Document nX6GQg80KQGYDe5kaRb2vkDm

/ Paper MORTALITY OF WORKERS AT THE HANFORD SITE: 1945-1986 Ethel S.Gilbert,*Ellen Omohundro,+Jeffrey A. Buchanan,* and Nancy A. Hoke? tbsrract-updated analyses of mortality of workers at the Haford site provide little evidence of a positive correlation ,I cumulative occupationalradiation dose and mortality from .&emia and from all cancer except leukemia. Estimates of :he excess relative risk per 10 mSv were negative for both &ease categories, but these estimates are consistent both sith no risk and with estimates obtained through extrapolaion from high-dose data. For all cancer except leukemia, the Jpperlimit for a two-sided 90% confidenceintervalwas about 15times the prediction of the BEIR V model, but several mes the estimate recommended by the ICRP 60 committee. For leukemia, the comparable upper limit was very close to tat predicted by either BEIR V or ICRP 60. The all-cancer isk estimate, from a recent report on updated analyses of uta for Oak Ridge National Laboratory workers, was aongly rejected based on the Hanford data. Of 24 specific mcer categories evaluated, only cancer of the pancreas and Hodgkin's disease showed positive correlations with radiation !ose that approached statistical significance with one-tailedp dues of 0.07 and 0.04, respectively; these correlations are nterpreted as probably spurious. For multiple myeloma, for ahich a correlation was reported previously, the p value was 1.10. However, a significantcorrelation (p< .OS) was obtained *hen analyses were expanded to include deaths with multiple n!eloma listed on the death certificate but not considered to Y the underlying cause, when analyses were expanded to =dude deaths occumng in Washington State during the time Xnod 1987-1989, or when a 2-y latency period (instead of 10-v) was assumed. Hdth Phys. 64(6):577-590; 1993 Key words: mortality; cancer, radiation dose; risk estimates INTRODUCTION Oak Ridge National Laboratory (ORNL) workers was found to be about 10 times the estimate obtained through linear extrapolation from Japanese atomic bomb (A-bomb) survivor data (Wing et al. 1991). Be- cause this finding was different from the result based on earlier data, and because the effect was strongest when a minimal latent period of 20 y was assumed, Wing et al. (1991) suggested that the longer follow-up period for this cohort might account for the seeming discrepancy between the ORNL findings and those from several other worker studies. Results of analysesof mortality of Hanford workers for the time period 1945-198I were reported by Gilbert et al. (1989a) and provided little evidence of a correlation of radiation dose and overall cancer mortality. The purpose of this paper is to update these analyses to include an additional5 y (1982-1986)and, particularly, to investigate' whether the longer follow-up period might modi@ results in a manner similar to that observed in ORNL workers. An additional objective is to provide a more rigorous comparison with estimates obtained through extrapolation from high-dose data; at the time the 1989 paper was prepared, risk estimates based on revised Japanese A-bomb dosimetry were not available, and the report of the BEIR V committee (National Academy of Science 1990) and recommendations of the International Commission on Radiation Protection (1991) had not been published. The earlier paper also included supplementary analyses including deaths occumng in the state of Washington for 19821985. In this paper, similar supplementary analyses include Washington deaths for 1987-1989. k x r s of several studies of nuclear workers have been -:Ported (Beral et al. 1985, 1988; Smith and Douglas 986: Wing et al. 1991; Kendall et al. 1992), and for '.?e most part have indicated effects that are consistent -&lth risk estimates that have been obtained through Trapolation from highdose data. Recently, however, ~n estimate based on updated data (through 1984) on 'j. -* P$cific Northwest Laboratory, P.O.Box 999, Richland. WA .'>?: Hanford Environmental Health Foundation, P.O.Box 100. ',chiand. WA 99352. ' (.\lanuscript received 30 June 1992; revised manuscript received December 1992, accepted 15 January 1993) '-'917-9078/93/$3.00/0 'dpyright 0 1993 Health Physics Society THE STUDY POPULATION The study population, the radiation exposure data, and the methods used to determine vital status and cause of death, were described by Gilbert et al. ( I989a), which is referred to hereafter as the "1989 paper" or as "1989 analyses." These elements are described only briefly here, with emphasis on modifications that have been made. As in the 1989 analyses, the study population consisted of all workers, other than those employed only in construction activities, initially employed at the Hanford site from 1944-1978 by U.S. Department of Energy contractors. Whether a worker was a member 577 578 Health Physics of the study population was determined through the use of employment histories, which have been maintained by the Hanford Environmental Health Foundation since the mid- 1960s when the study was initiated. Since the 1989 publication, extensive efforts have been made to refine the database used for statistical analyses, and these efforts are described in detail by Gilbert et al. (1992). Here it is noted only that these efforts resulted in the removal of 265 workers primarily because current records indicate they did not actually start employment at Hanford (they had appeared in the employment histories because they had participated in preemployment physical examinations), and the addition of 456 workers who did not have employment history records at the time our earlier data base was prepared. Some correctionsto dosimetry and employment data used in previous analyses were also made. To investigate the possibleeffectsof these modifications,the main analysis presented in Table 3 of the 1989 paper was redone, using identical methodologybut with the modified data. The results differed very little from the published results based on old data. Two Hanford workers who were involved in incidents in which more than 250 mSv was received on a single occasion were also excluded from the study population used in analyses in this paper; these workers died of circulatory diseases in 1987 and 1988, respectively. Unlike analyses in the 1989paper, internal comparisons by dose in this paper were restricted to workers employed at Hanford for at least 6 mo. This restriction was made for comparability with past and future analyses of combined mortality data on workers at Hanford, Oak Ridge National Laboratory, and Rocky Flats Weapons Plant (Gilbert et al. 1989b), where it was judged more appropriate to exclude extremely shortterm workers. Vital status Vital status was established by periodic searches of earnings and benefits files of the Social Security Administration (SSA) (1944-1986), by the use of the U.S. National Death Index (NDI) (1979-1986), and through linkages (of Social Security numbers and names) with June 1993, Volume 64, Number 6 the State of Washington vital statistics computen~~, files for 1968-1989 and the State of California VIS statisticscomputerized filesfor 1960-1986.In our !gkpaper, we determined that the proportion of deaths Ithe State of Washington (where mortality ascenal,. ment should be close to 100%complete),also identlfi, through the SSA or NDI, was 97%. A study to gal: better information on the adequacy of follow-up : currently underway. For now, we note that the intenally based comparisons emphasized in this paper ar: less likely to be biased because of inadequate follon-u: than comparisons of death rates with those of ttlr general population. The National Center for HeallStatistics codes all medical conditions recorded 0: death certificates and also assigns the underlying cauLr of death. External radiation exposure data External dose estimates are expressed in milliseji. ert (mSv). Because exposures were primarily to ]Oh linear-energy-transfer gamma radiation, the absorb; dose (mGy) and dose equivalent (mSv) were similar fc: most workers. Thus, we refer to "dose" even thoug "dose equivalent" is more accurate terminology. A;. though some members of the study population als, performed constructionwork either prior or subsequer: to their employment in operations work, data on thej, doses had not been extracted for files used to perfor 1989 analyses of the Hanford data; thus, the 196analyses were based only on dose received in operatior. work. In this paper, dose incurred while performin; such construction work was included but, for comparison with previous results, supplementaryanalyses ne? performed with this dose excluded. Table 1 shows the number of subjects in the stuC:. population by gender, by whether or not they we:. monitored for external radiation, and by whether Y not they were employed at the Hanford site for at l e 5 6 mo. It also shows the total person-Sv and the a v e n s dose accumulated through 1985by monitored work? Of the 861 person-Sv recorded for the total populatio? 