ML20134C827

Discusses Recent Publication Confirming linear-no Threshold Theory

ML20134C827
Person / Time
Issue date:	07/18/1995
From:	Pollycove M NRC
To:	Cool D NRC OFFICE OF NUCLEAR MATERIAL SAFETY & SAFEGUARDS (NMSS)
Shared Package
ML20134B716	List: ML20134B713 ML20134C768 ML20134C782 ML20134C786 ML20134C791 ML20134C792 ML20134C798 ML20134C801 ML20134C806 ML20134C827 ML20134C830 ML20134C837 ML20134C840
References
ACRS-GENERAL, FACA, NACNUCLE, NUDOCS 9610080442
Download: ML20134C827 (14)

Text

_ _ _ _ . _ _ __ _ -. _ __ .. _ . _ _ _ _ _ . _ _ ..... _ . _ _ _ _ _ . _ _ . _ _ _ _ . .

I I

pueo u j

=

f 2> *. UNITED STATES

\

l I

• j f

NUCLEAR REGULATORY COMMISSION WASHINGTON, D.C. M1 I

l

'+9 g

• • • • • July 18, 1995 MEMORANDUM TO: Donald A. Cool, Director Division of Industrial and MedicalNuclearSafey,NMSS FROM: Myron Pollycove / ,b Visiting Medical Fellow

SUBJECT:

RECENT PUBLICATION " CONFIRMING" LINEAR-NO THRESHOLD THEORY 1

One cannot significantly contradict the linear no-threshold theory by l restricting data analysis only to risk increments. The diverse adaptive responses to radiation documented in UNSCEAR 1994 repeatedly demonstrate .

significant risk decrements. Attached is the recent publication identified by Shirley Fry (one of the authors) as the one, cited in Sweden to Bob Bernero, that confirms the linear no threshold theory with no dose rate reduction at low dose.

Risk decrements are excluded a oriori from this publication so that one-sided P tests are used throughout. Even so, the data demonstrate no evidence of  !

association between radiation dose and mortality from all causes or from all I cancers includina mortality from leukemia other than chronic lymphatic leukemia (CLL). The lack of a significant number of leukemia deaths is circumvented by a trend analysis that ignores the 4 dose ranges with fewer observed than expected deaths (4 of 7) and uses a computer simulation of 5000 samples instead of the 6/2.3, 8/6.4, and 19/17.2 observed / expected leukemia deaths observed in each of the 3 remaining interspersed dose ranges.

The U.S. Nuclear Shipyard Worker (NSW) study reported in 1991 and in UNSCEAR 1994 is omitted from this 1995 report of cances ..ortality among 95,673 monitored nuclear industry workers in the U.S., U.K, and Canada. The Johns Hopkins NSW study of almost 700,000 workers includes 106,851 monitored nuclear workers in the data base and was chaired by Arthur Upton. The NSW study demonstrated a highly significant -0.24 risk decrement (-16 SD) of standardized mortality from all causes, no increase of leukemia or lymphatic and hematopoietic cancers, and the absence of any healthy worker effect.

Attachments: As stated cc: H. Thompson, EDO R. Bernero J. Glenn, RES C. Paperiello, NMSS 9

9610080442 960326 PDR ACRS GENERAL PDR

i 4

4 sointhn utir 4Rt H 142. ll\l.1)1 LW4:

i i Effects of Low Doses and Low Dose Rates of External

! lonizing Radiation: Cancer Mortality among Nuclear j Industry Workers in Three Countries  :

1 i

l I E. Cardis. E. S. Gilbert.' L. Carpenter / G. Howe!' I. Kato." B. K. Armstrong/ V. Beral! G. Cowper.' A. Douglas /

J. Fix? S. A. Fry! J. Kaldor.' C. Lasd." L. Salmon.' P. G. Smithf G. L. Voelz' and L. D. Wiggs' i

• Internatwnel Agencv for Research on Cancer. Lyon. France:'Pacaric Nankwest Laboratorres. Richland. Washmston:'Depenment of Pubhc l Health and Prsmarv Carr. Oxford. Unsted Kangdom:'NCIC Epsdemiologv Unu. Unwersutv of Toronto. Canada:'Austrahan Insruute of l Heakh and Welfare. Canberra. Auuraha:'imperual Cancer Research fund. Oxford. Unned Ksngdom:'Atomsc Energy of Canana. Deep Ro er. Ontarw. Canada:' London Sshool of Hsgsene and Tropocal Medscsne. London. Unned Ksngdom: Oak Rodge instuute for l

\ Science and Educatwn. Oak RsJge. Tennessee: %arwnal Centre on HIV Epsdemwlogy and Chnscal Research. 5sdnes. Austraha.

• Atomsc Energy Autharttv. HarmeII. Osdcot. Unned Kwesham; and 'Los Alamos Na:wnal Laboratorv. Los Alamos. Vem %Ieseco l dose in the Hanford (U.S.) and Seilaneld (UK) cohort [ Th j Cardis. E. Gilbert. E. S Carpenter. L., Howe. G.. Kato, l Ann. escess retetise risk (ERR) estimates for all cancers excluding I

strong. B. K., Beral. V Cow per. G Douglas. A.. Fis J., Fry. S. A., leukemia, and leukemia excluding CLL. the two main grouptngs j Kaldor. J., Lave. C.. Salmon L., Senith. P. G.. Voelz. G. L. and of causes of death for which risk estimates have been derned j' Wiggs. L. D. Effects of Low Doses and Low Dose Rates of Esternal from studies of atomic bomb survivors, were -0.07 per Sv [909 lonizing Radiation

Cancer Mortality among Nuclear Industry i:onAJence interval (Cl): -0.4. 0.3l and 2.18 per Sv (90'4 CI: 0.1.

