Click for next page ( 74


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 73
12 Discussion In the last chapter of this report of the Five Series Study, we first discuss the principal limitations of the available data. We then discuss our findings (as pre- sented in Chapter 11) in light of the limitations and strengths of these data and the findings that others have reported in studies of atomic veterans. LIMITATIONS Fact-of-Death and Cause-of-Death Ascertainment Death rates for both the participant and the referent cohorts were generally lower than those for the U.S. white male population, resulting in standardized mortality ratios that were nearly all less than 1.0. In part, these low SMRs are due to the "healthy soldier" effect (Seltzer and Jablon, 1974, 1977; see discus- sion in Appendix C), but underascertainment of fact and cause of death also contributed to lower SMRs. Among the shortcomings of this analysis are inequalities in the follow-up of participant and referent deaths for which we can assign the cause. For example, we estimated that BIRLS notes roughly 91.8 percent of participant deaths and 90.7 percent of deaths among the referents. In addition, we obtained underlying causes for 95.5 percent of participant deaths but only 92.7 percent of referent deaths (see Chapter 8~. The cumulative effect of these differences is an underascertainment of deaths for which we can assign the cause. For participants, the cumulative as- certainment of deaths with cause is estimated to be 87.7 percent (.918 x .955), and for referents, 84.1 percent (.907 x .927~. The net effect of this underascer- tainment of deaths by cause is to underestimate participant SMRs by roughly 12 73

OCR for page 73
74 THE FIVE SERIES STUDY percent and referent SMRs by roughly 16 percent. Although we have not cor- rected the SMRs in our tables for this estimated underascertainment, one could multiply participant SMRs by 1.14 (1 . .877) and referent SMRs by 1.19 (1 . .841) to obtain such an adjustment. Thus, for example, an SMR for participants of 0.75 would be adjusted to 0.86, while the same SMR for referents would yield an adjusted value of 0.89, both values more in keeping with SMR esti- mates in other military and occupational cohorts. Further, we point out that such adjustments will not only affect the SMRs, but also the risk estimates we have made. We show the potential effect of under- reporting by working through an example, step by step. First, assume that the deaths with unknown cause are distributed in the same fashion as the deaths with cause. Then, taking all cancer mortality as an example (using numbers from Table 11-2), there would have been 231 (i.e., t842 . 18,498] x 5,081) additional deaths observed among participants. The revised SMR is now .78 (i.e., [5,081 + 231] - 6825~. For the referent cohort, similar calculations yield a revised SMR of .80, compared to the original value of .74. Although the calculations above are explicitly made for SMRs, it turns out that (data not shown) the ratio of SMRs is a fairly good empirical approximation to the hazard ratio in this study, probably due in part to the fact that the participant and referent cohorts were frequency matched on age, branch of service, time of service, and paygrade. Thus, although we cannot directly adjust the hazard ratio because we do not have sufficient information, we can use the ratio of the adjusted SMRs to approximate an adjusted hazard ratio. For all cancer deaths, the partici- pants have an adjusted SMR of 0.78, compared to an unadjusted value of 0.74, that is the adjusted SMR is 1.05 times bigger than the unadjusted. For referents, these values are .80 and .74, so that the adjusted SMR is 1.08 times larger. The ratio of the two adjusted SMRs incorporates both of these factors: 0.97 = (.74 x 1.05) - (.74 x 1.08~. When rewritten as (.74 - .74) - (1.05 - 1.08), it is clear that the adjusted SMR ratio is the original SMR ratio times an adjustment factor of 0.97 (i.e., 1.05 - 1.08~. Within rounding error, this is [1 . .955] divided by [1 - .972l, these two quantities being the reciprocals of the percentages of missing causes of death noted above. This further suggests that if we wish to adjust risk ratios (either SMR ratios or hazard ratios) for both missing cause and for unascer- tained deaths, we should use an adjustment factor of 0.96 (i.e., 1.14 - 1.19; see above). Thus, an estimate for the all cancer hazard ratio adjusted for missing cause of death and for unascertained deaths would be 1.023 x 0.96 = 0.982. Similar cal- culations yield estimates for adjusted hazard ratios of 1.094 for leukemia minus CLL and 2.232 for thyroid cancer. The effect of this adjustment is always to re- duce the size of the original hazard ratios. However, because the same shrinkage factor~.9~would be applied uniformly to all hazard ratios, we have not dis- played adjusted hazard ratios separately. Stated more generally, the mortality ascertainment was slightly more com- plete for participants than for referents. This could have contributed to our find- ings of increased mortality risk among participants. However, we note that all