26 person-Sv were received in construction \wrr About 99% of the total dose was received by ~ ~ r k ? Table 1. Number of workers, number of monitored workers, total dose, and average cumulative dose through 1985 by gender and length of emulovment. Number of workers Number of monitored workers Total person-Sv Average cumulative dose" (mSv) All workers Male Female Employed 6+ mo Male Female Employed <6 mo Male Female 44,154 3 1,486 12,668 36,439 25,998 10.44 I 7,715 5,488 2,227 36,97 1 28,086 8,885 32,643 24,672 7.97 1 4.328 3,414 914 861.O 814.5 46.5 854.1 808.1 46.0 6.9 6.4 0.5 23.3 29.0 5.2 26.2 32.8 5.8 1.6 1.9 0.5 Based on workers monitored for external radiation. 7 Mortality at the Hanford site 0 E. S. GILBERT ET AL. 579 ;gployed at least 6 mo, and 95% was received by male ;orkers. The additional 5y of follow-up increased the .Jtal person-year-Sv (calculated with a 10-y lag) by )Dout 50% (from 5,300 to 8,000). %ioeconomic status Analyses in this paper were based primarily on 2ternal comparisons by level of radiation exposure. A vssible source of bias in these comparisons is that lose workers performing jobs involving radiation exysure may have different socioeconomic characteriscs than workers performing otherjobs. For this reason, yevious analyses of the Hanford data have made use )f a general occupational category determined from 3ureau of Census codes (1971), and some studies of -uclear workers in the UK have adjusted analyses for m a l class ( B e d et al. 1985, 1988). Since our 1989 paper, job category data for HanJrd workers have been used to define a socioeconomic mable that is similar to the social class variable used -Ithe UK studies, which was also based on job category 2ta. We make use of four general socioeconomic catsones, as defined in Table 2. Workers were assigned I J the general category in which they had spent the ingest period of time from initiation of Hanford em:ioyrnent through the end of 1985. About 72% of the :udy population spent their entire employment period -I Hanford in a single general category, while 88% x n t at least 75% of their employment period in a ngle category. From Table 2, it can be seen that about f -'cC of the total personSv was received by workers 1 -signed to the skilled and semiskilled manual category, thile only about 3.5% was received by workers assigned the clerical and unskilled categories. All four catemes included substantial numbers of workers with -:? little occupational dose; the proportionsof workers tho had accumulated total doses e 1 0 mSv were 6596, -'647.%, and 84% for the four respective categories ' Xmn in Table 2. STATISTICAL METHODS The methods used to analyze these data have been described elsewhere (Gilbert and Buchanan 1984; Gilbert 1989; Gilbert et al. 1989a and b); a nontechnical discussion of methods for analyzing data from studies of nuclear workers is given by Gilbert (1991). Here, we provide only sufficient detail to indicate the specific approach used in this paper. External comparisons were made by calculating standardized mortality ratios (SMRs) based on age- gender-calendar year-specific death rates for the U.S. using software developed by Monson (1974), which currently includes U.S.death rates through 1985. How- ever, because good dose measurements were available, and because dose varied considerably among workers, as in previous analyses, we emphasize analyses based on internal comparisons of death rates by level of radiation dose. Such comparisons are less subject to bias, and more likely to detect risks resulting from radiation exposure than are external comparisons. The trend test statistic used to evaluate the association of cumulative radiation dose with death from several causes is sensitive to an increase in death rates with increasing exposure. The trend test was based on individual dose values rather than scores for dose categories. In addition, observed and expected deaths were calculated for five dose categories using the MantelHaenszel method (Mantel and Haenszel 1958). These calculations were made using the computer software MOX (Gilbert and Buchanan 1984). In all cases in which the test statistic exceeded 1.645, the one-tailed p value was estimated using computer simulations based on 10,000 samples, as described in Gilbert (1989). Because of the highly skewed exposure distribution, this procedure provides a more accurate assessment than would the usual asymptotic normal approximation. These simulations were based on eleven dose categories (0-, 5-, lo-, 20-, 50-, loo-, 150-, 200-, 300-, 400-, and 500+ mSv) rather than on the individual dose -out& - - Table 2. Number of workers, total dose, and average cumulative dose through 1985among monitored workers employed at least 6 mo at the Hanford site by longest general socioeconomiccategory. General socioeconomic category Corresponding social class" Number of workers Total person-Sv Average cumulative dose (mSv) 1 Professionaland technicalb Clerical' I and I1 IIINM 14,028 6,487 248.2 23.4 17.7 3.6 t Skilled and semiskilled man- IIIM and IV 10,875 i uald i Unskilled manual' V 1,253 * These are the correspondingsocial classes used in the UK. 576.0 6.5 53.0 5.2 Professional, technical, and managerial workers with Bureau of Census Codes 001-245, and 265. NM = nonmanual.Clerical workers with Bureau of Census Codes 280, and 301-395. M = manual. Skilled and semiskilled manual workers with Bureau of Census Codes 401-575, 590,600-726, 755, 912- 916, and 961-995. The Bureau of Census codes were expanded for use with the Hanford study. Special codes developed were 590 for supervisory manual workers, 69 1 for radiation monitors, 692 for process operators, 693 for reactor operators, 694 for utility operators (unspecifiedas to type), 695 for power operators, and 696 for other operators. Unskilled workers with Bureau of Census Codes 740-954. except 912-916. I ? 580 Health Physics values. Latency time periods were allowed for by conducting analyses based on cumulative dose calculated with 2-, lo-, and 20-y lags. For all analyses, stratification variables included gender, age (25-29, 30-34, single-yearintervals forages 35-80, 80-84, and 85+), and calendar year (19451949, 1950-1954, 1955-1959, 1960-1964, 1965-1969, 1970-1974, 1 975-1979, 1980-1 98I , 1982-1984, and 1985-1986). Unless stated otherwise, analyses also included stratificationon number of years monitored (14,5+) and on the four socioeconomiccategories shown in Table 2. Number of years monitored was selected because of a demonstrated effect of this variable in past analyses of the Hanford data (Gilbert 1989), and for consistency with previous analyses. Analyses of the newly defined socioeconomic categories indicate that this variable is an important predictor of risk. For example, relative risks (with 90% confidence intervals)for all cancer, expressed relative to the professional and technical category, were 1.1 (0.9, 1.3) for clerical workers, 1.3(1.1, 1.4)for skilled and semiskilled workers, and 1.5 (1.2, 1.8) for unskilled workers. These results, along with the association with radiation dose shown in Table 2, indicate that socioeconomic status is a potentially important confounder. Workers were entered into the analysis at the beginning of the year following the initial employment year plus five, or the first year of monitoring, whichever occurred later. For example, a worker who initiated employment in 1950, and received doses of 10 mSv, 5 mSv, and 25 mSv in 1950, 1951,and 1952,respectively, would begin contributing person-years at the beginning of 1956. With a 10-y lag, this worker would be assigned 0 mSv for each of the years 1956-1960, 10 mSv for the year 1961, 15 mSv for the year 1962, and 40 mSv for 1963 and all succeeding years of follow-up. The first 5 y of follow-up were excluded to allow for the possibility of an especially strong healthy worker effect present during this time period, and because cancers occurring during this period are highly unlikely to be the result of occupational exposure. It is noted that when doses are lagged for 5 or more years (as in most analyses in this paper), excluding the first 5 y is equivalent to stratification on follow-up period with two strata, <5 y and 2 5 y; this is true because all workers have zero dose for the first 5 y of follow-up