Workers in Three Countries. Radiet. Res. 142. 117-132(1995). 5.7), respectively. These values correspond to a relative ris' -'

O.99 for all cancers escluding leukemia and 1.22 for leul 1 Studies of the mortality among nuclear industry workforces escluding CLL for a cumulative protracted does of 100 mSv

have been carried out, and nationally combtned analyses per- pared to 0 mSv. These estimates, which did not differ signifn.

formed. in the U.S the UK and Canada. This paper presents the ly across cohorts or between men and women, are the most com-

results of inter =maiamany combined analyses of mortality data on prehensive and precise direct estianates of cancer risk associated

95.673 workers (f.5.4% amen) monitored for enternal exposure to with low dose protracted exposures obtained to date. Although

lonizing radiation and employed for 6 months or longer la the they are h than me linew animous obtamed from nudies of nuclear industry of one of the three countries. These analyses atonus bosab survivors, they are compatible with a range of possi.

were undertaken to obtain a more precise direct assessment of the bilities, froen a reduction of risk at low doses, to risks twice those j; carcinogenic effects of protracted low level exposure to external

• on which current radiation protection recommendations are i predominantly y, radiation. The combination of the data from based, uw.all, the reamles of this study do not suggest that cur.

l the various studies increases the power to study associations rent radimetam risk amelemmens for cancer at low levels of exposure 1 between radiation and speciac cancers. The combined analyses ,,, " .m ' g,,,,,,, ,, ,,,,,,,,,,,,,,,,,,,,

covered a total of 2.124J26 pereen years (PY) at risk and 15A25 deaths,3,976 of whick.were due to cancer. There was no evidence

of an association between radiation dose and mortality froan all INTRODUCTION

causes er from all cassers. Mortality freut leukemia, escluding chronse lgi A laubenda(CLL)-4he cause of death soom Current estimates of canccr risk associated with external i

strongly and ===i=aM F toisted to rad 1=ela= does in studies of exposure to low-linear energy transfer (LET) ionizing

}

atomic beesh survivors and esher populations esposed at high i dose ram =g=m==dy ====e'=a-d wkh cuenulative enternal Low LET radiations: y and X rays m the range 100 to 2500 ket.

radiation dose - P valene a 0.046; 119 deaths). Among

' Abbreviations used: AEA. Atomic Energy Authonty: AECL.

the 31 other spec pas of cancer studied, a signincant associa- Atomic Energy of Canada Ltd.: AWE. Atomic Weapons Estabhshment.

tion was observed only for :=lticle mveioma ( *alue = CLL. chronic lymphocytic leukemia: CML. chronic mveload leukemia.

0.037; 44 deaths), and this was a.tr butable pn y to the asso. DDREF. dose and dose rate effectiveness factor: ERR. etcess relatne

ciations reported previously between this disease and radiation nsk; 1 ARC. International Agency for Research on Cancer

ICRP. Inter-

' national Commission on Radiological Protection: LET. hnear eneres

'n transfer: ORNL Oak Ridge National Laboratorv: RERF Rad j

'To whom correspondence thould be addressed at international Effects.Research Foundation: RR. relative risk: SES. wcio-e Agency for Research on Cancer 150 Cours Albert Thomas. 6937: Lvon status: UNSCEAR. United Nations Scientific Committee on the l

Cede 08. France. of Atomic Radiation.

' E b h'# m w -n u v k NCT$U$ Y1oY Y'O- ct" "*** ~"* ~ ~

• 4 ma unsohad WMecu5- All nghts of reproduciton an ang Mrm reger fJ

/00

118 c ARDIS TT.it radiation are derised pnmarily from studies of the mortality mates of cancer risk in populations recening protracted af atomic bomb survivors in Hiroshima and Nagasaki and low-dose exposure to ioni7ing radiation and compare these of patients irradiated for therapeutic purposes (/-4). Both estimates with those densed from high-dose studies ((2) to these groups were exposed pnmarily at high dose rates. increase the statistical power to study associations between l Radiation protection recommendations for environmental radiation dose and single cancer types; and (3) to under- l and occupational exposures have generally been based on stand similanties and differences between studies.

the use of these estimates in conjunction with models to The estimates of risk per unit radiation dose for all can- l extrapolate the effects of such acute (or short term) high- cers excluding leukemia and leukemia excluding chronic l level exposures to the relatisely low dose, low-dose rate lymphocytic leukemia (CLL) obtained from the interna-exposures of environmental and occupational concern. and tional combined analyses have been published elsewhere across populations with different baseline cancer risks (4). (33. 34) and compared with estimates derised from analysis These models are. inevitably. subject to uncertainties. of data on atomic bomb survivors. The current paper A direct assessment of the carcinogenic effects of long- presents a detailed comparison of the estimates for the term. low-level radiation exposure in humans can be made combined worker population with risk estimates derived from studies of cancer risk among workers in the nuclear from high dose rate studies, comparisons of risk estimates industry.' Many of these workers have received low, above. across facilities, together with the results of anal)ses of l brckground doses of ionizing radiation, predominantly cause-specific mortality and radiation dose.

from external y-ray exposures. and their radiation doses have been carefully monitored over time through the use of MATERIALS AND METHODS personal dosimeters. Published studies have cosered cohorts of nuclear industry workers m the United States of '"'"""*"*#"'#"""'"'"

America (U.S.). United' Kingdom (UK) and Canada' "'"""""d"'"""*d")***"*""*"'#"*"'#

pner to W were considered for melusion m the combmed analyses n (5-20). Most of these studies hase provided little esidence thes met the solio.. nit criteria t.i.4 t il members of the cohort had of dose-related increase in all cancer mortality. although poicnual ior whole.hous esternal esposure to ionams radiation throuan statistically sieniticant associations between mortality from emplosment in the nuclear industry;(2) monitonns of radiation espo-11 cancers co'mbined and cumulative radiation dose were sure had been carned out routmely (by use of personal dosametero on bserved in two studies: of Oak Ridge National Laboratorv '"""'D'""*d'*'""*8'*d'*""'"#"'"*"*d"""

kept. (3) esumates of whole-body dose from external exposures were ORNL) employees in the U.S. (24) and of the employees as,itable for individual =orkers hkely to be exposed on a yearly basis.

of the Atomic Weapons Establishment ( AWE)in the UK <o mrormauon on a mmimum sei of vanables =as avadable for all mde (16). The statistical power of individual studies was low. viduals m the cohort;($) the rnechanisms of follow-up were not selective however. and in most cohorts the confidence intervals of ga nampie. nstncted current w rkers), and 90% or bette ascer-th; nsk estimates were compatible with a range of possibili- tamment of vital status and of cause of death was possible:(6) mforma-tion was ava.lable on monnonns pohcies and practices over ume.

tits, from negative effects to risks an order of magnitude Ten cohoris mei these criteria. one (5) was excluded because expo-greater than those on which current radiation protection sure was mamly internal and two to. 7) because. for logistic reasons. the recomrMtions are based. Analyses of specific types of 'nsenugaior couiJ .w r*vicipate m the combmed annivses The seven C!ncer were also carried out in most studies. no consistent coho maded m 'h* ' =6'a'd *a*'r ' "P"'*d ** P' Y '"'

pittern of increase for r.::v single cancer type was observed Hanford site (21,27). Oak Ridge Nauonal Laboratory (ORNL)(17. 20 and Rock) Flats nuclear weapons plant (18) m the U.S.. the Setla6 eld tcross all cohorts. plant of Bntish Nuclear Fuels (11.19). the Atomic Energy Authority in 1988, the investigators responsible for the published i AE A)(9. 26) and the Atomic Weapons Establishment ( AWE)(161 m studies agreed to conduct combined analyses of data from the UK; and Atomic Energy of Canada Ltd. ( AECL)(14. 29) They such studies through the international Agency for Research '""d'd**' " st j $0,000 workers (m sily men) m the three on Cancer (I ARC) to maximize their informativeness. At countnes with an average duration of follow up of approximately 2 wears. The tvpes of activities carned out m the facihties included are that time, national analyses of data were already planned in s'hown m Tatste 1.

the UK and the U.S., and these have been published The enethods used to identify the members of the cohorts. to collect (30-32). The objectives of the international combined dosimetnc mformauon. to carrv out mortahty follow-up and to ascertam analyses wcre:(1) to increase the precision of direct esti- the cause f death of workers vaned between (and sometimes withini countries They are summarued elsewhere t34).

