A-bomb data: detection of bias in the Life Span Study cohort.

By drawing a distinction between A-bomb survivors with and without bomb-related injuries, it was possible to see that instead of the Life Span Study (LSS) cohort being a normal, homogenous population, there were significant differences between survivors with and without multiple injuries, and that these differences occurred largely among survivors who were under 10 or over 50 years of age when exposed. There also was a concentration of A-bomb-related injuries among survivors who eventually developed leukemia. So it is possible that deaths before 1950 had left the LSS cohort permanently biased in favor of persons who had high levels of resistance to all (early and late) effects of radiation. It is also possible that the high proportion of leukemia cases among the deaths of A-bomb survivors from 1950 to 1970 were because the radiation caused an initial leukocytosis followed by loss of immunologic competence.


Introduction
In spite of the huge population losses sustained by Hiroshima and Nagasaki, Japan, between August 1945 and October 1950, the noncancer death rate of the Life Span Study (LSS) cohort remained close to expectations based on national statistics and (unlike the cancer death rate) did not exhibit evidence of a linear trend with dose (1). As a result of these observations "the use of A-bomb data for risk assessment is generally predicated on the assumption that the survivors, apart from their radiation dose, are representative human beings" (2).
According to this hypothesis, cancer was the only late effect of the A-bomb radiation and neither division of the exposed population into deaths before and after 1950 nor division of the LSS cohort into survivors with and without bomb-related injuries would have affected levels of sensitivity to this late effect of the radiation (current hypothesis, Figure 1). There are, however, several analyses of LSS data by Stewart and Kneale (3)(4)(5) that are difficult to reconcile either with the assumption of no selection effects of the early deaths or with that of no late effects of the radiation apart from cancer.
Though lacking full access to all the records assembled by the Radiation Effects Research Foundation (RERF), Stewart and Kneale have a diskette version of LSS data (RERF, Hiroshima, Japan) and observed that, first, for all causes of noncancer deaths except cardiovascular accidents there is a biphasic dose-response curve whose lowest point is close to 1 Gy (3). Second, the proportion of high-dose (over 1 Gy) survivors is much smaller for the youngest and oldest of five exposure age groups (under age 10 or over age 50) than for intervening age groups (4). Third, for survivors with two or more bomb-related injuries, the dose-response curve for leukemia and other neoplasms is exceptionally steep and this finding is true primarily for those exposed before age 10 or after age 50 (5).
On the strength of these observations Stewart and Kneale came to two conclusions. First, compared with other A-bomb victims, persons who died before 1950 as well as survivors with bomb-related injuries were exceptionally sensitive to all (early and late) effects of radiation (alternative hypothesis, Figure 1). Second, the relatively high levels of sensitivity to cancer effects of radiation regularly observed by the RERF among persons who were under 30 years of age  I I III when exposed (6) probably were the result of older persons succumbing more often to acute effects of high doses of radiation.
If these conclusions are true, it may one day be necessary to replace the current hypothesis depicted in Figure 1 with the alternative hypothesis. Such a change would have important implications for risk assessment. The latest analysis of LSS data by Stewart and Kneale (5) is briefly summarized below with regard to the 5965 deaths from neoplasms during the 1950 to 1985 period.

Materials and Methods
Stewart and Kneale began with the same sample of LSS data as the Biological Effects of Ionizing Radiation V committee (7); they also used the same tests (Poisson regression analysis) of dose-related effects for cancer and other causes of death. Instead of assuming that the cohort of 75,991 survivors was a single homogeneous population, however, they restricted their analysis to 74,042 survivors who had records of bomb-related injuries (Tables 1  and 2) and observed the effects of treating these survivors by considering the survivors either as a single cohort or as a mixture of two or three distinct cohorts (Table 3). Stewart and Kneale also followed tests of dose-related effects of the radiation (Table  4) by chi-square tests of exposure age effects and cohort homogeneity ( Table 5).

Results
Regression analysis of the six cohorts listed in Table 3 yielded both evidence of a dose-related cancer risk and evidence that this risk was a) influenced by age when exposed to the radiation, and b) appreciably smaller for the 64,758 survivors who had no bomb-related injuries than for the 2601 survivors with multiple injuries. Because in this respect the small group had more in common with the nonsurvivors than the large group, it is possible that the usual source of risk estimates for cancer effects of radiation (i.e., the LSS cohort of A-bomb survivors) is biased in favor of persons who had exceptionally high levels of resistance to late as well as early effects of radiation (alternative hypothesis, Figure 1).

Discussion
By showing that the normal noncancer death rate of the LSS cohort was associated with two dose-related factors, Stewart and Kneale's analysis (3) identified both a reason why the usual effect of an excessively high death rate (i.e., a reduced death rate caused by survival of the fittest) was not a feature of LSS data, and a reason why the death rate for blood diseases other than leukemia remained both higher than normal and strongly dose related long after 1950 (residual effect of the marrow damage that caused thousands of deaths from aplastic anemia before 1950). Then came the 1993 analysis of RERF publications as well as LSS data (4) with evidence that there was a conspicuous shortage of first trimester exposures in the cohort formed from live births between August 1945 and June 1946 (the in utero cohort). The analysis also established that there was a shortage of high doses (over 1 Gy) in the LSS cohort among persons who were under 10 or over 50 years of age when exposed. Four years later these findings were followed by an analysis of bomb-related injuries (5) that by showing that one requirement of the alternative hypothesis in Figure 1 could be met suggested that deaths before 1950 left the   survivors biased in favor of persons who were exceptionally resistant to all causes of death, including radiogenic cancers. These analyses, therefore, suggest the following conclusions: * Although some members of the LSS cohort may still be experiencing infection sensitivity effects of marrow damage, others may outlive their nonexposed contemporaries (healthy survivor effect of deaths before 1950). * The exceptionally high frequency of bomb-related injuries among persons who eventually developed leukemia ( Table 2) possibly was the result of the radiation initially causing both a leukocytosis and loss of immunologic competence, then these early changes leading to extra deaths from cases of myeloid leukemia with exceptionally short latent periods (8). Under this assumption the special leukemogenic effect observed in A-bomb survivors and radiotherapy patients (9) would have no counterpart in the exclusively low-dose situations  Table 3 for explanation of cohort subgroups. Equivalents of subgroup A of LSS cohort include: subgroups B + C, D + E, or B + D + F. *p < 0.05.
resulting from either background radiation or occupational exposures to gamma radiation (10). Whether or not it becomes necessary to replace the current hypothesis with the alternative hypothesis (Figure 1), it would be interesting to observe the effects of adding to the data collected by the RERF other records of A-bomb survivors (from their special hospitals and clinics) and using the pooled data to study factors associated with different levels of sensitivity to the carcinogenic effects of radiation.