and, thus, there is no information available for internally based comparisons. With the shorter 2-y lag period, leukemia is the disease of primary interest;there were three leukemia deaths duringthe first 5 y of followup, but none had doses exceeding 5 mSv. The choices just given are those that are emphasized on an a priori basis, but analyses based on several alternative choices are also presented. For leukemia, we emphasize analyses based on the 2-y lag period, while for other cancers, we emphasize the 10-y lag period; these are the choices used in most radiation risk assessment efforts. Estimates and confidence limits for cancer risks June 1993, Volume 64, Number 6 per unit of dose were calculated based on a rnoit.1 which the relative risk was assumed to be of the iorT 1 + pZ, where Z is the cumulative dose, and J ,. expressed as a percent increase per 10 msv. Thrs5 estimates were obtained by maximizing the likelih~~, function with confidence limits obtained by use of lf:l score statistic(Gilbert 1989).The score statistic is CaSirto apply if based on grouped rather than indi\i& dose measurements; for this purpose, we used t h e : dose categories previously indicated. For leukemia. I simulation procedure was used to determine confiden;; limits as described in Gilbert (1989). Ninety perter: confidence intervals are emphasized, but some resui:. are presented with both 90 and 95% confidence inre:. vals. RESULTS Comparison with U.S. rates In Tables 3 and 4,SMRs are presented for mor, tored and unmonitored male and female worker workers employed <6 mo were included in these cz culations. The all-cause SMR was 0.82, compared I 0.79 reported in 1989. The SMR for all mahgna: neoplasms was 0.86, compared with 0.85 reported I1989. The SMRs for diseases of the musculoskeletal s! tem, for cancer of the pancreas in unmonitored mal: and for all other solid tumors in unmonitored male were elevated based on earlier data and remained so .- these updated analyses. The previously elevated S51F for lung cancer in females was no longer elevated Tr. resultsin Tables 3 and 4generallyshow a strong healtr worker effect. The highest SMR for specific cancers .- all workers was 1.03 for the "all other solid tumo-' category. Radiation exposure analyses Table 5 shows detailed results of analyses rh:. examine the trend in death rates from many diseav with increasing exposure. Table 6 shows relative n.r;.. by dose category for cancers, noncancers, leukerniand multiple myeloma. For noncancers, the directicof the correlation of radiation exposure and mortal:: was negative with a stronger correlation for the 2-! k than for the 10-y lag. Deaths from accidents. poiso:ings, and violence showed a particularly strong ne_ea1''*. correlation. The categories of respiratory diseases (t' cluding pneumonia) and cirrhosis were examined. provide insights regarding possible differential smokl:: and drinking habits by exposure category. Neither <these disease categories was strongly correlated-::'1 radiation dose. The direction of the correlations for both all-n' cers and leukemia was negative whether analyses \e based on 2-y or 10-y lags. In 1989 analyses, the C 6 : ' lation for all cancers in females was positive and 5'proached statistical significance, but this correlar!:. was no longer close to statistical significance. Res: .' Tab. mor Cau! - All c Infec 13 Mali; Benil ( 21 Endc ti0 Disei or4 Men) Dise; ser Dise: (35 Disei (4f Diser (52 Diser (58 Dim tiss Disez ten 731 Syw (78 Accic (80 -Causc a The year-! on th ICD The! 1950. in the * Sign Mortality at the Hanford site 0 E. S.GILBERT ET AL. Table 3. Standardized mortality ratiosa (SMRs)and observed deaths (OBS)for major causes of death in monitored and unmonitored white male and female Hanford Site workers from 1945-1986. Males Females All workers u:oT,eodni-Monitored Unmonitored Monitored Cause of death (eighth revision ICDb code) SMR OBS SMR OBS SMR OBS SMR OBS SMR OBS All causes of death 0.81 6,678 0.90 1,441 0.76 Infective and parasitic disease (000- 0.28 30 0.25 6 0.27 136) Malignant neoplasms (140-209) 0.86 1,508 0.90 280 0.81 Benign and unspecified neoplasms 0.54 12 0.74 3 0.00 (210-239) Endocrine, metabolic, and nutri- 0.91 126 0.94 24 0.64 tional diseases (240-279)' Diseases of blood and blood-forming 0.42 8 0.99 4 0.94 organs (280-289) Mental disorders (290-315)' 0.7 1 33 1.21 9 0.89 Diseases of the nervous system and 0.86 67 0.50 7 1.02 sense organs (320-389) Diseases of the circulatory system 0.82 3,377 0.87 729 0.65 (390-458) Diseases of the respiratory system 0.86 449 1.01 106 1.03 (460-519) Diseases of the digestive system 0.67 250 1.06 70 1.03 (520-577) Diseases of genito-urinary system 0.57 64 0.84 22 0.21 (580-629) Diseases of skin and subcutaneous 0.16 1 0.79 1 0.00 tissue (680-709) Diseases of the musculoskeletal sys- 1.30 18 1.21 3 2.06` tem and connective tissue (710- 738) Symptoms and illdefined conditions 0.5 1 45 0.99 15 0.63 (780-796) Accidents, poisonings, and violence 0.82 591 1.09 125 1.09 (800-999) Cause unavhable 85 34 713 0.82 620 0.82 9,452 3 0.39 4 0.28 43 224 0.87 183 0.86 2,195 0 1.35 5 0.58 20 17 0.67 14 0.86 181 3 0.76 2 0.59 17 4 1.47 13 0.73 5 0.82 7 0.82 51 94 244 0.70 223 0.81 4,573 47 1.16 42 0.91 644 43 1.06 35 0.78 398 3 0.64 8 0.59 97 0 2.41 3 0.49 5 9 0.94 3 1.38` 33 6 0.29 69 1.35 22 2 OS7 58 0.90 26 68 843 167 a The SMR is the ratio of observed to expected deaths, where expected deaths are calculated from age-specificand calendar year-specific mortality rates for U.S.white males or females. The SMRs are corrected for those deaths with no certificates on the assumption that the distribution of causes is similar for those with and without certificates. ICD = International Classification of Diseases, Eighth Revision. `These SMRs are based on the years 1950-1986 since U.S. mortality rates for these categories were not available prior to 1950. For this reason, and because separate calculations for ICD codes 740-759 (congenitalmalformations) are not included `in the Monson sohare, the sum of deaths from specific causes is less than the total. Significantly elevated at the 0.05 level based on a one-tailed test. 58 1 1 'fseparateanalyses for cancers that have been linked ilth smoking (Doll and Pet0 1981) and those that have 10t been so linked were similar; although, the correla' `.onfor the smoking-related cancers was more positive t `hn for the remainder. 1 Twenty-four separate cancer categories were ana:zed. For both the 10- and 2-y lags, the numbers of %it.ve and negative trend test statistic were approxi?ately equal. It is evident that with 24 categories and --Y lag periods, some categories are likely to demon:rate individual one-tailed p values of 4.05by chance he. ~ i Multiple myeloma, which has previously been the type most strongly linked with radiation dose in did not exhibit a statistically signifiwith the 10-y lag (p = .IO), but did a correlation with the 2-y lag (p = .008). In 1989 analyses, the category of other female genital cancers (the combined categories of cancer of the cervix, uterus, and ovary) demonstrated a correla- tion of borderline statistical significance. With the up- dated data, the trend test statistics for this combined category were 1.13 for a 10-y lag and 0.66 for a 2-y lag, neither close to statistical significance. Table 5 shows a statistically significant correlation for Hodgkin's disease with both 10- and 2-y lags, and cancer of the pancreas shows such a correlation with the 2-y lag. Neither disease showed a statistically significant correlation in 1989 analyses, but cancer of the pancreas exhibited a statistically significant correlation in much earlier analyses (Gilbert and Marks 1979). Cancer of the liver shows a correlation of borderline statistical significance with a 2-y lag, but not with a 10- Y 1%. 