%roughout the paper. the term nuclear industrv is used to refer 'Throughout this paper, the term -high-dose studies" is used. for sim-to facihties engaged m the production of nuclear power, the manufacture pbcitv. to refer to the studies curreritiv used m formal radiauon risk of nuclear weapons. the enrichment and processing of nuclear fuel, or assessment. namely the atomic bomb survivors, and panents irradiated restor or weapons research. Uranium msnmg is not mcluded for therapeuuc purposes. Not all subjects m these studies receised high

'O M Matanoski. Heaith Effects of Low level Radtauon m Shipyard doses. but the exposure rates are assumed to be greater than would not.

3rkers Report .o DOE.1991. mallv he espesenced in the occupauonal and general environment 101

LOW DOSE RADIATION AND CA.NCER. A THREE CObTRY STL Di 1W TABLEI mas thourht hkel> to be small. although recorded Jose probabh m r-Type of Activities Carried Out in the Facilities included '"'"*t'd dose the bone marrow shghtly. A small minonty os morur-in the Combined Analyses during the Study Period Ncened a substanual p ni n blW f their dose from esposure u photons ot lower energies or neutrons or trom intake ut radionuclido y g it has not been possible in general to estimate these doses adequatrip but efforts were made to identify such workers and. to the extent powi-U.S Hanford Nuclear reactors. reprocessing. waste ble. exclude them from selected subsidiarv analyses. designated in the treatment. pun 6 canon of plutonium current paper as analyses based on the restncted dosimetry popul.a.

U.S. Rocky Flats Plutonium me. pons non Further details on the ident Acauon of these workers are ;nen M U.S. ORNL* Research and development plant, reactors

• ""d # U*D"*'"

UK Sellaneld Nuclear reactors. replacement. S'8"8"8# M'd'8d3 reprocessang. waste treatment. In general. the statistical methods used were similar to those whicn fast reactor fuel fabncation have been used for the nauonal combined analyses (30 .U) and msobed UK A EA* Research and development. internal compensons of mortality within cohorts by level of external reactor processing radiation doses. For each worker, cumulative dose and person years st UK AWE' Weapons research risk were accumulated over time from his or her date of entry in the Canada AECL* Nuclear reactor.research and study (denned as the later of date of start of employment plus 6 month, related technologies and date of Arst monitonng) to his or her date of exit (denned as the car.

liest of date of death.date of loss to follow up and date of end of follom-

• ORNL = Oak Ridge Nanonal Laboratory: AEA = Atomic Energy up in the appropnate cohort). Person years at risk and deaths mere Authonty: AWE = Atomic Weapons Estahtishment: AECL = Atomic strat:6ed by levels of potential confounding vanables (see belo* L Energy of Canada Ltd_ Observed (0) and espected (E) numbers of deaths were calculated tw dose categories for 47 underlying causes of death denned in the Annen. The espected numbers of deaths were calculated assuming that. mithin a stratum denned by lesels of the confounding sanable.

Defininon of the Santi Popedarmn the mortality rate in each dose category mas the same as that of the For the comNned anal)ws. the study populauon was restncted to the entire stratum. i e.. that the cause of death under study was not assoel-95.673 workers who were monitored for esternal radiation bs the use of sted with esposure, personal dosimeters and who were emp6oyed in any of the panicipating The score test statistic (38) hamad en ths linear relative nsk model facilities for at least a months. Workers with short duranons of ernplos- mas used to test for trends in mortahiy across 11 dose categones (<4. 4 .

ment were escluded as it was thought they might not be comparable to 10 . 20 . 50 ,100 . IS. 200 300 400- and 500- mss). As there was longer-term workers in many aspects related to cancer risk. Nineteen no reason to suspect that esposure to radiasace would bc ==ned with workers who ma) hase received a high dose-rate exposure were also a decrease is nok of any speedc type of cancer, one-sided tests are pre-identined and excluded from the analyses. The specific entenon chosen sented throughout. Because of the skeeness of the donc distribution, use for this exclusion was the fact of having had at least one annual dose of of the normal approximation may exaggerate statistical signincance for 250 mSv or more, the enterion in the U.S for a radnauon ancident of diseases with small numbers of deaths For thss reason. for leukemia medical signincance (J6). excludsag CLL muttaple aveloma and all cases where the test stausue Dosusierrv exceeded 1.28 (correspoeding to a one tailed F value of 0.10) and the The recording of indnidual radianon dose was done to ensure number of deaths was less than 30, the P value presented was esumated compliance with radiation protection guidelines in force at the time using computer sunulassons (39) based on 5000 samples, rather than the and not for epidemiological purposes. The accuracy and precision of nwmal appranmation.

individual dose esumates therefore varied with time, place and radia. To aHow fw a poss Me latent pW Mwun an expmun and m a consequences cumulative doses were tagged oy 2 years for leukemia A and 10 years for other causes of death as follows: with a lag of s years.

persons experwnced in dosametry and radiation pro-scetion from each participatang country (G. Cowper Cand J. Fix. U.S.; annual doses were included in the calculation of the cumulative dose at time r if they had been received in or before time t - 1. Person years 8 was ut W 80 study assencal inc Pracuces in h canous facihtaes with the asas of(1) identifying sources of error and lack were attributed to the category of dose accumulated by that time of comparabihty in individusi done estimates. (2) estimaung the magni- Doses received off-site-i.e. in facilities other than those included in the combined analyses-were treated identically. In particular. a sub-tude of these errors and (3) e _ ;the extent to which recorded g , , ,, g, p o p doses approximated doses y speede organs. Detailed results of this facilit es entered the follow-up with the w....,~6 ding dose-when no study an upwted eluwhere W lag was used-or witti that dose cumulated up to years previously Overall the matonty of the skiess to weders was predonunantly from exposure to higher-energy (100 kev to I MeV) photons, and it was when a lag of years was used.

