young and the very old would have been susceptible to many early causes of death (Stewart, 1997). Donald Pierce of the University of Oregon and colleagues have suggested that it is unlikely the radiation–mortality association was reduced by more than 20 percent (Pierce et al., 2007). It has long been known that conditioning on variables affected by the study exposure can create problems of generalizability.

Accounting for and protecting the vulnerable

More than six decades after the bombings of Hiroshima and Nagasaki, the Life Span Study continues to provide valuable evidence and yield important new insights on the health effects of radiation exposure. It is a study of remarkable survivors of a horrific event. Conditioning on long-term survival of an atomic bombing may have a variety of effects on risk estimates, and consequently a study of atomic-bomb survivors may offer a distorted picture of the risks to be expected from radiation exposures occurring in other situations.

Questions about radiation effects on those most vulnerable to early mortality following the bombings are particularly difficult to address with the Life Span Study. A striking example of this comes from considering those who were exposed to ionizing radiation in utero. There is no appreciable excess risk of cancer among children known to have been irradiated in utero among the atomic-bomb survivors, although such estimates are necessarily imprecise (given the small number of children irradiated in utero counted among the survivors) (Doll and Wakeford, 1997). Yet when one looks to epidemiological evidence outside of the atomic-bomb survivor studies, one finds a literature that indicates a notable excess risk of cancer among those children exposed in utero to ionizing radiation (El Ghissassi et al., 2009). The largest and most informative of these studies, the Oxford Survey of Childhood Cancer, suggests that in utero exposure to a dose on the order of 10 millisieverts leads to an excess relative risk of childhood cancer of about 50 percent (Bithell, 1993).

Regulators and policy makers would be well-served by supplementing the information obtained from the Life Span Study with data from the Oxford Survey and other research into the effects of radiation on vulnerable populations. By design, the study of Japanese survivors fails to offer information regarding cancer risks in the years immediately after the bombings, and it fails to look at people other than those robust enough to survive the attacks; these aspects of the study have implications both for its internal validity and the generalizability of its results. But all is not lost. Indeed, there is a good starting point to address these limitations: that is, to conscientiously incorporate radiation risks—and to ensure adequate standards for radiation protection—for the people who may be most susceptible to the effects of radiation exposure, including the very young and the very old.