582 Health Physics June 1993, Volume 64, Number 6 Table 4. Standardized mortality ratios" (SMRs)and observed deaths (OBS)for specific cancer types in monitored and unmonitored white male and female Hanford Site workers from 1945-1986. Males `::ti-Monitored Females :'-Monitored All workers Type of cancer (eighth revision ICDb code) SMR OBS SMR OBS SMR OBS SMR OBS SMR OBS Buccal cavity and pharynx (140-149) 0.8 1 42 0.66 6 0.00 0 1.08 3 0.76 51 Digestive organs and peritoneum ( 150- 0.87 414 0.80 72 0.74 46 0.75 37 0.84 569 159) Esophagus ( 150) Stomach (1 51) Colon ( I 53) Rectum (154) 0.84 35 0.56 0.84 70 0.46 0.88 141 0.71 0.72 35 0.41 4 0.00 8 0.82 21 0.70 4 0.52 0 1.75 6 0.66 19 0.84 3 0.85 3 0.80 42 4 0.77 88 18 0.84 199 4 0.67 46 Liver and gall bladder (155-156) 1.01 37 0.86 6 0.85 5 0.64 3 0.94 51 Pancreas ( 157) Respiratory system (160-163) Iarynx (161) Lung ( 162) Bone (170) Skin (172-173) 0.9 1 84 1.57d 26 0.85 I O 0.56 5 0.96 125 0.80 491 0.93 93 1.02 41 0.97 28 0.83 653 0.60 15 1.36 6 1.19 1 1.61 1 0.74 23 0.80 470 0.92 87 1.01 39 0.97 27 0.84 623 0.44 3 0.74 1 1.13 1 0.00 0 0.51 5 0.76 25 0.97 5 0.94 4 0.66 2 0.80 36 Female breast (174) All uterus ( 180-182) 0.95 60 0.89 42 0.92 102 0.33 7 0.99 17 0.63 24 Other female genital organs (183-184) 0.78 16 1.02 16 0.89 32 Prostate (185) Testes and other male genital organs 0.95 116 1.00 27 0.7 1 6 1.44 2 0.96 143 0.82 8 (186-187) Bladder (188) 0.68 34 0.67 7 0.36 1 1.34 3 0.69 45 Kidney (189) Eye (190)c 0.94 41 1.22 1.43 2 0.00 9 0.72 0 0.00 3 0.63 0 0.00 2 0.95 55 0 0.97 2 Brain and other central nervous system 0.96 48 0.52 4 0.78 6 0.53 3 0.86 61 (1 91-1 92) Thyroid (193)c 0.6 1 2 1.74 1 1.19 1 0.00 0 0.75 4 All other solid tumors (171, 194-199) 1.02 126 1.47d 31 0.81 16 0.80 12 1.03 185 All lymphatic and haematopoietic can- 0.96 158 0.76 22 0.92 22 1.01 18 0.93 220 cer (200-209) Lymphosarcomaand reticulosarcoma 1.04 32 0.37 2 0.92 4 1.48 5 0.98 43 (200)' Hodgkin's disease (201) 0.93 16 1.03 3 1.26 3 0.56 1 0.94 23 Multiple myeloma (203) Leukemia and aleukemia (204-207) 0.93 21 1.29 0.84 56 0.74 5 0.83 9 0.76 4 0.55 7 1.16 2 0.91 8 0.84 32 80 Other lymphatic tissue (202-203,208- 0.99 50 0.98 8 1.02 8 0.70 4 0.97 70 209)c `The SMR is the ratio of observed and expected deaths, where expected deaths were calculated from age-specific and calendar year-specific mortality rates for U.S.Caucasian males or females. The SMRs were corrected for those deaths with no certificates on the assumption that the distribution of causes was similar for those with and without certificates. ICD = International Classification of Diseases, Eighth Revision. These SMRs are based on the years 1950-1986 since U.S.mortality rates for these cancer types were not available prior to 1950. Significantlyelevated at the 0.05 level based on a one-tailed test. Table 7 shows excess relative risk estimates based on several categories of cancer, including the broad categories chosen for risk modeling by the BEIR V Committee (National Academy of Science 1990). Most estimates were very close to zero and, in all cases, confidence limits included zero risk. Confidence limits were especially wide for cancer in females because of the smaller number of females and the limited dose received by females(see Table 1). The upper confidence limits based on analyses of cancer linked with smoking were very close to those based on all cancers; analyses of cancers not linked with smoking are possibly less subject to bias than analyses of all cancers. Analyses comparing death rates in workers with two levels of confirmed plutonium depositions !<-` 74+ Bq) to those of other workers without such der - sitions were also performed. Both the approach a d 1.: results of these analyses were similar to those repoflr.: in the 1989 paper, with no evidence of a posl1``. correlation of plutonium depositions with cancer d C rates. The relative risks (with 90% confidence Ilm::. for all cancers of those with c74 Bq and those w`74+ Bq relative to workers with no plutonium dW-.''tions were 0.7 (0.4, 1.2) and 0.9 (0.5, 1.6), respect]\' Alternative radiation exposure analyses Certain choices in the manner that dose-resP@? analyses are conducted are necessarily arbitrap Table -mo. E AI1 cau Cause All can Malt Fem, Smo Resic Bucc EW ston COlO Rect Liver Gail1 Panc LarY Lunl Bont Fem, Cerv Ova1 Pros1 Blad, Kidr Braii sy! Thy1 Non. Ph Hdl Mull Chrc mi Leuk Leuk (2- All nor Circi Rep Pn Cirrt Extei Person, 'The ti standai with a ~ Expec gender, The SI Based 0.065 1; were es for mu -vpes in --ken -OBS 51 569 42 88 199 46 51 125 653 23 623 5 36 102 24 32 143 8 45 55 2 61 4 185 220 43 23 32 80 70 :ific and :ths with prior to s.Mortality at the Hanford site 0 E. GILBERT ET AL. Table 5. Results of analyses of external dose in monitored workers employed at the Hanford Site for at least 6 mo. Except where noted, this is based on a 10-y lag. Trend test statistic' Exposure lagged for: Observed and expected deaths by exposure category (mSv) Cause of death 1Oy 2 y e Obs/Expb 10Obs/Exp 50- 100- 200+ Obs/Exp ObsIExp ObsIExp All causes -0.90 -1.91 4,34 114,267.7 1,333/1,378.9 2 131227.9 1581169.8 1551155.7 Cause unavailable -1.28 -1.32 6 1159.0 22119.5 013.2 212.1 112.1 All cancer Male Female Smoking-linked cancers' Residual Buccal . Esophagus Stomach Colon Rectum Liver Gallbladder Pancreas Larynx Lung Bone Female breast Cervix and uterus ovary Prostate Bladder Kidney Brain and central nervous system Thyroid Non-Hodgkin's lym- phoma Hodgkin's disease Multiple myeloma Chronic lymphatic leuke- mia (CLL) Leukemia excluding CLL Leukemia excluding CLL (2-Ylag) -0.29 -0.29 0.02 0.10 -0.49 -1.45 0.14 0.18 -0.35 0.40 1.37 -1.27 1.59d -0.62 0.16 -0.68 -0.05 0.96 0.79 -1.15 -0.34 0.72 -0.90 -0.42 -0.85 1.80' 1.54' -0.34 -0.8 1 -0.50 -0.54 0.28 -0.0 I -0.67 -1.58 -0.38 0.37 -0.63 0.14 1.93' -1.30 2.36* -0.94 0.01 -0.70 0.58 0.29 0.6 1 - 1.51 -0.66 0.60 -0.66 -0.46 -1.14 2.38' 2.23' -0.45 -0.85 9751982.9 8161818.8 1591164.2 380/389.0 5951593.9 24/22.3 20118.6 43144.1 1031106.6 25122.9 15114.5 13110.8 5 1152.5 9/85 2561265.0 312.3 45146.5 315.0 12112.1 59161.7 17118.9 27123.3 28130.1 211.7 37139.1 14114.8 17117.0 6/56 27126.2 25123.7 3461334.6 3161310.7 30123.9 1541148.3 1921186.3 817.3 515.9 14112.7 36129.1 . 617.7 313.9 112.5 19118.3 313.0 1071104.2 010.4 715.3 310.8 1/1.5 32127.1 1118.7 217.9 16111.3 58160.1 55156.4 313.7 30126.9 28133.2 011.3 1/1.1 212.5 214.7 111.4 010.6 010.3 213. I 110.6 24119.2 010.1 110.9 010.2 010.2 514.4 111.3 211.4 112.4 111.0 17113.7 010.2 312.6 212.8 214.9 112.1 010.5 210.9 210.6 14111.7 14113.1 112.4 112.7 47144.9 43141.5 413.4 22120.7 25124.2 111.0 010.7 311.9 413.1 110.9 010.6 010.3 312.2 010.5 17115.0 010.1 111.0 010.06 110.2 213.4 111.0 311.2 211.7 010.1 111.8 110.3 110.6 010.3 111.7 312.0 40143.5 40142.7 010.8 19/20. I 21123.4 011.1 110.7 111.9 213.4 I/l.O 210.4 010.06 311.9 010.3 14114.6 010.03 010.3 o/o.oo 010.04 213.4 111.1 111.2 011.5 010.0 111.8 210.5 210.6 010.4 112.0 112.6 All noncancer Circulatory Respiratory excluding pneumonia Cirrhosis External -0.70 -0.48 0.10 0.46 -1.81 -1.78 -1.28 -0.16 0.39 -1.96 3,30513225.7 2,19312.120.6 194/208.2 86188.0 3741366.1 96511,024.8 1551164.6 1091122.8 1141110.1 6421699.7 1021113.0 76183.6 81177.0 96184.7 19112.8 8110.0 9110.3 22122.7 74175.9 613.5 212.8 11113.5 1119.9 312.0 216.6 Person-years 499.847 96,73 1 17,545 11,430 7,958 a The trend test statistic was calculated from individual doses. not the five exposure categories. It may be compared with a standard normal distribution to assess statisticalsignificance. However, statisticalsignificance may be exaggerated for diseases with a small number of deaths. See footnote d. Expected deaths were calculated from the experience of all workers in the study population, allowingfor age, calendar year, gender, number of years monitored, and general socioeconomic category. 'The smoking-linked cancers were respiratory cancer, buccal cancer, and cancer of the esophagus, pancreas, and bladder. Based on computer simulations, the one-tailed p values associated with the trend test with a IO-y lag were estimated to be 0.065 for cancer of the pancreas, 0.038 for Hodgkin's disease, and 0.10 for multiple myeloma. For the 2-y lag, these p values were estimated to be 0.056 for cancer of the liver, 0.026 for cancer of the pancreas, 0.029 for Hodglun's disease, and 0.030 for multiple myeloma. 583 5i4 * Health Physics June 1993. Volume 64. Number 6 Table 6. Relative risks" (with 90% confidence limits) by external dose category for monitored H,!.,: . . workers employed at least 6 m o (based o n a 2-y lag for leukemia and a 10-Y lag for other categories~. ' "2 Dose category (mSv) 0- All cancer I .oo All noncancer I .oo Leukemiab 1 .o -------- Multiple myelom;, I .o 10- 1.04(0.9, 1.2) 0.89 (0.8, 1.0) 0.8 (0.4, 1.6) 0.4 (0. I . I . 3 ) 50- 1.01(0.8, 1.3) 0.85 (0.7. 1.0) 0.3 (0.03, 1.6) 4.2 (0.7, 19) 100- 1.17 (0.9, 1.5) 0.83 (0.7, 1.O) 1.5 (0.4, 4.8) 5.9 (0.5.4I 200- 0.93 (0.7, 1.3) 0.96 (0.8, 1.2) 0.3 (0.02, 1.3) 21 (2.1.270) ,za The relative risks are the ratio of the risk for the indicated category relative to that of the lowest dose category ((1- I Excluding chronic lymphatic leukemia. Table 7. Excess relative risk estimates with 90%confidence intervals for several cancer categories (based on a 2-y lag for leukemia and a 10-y lag for other categories). Excess relative risk estimates [% (10 mSv)-'] 90%confidence 95% confidence intervals intervals All cancer -0.1%(<O,0.8%) (<O, 1.0%) Leukemia' - 1.1% (<O, 1.9%) (<O, 3.0%) All cancer except leukemia -0.0% (4,1.O%) (<O, 1.2%) Digestive cancer Respiratory cancer Female breast cancer Other cancer 1.1% (<O,3.1%) (<O, 3.7%) -0.0% (<O,2.4%) (<O,3.0%) -0.3% (<O,27%) (<O,39%) -0.9%(<O,0.7%) (eo, 1 . 