Estimates of excess relative nok (ERR) per Sv' were obtained using judged est se masunants across facihues and ume wen nasonabh Poisson regression. based on a modelin which the relatne risk

• as comparable. The bias in usang recorded doses as estimates of organ doses assumed to be of the form I + SZ. where Z is the cumulatne dose in St. The relative risk (RR) at a given dose level d compared to zero T Fix. L Salmon. G. Cowper and E. Cardis. A retrospecuve evalua-tion of the dosimetry employed in a :ombined epidemiological analysis.

Unpublished work. *Althougheit is recognued that very few workers recened dose as

'G. M. Kendall and L Salmon. Records of UK caposure to ionumg large as I Sv. this unit was chosen for comparabshty to results reporteJ radiation and their role in epidemiological studies. Unpub4:shed work. for high dose studies.

IoL

120 CARDIS ET At dose. a more commonly used risk measurement in epidemiology, can betacen cohorts. see below) and study population iHantord. ORNL be obtained by muluplying the ERR by the dose d and adding i The Rocky Flats. Sellaneld. Dounreav." South of England" and A ECLi W% con 6dence intersah (90% Cl) for the ERR were bawd on the insormation un ab was obtained for all cohorts except AECL Nr score statistic using the expected information as described by Gilbert Hanford. Job categorv data were used to de6ne four socio.cconomic at.

(39) and Gilbert er of (40). For leukemia con 6dence intervals were epones as desenbed by Gilbert er at (27) For ORNL. a sanable w a based on simulations as desen4ed by Gilbert (39). Model 6tting was sided indicating whether a worker was paid hour!y. weekly or monthh carned out using the computer software EPICURE Tests of homo. For Rocky Flats. mformation on educanonal lesel was used to deme a geneity of the ERR per Sv across a factor (such as facihtyl were three level ciassification (no or some high school. hieh school graduate obtained from the hkelihood ratio test statistic resulting from the com. or some college. college graduate or higher) Sella 6 eld workers were panson of desiances of models with and without inclusion of the rele- classihed only as "industnal" and non industnalf For the rest of the want factor. It is noted that the study had httle power to detect hetero. UK. the Bntish Registrar General's six category social class classihcation geneity of risk across facdaties.

was asailable: for the purpose of the analyses. these were combined to Two approaches were used for the formal comparison of risk esti- form four categones: I . II. ll!-manual. Ill-non manual and IV - V mates from the data for nuclear industry workers and from high dose. Adjustment on facihty or grouping of facihties (as in the UK) where a rate studies. Dese comparisons were restncted to men because the num- worker was employed was Jrned out to account for possible differences ber of exposed women in the workers cohorts was small.

in cancer risk by geographic location or across workplaces Workers First. risk estimates and con 6dence intervals from the data for the known to have worked in more than one facility were assigned to the workers were compared to estimates of risk among male atomic bornb facility of last employment (34).

survivors expowd between the ages of 20 and 60 years densed at l ARC As only hmned informauon was available on tobacco smoking or using data supphed by the Radiation Effects Research Foundation alcohol consumpoon in the participating cohorts, the relations %ps (RERF). The constant knear relatise nsk model used for densmg esti- between radiation esposure and mortahty from smoking related cancers mates for the workers was apphed to the RERF data. restncted to sub-Wh and from non mahgnant respiratort discaaes escluding pneumonia Jects with kerma dose below 4 Gv and attamed age hd sw 75 years. as were studied as indirect indicators of confounding by smoking. and that was done m the analyses of the C S National Academs of Sciences Com-with death trnm hser cirrhosis as a possible indicator of confounding h mittee on the Biological Effects of lonving RaJ; anon IBEIR V) t/) .sicohol consumption Analyses were adjusted for attained age tin 5-year intersaM. ealendar W informahon was asailable on radiation dose from natural back-penca (in 3 9eJr intersalO and eit), and based on nine dow eJtegories pround or medical esposures The tacit assumption was made that. with-to- 5 . $$ . QS .195 . 495 WS .1995 2W$- mss i. E,umates or bone in a tacihn. non.occupanonal radiauon dose level was mdependent ut marrow dose laefed by 2 years and of stomach dow lagged bs 10 scars occupanonal dose lescl. thus non occupauonal radiation dose w as not were used. respectnel), for esumating the nsk of letandmaa and that'ef alt *consrdered to be a contounder of the associanon between occupanonal tancers escluding leukemia The quakt) factor" for neutrons was taken radiauon dose and cancer risk. Differences m natural background radia.

o be 20. as in the BEIR V anahses t / t non leseis between geographical areas were taken mio account bs the Second. the models derived by the BEIR V Committee til for esu- adjustment on facihty desenbed above.

maung nsks of leukemia and respiratory. digestne and other cancers were apphed to the data for the workers. Specincally. the annual doses of male workers were weighted. for each year of the follow up. according to RESULTS the age at which they were recened and the ume since the dose was received, usmg the BEIR V coef6cients for men. The resulung nsk esti- Characteristics of the study population are shown. bv mates and conAdence mtervals were expressed as multiples of nsk under facility. In Table !!. The mean cumulative radiation dose m the BEIR V model. Respiratory. digestne and other cancers were then the combined cohort was 40.2 mSv per-worker and the col-analyred sirnultaneousts, stratifying on these three cancer groupings lective dose was 3.843.2 Sv. The total number of person-14/), to obtain an esumate of nsk for allcancers excluding leukemia as a years PY) at risk was 2.124.526. A total of 15.825 deaths multiplier of the BEIR V esumate. For leukemia. onh the linear term uf the model was used because, although the preferred BEIR V modelis Ecurr(ed during the ' study period. Women compn hnear quadraue in done, at low doses and dose rates the contnbuuon of than 15% of the workers and their mean cumulative dose the quadrauc term is nel bgable, was low (6.2 mSv) compared to the men's (46.0 mSv).

ne magrutude of the esurnates and conadence micrvais for the work-Overall, the distribution of doses was very skewed (Fig.1);

ers was. in addition. cc'mpased to the estimates for men aged 20 and close to 60% of subjects had cumulative doses below 10 above at the time of exposure denved by the United Nanons Scienunc Comtruttee on the Effects of Atorme Radiauon (UNSCE AR. ref. h using mSv. 80% below 50 mSv;less than 2% had doses greater a quahty factor of one for neustons, truncatmg doses at 6 Gv and straufy, than 400 mSv.

ing age at exposure m three categones. The UNSCEAR coefhcients Observed and expected number of deaths by cumulative sersed as a basis for the current recommendations of the Internanonal dose and trend test statistics for the 47 causes of death are Commasson on Radiological Protection (ICRP. ref. 4) presented in Table III. There was no evidence for an associa-Confounduvg facrors tion between radiation dose and all-cause (e = 0.23) or an-All analyses were adjusted through strau6 canon for set. attained age (by 5-year mtervals). calendar penod t by 5 year categones), socio-eco-nomic status (SES) within facihty (as mformanon asailable differed

" Note: In the data provided on UK workers. it was not posuble la distinguish AEA and AWE workers. To ad;ust for possible environmen-tal differences across the facihues of the UK AEA and AWE homeser

• Ouahty factor weightmg factor apphed to an absorbed dose to take information gas provided to separate workers from Scotland iDoun.

to account the type and energy of the radianon causing the dose. reav) from those of the South of England.

