6
Atomic Bomb Survivor Studies

INTRODUCTION

The Life Span Study (LSS) cohort consists of about 120,000 survivors of the atomic bombings in Hiroshima and Nagasaki, Japan, in 1945 who have been studied by the Radiation Effects Research Foundation (RERF) and its predecessor, the Atomic Bomb Casualty Commission. The cohort includes both a large proportion of survivors who were within 2.5 km of the hypocenters at the time of the bombings and a similar-sized sample of survivors who were between 3 and 10 km from the hypocenters and whose radiation doses were negligible. The LSS cohort has several features that make it uniquely important as a source of data for developing quantitative estimates of risk from exposure to ionizing radiation. The population is large, not selected because of disease or occupation, has a long follow-up period (1950–2000), and includes both sexes and all ages at exposure, allowing a direct comparison of risks by these factors.

Doses are reasonably well characterized and cover a useful range. Doses are lower than those usually involved in medical therapeutic exposures, but many survivors were exposed at doses that are sufficiently large to estimate risks with reasonable statistical precision. In addition, the cohort includes a large number of survivors exposed at low doses, allowing some direct assessment of effects at these levels. The exposure is a whole-body exposure, which makes it possible to assess risks for specific cancer sites and to compare risks among sites. Because of the use of the Japanese family registration system, mortality data are virtually complete for survivors who remained in Japan. High-quality tumor registries in both Hiroshima and Nagasaki allow the study of site-specific cancer incidence with reasonably reliable diagnostic data. In addition, the LSS cohort is probably less subject to potential bias from confounding than many other exposed cohorts because a primary determinant of dose is distance from the hypocenter, with a steep gradient of dose as a function of distance. Finally, special studies involving subgroups of the LSS have provided clinical data, biological measurements, and information on potential confounders or effect modifiers.

The LSS also has limitations, which are important to consider in using and interpreting results based on this cohort. The subjects were Japanese and exposed under wartime conditions and, in this sense, differ from various populations for which risk estimates are desired. To be included in the study, subjects had to survive the initial effects of the bombings, including the acute effects of radiation exposure, and it is possible that this might have biased the findings. Dose estimates are subject to uncertainty, especially that due to survivor location and shielding. The cohort provides no information on dose-rate effects since all exposure is at high dose rates. Estimates of linear risk coefficients tend to be driven by doses that exceed 0.5 Gy; although estimates based only on survivors with lower doses can be made, their statistical uncertainty is considerably greater than those that include survivors with higher doses. Even at higher doses, data are often inadequate for evaluating risks of cancers at specific sites, especially those that are not common (although, for many site-specific cancers, the LSS provides more information than any other study).

Because of its many advantages, the LSS cohort of A-bomb survivors serves as the single most important source of data for evaluating risks of low-linear energy transfer radiation at low and moderate doses. This chapter describes the LSS cohort and presents findings for leukemia and for solid cancers as a group. The most recent major publications on cancer mortality (Preston and others 2003) and incidence (Preston and others 1994; Thompson and others 1994) are emphasized, but papers addressing special issues such as the shape of the dose-response function are also considered. Results for cancers of specific sites, including results from the three publications just noted, are discussed along with material from various special studies. Risks from in utero exposure are discussed separately. Although cancer is the main late effect that has been demonstrated in the A-bomb survivor studies, several studies have addressed the effects



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 141
6 Atomic Bomb Survivor Studies INTRODUCTION ments, and information on potential confounders or effect modifiers. The Life Span Study (LSS) cohort consists of about The LSS also has limitations, which are important to con- 120,000 survivors of the atomic bombings in Hiroshima and sider in using and interpreting results based on this cohort. Nagasaki, Japan, in 1945 who have been studied by the Ra- The subjects were Japanese and exposed under wartime con- diation Effects Research Foundation (RERF) and its prede- ditions and, in this sense, differ from various populations for cessor, the Atomic Bomb Casualty Commission. The cohort which risk estimates are desired. To be included in the study, includes both a large proportion of survivors who were subjects had to survive the initial effects of the bombings, within 2.5 km of the hypocenters at the time of the bombings including the acute effects of radiation exposure, and it is and a similar-sized sample of survivors who were between 3 possible that this might have biased the findings. Dose esti- and 10 km from the hypocenters and whose radiation doses mates are subject to uncertainty, especially that due to survi- were negligible. The LSS cohort has several features that vor location and shielding. The cohort provides no informa- make it uniquely important as a source of data for develop- tion on dose-rate effects since all exposure is at high dose ing quantitative estimates of risk from exposure to ionizing rates. Estimates of linear risk coefficients tend to be driven radiation. The population is large, not selected because of by doses that exceed 0.5 Gy; although estimates based only disease or occupation, has a long follow-up period (1950– on survivors with lower doses can be made, their statistical 2000), and includes both sexes and all ages at exposure, uncertainty is considerably greater than those that include allowing a direct comparison of risks by these factors. survivors with higher doses. Even at higher doses, data are Doses are reasonably well characterized and cover a use- often inadequate for evaluating risks of cancers at specific ful range. Doses are lower than those usually involved in sites, especially those that are not common (although, for medical therapeutic exposures, but many survivors were ex- many site-specific cancers, the LSS provides more informa- posed at doses that are sufficiently large to estimate risks tion than any other study). with reasonable statistical precision. In addition, the cohort Because of its many advantages, the LSS cohort of A- includes a large number of survivors exposed at low doses, bomb survivors serves as the single most important source allowing some direct assessment of effects at these levels. of data for evaluating risks of low-linear energy transfer The exposure is a whole-body exposure, which makes it pos- radiation at low and moderate doses. This chapter describes sible to assess risks for specific cancer sites and to compare the LSS cohort and presents findings for leukemia and for risks among sites. Because of the use of the Japanese family solid cancers as a group. The most recent major publications registration system, mortality data are virtually complete for on cancer mortality (Preston and others 2003) and incidence survivors who remained in Japan. High-quality tumor regis- (Preston and others 1994; Thompson and others 1994) are tries in both Hiroshima and Nagasaki allow the study of site- emphasized, but papers addressing special issues such as the specific cancer incidence with reasonably reliable diagnos- shape of the dose-response function are also considered. tic data. In addition, the LSS cohort is probably less subject Results for cancers of specific sites, including results from to potential bias from confounding than many other exposed the three publications just noted, are discussed along with cohorts because a primary determinant of dose is distance material from various special studies. Risks from in utero from the hypocenter, with a steep gradient of dose as a func- exposure are discussed separately. Although cancer is the tion of distance. Finally, special studies involving subgroups main late effect that has been demonstrated in the A-bomb of the LSS have provided clinical data, biological measure- survivor studies, several studies have addressed the effects 141