OCR for page 73
DISCUSSION 75 cause mortality was actually lower among participants than referents. Nonethe- less, increased ascertainment of deaths with cause could have contributed to the increased cause-specific risks of death among participants. Although the in- creased proportion of causes of death among participants is of a much smaller magnitude (roughly 3 percent t4.5% participants, 7.5% referents]) than the in- creases in leukemia risks we observed (as much as 49% in land series partici- pants), it is possible that ascertainment for certain causes was more differential than the overall difference would suggest. In this study, mortality ascertainment was hampered by the lack of a na- tionwide records system that covered the entire study follow-up period. For ex- ample, the Health Care Financing Administration of the Department of Health and Human Services tracks Medicare and Medicaid benefit claims, but its data- base developed for reasons other than epidemiologic research~oes not pro- vide useful information for the years before 1980. Our reliance on data from the National Death Index was limited to deaths since 1979, the year the index was begun. VA records can give information only about those veterans who seek benefits from the VA. Difficulties remain even when records are available. For example, the cod- ing for cause of death on death certificates is not necessarily uniform across geographic regions or time periods, or across the various groups of personnel responsible for choosing which cause, among the overlapping possibilities, to formalize on a death certificate. Statistical Power Veteran concern about radiogenic cancer was a major impetus for this re- search. That leukemia, the cancer that is most consistently linked with radiation, is fairly rare is fortunate overall but presents an obstacle to a study of this kind. Only a study cohort four times the size of the one available would have been likely to identify the observed leukemia risk as statistically significant. The sample size presently available does not provide sufficient power to achieve statistical significance for risks of the magnitudes we observed. Other Possible Confounding Factors Military Service Characteristics Within the military, most personnel serve for a discrete time period and then proceed into the civilian arena of work and life. Others choose a military career and remain in the service. The two sets of personnel may have different charac- teristics. Career military personnel, similar to many but not all occupational groups, must maintain reasonably good health to remain in the military, making them a healthy cohort. However, if one's military occupation involves, for ex

OCR for page 73
76 THE FIVE SERIES STUDY ample, radiation exposure, longer service could involve higher cumulative doses. Had individual-level data been available on length of service and job categories (as proxies for potential hazardous exposures), we might have been able to identify differences between the participant and referent cohorts, if any, that may have confounded associations between participation and mortality. Other Lifetime Radiation Exposure We have no information on other lifetime radiation exposure to members of either cohort either before or after the time period of the atomic tests. Sources of dose include background radiation, medical procedures (diagnostic and thera- peutic), and occupational practices (civilian and military). Only if additional exposures were substantially unequal in the two cohorts could they create a bias. Contributing and Associated Causes of Death We did not analyze data on associated causes of death (i.e., those noted on the death certificate in addition to the underlying cause). In the case of leukemia minus chronic lymphoid leukemia, this was not an important shortcoming, since only two leukemia minus CLL deaths were listed as associated and not under- lying causes of death. There may have been other mortality outcomes, however, for which an analysis of associated causes would be more fruitful. Inadequate Dosimetry Although the oversight committee concluded that the dose data in their cur- rent form were unsuitable for epidemiologic analysis (IOM, 1995), it also con- cluded that carefully done custom dose reconstructions performed anew for se- lected participants using consistent methods could provide usable dose data. Such custom dose reconstructions, however, would be prohibitively expensive to carry out for the entire cohort of participants. A more efficient way to make use of custom dose data would be to undertake a further study using a nested case-control design (Rothman and Greenland, 1998~. Briefly, leukemia deaths among participants could be selected for study, along with a randomly sampled control group (also of participants). Radiation dose would be estimated for eve- ryone in the study using custom dose reconstructions, and the pattern of radia- tion dose among the leukemia deaths could be contrasted to the pattern among the selected participant controls. In particular, dose-response analyses could be undertaken using such a study design. Other research options include identifying incident cases of cancers. Such information would assist in understanding the association of radiation with non- fatal cancers. The absence of a national cancer registry would hamper any study