/o 3

l l

LOW. DOSE RADIATION AFD CANCER: A THREE-COUNTRY STL D) 13 TABLEII Distribution of Workers. Person Years and Collective D0se by Facility

• l UK other I Hanford Rocky Flats ORNL Sellafleid

. .._ . . . . . . . . . than Sellaneld AECL Toul Number of workers 32.595 e.638 6.591 9.494 29.000 11.355 95 A t Men 24.628 6.6.18 6.591 H.800 26.495 Women 7.967 8.591 . al. 45 )

0 0 692 2.505 2.764 t.t.9:s  !

Recruitment penod 1944-197M 1951-1979 194k1972 1947-1976 1 % I982 1956-1980 Follow up penod 1944-1986 1951-1979 1943-1944 1947-19MM 1946-1988 1956-1985 Number of deaths 6.445 587 1.246 2.027 4.629 891 L5.825 Number of cancer deaths 1308 109 104 544 1

1.272 239 3.976 Person years 781.549 100.022 173.730 233.(N0 637.925 198.210 2.124326 '

Collective dose (Sv) 877.2 241.8 141.4 1.309 6 958.6 314.6 3.843 2 Men 831.6 241 8 141.4 1.294.5 9M4 311.8 3.757 5 Women 45 6 0 0 15 1 22.3 2.8 83 8 h results are restneted to monitored workers who were employed at least 6 months in any of the participetsng facihties and exclude wo having received a done of 250 mSv or abose in a single year.

cancer mortality (P = 0.51). Among non-cancer causes of negative relationship between radiation dose and mortality death, mortality from circulatory diseases was significantly fron respira'ory diseases excluding pneumonia.and from associated with radiation dose IP = 0.045). There was a weak hser cirrhosis. These conditions were considered a l

25000a Hanford 25000- Rocky Flate 25000- Oak Ridge 20000' 20000- 20000 se e

i g1S000< 15000- 15000' '"

10000- 10000' "

10000a 3 .= me see amo ese ' ma am .= se e= i"'"***"

S000' # ""*

• S000<

f5000' /

O O '-

O 0 100 200 300 400 500 0 100 200 300 400 SOO O 100 200 300 400 600 dose (mSv) does(mSv) dose (mSv) 25000< Seele#ete 25000< Other UK 25000 Canade 20000< 20000< 20000 a "

g1SooO. -

1SOOO< '

1SOoo<

10000' 10000< 10000' 3 an. es. .==m.== an am em 5000< ""**"** *"**"** ' " * "

SOOO- S000-Ou 0 0

m 0 100 200 300 400 500 O 100 200 300 400eSOO O 100 200 300 400 500 dose (mSv) does (mSv) dose (mSV)

FIG.1. Distnbution of monitored workers by cumulatne dose.

/e4

122 CARDIS ET AL.

TABLE Ill 6 served and Expected Numbers of Deaths by Cause, All Facilities Combined Adjusted for Sex. Age. Calendar Period.

SES and Facility (Doses Are Lagged by 2 Years for Leukemia and 10 Years for Other Causes of Death)

Cumulatne dose (mSv) 0- 10- 2fb 50- 10G. 20G- 400- Total Tr'end' Causes of death OsE O-E O/E O/E OE OE OE Deaths (1 sided P)

Alic uses 9582/9506 3 1848/1915 M IMMon1909 9 989/986 5 ?02'719.2 5x6/549 2 238/238 2 !!.825 0 ?5 (0 22m All cancers 2317/2317.3 483/483.5 465,494 7 285/263 2 201/196 8 165'151.5 60689 3.976 -0.02 (0.508 All except leukemia 2234/2228 3 462/465 4 445/476 9 276/254 3 196/190.5 161/147.5 56673 3.830 -028(0 609)

Buccal and pharynx 51/42.3 10/8.0 G7.0 3/3.9 t/31 2/2.7 t /0.9 6A -1 10 (0 R64)

Esophagus 52,48 8 13/13.1 16/15.3 6/96 6n s 6/63 5/32 104 0 32 (0 375)

Stomach 158/155 1 34/31.7 29/35.2 24/20.9 13/15 3 11/11.1 6/5.7 275 -0 21 (0.582 p Small intestine 66.1 f t.6 1/l .8 t/1.0 1/l.0 110.4 0/0.1 12 0 36 (0.360)

Colon 220/204.7 36/42.8 32/41 3 22/21.9 19/15 4 9/11.6 5/53 343 -0 83(0.797)

Rectum 7477.7 15/14.2 2&l7.0 5/96 10/7.6 6/6.1 5/2.8 135 0 99(0 161)

Liver 24/21.4 3/41 4/3.7 0/1.7 &l.1 2/0.7 GO.2 33 0.01(0.4951 Bihary tract 23/21.4 3/3.8 4/3.2 0/ t.4 108 0<0 4 0/0.1 31 -0 86(0.8061 Pancreas i 15,1 I3.2 27/24.5 2G23 4 10/I l .6 7'86 7 6. A $/2.9 191 1.20(0 115i Nasal canty 6,6. 2 0/0.7 21.1 (M14 iwLt 102 0,0.1 9 043(0334)

Larvns 2401 A 3 '4.0 242 3,1 A l'13 Oro s 1,0.1 34 1.17(0 1221 Lung 676'689 4 144154 2 154161 2 111-87 6 15 67.0 64540 l&246 1.238 -0.28(0 6l0)

Pleura 4Mi . .t 2.5 534 3 11 31.6 2.1.4 a 1.0 20 -0 18 0.571)

Bonc 10 M.2 007 10m 004 iM)3 IH)4 0/02 Ii -1.2110.887)

