ELAINE RON, DALE L. PRESTON, AND KIYOHIKO MABUCHI

For several decades, mortality data have been the mainstay of cancer risk estimates derived from the Life Span Study (LSS) cohort of atomic bomb survivors. However, with the recent availability of comprehensive incidence data, the evaluation of cancer incidence is now possible. Solid tumor incidence data are obtained by matching the LSS cohort with the Hiroshima and Nagasaki Tumor Registries. These registries were established in 1958 and cover all residents of Hiroshima and Nagasaki. The LSS mortality data are obtained through the nationwide family registry system. Using the incidence data, 8,613 first primary solid cancer cases were identified between 1958 and 1987, compared with 6,887 solid cancer deaths reported on death certificates throughout Japan between 1950 and 1987. Of cancer patients identified through the tumor registries, 23% were alive in 1988, the underlying cause of death was not attributed to cancer for 14%, and the type of cancer recorded on the death certificate was different than in the tumor registries for 5%. Still, the incidence data generally confirmed the mortality findings. Significant excess risks were observed in both the incidence and mortality series for all solid cancers, for cancers of the stomach, colon, liver (when it was defined as primary liver cancer or liver cancer NOS on the death certificate), lung, breast, ovary, and urinary bladder. In general, the risk estimates were larger based on incidence data. No significant radiation effects were seen for cancers of the pharynx,

rectum, gallbladder, pancreas, nose, larynx, uterus, prostate or kidney in either series. A significant excess of esophageal cancers was observed in the mortality data, and significant excesses of non-melanoma skin cancers and cancers of the salivary gland and thyroid were seen in the incidence data. Incidence data allow a more precise description of radiation-induced cancer risks because the number of incident cases is considerably larger than the number of cancer deaths; incidence data provide the only reliable source of data for malignancies with relatively good survival rates, and diagnoses are more accurate and detailed. Mortality data are easy to obtain and are virtually complete for all of Japan starting in October 1950. The next 20 years will yield extremely important information on radiation exposure at early ages, and analyses of both incidence and mortality are needed to fully understand the role of radiation in carcinogenesis.

For several decades, mortality data have been the mainstay of cancer risk estimates derived from the Life Span Study (LSS) cohort of atomic bomb survivors (Shimizu et al., 1990). This cohort has been followed up by the Atomic Bomb Casualty Commission until 1974 and by the Radiation Effects Research Foundation (RERF) subsequently. Population-based cancer registries were established in Hiroshima and Nagasaki over 30 years ago, but problems with data collection have limited their use. During the 1980s, RERF worked to improve the registries' data quality. Previously collected data were reviewed, revised, and supplemented, and new methods for the systematic collection and management of future data were instituted. As a result, comprehensive incidence data became available for study (Mabuchi et al., 1994). Last year, four reports summarizing cancer incidence in the LSS were published (Mabuchi et al., 1994; Thompson et al., 1994; Preston et al., 1994; Ron et al., 1994a). The wide use of these data (UNSCEAR, 1994) demonstrates their valuable contribution to risk estimation, and ensures that they will play a larger role in the future research of RERF.

The LSS cohort includes 120,321 people; however, in this report, we did not include the 26,580 individuals who were not in Hiroshima or Nagasaki at the time of the bombings, or the 7,109 persons with unknown radiation doses. As described by Mabuchi et al. (1994) and Thompson et al. (1994), solid tumor incidence data are obtained by matching the LSS cohort with the Hiroshima and Nagasaki Tumor Registries. These registries were established in 1958 and cover all residents of Hiroshima and Nagasaki. The Hiroshima and Nagasaki registries use the same methods for ascertaining cases, the majority of which are identified by searching hospital records. These data are supplemented by abstracting information from the Hiroshima and Nagasaki Tissue Registries; the Nagasaki Prefectural Cancer

Registry; the Leukemia Registry; data from the RERF clinical, surgical and autopsy programs; notification by physicians; and review of national death certificates. Because the cancer incidence data are derived from the Hiroshima and Nagasaki Tumor Registries, follow-up for the incidence series begins January 1, 1958, and includes the 79,972 LSS members who were alive and not known to have had cancer before 1958 (Thompson et al., 1994) (Table 8.1).

The LSS mortality data are obtained, in three-year cycles, through the nationwide family registry system (koseki). For each deceased survivor, the underlying cause of death is abstracted from the death certificate. Mortality follow-up begins on October 1, 1950, and includes the 86,309 LSS members who were known to be alive on that date.

For the incidence series, follow-up ended at the first diagnosis of a primary cancer, or the date of death, or December 31, 1987, whichever came first. In the mortality series, follow-up ended at the date of death or December 31, 1987. There are more than 600,000 additional person years of follow-up in the mortality series than in the incidence series because follow-up began eight years earlier. At the time of the bombings (ATB), the mean age was 29 years for subjects in the mortality series, compared with 26.8 years for survivors in the incidence study, because many people who were older ATB died before the incidence follow-up began (Ron et al., 1994a).