1 %) All cancer in males All cancer in females -0.1% (eo,0.8%) (<O, 1.O%) 0.1% (<O, 15%) (<O,20%) Smoking-linkedcancersb Other cancers 0.1%(<O, 1.8%) (<O,2.2%) -0.3%(eo,0.8%) (<O, 1.1%) Excluding chronic lymphatic leukemia. The smoking-linkedcancers were respiratorycancer, buccal cancer, and cancer of the esophagus,pancreas, and bladder. determine the sensitivity of analyses to these choices, analyses based on several alternative approaches were conducted. Risk estimates and confidence limits for all cancers are presented in Table 8; additional details for two of the approaches are presented in Tables 9 and 10. Analyses in Tables 5-7 were based on the under- lying cause of death as has been done in previous Hanford analyses and in most analyses of data from other nuclear worker studies. However, some recent analyses of worker data have included cancers noted on the death certificate but not considered to be the underlying cause of death (Wing et al. 1991;Kendall et al. 1992). It should be noted that the statistical methods used for these analyses are based on the assumption that the time of death is in some sense a surrogate for the time of tumor occurrence; this assumption may be less appropriate for cancers that are not considered to be the cause of death. Analyses numbers 2, 10, and 12 in Table 8 were based on this alternative approach, while Table 9 pre- sents additional detail. For these analyses, cancers (and cancers of specific types) were counted if indicated anywhere on the death certificate.Potentialj). -~.... could contribute to cancer type; in fact, the analyses of morr there were no instance ::.-~:..-.-_--,,. death was included in more than one of Ip._ --.-_. subcategories included in Table 9. Note ;-.,. includes 159 more deaths in the all-carlcc: cL:7. , than does Table 5. The risk estimate for all-cancer remaint.2 :.-+ with this approach. With a 10-y lag, the upp,.- . . dence limit was the same as that obtained N:.:: cancers that were the underlying cause cjf dt2:, ._ included. There were no specific cancers tha! ..I . significant correlations that did not also sh,.., , correlations in analyses based on the undt.;l!ir., . __ of death. The risk estimate tables) was negative, with for an leukemia (no:.:5 upper confiden.: ....- (two-sided90% interval)of 1.5%per 10mS\..cc-.: -to the 1.9% value shown in Table 7. Table 9 indicates two additional cases \vitk - ple myeloma, one each in the top two dose cart; - Both these deaths were reported in the 1989 pa?- - had not been included in formal statistical acs .. The addition of these two deaths led to a more 5:;- cant correlation for multiple myeloma with both 1:. and 10-ylags. Table 9 also showsthree additions; 2:- . with cancer of the pancreas (one each in the i- and 50-mSv categories); these cases slightly in::--< the p values for this disease. There were no deaths 'Hodgkm's disease indicated on the death certificz:: -. not considered to be the underlying cause of if-'. There was one such death with liver cancer. bur :: * - in the lowest dose category and did not grea!!! ~ E X results from Table 5. Analyses in Table 10 augment the anal!? Tables 5-7 by including deaths occumng in ?he y-. of Washington for the time period 1987-1989. of linkage with death files in the State of W a s h W - these deaths can be ascertained more quickly t h . ~ those occumng in other states. The recent WashlG ,' deaths were included by using a proportional mor=. approach, which is described in the 1989 paper- ..* methods used in this paper to obtain risk estimalci - confidence limits are not appropriate for propon!': mortality data and, therefore, risk estimates inclus these recent Washington deaths were not calculats I'Table 10 includes 127 cancer deaths that we'! 'included in Tables 5-7. These additional deaths dlc Lag : I. 2. 3. 4. 5. 6. 7. 8. Lag : 9. 10. Lag : 11. 12. Mortality at the Hanford site 0 E. S.GILBERT ET AL. Table 8. Excess relative risk estimates (with 90% confidence limits) for all cancer based on several alternative analytic approaches(expressed as percent increase per 10 mSv). Analysis number Modification in analytic approach' Trend test statistics Excess relative risk estimate (with 90% confidence limits) Lag = I O y 1. None -0.29 - 0.15% (<O, 0.8%) 2. Includes cancers indicated on death -0.17 -0.08% (<O, 0.8%) certificate, but not underlying cause 3. Excludes dose received in construc- -0.16 -0.08% (<O, 0.9%) tion work 4. Includes monitored workers em- -0.42 -0.2 1% (<O, 0.7%) ployed <6 mo 5. Excludes workers with confirmed 0.0 1 internal depositions 0.01% (<O, 1.1%) 6. No control for socioeconomic sta- 0.73 tus 0.39% (<O, 1.4%) 7. No control for number of years -0.68 -0.3 1% (<O, 0.5%) monitored 8. Lag=2y 9. No control for socioeconomic status or for number of years monitored None 0.23 -0.50 0.1 1% (<O, 1.O%) -0.19% (<O, 0.5%) 10. Lag=20y 11. Includes cancers indicated on death certificate, but not underlying cause None -0.29 0.57 -0.10% (<O, 0.5%) 0.56%(<O, 2. I %) 12. Includes cancers indicated on death 0.53 certificate, but not underlying cause 0.5 1% (<O, 2.1 %) a Except as indicated, analyses included all monitored workers employed at least 6 mo, included dose received in both operations and construction work, were based on deaths considered to be the underlying cause of death, and were controlled for age, calendar year, gender, number of years monitored (1-4 vs. 5+), and general socioeconomiccategory. 585 Jange the results for all cancers greatly, and no new %is:ically significant correlations for specific cancers F-aluatedin Table 5 resulted. However, the correlations 3r both cancer of the pancreas and multiple myeloma *erestrengthened with the inclusion of these additional ::aths. There were five additional deaths from cancer (the pancreas including one in each of the two top :ose categories, and four additional deaths from mul:Piemyeloma, including one death in each of the three '3Pdose categories. There were no deaths from Hodg\in's disease occurring in Washington State during the 98-11989 time period. Table 11 shows results of analyses that include -ancersthat were not the underlying cause of death and 'lncers occurring in the State of Washington during he time period 1987-1989 (including 16 deaths in this *nod where the underlying cause was not cancer). "Ith a 10-y lag, this resulted in a p value of 0.01 for 'W iple myeloma and a p value of 0.06 for cancer of he mcreas. Unlike most analyses presented in this paper, 1989 analyses did not include dose received in construction work but did include workers employed <6 mo. Analyses numbers 3 and 4 in Table 8 indicate that these changes do not greatly affect results; although, the risk estimate and upper confidence limit were slightly lower when very short-term workers were included. It is noted that the risk of cancer for workers employed <6 mo relative to workers employed longer was 1.28; for all causes of death, the comparable relative risk was 1.26. Excluding workers with confirmed internal depositions (analysis 5 ) slightly increased both the estimate and the upper confidence limit. This exclusion also lessened the evidenceof a correlation with external dose for cancer of the liver, as the liver cancer death with the highest dose also had a confirmed plutonium dep- osition; this death was discussed in our 1989 paper. Analyses numbers 6, 7, and 8 show the effect of modifying the choice of controllingfactorsand indicate that not controlling for socioeconomic status increases .