OCR for page 73
DISCUSSION 77 of cancer incidence or prevalence. Alternative sources of information could in- clude geographically specific or disease-specific registries, health insurance claim data, and treatment data sources. DISCUSSION The data presented in Chapter 11, based on more than 5 million person- years of mortality follow-up, represent one of the largest cohort studies of mili- tary veterans ever conducted. Overall, no statistically significant differences in all-cause, all-cancer, or leukemia mortality between participants and referents are evident, although the participant risk of leukemia death is 14 percent higher than the referent risk. Across broad categories of noncancer deaths, participants and referents had the same mortality risk, except for death due to external causes, for which par- ticipants had a significantly higher risk (HR = 1.08; 95% C.I. 1.02-1.16; see Table 11-2~. Neither information about the nuclear tests nor current under- standing of radiobiology helps us to explain this observed higher risk. Similar estimates of excess mortality due to external causes, however, have been found in the study of British nuclear test participants (relative risk 1.06 for the initial follow-up through 1983 and 1.03 for 1984 through 1990; Darby et al., 1993b), New Zealand nuclear test participants (1.10; Pearce et al., 1997), as well as other military populations in Vietnam and the Persian Gulf (Kang and Bullman, 1996; Thomas et al., 1991; USDHHS, 1987; Watanabe and Kang, 1995~. Conversely, participants in Operation CROSSROADS had a lower risk for deaths due to accidents relative to their referent group (HR = 0.98) (Johnson et al., 1996~. In the following section of the report, we include discussion of the cancer findings that (1) relate to leukemia, a predetermined analytic endpoint; (2) are of interest because they relate to organs that are known to respond to radiation or have been identified in other studies of atomic test participants thyroid and lung cancers; or (3) are statistically significant nasal and prostate cancers. Leukemia For all leukemias and leukemia minus CLL we found increased, not statisti- cally significant, hazard ratios. Other studies of atomic veterans provide mixed evidence for radiogenic leukemia. In our earlier mortality study of Operation CROSSROADS participation (Johnson et al., 1996), we reported a not statisti- cally significant increase in leukemia deaths among participants relative to ref- erents. However, Darby and colleagues (1993a,b) found a statistically significant Based on current expert understanding of radiogenicity, we define leukemia throughout this chapter as ICD-9 codes 204 to 208 excluding 204.1, chronic Iymphoid leukemia. Other use of the term leukemia is noted in the text.

OCR for page 73
78 THE FIVE SERIES STUDY increase in leukemia deaths (RR = 1.75; 90% CI 1.01-3.06) among British par- ticipants in nuclear tests in Australia and the Pacific. The rate was higher (RR = 3.45; 90% CI 1.50-8.38) when limited to the earlier years of follow-up (Derby et al., 1988a,b). Similarly, Pearce and colleagues (1996, 1997) found a statisti- cally significant increase in leukemia deaths (RR = 5.59; 90% CI 1.04-41.7) among New Zealand test participants. Arguing against a radiogenic cause for the leukemia excess among Ameri- can atomic test series veterans is the finding from Watanabe and colleagues (1995) that highly exposed (>1,000 mrem) participants had no significant excess leukemia mortality. However, Watanabe's study used only sea series nuclear test participants and dose measures that the IOM advisory committee found to be inappropriate for epidemiologic analysis. We too found no increased risk of leu- kemia among sea series participants (HR = 0.92; 95% CI 0.67-1.27~. We found that land series participants, relative to land series referents, were at a statistically significant increased estimated risk of death due to leukemia (HR = 1.49; 95% CI 1.04-2.13~. For all causes of death, sea series participation was associated with a statistically significant lower mortality risk; land series participants and referents were essentially equal. However, we cannot rule out the possibility that the excess leukemia in land participants is due to chance. Although the data are far from definitive, our findings are broadly consis- tent with a radiogenic basis for the excess leukemia deaths observed. The rela- tive risks for leukemia deaths were highest for two acute leukemia subtypes- lymphoid leukemia excluding chronic lymphoid leukemia and myeloid leukemia excluding chronic myeloid leukemia (Table 11-4~. However, neither the pattern of leukemia deaths by follow-up period nor the pattern by age at test series par- ticipation serves to strengthen the evidence of a radiogenic relationship, al- though we had limited statistical power to analyze such patterns. We also made only preliminary investigations of leukemia minus CLL risk patterns by branch, paygrade, and participation status (see Appendix E). Thyroid Cancer Although thyroid cancer is one of the four cancers (leukemia, lung, and breast cancers are the other three) with strong evidence for radiation risk (Boice, 1996), the evidence relates almost entirely to childhood exposure. In fact, available evi- dence suggests that the adult thyroid gland is relatively insensitive to induction from radiation exposure (Hall et al., 1996; IOM, 1999, Thompson et al., 1994~. Although there were only 18 thyroid cancer deaths observed, we looked at the hazard of thyroid cancer deaths by age at first nuclear test participation. The results were not consistent with what we expected based on the literature: the hazard ratio (of participants relative to referents) was higher for those age 22 and older than for those less than age 22 at the time of participation. Explanations of the apparent increased rate of thyroid cancer deaths among participants aside from a radiation effect include increased surveillance among