Organ doses were computed as the sum of the gamma-shielded kerma and ten times the neutron-shielded kerma, as estimated by the 1989 version of DS86 (Roesch, 1987; Fujita, 1989). Kerma is expressed in Gy, and the weighted organ doses are expressed in sieverts. Site-specific analyses were based on estimated organ doses, except for skin cancers, for which we used kerma because no skin dose was calculated as part of DS86. Solid tumors as a group were analyzed using

intestinal doses. Persons with kerma estimates over 4 Gy were excluded from the analysis because the accuracy of these dose estimates is questionable (Pierce et al., 1990).

Risk estimates were calculated using Poisson regression methods. Analyses were based on general excess relative risk (ERR) models and time-dependent excess absolute risk (EAR) models. Background rates were modeled as a function of city, gender, attained age, and year of birth. A linear dose-response function was used as the standard model, although we tested for non-linearity. Maximum likelihood methods were used to calculate parameter estimates and 95% confidence intervals (Cox and Hinckley, 1974). The statistical methods are described in more detail in Thompson et al. (1994) and Preston et al. (1994).

Using the incidence data, 9,014 first primary cancer cases were identified between 1958 and 1987, compared with 7,308 cancer deaths reported on death certificates throughout Japan between 1950 and 1987. Of the incidence cases, 8,613 (95.5%) were solid tumors, and of the cancer deaths, 6,887 (94.2%) were attributed to solid cancers. When the mortality series was limited to deaths occurring during the incidence series time period (i.e., 1958–1987) 2,671 more incident cancer cases than cancer deaths were identified. Further restricting the mortality series to deaths occurring between 1958 and 1987 in Hiroshima or Nagasaki would result in 3,155 more incident cancer cases than cancer deaths.

The number of deaths from cancers of the digestive and respiratory systems, occurring in all of Japan between 1950 and 1987, was not much smaller than the number of cases in the incidence series. But, for cancers of the male genital and urinary system, the incidence series was at least twice as large as the total mortality series. The number of incident cases of cancers of the skin, female breast, and thyroid was more than three times greater than the number of deaths. Cancer ascertainment was not only more complete using incidence data, but diagnoses were more precise than those recorded on death certificates. For example, specific diagnoses for cancers of the uterus were available for 88.1% of the incident uterine cancers, whereas in the mortality series, 71.6% of cancers of the uterus were recorded as not otherwise specified (NOS) (Table 8.2).

When we compared the first primary cancer incidence diagnoses with the underlying causes of death, we found 23% of cancer patients were still alive in 1988 and, therefore, would not have been included in an analysis of the mortality data. In addition, the underlying cause of death was recorded as noncancer for 14% of the cancer patients. Another 5% had a different type of cancer listed on the death certificate than in the tumor registries, even when the comparison was made at the organ-system level. This means that only 58% of the cancer cases analyzed in the incidence series would be included in the mortality series with the cancer diagnoses listed in the cancer registries. Although statistics vary depending on the

cancer type, if death certificates were the only source of medical information, less than 50% of the cancer cases would have been classified as in the tumor registry for the majority of sites studied. For cancers with particularly poor survival, such as cancers of the esophagus, liver, pancreas, and lung, over 70% of the incident cancers would have been identified using death certificate diagnoses.

Even though there were considerable differences in the two data sources, the incidence data generally confirmed the mortality findings. Significant excess risks were observed in both the incidence and mortality series for all solid cancers. Based on incidence data, the excess relative risk at 1 sievert (ERR_1Sv) was about 40% larger and the excess absolute risk per 10,000 person-year sievert (EAR/10³ PYSv) almost three times higher than the risk estimates derived from mortality data (Table 8.3). Based on the incidence series, 504 excess cancer cases were predicted, compared with 304 excess cancer deaths based on the mortality data. Even during the last follow-up period (1976-1987) when more cancer deaths would be expected because of the older age of the cohort, 128 more excess cancer cases were predicted than cancer deaths.

Using either the incidence or mortality data, excess relative risks were approximately two times greater for females than males, although the level of significance was higher using the incidence data. The estimated excess relative risks for persons under 20 years ATB were substantially larger in the incidence series than

the mortality series, whereas for persons exposed to the bombings after age 20 the risks derived from incidence and mortality data did not vary as much. Since the excess relative risks for developing cancers of the breast, thyroid, and skin, which are often not fatal, are particularly pronounced among persons exposed to the bombings during childhood, this finding would be expected. During the first 19 years of follow-up, risks were considerably higher using incidence data than mortality data. In fact, using incidence data, risks were high in the first follow-up period, decreased in the second, and increased again in the last, whereas in the mortality data risks increased with increasing time since exposure.