OCR for page 141
142 BEIR VII of radiation exposure on other health outcomes including they are not available before 1958 and do not include sub- benign tumors and mortality from causes of death other than jects who have migrated from Hiroshima or Nagasaki.1 cancer. These are discussed at the end of the chapter. In gen- The Adult Health Study (AHS) is a resource for data on eral, the committee has summarized papers on cancer inci- health end points that require clinical data. The AHS cohort dence, cancer mortality, and noncancer mortality in the LSS is a 20% subsample of the LSS, oversampled to provide cohort that have been published since BEIR V (NRC 1990). greater representation of subjects in high-dose categories. This chapter is based on published material and does not Since 1958, AHS subjects have been invited to participate in include results of analyses conducted by the committee, biennial comprehensive health examinations at RERF. The which are described in Chapter 12. At the time of this writ- level of participation has been between 70 and 85% for those ing, detailed analyses of mortality data covering the period living in the Hiroshima and Nagasaki areas (Ron and others 1950–1997 and of incidence data covering the period 1958– 1995a). 1987 had been published. The committee’s analyses were based on the most recent DS02 dosimetry system, whereas Dosimetry most of the published analyses described in this chapter were based on the earlier DS86 dosimetry system (see discussion Most results presented in this chapter were based on the of dosimetry below for further comment). Preston and col- dosimetry system adopted in 1986 (DS86). The committee’s leagues (2004) recently evaluated the impact of changes in analyses, described in Chapter 12, are based on the revised dosimetry on cancer mortality risk estimates using mortality DS02 system, adopted in 2004. The DS02 system is the re- data through 2000; these results are summarized in the dis- sult of a major international effort to reassess and improve cussion of dosimetry. survivor dose estimates. This effort was initiated because reports in the early 1990s on thermal neutron activation mea- sured in exposed material (e.g., Straume and others 1992; DESCRIPTION OF THE COHORT Shizuma and others 1993) were interpreted as suggesting The full LSS cohort consists of approximately 120,000 that the then-current survivor dosimetry system (DS86) persons who were identified at the time of the 1950 census. might systematically underestimate neutron doses for It includes 93,000 persons who were in Hiroshima or Hiroshima survivors who were more than about 1 km from Nagasaki at the time of the bombings and 27,000 subjects the hypocenter. However, the revised estimates of neutron who were in the cities at the time of the census but not at the dose do not differ greatly from the DS86 estimates. The new time of the bombings. This latter group has been excluded dosimetry system also introduces improved methods for the from most analyses since the early 1970s because of incon- computation of γ-radiation doses and better adjustments for sistencies between their mortality rates and those for the re- the effects of external shielding by factory buildings and lo- mainder of the cohort. cal terrain features. Preston and colleagues (2004) analyzed mortality data on solid cancer and on leukemia using both DS86 and DS02 Health End Point Data dose estimates. They found that both the risk per sievert for Data on health end points are obtained from several solid cancer and the curvilinear dose-response for leukemia sources. Vital status is updated in 3-year cycles through the were decreased by about 10% by the dosimetry revision. legally mandated Japanese family registration system in They also found that parameters quantifying the modifying which deaths, births, marriages, and divorces are routinely effects of gender, age at exposure, attained age, and time recorded. This ensures virtually complete ascertainment of since exposure were changed very little by the revision. death regardless of where individual subjects reside in Ja- Table 6-1, based on Preston and colleagues (2003), shows pan. Death certificates provide data on the cause of death. the distribution of survivors in the LSS cohort by their esti- The Leukemia Registry has served as a resource for leuke- mated DS86 doses to the colon. The dose to the colon is mia and related hematological disease (Brill and others 1962; taken to be the γ-ray absorbed dose to the colon plus the Ichimaru and others 1978). In the 1990s, it became possible neutron absorbed dose to the colon times a weighting factor to link data from both the Hiroshima and the Nagasaki tumor 10. This weighted dose is denoted by d, and its unit sieverts;2 registries to the LSS cohort, which allows the evaluation of such estimates were available for 86,572 survivors. The cancer incidence (Mabuchi and others 1994). An advantage of the registry data, in addition to the inclusion of nonfatal cancers, is that diagnostic information is of higher quality 1Analyses of cancer incidence data have included an adjustment of per- than that based on death certificates. Both tumor registries son-years to account for migration (Sposto and Preston 1992). 2Use of the symbol Sv for the unit of d is an extension of the convention employ active approaches for case ascertainment and provide to use sievert as a special name of the unit joules per kilogram (J/kg) with high-quality data from 1958 onward. Published analyses regard to the effective dose or the equivalent organ doses (i.e., the dose based on these data cover the period 1958–1987 (Thompson quantities that contain the radiation weighting factor recommended by ICRP and others 1994). Limitations of the incidence data are that 1991).

OCR for page 141
ATOMIC BOMB SURVIVOR STUDIES 143 TABLE 6-1 Number of Subjects, Solid Cancer Deaths, and Noncancer Disease Deaths by Radiation Dose DS86 Weighted Colon Dose (Sv)a Total 0 (<0.005) 0.005– 0.1 0.1–0.2 0.2–0.5 0.5–1.0 1.0–2.0 2.0 Number of subjects 86,572 37,458 31,650 5,732 6,332 3,299 1,613 488 Solid cancer deaths (1950–1997) 9,335 3,833 3,277 668 763 438 274 82 Noncancer disease deaths (1950–1997) 31,881 13,832 11,633 2163 2,423 1,161 506 163 aThese categories are defined using the estimated dose to the colon, obtained as the sum of the γ-ray dose to the colon plus 10 times the neutron dose to the colon. SOURCE: Based on data from Preston and others (2003). 37,458 survivors (43%) with doses less than 0.005 Sv were attained age (a), and time since exposure (t). Not all vari- primarily survivors who were located more than 2.5 km from ables are included in all models; in fact, any two of the vari- the hypocenter. Only 2101 (2.4%) had doses exceeding 1 Sv. ables e, t, and a determine the third. Parametric models are Table 6-1 also shows the number of solid cancer deaths and used for the ERR and EAR. The most recent analyses of noncancer disease deaths in the period 1950–1997. solid cancer mortality (Preston and others 2003) have been based on models of the form STATISTICAL METHODS ERR or EAR = ρ(d) βs exp (γe) aη. (6-3) The material in the sections that follow draws heavily on results presented by Thompson and colleagues (1994) and Earlier analyses (Thompson and others 1994; Pierce and oth- Preston and colleagues (1994, 2003). Here, features of the ers 1996) were based primarily on ERR models of the form statistical methods that were used for most analyses in these papers are described. Readers should consult the source pa- ERR = ρ(d) βs exp (γe). (6-4) pers for details. In nearly all cases, analyses were based on Poisson regression using the AMFIT module of the com- The function ρ(d) is usually taken to be a linear or linear- puter software EPICURE (Preston and others 1991). quadratic function of dose, although threshold and categori- Most recent analyses have been based on either excess cal (nonparametric) models have also been evaluated. With relative risk (ERR)3 models, in which the excess risk is ex- the linear function, ρ(d) = βsd, and βs is the excess relative pressed relative to the background risk, or excess absolute risk per sievert (ERR/Sv), which provides a convenient sum- risk (EAR)4 models, in which the excess risk is expressed as mary statistic. The parameters γ and η measure the depen- the difference in the total risk and the background risk. The dence of the ERR/Sv on age at exposure and attained age. age-specific instantaneous risk is given either by Preston and colleagues (2003) and Thompson and col- leagues (1994) used parametric models for the background λ(c,s,a,b) [1 + ERR(s,e,a,t,d)] (6-1) risks. Some past analyses, such as those by Pierce and co- workers (1996) treated the background risk in ERR models or by including a separate parameter for each category defined by city, sex, age at risk, and year. Thompson and colleagues λ(c,s,a,b) + EAR(s,e,a,t,d) (6-2) did not fit EAR models; however, average EARs were esti- mated by dividing the estimated number of excess cancers where λ denotes the background rate at zero dose and de- by the total person-year-Sv. pends on city (c), sex (s), attained age (a), and birth year (b), Analyses of leukemia are based on bone marrow dose; and the excess may depend on sex (s), age at exposure (e), analyses of the combined category of all solid cancers are based on colon dose; and analyses of site-specific cancers are based on specific organ doses. Dose is expressed in 3The ERR is the rate of disease in an exposed population divided by the sieverts and is a weighted dose obtained as the sum of the rate of disease in an unexposed population minus 1.0. dose of γ-radiation and 10 times the neutron dose. This ap- 4The EAR is the rate of disease in an exposed population minus the rate proach is based on the assumption of a constant relative bio- of disease in an unexposed population. logical effectiveness (RBE) of 10 for neutrons. In most