OCR for page 73
DISCUSSION 79 participants, chance, and differential ascertainment of causes of death for the par- ticipant and referent deaths. In considering the possibility of increased detection of thyroid cancers among participants, we note the possible incentive for participants to seek screening tests more aggressively out of both a concern about prior radia- tion exposure and the knowledge that thyroid cancer is compensable under VA regulations. More cases identified could result in more diagnoses noted on the death certificate. Because thyroid cancer has a relatively low (10 percent) fatality rate, a few additional found cases could influence the study findings. Neither the Darby nor Pearce team found an increased risk of death due to thyroid cancer; Watanabe et al. did not present data on this site. However, John- son and colleagues reported an HR of 3.48 for thyroid cancer mortality among Operation CROSSROADS participants, not statistically significant, but still the highest relative risk reported. Lung Cancer Although evidence for the radiogenicity of lung cancer is strong (Boice, 1996), the well-documented association between smoking and lung cancer and our lack of data on smoking-make an interpretation of any association prob- lematic. Indeed, there is a particular interest in the interaction of the effects of tobacco smoke and radiation (Mettler and Upton, 1995~. We found no evidence of an increased risk among participants of death due to lung cancer (FIR 0.99; 95% CI 0.93-1.06~. This is consistent with the nonsig- nificant findings of all the other follow-up studies of nuclear weapons test par- ticipants: Darby and colleagues (1993a,b), 0.85 (90% CI 0.73-0.99~; Pearce and colleagues (Pearce, 1996; Pearce et al., 1997), 0.94 (90% CI 0.45-1.84~; Wata- nabe and colleagues (1995), 1.16 (95% CI 0.66-2.05) in the high-exposure group and 1.07 (95% CI 0.83-1.38) overall; and Johnson and colleagues (1996), 1.05 (95% CI 0.96-1.14~. Nasal Cancer We found an excess risk of mortality attributed to nasal cancer in this study, but other studies have not reported similar results. For example, no increase in nasal and pharyugeal cancer has been seen in Japanese atomic bomb survivors (Schull, 1995~. In their textbook on medical effects of ionizing radiation, Mettler and Upton (1995) listed the nasal sinuses as having low susceptibility to radia- tion-induced cancers compared to other sites listed as high or moderate. Johnson et al. (1996) did not look specifically at nasal cancers in their study of mortality associated with participation in the CROSSROADS nuclear test series. How- ever, going back to that dataset, we find a hazard ratio of 6.70 (95% CI 0.82- 54.49~.