The EAR for females was 1.8 times greater than for males in both the incidence and mortality series. The high risk for females was observed particularly among survivors exposed to the bombings before age twenty. However, the EARs increased with attained age more rapidly for males than for females. By about age 55 years, males had a higher EAR of cancer deaths than females, whereas this gender crossover was seen at about age 68 in the incidence data.

Using either a relative or absolute risk model, statistically significant radiation effects were demonstrated in both the incidence and mortality data for cancers of the stomach, colon, liver (when it was defined as primary liver cancer or liver cancer NOS on the death certificate), lung, breast, ovary, and urinary bladder (Table 8.4). In general, the ERR point estimates were larger in the incidence series than in the mortality. Particularly notable were the high risks observed for the incidence of cancers of the stomach, colon, and lung, compared with the mortality series. The EAR estimates also were larger when based on incidence data than mortality.

No significant radiation effects were seen for cancers of the pharynx, rectum, gallbladder, pancreas, nose, larynx, uterus, prostate, or kidney in either series. For cancers of the oral cavity and pharynx, the risk estimates based on the incidence data were positive, while those based on mortality data were negative. The opposite situation was observed for cancers of the uterus: negative risk estimates based on incidence data and positive estimates based on mortality data. However, the confidence bounds around these estimates were wide and overlapped.

A significant excess of esophageal cancers was observed in the mortality but not in the incidence data (Table 8.4). This is explained partly by the extremely high risks of death from esophageal cancer seen shortly after the bombings, i.e., before incidence data were available.

Non-melanoma skin cancers, and cancers of the salivary gland and thyroid were rarely evaluated in mortality studies, but we demonstrated a significant excess of each of these cancers in the incidence data (Table 8.4). There were so few deaths due to salivary gland tumors that we couldn't evaluate them using mortality data. Although the number of incident salivary gland cancers was also small (13 exposed and 9 non-exposed cases), the ERR was very high (ERR_1Sv = 1.77; 95% CI = 0.15-6.0) and was greater for persons exposed before age 20.

Based on 168 first primary cases of non-melanoma skin cancer, there appeared to be some evidence for non-linearity (p = 0.12 for upward curvature) in the dose

response. A spline model appeared to fit the data best. An inverse relationship between risk and age at exposure was also demonstrated. The ERR_1Sv was 7.2 for persons under age 10 ATB, 2.4, 1.2, and 0.27 for those 10–19, 20–39, and 40+ ATB, respectively. Based on only 36 deaths from non-melanoma skin cancer, the ERR_1Sv was approximately one-third of that observed in the incidence data.

There were only 49 deaths due to thyroid cancer compared with 225 incident cases. Of the 49 deaths, only 37 were the same cases as in the incidence series; 4 had other cancers in the tumor registry, 3 were not residents of Hiroshima or Nagasaki and 5 died before the tumor registries were established. Based on the mortality data, the ERR_1Sv was 0.09. In contrast, even at doses as low as 25 cGy, we observed a large excess risk in the incidence data and found a dramatic difference in the effect of radiation exposure during childhood and adulthood. The ERR for persons <10 ATB was 9.5 (95% CI = 4.1–18.9) but the risk for persons >20 ATB was 0.1 (<-0.2–0.8). Although survivors who were routinely examined

clinically as part of the Adult Health Study had 2.5 times more thyroid cancers than other LSS members, the slope of the dose response did not differ significantly between the examined and non-examined groups.

The radiation effects observed for non-fatal cancers strongly influenced the incidence-based risk estimates. Breast, thyroid, and non-melanoma skin cancers comprised 10.7% of the solid cancer cases in the incidence series, but only 3.8% in the mortality series. Using incidence data, the ERR_1Sv for these three cancer types was 1.86, i.e., about four times larger than for other incident cancers combined (Table 8.5). For breast, thyroid, and non-melanoma skin cancers, all females and individuals (males and females) exposed to the bombings before age 20 had excess relative risks that were considerably greater than for males or people above age 20 ATB. The ERR_1Sv for survivors under age 20 ATB was 5.53. When breast, thyroid, and non-melanoma skin cancers were not included in the incidence data, the excess relative risk estimates based on incidence or mortality were virtually the same.

Because the number of deaths specifically reported as due to primary liver cancer was small, a significant ERR was found only when deaths from primary liver cancers were combined with liver cancer deaths not designated as either primary or secondary. Using incidence data, we found that the ERR was higher when good diagnostic data were available. The ERR_1Sv was 0.49 (95% CI = 0.16–0.50) when all liver cancers were evaluated. But when the analysis was restricted to histologically confirmed cancers, the ERR_1Sv rose to 0.66 (95% CI = 0.11–1.44). The ERR_1Sv was 0.41 (95% CI = -0.14–1.27) for cases diagnosed clinically and 0.36 (95% CI = -0.11–1.11) for cases ascertained through death certificates.