OCR for page 141
144 BEIR VII analyses, the kerma5 doses are truncated at 4 Gy, equivalent and others 1991), which allows meaningful analyses of spe- to truncating organ doses at 3 Gy. Analyses by Preston and cific types of leukemia. Preston and colleagues evaluated colleagues (2003) and by Pierce and colleagues (1996) were patterns of risk by sex, age at exposure, and time since expo- adjusted for random errors in doses using an approach de- sure for four major subtypes of leukemia: acute lymphocytic scribed by Pierce and colleagues (1990) and based on the leukemia (32 cases), acute myelogenous leukemia (103 assumption of a coefficient of variation of 35% for the error cases), chronic myelogenous leukemia (57 cases), and adult in individual dose estimates. This adjustment generally in- T-cell leukemia (39 cases). Dose-response relationships were creases estimated risk coefficients by about 10%. Earlier seen for the first three but not for adult T-cell leukemia. The papers, such as analyses by Thompson and coworkers (1994) estimated numbers of cases in excess of background were and by Preston and coworkers (1994), did not include this 17.1 for acute lymphocytic leukemia, 29.9 for acute myelog- adjustment. enous leukemia, and 25.9 for chronic myelogenous leuke- For analyses based on tumor registry data, adjustments mia. The other major type of leukemia, chronic lymphocytic were necessary to account for migration from the two cities. leukemia, showed no excess, but it is infrequent in Japan. These are described briefly by Thompson and colleagues Results of analyses of all types of leukemia showed de- (1994) and Preston and colleagues (1994) and in more detail pendencies on sex, age at exposure, and time since exposure by Sposto and Preston (1992). similar to those for the mortality data and led to a model similar to that based on mortality data. Preston and col- leagues note that allowing overall modification by sex and Leukemia age at exposure in an EAR model did not significantly im- This section reviews analyses of mortality data for the prove the fit once time since exposure was included in the period 1950–1990 (Pierce and others 1996) and of incidence model, but that these factors significantly modified the time data for the period 1958–1987 (Preston and others 1994). since exposure effects. Specifically, risks for those exposed Leukemia mortality data for the period 1950–2000 were ana- early in life decreased more rapidly than for those exposed lyzed by Preston and colleagues (2004) and used to develop later, and the decrease was less rapid for women than for the committee’s models for estimating leukemia risks; these men. Analyses of specific leukemia types indicated that there analyses are described in Chapter 12. were significant differences in the effects of age at exposure Leukemia was the first cancer to be linked with radiation and sex and in the temporal pattern of risks. The shape of the exposure in A-bomb survivors (Folley and others 1952) and dose-response did not show statistically significant differ- has the highest relative risk of any cancer. Pierce and col- ences among the subtypes. leagues estimated that 78 of 176 (44%) leukemia deaths among survivors with doses exceeding 0.005 Sv were due to ALL SOLID CANCERS radiation exposure. Leukemia risks increased with dose up to about 3 Sv, with evidence of upward curvature; that is, a Analyses of cancers in this category, which excludes leu- linear-quadratic function fitted the data significantly better kemia and other hematopoietic cancers, are useful for pro- than a linear function. With this linear-quadratic function, viding summary information and models based on larger the excess risk per unit of dose at 1 Sv was about three times numbers than are available for cancers of specific sites (dis- that at 0.1 Sv. cussed below). The discussion in this section is based on For those exposed under about age 30, nearly all of the both mortality (Preston and others 2003) and incidence data excess deaths occurred before 1975, but for those exposed at (Thompson and others 1994). Mortality analyses were based older ages, the excess risk appeared to persist throughout the on 9335 solid cancer deaths that occurred during 1950–1997, follow-up period. The temporal trends also differed by sex, whereas incidence analyses included 8613 incidence cases with evidence of a steeper decline in risk for males than for occurring during 1958–1987.6 The incidence data do not females. Both the nonlinear dose-response and the complex include cases of subjects who migrated and were diagnosed patterns by age and time since exposure mean that simple with cancer outside of Hiroshima and Nagasaki; as noted models cannot adequately summarize leukemia risks. above, analyses were adjusted for migration. Preston and colleagues (1994) analyzed data from the leu- Preston and collegues estimate that 8% of the 5502 solid kemia registry. An important recent development in studies cancer deaths among those with doses exceeding 0.005 Sv of leukemia is the reclassification of leukemia cases by new were due to radiation, much lower than the corresponding systems and criteria (Matsuo and others 1988; Tomonaga percentage of 44% for leukemia. This percentage was 5Kinetic energy released in material. A dosimetric quantity, expressed in grays, that equals the kinetic energy transferred to charged particles per unit 6These numbers contrast with 10,127 solid cancer deaths occurring in mass of irradiated medium when indirectly ionizing (uncharged) particles, 1950–2000 and 12,778 incident cases of solid cancer excluding thyroid and such as neutrons, traverse the medium. If all of the kinetic energy is absorbed nonmelanoma skin cancer occurring in 1958–1998, the periods covered by “locally,” the kerma is equal to the absorbed dose. analyses conducted by the committee and described in Chapter 12.

OCR for page 141
ATOMIC BOMB SURVIVOR STUDIES 145 slightly higher for the incidence data, where 11% of 4327 and attained age, but also pay attention to a model in which cancers in the exposed were estimated to result from radia- the ERR varies only with age at exposure since the evidence tion exposure (Thompson and others 1994). For both the for this effect was stronger. mortality and the incidence data, risks of solid cancer in- Similar plots based on the committee’s analyses of cancer creased with dose up to about 3 Sv, with little evidence of incidence data are presented in Figures 12-1 and 12-2. These nonlinearity in the dose-response for doses in the 0–3 Sv data show similar patterns to those for mortality except that range. For mortality data, this is illustrated by Figure 6-1, the evidence for modification of the ERR by attained age taken from Preston and colleagues (2003). Estimates based was stronger with the updated incidence data than with the on only the low-dose portion of the mortality data are similar mortality data. to those based on the range from 0 to 2 Sv. For example, Preston and colleagues (2003) also present lifetime risk there was a statistically significant dose-response (p = .025) estimates for an LSS cohort member exposed to 1 Sv. These based on analyses restricted to the 0–0.125 Sv dose range, estimates were 18–22% for a person exposed at age 10, 9% with the ERR/Sv estimated to be 0.74 (SE = 0.38). This esti- for a person exposed at age 30, and 3% for a person exposed mate did not differ significantly (p > .5) from the estimate of at age 50. These estimates did not differ greatly from those 0.54 Sv–1 (SE = 0.07) based on the 0–2 Sv range (Preston based on earlier mortality data (Pierce and others 1996). and others 2003, Table 4). Figure 6-2 shows plots of the ERR and EAR for solid Additional Analyses Addressing the Shape of the Dose- cancer mortality by age at exposure and attained age. The Response Function ERR for females was about twice that for males, but the EARs were similar for the two sexes since baseline risks for Several additional papers address the shape of the dose- females are about half those for males. Both the ERR and the response function and evidence for risk at the lower end of EAR were found to decrease with increasing age at expo- the dose distribution; these include analyses by Kellerer and sure. The EAR increased with increasing attained age within Nekolla (1997), Little and Muirhead (1997), Hoel and Li age-at-exposure groups, while the ERR decreased with in- (1998), and Pierce and Preston (2000). These analyses take creasing attained age, especially for those exposed in child- advantage of the large number of survivors with lower doses hood. Preston and colleagues emphasize results based on a and investigate the possibility of a threshold, departures from model that allows the ERR to vary with both age at exposure linearity, and the degree to which effects might be overesti- FIGURE 6-1 Solid cancer mortality dose-response function averaged over sex for attained age 70 after exposure at age 30. The solid straight line is the linear slope estimate, the points are dose-category-specific ERR estimates, the dashed curve is a smoothed estimate derived from the points. Dotted curves indicate upper and lower one-standard-error bounds on the smoothed estimate. SOURCE: Reproduced with permis- sion from Preston and others (2003).