* 586 Health Physics June 1993. Volume 64, Number 6 Table 9. Results of analyses of external dose for selected cancer categories in monitored workers employed at the Hanford Site for at least 6 mo including cancers noted on death certificate, but not considered to be the underlying cause of death. Except where noted, this is based on a IO-y lag. Cause of death Trend test statistic' Exposure lagged for: l0Y 2Y Observed and expected deaths by exposure category (mSv) 0Obs/Expb 10- 50- loo- 200+ ObsIExp Obs/Exp Obs/Exp ObsIExp All cancer Pancreas Multiple myeloma Leukemiad Leukemiad (2-y lag) -0.17 1.45` 2.50' -0.98 -0.29 2.20' 2.95` -1.01 1,079/1,082.1 52154.0 17117.3 3313 1.3 3 1128.6 3811377.7 20119.3 216. I 15113.2 15114.6 69166.9 313.3 211.1 112.5 112.8 50149.6 312.4 210.7 111.8 312.1 46148.7 312. I 310.8 112.2 112.8 The trend test statistic was calculated from individual doses, not the five exposure categories. It may be compared with a standard normal distribution to assess statistical significance;however. statisticalsignificance may be exaggerated for diseases with a small number of deaths. See footnote c. Expected deaths were calculated from the experience of all workers in the study population, allowing for age, calendar year. gender. number of years monitored, and general socioeconomic category. Based on computer simulations, the one-tailed p values associated with the trend test with a 10-y lag were estimated to be 0.080 for cancer of the pancreas and 0.023 for multiple myeloma. For the 2-y lag, these p values were estimated to be 0.032 for cancer of the pancreas and 0.007 for multiple myeloma. Excluding chronic lymphatic leukemia. Table 10. the Hanfc Except wl i th risk estimate and upper confidence limits, while not controlling for number of years monitored has the opposite effect. Both variables can be demonstrated to affect cancer mortality, and thus we think that analyses that include adjustment for them (as in Tables 5-7) are less likely to be biased than analyses that do not include such adjustment. Analyses number 9-12 show results based on 2- and 20-y lags. With the 20-y lag, the risk estimate was positive and the upper confidence limit higher, but the correlation did not approach statistical significance. Hodgkin's disease was the only cancer category exhibiting a statistically significant correlation with the 20-y lag. The correlation for multiple myeloma was in the negative direction with this lag; that for cancer of the pancreas was not close to statistical significance 0, = .17). Table 5 shows 86 deaths without cause-of-death information. For 44 cases, these were deaths for which certificates have not been located. After most analyses in this paper had been completed, it was discovered that 42 deaths had certificates that had not been sub- mitted to the National Center for Health Statistics for assignment of the underlying cause of death. Of these, 33 were for deaths occumng in 1986. Although nosologic codes for the underlying cause of death were not assigned by certified nosologists in these cases, death certificatetext was manually reviewed for mention of cancer. Nine of the 42 certificates mentioned cancer, but in no case was leukemia, multiple myeloma, cancer of the pancreas, or Hodgkin's disease mentioned. To examine whether or not the addition of these nine deaths might substantially alter the all-cancer risk estimate, an analysis including all cancers mentioned on the death certificate (as in Table 9 and analysis 2 - Table 8)was conducted. This modified the risk estim+ and confidence limits in Table 9 very little (the m. estimate was slightly increased from -0.08% to -0 0: per 10 mSv, and the upper confidence limit was al\ slightly increased from 0.77% to 0.82% per 10 mS. Thus, it seems highly unlikely that the omission of tht. deaths would have substantially modified results p-: sented in this paper. DISCUSSION Hanford workers continue to show a strong healrr worker effect with death rates from most causes sc:stantially below those of the general U.S. populatix Comparisons of death rates by radiation dose \vitk:the Hanford cohort show no evidence of a corre1ati~for all causes of death, all cancers, or leukemia w:radiation dose. These results are similar to those ::- ported in 1989 and are consistent with results h': other nuclear worker studies discussed in the 1"`paper. Since the 1989 paper was prepared. additlo:reports of nuclear worker studies have been publish: (Beral.etal. 1988;Winget al. 1991;Kendallet al. 19.These newer reports also show death rates for n V causes that are substantially below national ratesthough leukemia death rates were found to be ele\a:f-. in ORNL workers (Wing et al. 1991). Dose-response analyses of workers at the l'`. Atomic Weapons Establishment (AWE) (Beral e: - 1988) and of workers at ORNL show statisticall! y i nificant correlations for all cancer and risk estirn2"= that are several times those obtained from extra;731"':-< from high-dose data. Data on Hanford workers Ck2` Mortality at the Hanford site 0 E. S.GILBERT ET AL. Table 10. Resultsof analyses of externaldose for selected cancer categoriesin monitored workers employed at the Hanford Site for at least 6 mo, including cancers occumng in the state of Washington from 1987-1989. Except where noted, this is based on a 10-y lag. Trend test statistic? Exposure lagged for: Observed and expected deaths by exposure category (mSv) Cause of death 10Y 2 Y e Obs/Expb 10- 50- 100- 200+ Obs/Exp Obs/Exp Obs/Exp ObsIExp All cancer Pancreas Multiple myeloma Leukemiad Leukemiad (2-y lag) -0.20 1.91' 1.99' -0.78 -0.38 2.57' 2.9W -0.84 1,023/1.031.5 54155.4 18118.2 29127.8 27125.2 3921381.4 19119.4 216.3 15113.2 15114.7 70171.1 213.6 311.3 112.7 113.0 56153.2 412.4 211.0 211.9 412.2 52155.8 412.2 31 1.2 112.3 112.9 The trend test statistic was calculated from individual doses, not the five exposure categories. It may be compared with a standard normal distribution to assess statisticalsignificance;however, statistical significance may be exaggerated for diseases with a small number of deaths. See footnote c. Expected deaths were calculated from the experience of all workers in the study population, allowingfor age, calendar year, gender, number of years monitored, and general socioeconomic category. Based on computer simulations, the one-tailed p values associated with the trend test with a IO-y lag were estimated to be 0.033 for cancer of the pancreas and 0.040 for multiple myeloma. For the 2-y lag, these p values were estimated to be 0.019 for cancer of the pancreas and 0.0 I 1 for multiple myeloma. Excluding chronic lymphatic leukemia. 587 Table 11. Resultsof analysesof external dose for selected cancer Categoriesin monitored workers employed at the Hanford Site for at least 6 mo including cancers occurring in the state of Washington from 1987-1989, and including cancers noted on the death certificate, but not considered to be the underlying cause of death. Except where noted, this is based on a 10-y lag. Trend test statistic' Exposure lagged for: Observed and expected deaths by exposure category (mSv) Cause of death 10Y 2 Y 0Obs/Expb 10- 50- 100- 200+ Obs/Exp ObsIExp ObsIExp ObsIExp All cancer Pancreas Multiple myeloma Leukemiad Leukemiad (2-y lag) 0.05 1.61' 2.67' -0.94 -0.04 2.28' 3.45' -0.99 1.131/1.135.3 56157.1 18118.6 35132.8 33130.2 4341431.7 20120.9 217.5 16114.8 16/16.2 82179.3 313.9 311.5 112.9 113.2 61159.1 412.6 311.1 212.0 412.3 60162.5 412.5 411.4 112.5 113.1 'The trend statistic was calculated from individual doses, not the five exposure categories. It may be compared with a standard normal distribution to assess statisticalsignificance:however, statistical significance may be exaggeratedfor diseases with a small number of deaths. See footnote c. Expected deaths were calculated from the experience of all workers in the study population, allowingfor age, calendar year, gender, number of years monitored, and general socioeconomic category. Based on computer simulations, the one-tailed p values associated with the trend test with a 10-y lag were estimated to be 0.062 for cancer of the pancreas and 0.01 1 for multiple myeloma. For the 2-y lag, these p values were estimated to be 0.028 for cancer of the pancreas and 0.003 for multiple myeloma. Excluding chronic lymphatic leukemia. 