Connective tinue 10'12.0 32.2 222 u,1 0 til M i06 IkO2 19 0 3610358)

, Melanoma 32311 34a 244 424 21.7 213 W)4 46 0.21 0.416i Female breast 69 'l 3 x59 239 416 i 1.1 ius2 n00 54 0.50(0.3086 0

Cervin uteri 7x4 10A 20.5 iw): 0/00 (ku 0 uro.0 10 0.63(0.2666 Other uterus "7 Y7 "' ' " 81 d7 11.4 102 000 040 0/0.0 10 1.7110 092 r Os arv IM.193 2 12 113 abo n 1 03 iMil 0i00 24 0 4910 312) i Prostate 127.1.16 4 47 35A 4370 21 18.0 13,12.5 11 11.1 15.2 256 -1.68 (0.953)

Tesus 1414.6 114 21.4 107 145 00.3 0/0.1 19 -026(0604)

Bladder 51;53 4 17/132 16,14 4 5Mi 5'6.7 6/54 4/30 104 0.62 (0 266)

Kidney 54507 7/10 4 11,II 2 759 6/46 3/.14 0/1.7 88 -1.03(0 848)

Brain and CNS 73/75 l 12/14.5 19/143 9/7.7 4/5.5 40.6 1/13 122 -0.24 (0.593)

Thyrod IGM 4 2/13 12.3 111 3 0/0.9 00.6 1/0.3 15 0.58(0381) lli-defined and secondarv 151/145.8 32/34 5 33< 33.9 13/16.7 12/12.8 11/9.4 5/4.0 257 063(0263)

Non.Hodgkin's lymphoma 76/81.6 24'l6 9 1.1/16 3 IIIM.5 Se6 2 6/4.1 0/1.5 135 -025(0.600)

Hodgkini '~me 3:314 2.38 434 ib t H 3,12 2 ' t .0 40.5 43 0.28 (0 390)

Multiple myeloma 28/26 6 3s52 1/4.7 5,2.7 32.1 2/1.9 2!0.8 44 187 (0 037F Ailleukemia 72/75.7 23/21.2 2021.8 12/11 3 9/78 4/5.5 6/2.6 146 1.43(0 076)*

All escept CLL je0/62.0l 19/,17.2 l14/17.4l l8/9.0l 8/6.4 (4/4.7l 6/23 119 1.85 (0.046)*

j CLL 12/13.7 4<4 0 644 4/22 1/l.5 00 9 003 27 -0.74 (0.771)

Acute leukemias 33/31 3 7/9.0 1179 9 3<5 3 3/36 A'2.6 2/13 63 0.82 (0.206) i 4

Other cancers 32/29.8 4/6.2 7 '6.0 5/3.0 0/2.5 2/1.7 0/0.9 50 -0.87 (0.808)

Smokmg related 969/968.9 214/217.0 208/225.4 138/122.6 95 S4.5 85n5 9 30/34.7 1.739 0.29 (0.386)

Non-smokmg.related 1348/1348.4 269/266.5 257/269 3 147/140.7 106/102.4 8G75 6 30/34.2 2337 -0 32(0.624)

Alt non-cancer causes 72tL5/7189.0 1365/1432 3 1415,1415.2 7Gu'23.2 501!522 4 421/397.6 178/169 3 11.849 0 90(0 184)

Non. malignant iumors 24/23.1 lie d 6,48 t /2.5 3/1.7 1/1.1 GO.4 38 -026(0.601)

Circulatory diseases 4689 4626.5 908/975.7 954/975 4 487504.4 372/366 6 313/282.6 132/123.9 7.855 1.69 (O G45)

Respiratory descases 421/447 3 117d06.0 129/102.5 53496 2507.1 30/30.7 12/13.8 787 -0 82 (0.795)

Liver estrhosts 12Gil6.3 l9/2! .6 18,17 6 148.4 5/6.2 4/4.5 0/1.4 176 -1.29 (0.902)

External causes 880/880.0 99/97 9 9t/923 53462 33/31.7 14/19 6 3/53 1.173 -129(0 901)

Unknown 95/84.0 23/19.8 1il16 3 2'71 35.1 2/3.7 1/1.1 137 -1.76 to %1)

• Trend est based on 11 done catesones; can be compared to a standard normal distnbuuon. Howe er. stausucal sigmar==ce may be exaggerated diseases with a small number of deaths;

• denotes simulated P values (see Methods).

• r

/ -4= . ,/

^

7m;dg;_

j

.2 m i

-,w a pp kw/M 1%~ q loi

F h

L LOW. DOSE RADIATION AND CANCER. A THREE. COUNTRY STL D) [y 4

TABLE IV Table IV shows estimates of excess relatise risk per Ss Excess Relative Risk (ERR) Eatimates per Sv. and of relative risk for 100 mSv compared to O mss for 90% CanMence Inservals (CI) and Relative mortality from all cancer, and from leukemia and leukemu Risk (RR) Estimases at 100 mSv for subtypes. The ERR of 1.22 per Sv for leukemia excluding All Cancers and leukemias CLL would correspond to an excess of 9.7 leukemia deaths (8% of all such deaths) attributed to radiation in this cohort Type of cancer desi S. mut sNS(*

• if the observed association were causal. The ERR per Ss was greatest for myeloid ieukemias. in particular chronic All cancers 3976 4 02 ( 4 34.0.35) 1.00 myeloid leukemia (CML). but the confidence intervals for Except leukemia 3830 4 07 (-0.39. 0.30) 0 99 each subtype overlapped.

( 13.'5 7)( The ERRS for all cancers excluding leukemia and A & LL' il -0.39 (c0.d 7.31* 0.91 leukemia excluding CLL are presented separately by sex CLL' 27 4 95 (<0.' 9.4)* 0.91 and by facility in Table V. The ERRS for men were verv CML N l 9 hij similar to those observed for the entire population. The r

_ _ . . - _ . - . . . _ . . . - _ . . . point estimates for women differeo from those for men. but

'S** M*'h*d5-the con 6dence intervals were wid6 and the formal test for AL e h t71 u em A e te r yeloid leukemia. homogeneity provided no evidence of a difference in risk CLL: chronse lymphoevtic leukemia: CM L: chronic mveloid leukemia. between men and women.