Histologic type was available from the tumor registries for 69% of the lung cancer cases. Approximately 45% were adenocarcinomas, 31% squamous-cell

carcinomas, and 12% small-cell carcinomas. Although we found no significant difference in the slope of the dose response by histological diagnosis, the ERR for small-cell carcinomas was extremely large (Table 8.6). Females had a higher ERR for adenocarcinomas than males, and the ERR for squamous-cell carcinomas increased with increasing age at exposure and decreased with increasing time since exposure.

Incidence data allow a more precise description of radiation-induced cancer risks because the number of incident cases is considerably larger than the number of cancer deaths, they provide the only reliable source of data for malignancies with relatively good survival rates, and diagnoses are more accurate and detailed because they are based on hospital, clinic or pathology records. For example, 75% of the cancers identified among the LSS members were verified histologically,

and another 4.4% were diagnosed based on direct observation, e.g., endoscopy, bronchoscopy, or surgery. For several cancer types, over 90% of the tumors were histologically confirmed. Information on method of diagnosis, tumor location and size, and treatment also may be available.

The cancer incidence results have generated questions and hypotheses about risk patterns for specific cancers. To further study these questions, detailed site-specific incidence studies including histologic review are underway. A salivary gland cancer incidence study, which included special searches for salivary gland tumors occurring outside the parotid gland, has just been completed (Land et al., 1996), and a paper on non-melanoma skin cancers will be submitted shortly. Studies of cancers of the nervous system, liver, ovary, and thyroid are in the data collection phase, and plans for other site-specific studies are currently being developed.

The primary weaknesses of the incidence data are that coverage is restricted to LSS members who reside in the Hiroshima and Nagasaki Tumor Registry catchment areas, and that solid tumors cannot be ascertained from before 1957, i.e., for the first 12.5 years after the bombings. Approximately 8% of the cancer deaths between 1958 and 1987 occurred outside Hiroshima and Nagasaki. To take this into account, a migration adjustment, based on information from the Adult Health Study, was made in the incidence analysis (Sposto and Preston, 1992). Thirteen percent of cancer deaths occurred between 1950 and 1958, before the tumor registries were established. These missed incident cases result in an underestimate of the EAR and could affect excess relative risk estimates if the distribution of cancer types differed from later years. Although a small excess risk of solid tumors has been observed in the LSS and the ankylosing spondylitis study five to ten years after the bombings, a latency period of ten or more years is generally accepted (UNSCEAR, 1994). Thus, while the lack of follow-up before 1958 is clearly a drawback, the problem may not be severe.

Mortality data are easy to obtain and are virtually complete for all of Japan starting in October 1950 (Shimizu et al., 1990). Mortality data are unavailable for 1945 to 1949 because the LSS cohort was selected from the 1950 Japanese census data (Beebe and Usagawa, 1968). Mortality data, however, only provide information on fatal cases, and since cancer may be detected long before death, time factors cannot be evaluated directly using mortality data. Furthermore, inaccurate or non-specific death certificate diagnoses (cancers erroneously reported as noncancers, cancer site misclassified or reported as ''ill-defined" or "not otherwise specified," and noncancers recorded as cancers) limit results regarding specific cancer sites (Glasser, 1981; Cameron and McGoogan, 1981; Mollo et al., 1986).

In a study comparing 5,000 autopsy and death certificate diagnoses in the LSS, 9% of deaths reported as due to cancers by the death certificates were not cancer deaths as judged by autopsy, and 25% of cancers diagnosed at autopsy were missed on death certificates (Ron et al., 1994b). Sposto et al. (1992) report that these errors result in a 12% underestimate of the ERR and a 16% underestimate of the EAR. Clinical information often is not available at the time a death certificate is written.

The underlying cause of death as listed on death certificates often overrepresents the number of liver cancers, probably because metastatic cancers were misclassified. The failure to distinguish between primary and secondary liver cancers also makes reliance on mortality data questionable for this site (Ron et al., 1994b).

The finding of a lower ERR in the mortality data, when compared with incidence data before 1965, may be a result of a latency period for developing radiation-induced cancer. Since only mortality data are available before 1958, the mortality risk estimates would be diluted by a high proportion of spontaneous cancers. Furthermore, because people generally are diagnosed with cancer years before dying from it, radiation risks will be detected earlier when incidence data are analyzed.

Because individuals who were exposed to the bombings during childhood are only now entering the ages for spontaneous cancers to develop, the next 20 years will yield extremely important information on radiation carcinogenesis. Although much has been learned from mortality data over the last few decades, analyses of both incidence and mortality data are needed in the future.