10 not support the high-risk estimates obtained from ?RXL and AWE data. The Hanford-based upper con- .dencelimit of about 1% per 10 mSv is well below the ?lmates of 7.6% (AWE) and 3.3% (ORNL) obtained I ' ':Om .?!el cthaensebestsutdroiensgwlyitrhejaec1te0d-y(plag<, a.0n0d1e)swtimithattehseaHt tahnis- ' '3rd data. The ORNL estimate of 4.9%, obtained with 70-y lag, can also be strongly rejected (p < .OO 1) when '"lznforddata are analyzed with this lag. However, the AWE and ORNL risk estimates carry ir? uncertainties, and the three studies may not be inconsistent with one another. Confidence limits based on the AWE data are presented by Beral et al. (1988) while confidence limits based on the ORNL data are given in Gilbert (1992). Combined analyses of Hanford and ORNL data are currently underway and will address this question more rigorously. Combined international analyses are also being conducted (Cardis and Kaldor 1989). The AWE correlation resulted primarily from a correlation for lung cancer, and nonmalignant respiratory disease showed a similar correlation. The ORNL 588 Health Physics correlation was also strongest in cancers that have been linked with smoking (Gilbert 1992).Thus, a smokingrelated bias may have contributed to these correlations. Although the ORNL study found an excess of leukemia in comparison to the U.S. population, leukemia was not found to be correlated with occupational radiation dose in either AWE or ORNL workers. The total person& for AWE workers was 73 while that for ORNL workers was 144, compared with 861 person-Sv received by Hanford workers. In terms of total personSv, both the AWE and ORNL studies are quite small compared to those reported in other studies such as Hanford, UK Atomic Energy Facility with 660 personSv (Beral et al. 1985), and the Sellafield Plant with 1,250 person-Sv (Smith and Douglas 1986). Kendall et al. (1992) reported a weak positive association with radiation' dose for all cancers (onetailed p = -10)and a stronger association for leukemia (p = .06) in the UK National Registry for Radiation Workers (NRRW). Risk estimates for both disease categories were higher than those obtained through extrapolation from high-dose data, but confidence intervals indicated consistency with high-dose predictions. The Hanford-based confidence limits include the NRRW all-cancer risk estimate. The NRRW leukemia risk estimate was higher than the Hanford-based upper confidence limit, but leukemia confidence limits for the two studies overlap. The NRRW study includes many of the same workers evaluated in the AEA, AWE, and Sellafield populations. The total personSv for NRRW workers was 3,200; nearly four times that for Hanford. In previous analyses of the Hanford data, multiple myeloma was the cancer most strongly linked with radiation dose. The one-tailed p value obtained in the analyses presented in Table 5 with a 10-y lag was 0.10, compared with a p value of 0.002 obtained in comparable analyses based on data through 1981, and published in 1989. Ten of the 23 deaths shown in Table 5 occurred between 1982 and 1986, with one each in the 50-99 mSv and 200+ mSv categories. One death in the 200+ mSv category occurred in the state of Washington and had been included in 1989 supplementaryanalyses including recent Washington deaths. A test for consistency in results before and after 31 December 1981 yielded a p value between 0.1 and 0.2. However, significantcorrelationswere obtained a) when analyses were expanded to include deaths with multiple myeloma listed on the death certificate but not considered to be the underlying cause (p = .02); b) when analyses were expanded to include deaths occurring in Washington State in the time period 1987-1989 (p = .04); or c) when a 2-y latency period (instead of 10-y) was assumed (p = .03). When a) and b) are combined. the p value was 0.01. However, none of these correlations is as striking as that observed earlier and could be interpreted as spurious given the large number of disease categories examined. Furthermore, the approaches used in analyses presented in Tables 8, 9, and 10 may be particularly subject to bias. For June 1993, Volume 64, Number 6 example, the multiple myeloma correlation in Hanforc workers has been widely publicized, and it is POssib,: that a multiple myeloma that was not the cause of deal, would be more likely to be noted on the death certifica?, of a worker known to be a radiation worker than fc: other workers. Nevertheless, the continued appearance of mulll. ple myeloma deaths in Hanford workers with high U o b argues against readily dismissing this finding. Thi NRRW study just noted also shows a correlation fQy multiple myeloma (p = .06), which largely resulle; from Sellafield workers included in the study, for whol: a correlation had been previously reported (Smith an: Douglas 1986).Multiple myeloma has also been clear],, linked with radiation dose in A-bomb survivors. ai. though the magnitude of the risk estimate would no: lead to the expectation of a correlation at low doss (Shimizu-et al. 1990). With a 2-y lag, cancer of the pancreas showed : significant correlation with radiation dose in Hanfor; analyses based on data through I April 1974 (Gilbez and Marks 1979) and on data through 1 May 19-(Gilbert and Marks 1980). However, the correlation dl: not approach statistical significance with a 10-y lag 1r these earlier analyses. In later analyses, including d a t through 1978 (Tolley et al. 1983) and through 1%. (Gilbert et al. 1989a), the correlation had declined 1, nonsignificant levels. Analyses in this paper indicate correlation for cancer of the pancreas that was statist:. cally significant with the 2-y lag (one-tailed p = .Oi and that approached statisticalsignificancewith the I iy lag (p = .07). Results were similar with the additio: of cancers not considered to be the underlying cause i' death, of cancers occumng in the state of Washingto: in the time period 1987-1989, or of both. Death certificate diagnosis of cancer of the par.. creas is known to be inaccurate. It was discovered eark that one of the high-dose deaths in the Hanford analyx. was probably due to cancer of the stomach (Gilben an; Marks 1980). Removal of this death would reduce rn: evidence for a correlation. For consistency, all death! are categorized according to information on the deal' certificate and, thus, this death has remained in cancer of the pancreas category. Evidence for radiation-induced cancer of the pa:. creaswas reviewed by the BEIR V Committee(Nations Academy of Science 1990, p. 334), who summanzfc their conclusions as follows: An association between cancer of the pancreas and previous irradiation, suggested by several reports in the past, has not been confirmed in more recent and thorough studies of irradiated human populations. The pancreas appears, therefore. to be relatively insensitive to radiation carcinogene- sis. . . . Cancer of the pancreas was not among the sitesi:' which the ICRP (199 1) provided estimates and we@ ing factors and has not been linked with radian?exposure in nuclear workers studies other than Har- Mortality at the Hanford site 0 E. s. GILBERT ET AL. 589 .jrd. Because there is clearly no strong a priori reason jr expecting a correlation for cancer of the pancreas -,,I not other types of cancer, it seems appropriate to ,lterpret the Hanford finding as probably spurious. The correlation observed for Hodgkin's disease is :\en more likely to represent a spurious finding. The 3 ~ 1 RV report (National Academy of Science 1990, p. : 7 9 ) devotes a single sentence to this disease, stating "For Hodgkin's disease, the data are reasonably :cinsistentin showing no excess in irradiated popula:Ions.'' Hodgkin's disease has not shown a correlation iith radiation exposure in any of the other nuclear .\orkerstudies. A major objective of studies of nuclear workers is :J provide a direct evaluation of effects of exposure at JW doses and-dose rates for comparison with risk &mates that have been recommended for use in ra- $tion protection, for example, estimates provided by :he International Commission on Radiological Protec::on ( 1991) and by the National Academy of Science's 3EIR V Committee (1990). These risk estimates have .yen obtained by extrapolation from data on popula:ions exposed at high doses, such as the Japanese A- x m b survivors. The ICRP based its lifetime risk estimates for .:ukemia and all-cancer except leukemia on risk esti;lares presented in a report of the United Nations Scientific Committee on the Effects of Atomic Radia::on (UNSCEAR 1988). Based on analyses of Japanese 4-bomb survivor data, UNSCEAR presents linear ex:ss relative risk coefficients for leukemia of 3.7% and ;.ScI per 10 mSv for males and females, respectively, :sposed in adulthood. The comparable coefficients for dl cancer except leukemia were 0.24% and 0.46% for mles and females, respectively, exposed in adulthood. `or exposure received at low doses and dose rates, the :CRP recommended reducing the estimates provided -!UNSCEAR by a factor of 2. For leukemia, the risk estimate based on Hanford +ta (see Table 7) was negative and the confidence imits included zero, indicating that the possibility of 70 risk at low doses and dose rates cannot be excluded. a'ith the 90% interval, the upper confidence limit of :.'% per 10 mSv is about one-half the linear estimate ,resented in UNSCEAR and indicates that a risk estim e 21.9% can be rejected at the 0.05 level. This :suit could be interpreted as indicating that the revised QP model probably has not seriously underestimated tukemia risks. For all cancer except leukemia, the Hanford-based s t h a t e was again negative. The upper confidencelimit .Y. i.O% per 10 mSv was about four times the - XSCEAR coefficient for males and about eight times `he estimate reduced by the factor of 2 recommended :!the ICRP. To account for the larger UNSCEAR risk -'oeffcient for females, we performed an analysis in .