' Lower bound would lead to negative relauwe risks en the low-dose The ERR for all cancers except leukemia ranged from ra23e (<300 mSvi. -l.63 per Sv (90% CI <0. 0.5) at Rocky Flats to 1.66 per Sv (0.04. 4.4) at ORNL. The confidence intervals were wide.

howeser. and the test for homogeneicy indicated that these ,

indicators of possible differences in smoking and alcohol con. differences could be due to chance Auctuation. For I sumption across dose groups. There was, in adilition.'little seukemia excluding CLL. the ERRS ranged from -106 to l evidence of an association between cumulative dose and 48.4 per Sv across facilities. Here again. the confidence i mortality from smoking-related cancers as a group [de6ned intervals were wide and there was only weak evidence for I rs cancers of the oral cavity and pharynx. esophagus, pan- differences in risk across facilities (P = 0.08),

creas. larynx lung. bladder and ren2l pelvis (42)l (P = 0.39). Table VI summarizes the results of comparisons with Among individual cancer types. mortality from leu- estimates derived from high-dose studies based on the ktmia excluding CLL was signi6cantly related to radiation approaches outlined above. All estimates from high-dose dose (P = 0.046), as was mortality from' multiple myeloma studies are based primarily on the follow-up of the atomic (P = 0.037). bomb survivors. For all cancers excluding leukemia, the TABLE V Number of Deaths and Excess Relative Risk (ERR) Estimates per Sv and 90% CaaM*=ce Intervals (Cl) for All Cancers (Excluding feukemia) and Leukemia (Excluding CLL) by Sex and Facility

_All cancers escluding leukemaa _taukemsa escludsag CLL Subpopulation Number of deaths ERR per Sv 90% Cl Number of deaths ERR per Sv 90% C!

Sex Male 3522 -0 07 (-0 4. 0.3) 109 ' 2.21 (01. 5 8f Female 3(5 0.97 (<0 9? 8.2) 10 -2.67 (<0?127f x for homogeneity (I df) 0.11 P = 0.74 0.07 P = 0.79 Facility Hanford 1452 4 22 (<t1* 0 6) 47 4 90 (<0? 2.9 f Rocky Flats 104 1.63 (c0? 0.5) 4 4.0s (<-0 0. 54 2f ORNL 230 1.66 (004.44) 18 -1.06 (c0? d 8f SellaAeld 533 -0 03 q <0? 0.51 10 43.50 01. > too r Other UK 1227 -0 40 (<0f0.7) 35 1.50 (<0? 14 3 f Canada 234 0.13 ( <0.* 2.16 5 48.40 (2.8. > 100 f x2for homogeneity (5 d/) 4 86 P = 0 43 9.91

. . _ - _ . . . . , . . . . _ . P =__0 0R

~

• Simulated con 6dence meervals (see Methods).

• Lower bound would lead to negative relauwe nsks an the low-dose range a c500 mSv).

10 4

124 CARDIS ET AL.

TABLE VI strategy on estimates for the workers. Thev include (a)

Comparison of Excess Relative Risk (ERR) Estimates per analyses based on all workers whether thev we're monitored Sv'(and 90% Canadence Intervals) tretween Nuclear or not. all monitored workers (including' those employed Workers, Atomic Bomb Survivors and Other less than 6 months). and all workers in the restricted Published Fadmaa= of Risk from dosimetry population (see Methods above):(b) analyses in High Dose Studies: Men Only which cancer was treated as the cause of death if it 4as either the underlying or an associated cause of death; and oc iag Uu*kemn escudIng[t.L (c) analyses using different treatments of potential con.

ERR ERR ng acton Populanon per Sv 90% Ci per Sv 90% Cl Except in the following cases. the variations in analytical Nuclear.orkers data * -0.07 Fo n.o.30) 2.18 to.13. 5 7r approaches presented in Table VII had little effect on the

4. bomb.' haear o.1s t o.05. 0.34) 3 67 (2.0. 6 5) risk estimates. When cancer as an associated cause of death A bomb.' L-Q' - -

1.42 UNSCEAR 0.24 (co. 6.5) was included. the ERR for all cancers excluding leukemia 37 -

increased from -0.07 to 0.01 per Sv (90% CI: -0.3. 0.4). For F Mulupher/ 90% Ct Muinpher 90% CI leukemia excluding CLL. the effect was to reduce the ERR from 2.21 to 1.78 per Sv (90% CI:-0.1.5.1). Adjustment for

b. linar . . iy duration of employment at Hanford only (the only facility where an association between all-cancer mortality and

'Esumates of organ dose and I cm depth Jose were used respectnely duration of employment was demonstrated

'"*[h'd*'*""**"'"'*"*"d""C'"'*d") ties increased the ERR for all cancers excluding leukemia

'Ad usted i for age. SES. facihn and calendar ome to 0.03 and 0.08 per Sv. respectively. For leukemia exclud-

' Simulated conndence intersal ing CLL. adjustment at Hanford increased the estimate to

'A bomb: dau nor aiomic tomo unnort adjuwd roi age, cii> .nd 3.63 per Sv (90% CI: 0.8. 8.7). while adjustment in all facili-calendar ame-

' Based on the knear term of a knar.quadraue (L-On Jose-response t es reduced it to 1.72 per Sv (90% CI: <0. 5.5). Not adjust-

• . ina for SES in any facilitv increased the ERR for all cancers "I,$,',,7,$,*,,$p",,'[d g a mulople of the high dow excluding leuker'nia to 6.2 per Sv (90% CI:-0.2. 0.6) and IRR. for example the leukemia risk esumate is u.71 urses the Rela v had little effect on the ERR for leukemia excluding CLL.

snmate =nh conadence mienal rangmg from -o tw umes to twice the BEIR V esumate. Analyses aimed at assessing the in6uence of potential effect modi 6ers (attained age, age at exposure and time since exposure) and analyses using alternative lag periods (0,5.10.15 and 20 years) are presented in Table VIII.

excess relative risk obtained for male nuclear workers was Attained age, age at exposure and time since exposure had less than that estimated at IARC from data for the atomic little effect on the ERR for all cancers excluding leukem bomb survivors. The con 6dence interval was wide, however. or for leukemia excluding CLL and ranged from -2.2 to 1.7 times the estimate for the atom.

As the lag period used went from zero to 20 years the ic bomb survivors. The Eit R for male workers was estimat. lagged cumulative done decreased and the ERR for all can-ed to be -0.17 times the BEIR V estimate with a confidence cers excluding leukemia increased monotonically from interval ranging from -0.8 to 0.6 times the BEIR V estimate.