OCR for page 141
146 BEIR VII FIGURE 6-2 Primary descriptions of the excess risk of solid cancer mortality. Left panel: fitted sex-averaged ERR estimates using both attained-age-declining (solid black line) and attained-age-constant (dashed lines) forms, for age-at-exposure groups 0–9, 10–19, 20–39, and 40+. ERR estimates for women are about 25% greater, and ERR estimates for men 25% lower, than the values shown. Right panel: fitted EAR estimates for the same dose groups. There is no evidence of significant sex differences in the fitted EAR. SOURCE: Reproduced with permission from Preston and others (2003). mated based on linear extrapolation from high to low doses. the “overestimation factor.” This result might be interpreted The committee discusses the analyses by Pierce and Preston as indicating that the maximum DREF that is reasonably (2000) because these are the only analyses that include up- compatible with the A-bomb survivor data is unlikely to be dated cancer incidence data. greater than 2. In addition, Pierce and Preston (2000) evalu- Pierce and Preston (2000) investigated solid cancer risks ated threshold models in which the risk was zero up to a at low doses using cancer incidence data for 1958–1994, thus given threshold and then increased linearly. They estimated adding 7 years of data to that available in previously pub- the threshold to be 0 Sv with an upper confidence limit of lished incidence data analyses. Because experimental data 0.06 Sv. Evidence of a statistically significant dose-response have indicated that the RBE of neutrons decreases with in- was found in the dose range 0–0.10 Sv. creasing dose, the RBE was assumed to be a function of Pierce and Preston (2000) warn against overinterpretation dose, with a value of 40 at very low doses that decreased to of the minimum dose at which evidence of a significant dose- about 8 when the neutron dose reached 0.02 Gy (where the response is found, indicating that “in the presence of avail- gamma dose was about 2 Gy). Because of evidence that sur- able data, it is neither sound statistical interpretation nor pru- vivors located more than 3000 m from the bombings had dent risk evaluation to take the view that the risk should be higher cancer rates than other survivors estimated to have considered as zero in some low-dose range due to lack of zero doses, these distally located survivors were omitted statistical significance when restricting attention to that from the analyses described below. This exclusion had little range.” They further call attention to the large potential for effect on analyses based on the full dose range, but did affect bias due to confounding in analyses based on low doses, analyses directed specifically at low-dose effects. noting particularly that A-bomb survivor results in the low- In analyses based on the range 0–2 Sv, Pierce and Preston dose range are influenced by whether or not distally located (2000) found little evidence of nonlinearity in the dose-re- survivors are included. sponse except for a small elevation in risk over linearity in the 0.15–0.3 Sv range. They estimated a curvature param- Other Analyses eter θ, defined as the ratio of the quadratic and linear coeffi- cients for gamma dose, and found that the upper 95% confi- The A-bomb survivor data have been combined with data dence limit for θ was 0.75 Gy–1. At this value, the linear from cohorts of persons exposed for medical reasons, prima- coefficient was estimated to be a factor of 1.9 smaller than rily for the purpose of further exploration of the modifying that obtained from a strictly linear model, and the factor 1.9 effects of age at exposure, attained age, and time since expo- (i.e., the dose rate effectiveness factor, DREF) was termed sure (Little and others 1998, 1999a, 1999c; Little, 1999).

OCR for page 141
ATOMIC BOMB SURVIVOR STUDIES 147 Although these analyses provide valuable information on the Preston and colleagues (2003) used common models for comparability of risks and of modifying factors in different expressing risks for cancers at different sites. Specifically, cohorts, the results for the A-bomb survivor cohort itself gen- 15 sites were analyzed with parameters expressing the modi- erally confirm the findings reported earlier in the chapter, fying effects of age at exposure and attained age set equal to and they are not discussed further here. Biologically based those for all solid cancers. Results of these analyses are sum- models have also been applied to the A-bomb survivor data marized in Figure 6-3, which shows the ERR/Sv for expo- (Kai and others 1997; Pierce and Mendelsohn 1999). sure at age 30 and attained age 70. Except for sex-specific cancers, estimates are averaged for the two sexes. Preston and colleagues (2003) note that the variability in this plot is SITE-SPECIFIC CANCERS generally consistent with what would be expected if the true Because the exposure of A-bomb survivors was whole- site-specific ERRs were all equal to that for all solid cancers. body exposure, studies of the LSS cohort afford the opportu- More detailed analyses of the five most common types of nity to compare cancer risks by site. Inferences for site-spe- solid cancer (stomach, colon, liver, lung, and female breast) cific cancers are based on smaller numbers than those for all were conducted. With ERR models, the age-time patterns solid cancers and involve smaller ERRs than leukemia. This were similar for these sites, although the decrease in risk often means that there is considerable uncertainty in quanti- with attained age was more rapid for colon cancer. With EAR fying risk, in evaluating modifying factors, and even in models, statistically significant departures from the solid determining whether or not there is a dose-response relation- cancer temporal model were found for lung cancer, which ship. Although it is likely that radiosensitivity varies across increased more rapidly with attained age than other solid sites, it is often not possible to separate true differences from cancers, and breast cancer, which decreased more rapidly chance fluctuations. Cancers at some sites may fail to exhibit with age at exposure than other solid cancers. associations because of small numbers of cases and diagnostic Data from the Hiroshima and Nagasaki tumor registries misclassification, which is more problematic for mortality are preferable to mortality data for evaluating site-specific data than for incidence data. risks. These data have the major advantages of including FIGURE 6-3 Estimates of the site-specific solid cancer mortality ERR with 90% confidence intervals and one-sided p-values for testing the hypothesis of no dose-response. Except for sex-specific cancers (breast, ovary, uterus, and prostate) the estimates are averaged over sex. All estimates and p-values are based on a model in which the age-at-exposure and attained-age effects were fixed at the estimates for all solid cancers as a group. The light dotted vertical line at 0.0 corresponds to no excess risk; the dark solid vertical line indicates the sex-averaged risk for all solid cancers. SOURCE: Reproduced with permission from Preston and others (2003).

OCR for page 141
148 BEIR VII nonfatal cancers and of more accurate diagnostic informa- In addition to analyses by Thompson and colleagues tion with data on histological types of cancer. Results based (1994), several papers provide further analyses that, in some on analyses by the committee of updated incidence data cases, give more attention to histological type and, in other (1958–1997) are discussed in Chapter 12. cases, are based on case-control studies that include data on Thompson and colleagues evaluated cancer incidence possible modifying factors that were not available for the data from 1958 to 1987 for the cancer sites shown in Fig- full cohort. These results are summarized below for selected ure 6-4 and Table 6-2. For each site, they evaluated whether cancer sites. there was a significant association with dose, whether there were departures from linearity, and whether risks were modi- Female Breast Cancer fied by city, sex, age at exposure, attained age, or time since exposure. In a case-control interview study nested within the LSS Of the cancer sites shown in Figure 6-4 and Table 6-2, cohort and including cases occurring in 1950–1985, Land the largest ERR/Sv was for breast cancer. Relatively large and colleagues (1994b) investigated known risk factors for values were also seen for nonmelanoma skin cancer and for breast cancer: age at the time of a first full-term pregnancy, cancers of the ovary, urinary bladder, and thyroid. In addi- number of children, and cumulative period of breast-feed- tion to these sites, the 95% confidence intervals excluded ing. The influence of these factors on breast cancer risks in zero for cancers of the stomach, colon, liver, and lung. It women in the LSS cohort was similar to that found in other should be noted that the size of the ERR/Sv may be affected studies. The relationship of these factors and radiation expo- by the size of the baseline risk. These ERRs/Sv were ob- sure was reasonably well described by a multiplicative model tained from a model with no modifying factors and are not (in which known risk factors for breast cancer do not modify strictly comparable to those based on mortality data and the ERR/Sv), whereas an additive model could be rejected. shown in Figure 6-3, which included modifying factors and Preston and colleagues (2002a) conducted pooled analy- were intended to be applicable to a person exposed at age 30 ses of breast cancer incidence in eight cohorts. Analyses from at attained age 70. this paper based on the LSS cohort alone that included cases FIGURE 6-4 Excess relative risk at 1.0 Sv (RBE 10) for solid cancer incidence and 95% confidence interval, 1958–1987. SOURCE: Reproduced with permission from Thompson and others (1994).