OCR for page 73
80 THE FIVE SERIES STUDY Prostate Cancer Prostate cancer is not generally thought to be caused by radiation (Mettler and Upton, 1995), and no increase in prostate cancer has been seen among Japa- nese atomic bomb survivors (Schull, 1995~. Although data from the ankylosing spondylitis cohort (Derby et al., 1987) pointed to an early excess of prostate cancer, the authors noted that prostate cancer and ankylosing spondylitis can be confused, due to the presence of back pain from prostate cancer metastases to the spine. In addition, there have been a number of studies of occupational co- horts, with varying results (see Mettler and Upton, 1995; National Research Council, 1990~. Studies of nuclear weapons test participants have also yielded varying re- sults. Darby and colleagues (1993a,b) found a relative risk of 0.93 (90% CI 0.62-1.41~; Pearce and colleagues (Pearce, 1996; Pearce et al., 1997), 0.35 (90% CI 0.02-2.084; Watanabe and colleagues (1995), 1.46 (95% CI 0.3~6.31) in the high-exposure group and 1.41 (95% CI 0.71-2.80) overall; and Johnson and colleagues (1996), 0.77 (95% CI 0.61-0.97~. Thus, except for Johnson, who found a statistically significant deficit in prostate mortality risk, and the present study, which found a statistically significant excess risk (HR 1.20; 95% CI 1.03- 1.40), all other studies of nuclear weapons test participants have found no statis- tically significant excess or deficit in risk. Given the generally negative prostate findings reported in other studies, we urge caution in the interpretation of our findings. Moreover, a complicating factor in the study of prostate cancer is the large proportion of undiagnosed prostate cancers (Mettler and Upton, 1995), which could have affected our re- sults. Specifically, participant concerns about the possibility of cancer being caused by their participation may have led to more intensive follow-up, with a concomitant increase in prostate cancer discovered, and subsequently, in re- ported deaths due to prostate cancer. CONCLUDING COMMENTS Having described the difficulties faced in carrying out this study most of which are shared by other studies involving insufficiently recorded exposure and endpoint information, we here recapitulate some of its general advantages. First, in contrast to the earlier study by the Medical Follow-up Agency (Robinette et al., 1985), we now have more confidence that the five series participants have been properly identified (see Appendix D). The present study also makes use of a military referent cohort, rather than relying solely on standardized mortality ratios based on U.S. population controls. There is also a longer mortality follow- up period. In comparison to other studies of nuclear test participants, the Five Series Study's inclusion of more than 68,000 participants surpasses in size any previous research.

OCR for page 73
DISCUSSION 81 It is unlikely that another cohort study of this type and magnitude would provide more precise answers than this, because any atomic veteran study of this kind would face the same methodologic problems namely, inadequate expo- sure data and imperfect mortality ascertainment that we encountered in this Five Series Study. Other research strategies, using better-defined dosimetry data, might be those that focus on specific diseases and more detailed individual-level exposure information. The size, length of follow-up, and persistence of data collection efforts in- volved in this Five Series Study have helped to assure us that the findings we report are valid. The weak associations observed and the varied consistency with other studies, however, make interpreting these findings difficult. We can state that the participant group as a whole did not experience wide- spread early death. Even for leukemia, for example, there were an estimated 25 excess deaths in the participant cohort. That might be a comfort to those veter- ans who are not sick and to their families. The report findings do not rule out, however, possible increased risk among distinct subgroups of test participants.

OCR for page 73
THE FIVE SERIES STUD)Y PREP UBLICATION COPY FOR PUBLIC RELEASE I IAM OCTOBER 20, 1999 CONCLUDING COMMENTS Having described the difficulties faced in carrying out this study most of which are shared by other studies involving insufficiently recorded exposure and endpoint information, we here recapitulate some of its general advantages. First, in contrast to the earlier study by the Medical Follow-up Agency (Robinette et al., 1985), we now have more confidence that the five series participants have been properly identified (see Appendix D). The present study also makes use of a military referent cohort, rather than relying solely on standardized mortality ratios based on U.S. population controls. There is also a longer mortality follow-up period. In comparison to other studies of nuclear test participants, the Five Series Study's inclusion of more than 6S,000 participants surpasses . . - in size any previous researc n. It is unlikely that another cohort study of this type and magnitude would provide more precise answers than this, because any atomic veteran study of this kind would face the same methodologic problems namely, inadequate exposure data and imperfect mortality ascertainment-that we encountered in this Five Series Study. Other research strategies, using better-defined dosimetry data, might be those that focus on specific diseases and more detailed individual-level exposure information. The size, length of follow-up, and persistence of data collection efforts involved in this Five Series Study have helped to assure us that the findings we report are valid. The weak associations observed and the varied consistency with other studies, however, make interpreting these findings difficult. We can state that the participant group as a whole did not experience widespread early death. Even for leukemia, for example, there were an estimated 25 excess deaths in the participant cohort. That might be a comfort to those veterans who are not sick and to their families. The report findings do not rule out, however, possible increased risk among distinct subgroups of test participants. PP\CHARllBR 12 8