`_hkh risks were expressed as multiples of the - YSCEAR model and which involved weighting doses -:CL:yed by females by a factor 0.46/0.24 = 1.92 times doses received by males. This analysis also yielded an upper limit about four times the UNSCEAR prediction (or eight times the ICRP prediction). We also analyzed the Hanford data for consistency with predictions of the models recommended by the BEIR V Committee (National Academy of Science 1990). The BEIR V models incorporate dependencies of the excess relative risk on age at exposure, time after exposure, and gender, and were primarily based on analyses of the Japanese A-bomb survivor data (although the female breast cancer model used other data sets). Separate models were determined for leukemia, digestive cancer, respiratory cancer, female breast can- cer, and other cancers. Hanford data were analyzed by expressing estimates and confidence limits as multiples of risks predicted under the BEIR V models, and this involved weighting each annual dose according to age, time from the effect, and gender. All cancers except leukemia .were analyzed in combination and, in this case, the model involved weighting by the type of cancer as well. Using this approach, the leukemia risk estimate was -0.6 times the BEIR V predictions, and the upper confidencelimit (two-sided90% interval) was 0.8 times the BEIR V predictions. The BEIR V leukemia model included reduction of risks by a factor of 2 for low doses; thus, this result is very similar to the result previously noted for the ICRP model. For all cancers except leukemia, the risk estimate was almost exactly zero and the upper confidencelimit was 1.5 times BEIR V predictions based on a linear model with no reduction for low dose rates. The factor of 1.5 is smaller than the factor of 4 resulting from comparison with UNSCEAR 1988 linear estimates. This difference results primarily because the BEIR V respiratory cancer model predicts large risks in workers; this occurs because the BEIR V model transports a relative risk coefficient from the Japanese A-bomb survivors where baseline risks for respiratory cancer are low, and because the model includes a decline with time since exposure and assigns the highest risks to the time period when workers are receiving most of their exposure. The confidencelimitsjust discussed do not include uncertainty resulting from unidentified confounding factors that may have biased results or uncertainty in the dose estimates used in the study. In particular, the recorded dose may overestimate bone marrow dose and doses to other organs, which have been the basis of risk estimates obtained from recent analyses of Japanese Abomb survivor data. A detailed evaluation of Hanford historical dosimetry has recently been conducted (Wilson et ai. 1990). Further work to evaluate the relationship of recorded dose and doses to various organs, and to examine the effect of dosimetry biases and uncertainties on dose-responseanalyses, is planned. Current data from Hanford indicate that low-level radiation exposure risks are consistent with no risk, with predictions by the ICRP or BEIR V, and with risks that are several times these predictions. These results $0 - Health Physics appear to be consistent with those of most other nuclear worker studies, but two studies of much smaller cohorts have indicated that risks may be larger than predictions based on high-dose data. In order to obtain the best assessment of data from all occupational studies, it is important to conduct analyses based on combined data. Analyses of combined U.S. data, which will update results in Gilbert et al. (1989b), are in progress. Combined analyses of international data from the U.S., UK, and Canada are also underway with the International Agency for Research on Cancer serving as the coordinating agency (Cardis and Kaldor 1989). Acknowledgements-Thiswork was supported by the US.Depart- ment of Energy under contract DE-AC06-76RLO-1830.The authors acknowledge the vital statistics ofices of the individual states as the source of death record data and appreciate the ofices' technical support of this research. The authors are solely responsible for the data analyses and the interpretation of the results. REFERENCES Beral, V.; Fraser, P.; Carpenter, L.; Booth, M.; Brown, A.; Rose, G. Mortality of employees of the Atomic Weapons Establishment, 1951-1982. Br. Med. J. 297:757-770; 1988. Beral, V.; Inskip, H.; Fraser, P.; Booth, M.; Coleman, D.; Rose, G. Mortality of employees of the United Kingdom Atomic Energy Authority, 1946-1979. Br. Med. J. 291:440-447; 1985. Bureau of the Census. Alphabetical index of industries and occupations, 1970 census of the population. Washington, DC: U.S.Government Printing Ofice; 1971. Cardis, E.; Kaldor, J. Protocol of combined analyses of cancer mortality among nuclear industry workers. Lyon, France: International Agency for Research on Cancer (IARC); Internal Report No. 89/005; 1989. Doll, R.; Peto, R. The causes of cancer. New York Oxford Medical Publications, Oxford University Press; 1981. Gilbert, E. S. Issues in analyzing the effects of occupational exposure to low levels of radiation. Statistics Med. 8: 173187; 1989. Gilbert, E. S. Studies of workers exposed to low doses of external radiation. In: Occupational medicine: State of the art reviews. Vol. 6, No. 4, October-December 1991. Philadelphia: Hanley & Belfus, Inc.; 1991:665-680. Gilbert, E. S. Mortality of workers at the Oak Ridge National Laboratory. Health Phys. 62:260-26 1; 1992. Gilbert, E. S.; Buchanan, J. A. An alternative approach to analyzing occupational mortality data. J. Occupational Med. 26~822-828;1984. Gilbert, E. S.; Marks, S. An analysis of the mortality of workers in a nuclear facility. Radiat. Res. 79:122-148; 1979. Gilbert, E. S.; Marks, S. An updated analysis of mortality of workers in a nuclear facility. Radiat. Res. 83:740-741; 1980. June 1993, Volume 64, Number 6 Gilbert, E. S.; Buchanan, J. A.; Holter, N. A. Descfi-.:,,, , the process used to create 1992 Hanford mortali:! x-. database. Richland, WA: Pacific Northwest Lahora:,; Report No. PNL-8449; 1992. Gilbert, E. S.; Petersen, G. R.; Buchanan, J. A. Momi::-, workers at the Hanford site: 1945-1981. Health p;, 56:ll-25; 1989a. Gilbert, E. S.; Fry, S. A.; Wiggs, L. D.; Voelz, G. L.: cr-: D. L.; Petersen, G. R. Analyses of combined r.3data on workers at the Hanford Site, Oak Ridge \ a ~ ~ ; ~ Laboratory, and Rocky Flats Nuclear Weapons piir Radiat. Res. 120:19-35; 1989b. International Commission on Radiological Protection. p,?, ommendations of the International Commission on i, diological Protection. ICRP Publication 60; Ann. IC?,: 21:1-201; 1991. Kendall, G. M.; Muirhead, C. R.; MacGibbon, B. H.: 0.b~ gan, J. A.; Conquest, A. J.; Goodill, A. A.; Butland. B.i. Fell, T. P.; Jackson, D. A.; Webb, M. A.; Haylock. R. i E.; Thomas, J. M.; Silk, T. J. Mortality and occupatio, exposure to radiation: First analysis of the National Rectry for Radiation Workers. Br. Med. J. 304:220-225: 19;: Mantel, N.; Haenszel, W. Statisticalaspects of the analys!r : data from retrospective studies of disease. JNCI 22:-:.. 48; 1958. Monson, R. R. Analysis of relative survival and proportlc:, mortality. Computer Biomedical Res. 7:325-332: 19-: National Academy of Sciences. Committee on the Biolg:: Effects of Ionizing Radiation (BEIR V). Health effecli exposure to low levels of ionizing radiation. Washing:: DC:National Academy Press; 1990. Shimizu, Y.; Kato, H.; Schull, W. J. Studies of momhi: . A-bomb survivors.9. Mortality, 1950-1985:Part 2. Caz: mortality based on the recently revised doses (DSSRadiat. Res. 121:120-141;1990. Smith, P. G.; Douglas, A. J. Mortality of workers a! .: Sellafield plant of British Nuclear Fuels. Br. MeC. . 293:845-52; 1986. Tolley, H. D.; Marks, S.; Buchanan, J. A.; Gilbert. E. S further update of the analysis of mortality of workers 1.: nuclear facility. Radiat. Res. 95:2 11-213; 1983. United Nations ScientificCommitteeon the Effects of Atoz Radiation. Sources, effects, and risks of ionizing radiau:: Report to the General Assembly, with annexes. New YC:? United Nations; 1988. Wilson, R. H.; Fix, J. J.; Baumgartner, W. V.; Nichols. L 1 Description and evaluation of the Hanford personnel 2- simeter program from 1944 through 1989. Richland. u Pacific Northwest Laboratory; Report No. PNL-:": 1990. Wing, S.; Shy, C. M.; Wood, J. L.; Wolf, S.; Cragle. D.,1 Frome, E. L. Mortality among workers at Oak R1: National Laboratory: Evidence of radiation effects in :!. low-up through 1984.J. Am. Med. Assoc. 265:1397-1` - 1991. World Health Organization. International classificatior. : disease. 8th Rev. Geneva:WHO; 1965. 8g