OCR for page 141
ATOMIC BOMB SURVIVOR STUDIES 149 TABLE 6-2 Summary of Risk Estimates for Solid Cancer Incidence by Cancer Site or Organ System Cancer Site or Organ System Percentage of Total Casesa ERR1Sv EAR per 10,000 PY-Sv AR,b % Total solid tumors 100.0 0.63 (0.52, 0.74)c 29.7 (24.7, 34.8) 11.6 (10.2, 14.3) Oral cavity and pharynx 1.5 0.29 (–0.09, 0.93) 0.23 (–0.08, 0.65) 9.1 (–3.0, 25.9) Digestive system 55.7 0.38 (0.25, 0.52) 10.4 (7.0, 14.0) 7.8 (5.3, 10.6) Esophagus 2.1 0.28 (–0.21, 1.0) 0.30 (–0.23, 1.0) 6.5 (–5.0, 22.5) Stomach 30.9 0.32 (0.16, 0.50) 4.8 (2.5, 7.4) 6.5 (3.5, 10.5) Colon 5.3 0.72 (0.29, 1.3) 1.8 (0.74, 3.0) 14.2 (5.9, 23.9) Rectum 4.1 0.21 (–0.17, 0.75) 0.43 (–0.35, 1.5) 4.4 (–3.6, 14.6) Liver 6.8 0.49 (0.16, 0.92) 1.6 (0.54, 2.9) 10.9 (3.6, 19.4) Gallbladder 3.4 0.12 (–0.27, 0.72) 0.18 (–0.41, 1.1) 2.2 (–5.1, 13.1) Pancreas 2.8 0.18 (–0.25, 0.82) 0.24 (–0.36, 1.1) 3.5 (–5.2, 15.3) Respiratory system 11.9 0.80 (0.50, 1.2) 4.4 (2.9, 6.1) 16.3 (10.6, 22.6) Trachea, bronchus, and lung 10.1 0.95 (0.60, 1.4) 4.4 (2.9, 6.0) 18.9 (12.5, 26.0) Nonmelanoma skin 2.0 1.0 (0.41, 1.9) 0.84 (0.40, 1.4) 24.1 (11.5, 38.6) Female breast 6.1 1.6 (1.1, 2.2) 6.7 (4.9, 8.7) 31.9 (23.2, 41.1) Uterus 8.4 –0.15 (–0.29, 0.10) –1.1 (–2.1, 0.68) –3.3 (–6.4, 2.1) Ovary 1.5 0.99 (0.12, 2.3) 1.1 (0.15, 2.3) 17.7 (2.4, 37.3) Prostate 1.6 0.29 (–0.21, 1.2) 0.61 (–0.46, 2.2) 7.0 (–5.3, 25.5) Urinary organs and kidney 3.8 1.2 (0.62, 2.1) 2.1 (1.1, 3.2) 22.3 (11.8, 34.2) Urinary bladder 2.4 1.0 (0.27, 2.1) 1.2 (0.34, 2.1) 16.3 (4.8, 30.1) Kidney 0.8 0.71 (–0.11, 2.2) 0.29 (–0.50, 0.79) 15.2 (–2.6, 41.3) Nervous system 1.5 0.26 (–0.23, 1.3) 0.19 (–0.17, 0.81) 5.7 (–5.3, 24.5) Thyroid 2.6 1.2 (0.48, 2.1) 1.6 (0.78, 2.5) 25.9 (12.4, 40.7) a254 solid cancers of other and ill-defined sites are included in the total solid tumors category. bAR is the attributable risk, which in this case is the percentage of cases in exposed survivors attributed to radiation exposure. cValues in parentheses are the 95% confidence limits. SOURCE: Thompson and others (1994). occurring in the period 1958–1993 showed a clear decline in Thyroid Cancer the ERR/Sv with either age at exposure or attained age when Like breast cancer, thyroid cancer risks are described well evaluated separately. The EAR was also found to decrease by a linear dose-response function and also show a strong with age at exposure, but to increase with attained age at dependence on age at exposure. In fact, there is little evi- least up to age 50. These analyses, as well as earlier analyses dence of a dose-response for persons exposed in adulthood by Tokunaga and colleagues (1994) and by Thompson and (Thompson and others 1994; Ron and others 1995a), while coworkers (1994), found that the dose-response for breast the ERRs/Sv for those exposed as children were large (9.5 cancer was well described by a linear function. Tokunaga for persons exposed under age 10, and 3.0 for those exposed and colleagues (1994) also report a strong attained age ef- at ages 10–19; Thompson and others 1994). Although sev- fect, with an ERR/Sv of 13 for breast cancer occurring be- eral other cohorts provide data on thyroid cancer risks from fore age 35 compared to an ERR/Sv of about 2 for breast external radiation exposure in childhood (Ron and others cancer occurring after age 35. 1995a), the LSS cohort is the only cohort providing much Land and colleagues (2003) reported on an incidence sur- information on thyroid cancer risk from external radiation vey of breast cancers diagnosed during 1950–1990. As in exposure in adulthood. previous analyses a strong linear dose-response was found. A modified isotonic regression approach, which required only that the ERR/Sv be monotonic in age, was used to evalu- Salivary Gland Cancer ate in detail the modification of the dose-response by age at Because some types of salivary gland tumors are not exposure and attained age. The abstract notes that “exposure readily identified by the conventional disease classification before age 20 was associated with higher ERR1Sv compared codes used by tumor registries, a special evaluation that in- to exposure at older ages, with no evidence of consistent cluded pathology reviews of both benign and malignant sali- variation by exposure age under 20. ERR1Sv was observed to vary gland tumors was undertaken by Land and colleagues decline with increasing attained age, with by far the largest (1996). This resulted in an estimated ERR/Sv of 3.5 (95% CI drop around age 35.”

OCR for page 141
150 BEIR VII 1.5, 7.5) for malignant tumors, higher than any of the ERR/ this may have been because the number of cancers of the Sv shown in Table 6-2, although very uncertainly estimated. latter type was small. The ERRs/Sv was 0.7 (0.1, 1.7) for benign tumors. Most of It has been estimated that more than 60–75% of HCC the dose-response for malignant tumors resulted from mu- cases in Japan are related to chronic hepatitis C infection and coepidermoid carcinoma with an ERR/Sv of 8.3 (2.5, 29.6), that 20–25% are positive for hepatitis B surface antigen whereas most of the dose-response for benign tumors re- (Fujiwara and others 2000). Neriishi and others (1995) re- sulted from Warthin’s tumor with an ERR/Sv of 3.1 (0.6, ported a radiation dose related increase in the prevalence of 10.3). hepatitis B surface antigen in atomic bomb survivors. Fujiwara and colleagues (2000) did not find such a relation- Stomach Cancer ship for hepatitis C infection, but their data suggest that the radiation dose-response for chronic liver disease was greater This site merits special comment primarily because stom- for survivors who were positive for hepatitis C antibody than ach cancer is the most common type of cancer in Japan and, for survivors who were negative. specifically, in the LSS cohort. Based on cancer incidence data evaluated by Thompson and colleagues (1994), stom- Lung Cancer ach cancer had a relatively small but precisely estimated ERR/Sv of 0.32 (0.16, 0.50). The ERR/Sv for females was Next to stomach cancer, lung cancer was the most com- about three times that for males, and the ERR/Sv decreased mon cancer in the LSS cohort. This cancer showed a strong with increasing age at exposure. Nearly one-third (31%) of sex association with the ERR/Sv for females about four times the solid cancer cases included in the incidence data were as large as that for males based on the incidence data evalu- stomach cancers, so this cancer potentially has a strong im- ated by Thompson and colleagues (1994), which probably pact on overall solid cancer results. However, analyses of reflects at least in part the larger baseline risks for males. solid cancer mortality data with stomach cancer excluded Lung cancer also deviated from the usual pattern of decreas- resulted in parameter estimates that were similar to those ing risk with increasing age at exposure. Instead, lung cancer obtained for all solid cancers (Preston and others 2003). risks appeared, if anything, to increase with increasing age at exposure, although, based on the incidence data, this trend Liver Cancer was not statistically significant. Recently, Pierce and coworkers (2003) evaluated the joint Liver cancer is one of the most frequently occurring can- effects of smoking and radiation on lung cancer incidence cers in Japan and the third most common cancer (after stom- through 1994 in a subset of about 45,000 members of the ach and lung) in the LSS. Liver cancers reported on death LSS cohort for whom both radiation dose and smoking data certificates might in fact be cancers originating in other or- were available. The smoking data were obtained from mail gans because the liver is a frequent site for metastatic cancer. surveys of the LSS cohort and clinical interviews of mem- This can be a problem even for tumor registry data, since bers of the AHS conducted during 1963–1993. Pierce and some cases were based only on death certificate information. colleagues (2003) found that the effects of smoking and For this reason, Cologne and colleagues (1999) conducted a radiation were significantly submultiplicative and consistent study of primary liver cancer based on extensive pathology with an additive model. They note that the aging of the cohort review of known or suspected cases of liver cancer. This and higher smoking levels among more recent birth cohorts study showed a clear dose-response with an estimated ERR/ resulted in a stronger basis for evaluating the joint effects of Sv (with 95% CI) of 0.81 (0.32, 1.43). The ERRs/Sv for smoking and radiation than in previous analyses by Kopecky males and females were very similar (0.81 and 0.78, respec- and colleagues (1986), Prentice and colleagues (1983), and tively), in contrast to findings for many other cancers, and the National Research Council (NRC 1988); these earlier somewhat remarkable given that background rates for males investigations were unable to distinguish between additive were about three times those for females. The modifying and multiplicative effects. Pierce and colleagues (2003) also effect of age at exposure was also different from that for found that adjustment for smoking substantially reduced the other cancers, with excess risk peaking for those exposed in female-to-male ERR/Sv ratio; about 85% of the men and their twenties, but little evidence of excess risk for those 16% of the women were smokers. With adjustment for smok- exposed under age 10 or over age 45. ing, there was evidence of a decline in the ERR/Sv with in- Of the 364 cases analyzed, there were 307 hepatocellular creasing attained age (comparable to other solid cancer sites), carcinomas (HCCs), 53 cholangiocarcinomas, two mixed but no evidence of modification by age at exposure. hepatocellular-cholangiocarcinomas, and one each of he- patoblastoma and hemangiosarcoma. This is in contrast to Skin Cancer liver cancers associated with Thorotrast exposure, which are dominated by cholangiocarcinomas and hemangiosarcomas. Ron and colleagues (1998b) conducted a detailed study Cologne and colleagues found no difference in the dose-re- of skin cancer that included pathologic review of cases. Basal sponse for HCC compared to cholangiocarcinoma, although cell carcinoma (80 cases) was found to be associated with

OCR for page 141
ATOMIC BOMB SURVIVOR STUDIES 151 radiation dose with some evidence of nonlinearity in the BENIGN NEOPLASMS dose-response, but with no evidence of an interaction with Studies addressing benign neoplasms have generally been ultraviolet radiation. No dose-response association was based on either the AHS or the tumor registries. Fujiwara found for squamous cell carcinoma (69 cases). The relation- and colleagues (1992) used the AHS to investigate hyper- ships with dose for melanoma (10 cases) and Bowen’s dis- parathyroidism in Hiroshima survivors. About 4000 indi- ease (26 cases) were not statistically significant, but esti- viduals with DS86 doses were tested for hyperparathyroid- mates of the ERR/Sv were large. ism, and a dose-response relationship was found (p < .001). The estimated relative risk at 1 Gy was 4.1 (95% CI 1.7, Central Nervous System Cancers 14.0), and a decrease in relative risk with increasing age at exposure was suggested. The authors concluded that doses See discussion of central nervous system tumors at the lower than those used in radiotherapy might induce this dis- end of the section “Benign Neoplasms.” order. Nagataki and colleagues (1994) used Nagasaki AHS data to investigate thyroid diseases in 2587 subjects with Lymphoma diagnoses based on uniform procedures including ultrasonic scanning. Significant dose-response relationships were ob- Analyses of mortality data by Pierce and colleagues served for all solid nodules (females), adenoma, and nodules (1996) showed no evidence of an association for lymphoma; without histological diagnosis (females). An association was with the mortality data, it was not possible to distinguish also found for autoimmune hypothyroidism, one of the non- between Hodgkin’s and non-Hodgkin’s cases. Lymphoma neoplastic end points investigated. However, the dose- was not included in more recent mortality analyses. The in- response for hypothyroidism was not monotonic; risk in- cidence data included 210 lymphoma cases, of which 22 creased to about 0.7 Sv and then decreased. were Hodgkin’s and 188 were non-Hodgkin’s. A statistically Ron and colleagues (1995b) used data from the Hiroshima significant dose-response was found for males, but not for and Nagasaki tumor and tissue registries to evaluate benign females, for whom the estimated ERR/Sv was negative tumors of the stomach, colon, and rectum for 1958–1989. A (Preston and others 1994). total of 470 cases with histologically confirmed benign gas- trointestinal tumors (163 stomach, 215 colon, and 92 rec- Multiple Myeloma tum) were identified. A positive dose-response relationship was observed for stomach tumors, with an estimated ERR/ Multiple myeloma exhibited a statistically significant Sv of 0.52 (95% CI 0.01–1.43), similar to that for stomach dose-response based on the mortality data (Pierce and others cancer. There was little evidence of dose-response for either 1996), but incidence data showed little evidence of such an colon or rectal tumors. association (Preston and others 1994). The discrepancy in Tokunaga and colleagues (1993) investigated prolifera- these findings appears to be due to deaths with questionable tive and nonproliferative breast disease using breast tissue diagnoses and second primary tumors that were included in samples from 88 high-dose and 225 low-dose autopsy cases the mortality analyses, but not the incidence analyses. of members of the LSS cohort. Both proliferative disease in general and atypical hyperplasia in particular were found to CANCERS RESULTING FROM EXPOSURE IN UTERO be positively associated with radiation dose, with the stron- gest association for subjects who were 40–49 years of age at Delongchamp and colleagues (1997) analyzed data on exposure. The authors hypothesized that this finding might cancer mortality among atomic bomb survivors who were be “related to the age dependence of radiation-induced breast exposed either in utero or, for comparison, during the first cancer, in that potential cancer induced in this age group by 5 years of life. These analyses covered the period 1950– radiation exposure may receive too little hormonal promo- 1992, adding an additional 8 years of follow-up to data avail- tion to progress to frank cancers.” able to the BEIR V committee (Yoshimoto and others 1988). Kawamura and colleagues (1997) conducted a study of Analyses were restricted to cancers occurring between the uterine myoma based on ultrasound examination of 1190 ages of 17 and 45. Ten cancers were observed in the cohort female AHS participants in Hiroshima. The reason for con- exposed in utero, and a significant dose-response was ob- ducting this study was concern that the previously identified served with an estimated ERR/Sv of 2.1 (90% CI 0.2, 6.0). dose-response associations (Wong and others 1993), dis- This estimate did not differ significantly from that observed cussed below, might have resulted from bias in case detec- for survivors exposed during the first 5 years of life. An un- tion. This study resulted in an estimated ERR/Sv of 0.61 usual aspect of the finding was that 9 of the 10 cancers oc- (95% CI 0.12, 1.31). It was judged unlikely that bias could curred in females, and significant differences between the explain the association. In earlier analyses by Wong and col- sexes persisted even when the three female cancer sites leagues, time since exposure was found to be a significant (breast, ovary, and uterus) were excluded. modifier for uterine myoma, with younger survivors show-

OCR for page 141
152 BEIR VII ing a decrease with time and older survivors showing an mal and distal survivors (although results without the adjust- increase with time. ment are also presented). Preston and colleagues (2002b) investigated tumors of The estimated ERR/Sv for noncancers based on a linear the nervous system and pituitary gland based on cases ascer- model with no dependence on age at exposure or sex was tained through the Hiroshima and Nagasaki Tumor and Tis- 0.14, generally lower than that for all solid cancers (where sue Registries and through medical records from RERF and the ERR/Sv depends on age and sex). There was no evidence major medical institutions in Hiroshima and Nagasaki. His- of a statistically significant dependence on either age at ex- tologic diagnoses were obtained by having four pathologists posure or sex, but the data were compatible with effects simi- independently review slides and medical records. The ma- lar to those estimated for solid cancers. A linear dose-re- jority of the 228 central nervous system tumors included in sponse function fitted the data well, but it was not possible to the study were benign. A statistically significant dose-re- rule out a pure quadratic model or a model with a threshold sponse association was observed for all nervous system tu- as high as 0.5 Sv. Similar to Shimizu and colleagues (1999), mors with an estimated ERR/Sv of 1.2 (95% CI 0.6, 2.1). significant dose-response relationships were found for heart The ERR/Sv was highest for schwannomas (4.5; 95% CI disease, stroke, respiratory disease, and digestive disease. 1.9, 9.2), but the dose-response for all other central nervous There was no evidence of radiation effects for infectious dis- system tumors evaluated as a group was also statistically eases or all other noncancer diseases in the group evaluated. significant. The dose-responses for all nervous system tu- Lifetime noncancer risks for people exposed to 1 Sv were mors and for schwannomas were both statistically signifi- estimated to be similar to those for solid cancer for those cant when limited to subjects with doses of less than 1 Sv, exposed as adults, and about half those for solid cancer for and there was no evidence that the slope for this low-dose those exposed as children. Because baseline risks for the non- range was different from that for the full range. Modification cancer category evaluated are greater than those for all solid of risk by sex, age at exposure, and attained age was also cancers, even the relatively small ERR/Sv leads to a fairly investigated. large absolute lifetime risk. Because small ERRs can easily arise from bias, Shimizu and colleagues (1999) evaluated several potential sources of bias, including misclassification of cause of death, confound- NONNEOPLASTIC DISEASE ing, and cohort selection effects. Although Preston and co- workers (2003) discuss cohort selection effects in detail, they Findings Based on Mortality Data did not reevaluate other sources of bias. The committee sum- A statistically significant dose-response relationship with marizes the discussion provided by Shimizu and colleagues mortality from nonneoplastic disease in A-bomb survivors in the remainder of this section. was demonstrated by Shimizu and colleagues (1992) based With regard to misclassification, they note that Sposto on mortality data for 1950–1985. The addition of five years and coworkers (1992) investigated the possibility of bias of mortality data (through 90) strengthened the evidence for from this source using mortality data through 1985. These this effect and allowed a more detailed evaluation (Shimizu investigators used estimated age-dependent misclassification and others 1999). In these analyses, statistically significant probabilities obtained from RERF autopsy data to conduct associations were seen for the categories of heart disease, analyses that corrected for misclassification and found that stroke, and diseases of the digestive, respiratory, and hemato- estimates for noncancer mortality were reduced by 20%, but poietic systems. remained highly statistically significant. Shimizu and Preston and colleagues (2003) updated these results and colleagues (1999) used mail survey and interview data to present analyses of deaths from all causes excluding neo- examine the possible effect of several potential confounders plasms, blood diseases, and external causes such as acci- including educational history and smoking. Although most dents or suicide. They give considerable attention to the fact of the factors evaluated were found to affect noncancer mor- that for a few years after the atomic bomb explosions, tality, they were not found to be associated strongly with baseline risks for noncancers in proximal survivors (within dose. Analyses adjusted for various confounders, based on 3000 m of the hypocenter) were markedly lower than those survivors with available data, resulted in ERRs/Sv that were in distal survivors. They refer to this as the “healthy survivor very similar to the unadjusted values. effect” and note that it could lead to distortion of the dose- Shimizu and colleagues (1999) also evaluated noncancer response, particularly in the early years of follow-up. They diseases of the blood, benign neoplasms, and deaths from also note that a small difference (2%) in baseline risks for external causes. Because these categories were not reevalu- proximal and distal survivors persisted in later years, which ated by Preston and coworkers (2003), the committee sum- they consider likely to be due to demographic factors such as marizes these findings. The ERR/Sv for the 191 deaths from urban-rural differences. They address this potential source noncancer diseases of the blood was estimated to be 1.9 (90% of bias by conducting analyses restricted to the period 1968– CI 1.2, 2.9), larger than the estimated values for most solid 1997 and by including an adjustment for differences in proxi- cancers. The accuracy of death certificate diagnosis is known

OCR for page 141
ATOMIC BOMB SURVIVOR STUDIES 153 to be poor for this category and likely to include many sociation for myocardial infarction based on all of the data misclassified leukemias and malignant lymphoma deaths. (p = .02), with an estimated ERR/Sv of 0.17 (95% CI 0.01, Among 128 deaths for which additional diagnostic informa- 0.36). The association remained significant when analyses tion was available, there were 57 nonneoplastic disease were adjusted for various risk factors including blood pres- deaths. When these deaths were analyzed separately, the re- sure and cholesterol. Positive dose-response relationships sulting ERR/Sv was 2.0 (90% CI 0.6, 4.4), nearly identical were also found for several other end points of atherosclero- to that based on the full 191 deaths. Analyses suggested that sis, which the authors interpreted as supporting a real asso- the effect was limited to nonaplastic anemias (29 cases), ciation between radiation exposure and atherosclerosis. since the estimate for aplastic anemias (31 cases) was essen- Kodama and colleagues (1996) confirmed previously identi- tially zero. There was also a suggestion of a strong dose- fied radiation associations for uterine myoma, hyperparathy- response based on 13 deaths from myelodysplastic syn- roidism, and chronic liver disease with an ERR/Gy of 0.46 drome, a neoplastic disease thought to be a precursor of acute (0.27, 0.70), 3.1 (0.7, 13), and 0.14 (0.04, 0.27) for the three myelogenous leukemia. respective end points. Although the data evaluated by Shimizu and colleagues Wong and colleagues (1999) used AHS data to examine (1999) included 379 deaths attributed to benign neoplasms long-term trends in total serum cholesterol levels over the or neoplasms of unspecified nature, only 31 deaths were spe- 28 years from 1958 to 1986. Dose-response relationships for cifically indicated on the death certificate as being due to the increase in cholesterol levels over time were demon- benign neoplasms. There was no convincing evidence of a strated for women in general but only in the youngest birth dose-response for these 31 deaths. cohort (1935–1945) for men. Age, body mass index, city, With regard to deaths from external causes, suicide rates and birth year were considered in the analyses, and some showed a statistically significant decline with increasing analyses were adjusted for cigarette smoking. These results dose, whereas no evidence of a dose-response relationship may partially explain the dose-response relationship for was found for deaths from other external causes. coronary heart disease that has been observed in other stud- ies of atomic bomb survivors. Findings Based on the Adult Health Study (AHS) or on Autopsy Data LIFE SHORTENING Wong and colleagues (1993) evaluated the relationship Cologne and Preston (2000) investigated life shortening between exposure to radiation and the incidence of 19 non- in the LSS cohort using mortality data through 1995. Al- malignant disorders using data from the AHS cohort for though dose-related increases in both cancer and noncancer 1958–1986. They found statistically significant positive mortality imply that longevity is also related to dose, earlier dose-response relationships (p < .05) for thyroid disease papers addressing these effects (Pierce and others 1996; (p < .001), chronic liver disease and cirrhosis (p = .007), and Shimizu and others 1999) did not specifically attempt to uterine myoma (p < .001). In addition, myocardial infarction quantify the degree of radiation-induced life shortening, an showed a significant dose-response for 1968–1986 among end point that reflects the effects of both cancer and non- those who were under 40 years of age at exposure (p = .03). cancer mortality. The investigation of longevity was under- Statistically significant relationships were not detected for taken in part because of earlier reports in both the scientific hypertension, hypertensive heart disease, ischemic heart literature and the press that certain atomic bomb survivors disease, occlusion and stenosis of precerebral and cerebral had greater-than-average life expectancy. arteries, aortic aneurysm, stroke, cataract, gastric ulcer, A clear decrease in median life expectancy with increas- duodenal ulcer, viral hepatitis, calculus of kidney and ureter, ing radiation dose was found. Among cohort members with cervical polyp, hyperplasia of prostate, dementia, and estimated doses between 0.005 and 1.0 Gy, the median loss Parkinson’s disease. Modification of the ERR/Sv by sex, of life was estimated to be about 2 months, while among city, age at exposure, and time since exposure was also in- cohort members with estimated doses of 1 Gy or more, the vestigated for those end points that showed overall associa- median loss of life was estimated to be about 2.6 years. The tions. Age at exposure was found to be a significant modifier median loss of life among all cohort members with doses of risk for thyroid disease (decreasing ERR/Sv with increas- estimated to be greater than zero was about 4 months. ing age); modifying effects for uterine myoma are discussed Cologne and Preston (2000) present estimates of life ex- above (“Benign Neoplasms”). pectancy for groups defined by dose. For those with zero Kodama and colleagues (1996) reviewed results of stud- dose, separate estimates are presented for groups defined by ies addressing noncancer diseases and their relationship to distance from the hypocenter, including estimates for those radiation exposure in A-bomb survivors. They also update who were not in the city (>10 km from the hypocenter). Al- some of the analyses by Wong and colleagues (1993) to in- though the relative mortality for all nonzero-dose groups clude data through 1990, but do not present nearly as much compared to the combined in-city, zero-dose group was 1.0 detail as the latter. They found a statistically significant as- or greater, results for those in the lowest-dose category

OCR for page 141
154 BEIR VII (0.005–0.25 Gy) were somewhat dependent on the choice of portant for evaluating site-specific cancers. Although pub- comparison group. Cohort members in this low-dose cat- lished evaluations described in Chapter 6 are based on DS86 egory had a median life expectancy that was shorter than dosimetry, a revised DS02 system—the result of a major that of zero-dose survivors who were within 3 km of the international effort to reassess and improve survivor dose hypocenter (229 d), shorter than the not-in-city group (365 estimates—has recently become available and was used to d), but slightly longer (52 d) than survivors located 3 km or develop BEIR VII risk models. An initial evaluation indi- more from the hypocenter. These results do not support the cates that this revision will slightly reduce risk estimates. hypothesis that life expectancy for atomic bomb survivors The more extensive data on solid cancer that are now exposed at low doses is greater than that for comparable un- available have allowed more detailed evaluation of several exposed persons. issues pertinent to radiation risk assessment. Several investi- gators have evaluated the shape of the dose-response, focus- ing on the large number of survivors with relatively low SUMMARY doses. These analyses have generally confirmed the appro- The LSS cohort of survivors of the atomic bombings in priateness of linear functions to describe the data. The modi- Hiroshima and Nagasaki continues to serve as a major source fying effects of sex, age at exposure, and attained age have of information for evaluating health risks from exposure to also been explored in detail using both ERR and EAR radiation, and particularly for developing quantitative esti- models. The ERR/Sv has been found to decrease with both mates of risk from exposure to ionizing radiation. Its advan- increasing age at exposure and increasing attained age, and tages include its large size, the inclusion of both sexes and it now appears that both variables may be necessary to pro- all ages, a wide range of doses that have been estimated for vide an adequate description of the data. By contrast, the individual subjects, and high-quality mortality and cancer EAR shows a sharp increase with increasing attained age incidence data. In addition, the whole-body exposures re- and a decrease with increasing age at exposure. ceived by this cohort offer the opportunity to assess risks for The availability of high-quality cancer incidence data has cancers of a large number of specific sites and to evaluate resulted in several analyses and publications addressing spe- the comparability of site-specific risks. The full LSS cohort cific cancer sites. These analyses often include special patho- consists of approximately 120,000 persons who were identi- logical review of the cases and sometimes include data on fied at the time of the 1950 census. However, most recent additional variables (e.g., smoking for evaluation of lung analyses have been restricted to approximately 87,000 survi- cancer risks). Papers focusing on the following cancer sites vors who were in the city at the time of the bombings and for have been published in the last decade: female breast cancer, whom it is possible to estimate doses. Special studies of sub- thyroid cancer, salivary gland cancer, liver cancer, lung can- groups of the LSS have provided clinical data, biological cer, skin cancer, and central nervous system tumors. Special measurements, and information on potential confounders or analyses have also been conducted of cancer mortality in modifiers. survivors who were exposed either in utero or during the Mortality data for the period 1950–1997 have been evalu- first 5 years of life. ated in detail, adding 12 years to the follow-up period avail- Health end points other than cancer have been linked to able at the time BEIR V (NRC 1990) was published. The radiation exposure in the LSS cohort. Of particular note, a longer follow-up period not only increases statistical preci- dose-response relationship with mortality from nonneoplas- sion, but also allows more reliable assessment of the long- tic disease was demonstrated in 1992, and subsequent analy- term effects of radiation exposure, including modification or ses in 1999 and 2003 have strengthened the evidence for this risk by attained age and time since exposure. Importantly, association. Statistically significant associations were seen cancer incidence data from both the Hiroshima and the for the categories of heart disease, stroke, and diseases of the Nagasaki tumor registries became available for the first time digestive, respiratory, and hematopoietic systems. The data in the 1990s. These data not only include nonfatal cancers, were inadequate to distinguish between a linear dose-re- but also offer diagnostic information that is of higher quality sponse, a pure quadratic response, or a dose-response with a that that based on death certificates, which is especially im- threshold as high as 0.5 Sv.