-0.21 per Sv (90% CI:-0.4 to 0.1) to 0.14 per Sv (90% CI:

i It was also less than the UNSCEAR estimate. -0.5. 0.9) (Table VIII). The ERR for leukemia excluding The risk estin:ste for leukemia excluding CLL obtained CLL also increased from 1.89 per Sv (90% CI: 0.0. 5.2) to from the data for nuclear workers was greater than the esti- 5.13 per Sv (90% CI: 0.9.13.2).

mate based on a linearquadratic model and less than that based on a linear relative risk model obtained by reanalyz-ing data for male atomic bomb survivors at IARC (Table , DISCtJSSION VI). The con 6dence interval for the estimate for the work. This study combined mortality data from seven previ-ers was relatively wide, however, and ranged from 0.04 to ously published studies of nuclear industry workers in three {

l.6 times the linear estimate for the atomic bomb survivors. countries. The studies selected for inclusion in these analy- '

The ERR for male workers was estimated to be about 0.7 ses met a series of stringent quality criteri. Mned by the times the BEIR V estimate with a confidence interval rang. Study Group in the planning phase of the study. In addi-ing from -0.04 times to twice the BEIR V estimate. It was tion. efforts were made to ensure comparability of available  !

also less than the UNSCEAR estimate. radiation dose estimates across facilities and over time. l Table VII presents the results of additional analyses The estimates presented here are the most precise and signed to assess the impact of the choice of analytical comprehensive yet to have been obtained directly from foi

. . . _ _ . . - -. - _ _ ~ ~ - . . . ..__ . _.- .. - _ - - . - ~ . --- . .-

1 I

i LOWDOSE RADIATION AND CANCER. A THREE COUNTRY STtlDY [2f TABLE VII Number of Deaths and Excess Relative Risk (ERR) Estimates per Sv and 90% ConAdence Intervals (CI) for all Cancers (Excluding leukemia) and Leukemia (Excluding CLL) Using Alternative Analytical Strategies

, _ , Allcancers escludmgleukemia , _ Leukemia escluding CLL Number of Number of Subpopulation deaths ERR per Sv 90% Cl deaths ERR per Sv 90% Cl Standard 383n 4 07 (-0 4. 0.3) 2.I8 119 to 1. 5.7r Study populations l Total workers' 6444 -0.05 (-0.4. 0.3 ) 187 2.26 (0 3. 5 6F All monitored' 4180 -0.08 (-0.4. 0.3 ) 133 2.09 t 0.1. 5.2 r Restricted dosametry' 3455 -cot (-0.5. 0.5 ) 108 2.05 (-0.1. 6 4 r including associated causes' 4113 0.01 ( -0.3. 0.4 ) 127 1.78 (41. 5 t F ENect of SES No adjustment 3830 0.20 (-0.2. 0.6) t19 2.28 (0.2. 5.7 f ENect of duration of employment Adjusted-all facahtees 3830 0.08 (-0.3. 0.5 ) 119 1.72 (-0.3. 5.$ r Ad)usted-Hanford 3830 0 03 (43.04) 119 3.63 (0.8. 8 7 f a

• Simulated conndence intervals (see Methodst

• Includmg non.momtored workers m the zero-dose category.

' Including monstored workers employed less than a months.

'Excludmg workers judged to hase potential for sutMtantial dose from neutrof% anda.r miernal esposures (see Methods t

'Includmg cancer deaths hued on the death certineate as either undern mg or .moeiated uuses of death.

i ,. ,a. .. 3.-

populations with low level protracted exposure to mainly ICRU (J3) for radiation protection]. in most cases available y radiation. By contrast. the estimates obtained from the dose estimates overestimated organ doses by several per-atomic bomb survivors are innuenced by subjects receiving cent. The Dosimetry Committee judged that bone marrow doses of 500 mSv or more and relate to acute exposure doses were overestimated by about 20%. implying that the over a very short time. Our analyses imply that the esti- present leukemia risk estimate and con 6dence limits may mites obtained from studies of high-dose-rate exposures be underestimated by 20%. Random errors in dose esti-are unlikely to underestimate substantially the actual risk mates are likely to further bias the risk estimates down-at low doses and low dose rates. A signincant increase in ward. compared to estimates from high-dose studies which leukemia risk was observed in this study. at relatively low have been based on organ doses.

dose levels: the risk estimate was intermediate between the Snme workers in the UK and the U.S. were known to lintar and linear-quadratic estimates from studies of atom- have been employed in more than one study facility within ic bomb survivors. those countries. Efforts were made to reconstruct their

. , detailed employment and exposure history, particularly in Uncertamnes the UK (32). Doses incurred after ternunation of employ-Several points must be kept in mind when making com- ment in one of the study facilities were not generally avail-paruons of these worker-based risk esumates and con 6dence able, however and it is difncult to assess the impact of these bounds with those based on studies of high dose-rate expo- missing doses. .

sures. The most important are possible biases and uncertain- The ascertainment of vital status was 92-100% complete.

ties in dose estimates. errors in outcome data and inadequate and ascertainment of cause of death was 98-100% which adjustment for confounders. was required to meet the criterion for inclusion in the com-The risk estimates obtained for the restricted dosimetry bined analysis. Any misclassincation of vital status or of population." i.e. excluding workers who could be identified underlying cause of death is unlikely to have been related as having received substantial doses from radiation other to radiation dose and. if present, would tend to result in a than high-energy photons (Table VII). did not differ substan- small bias of estimates towarQ the null-tiilly from those based on the standard approach although In this study, adjustment for SES had a strong effect on the uncertainty in the risk estimates increased slightly. the risk estimate for all cancers excluding leukemia. As Although it appeans that. for the majority of workers in the type and tietail of information available from each thIse facilities the dose estimates were compatible with facility varied substantially and as no information was I cm depth dose [the quantity currently recommended by available for AECL workers,it is not possible to exclude lot

126 CARDIS ET AL.

TABLE VIII Excess Relative Risk (ERR) Estimates per Sv and Confidence Intervals (Cl) for All Cancers Excluding Leukemia and

( bukenua F-ha%g CLL by Attained Age, Age at Exposure Time zince Exposure and Different Lags All cancers escluding leukemia Leukemia eteluding CLL Charactenstics ERR per Sv 90% Cl ERR per Si 90% CI Standard al.09 (4 4. 0..t) 2.1 N (01.57)

Attained age

<65 0.07 (-0.5. 0.7) 0.35 (<0? 3.6i 65 75 -0.03 i 0.s. 0.3) 7 26 ( <0? !8.4) 75+ 0.22 (-1.0. I 4) 5 51 (<0? 22.6) x for homogenesty(2df) 0.84 P = 0.60 2.61 P = 0.27 Age at exposud

<35 0.43 (- l . i .1.9 ) -1.86 (<0? 10.5) 35-50 -035 (-01. 3.0) 3.37 (<0? 5.7 6 50+ 0.17 (-0 6.1.0) 2.20 (<0? 8.3) x2for hornogeneny (2 df) 0 89 P=064 0.19 P = 0.91 Time since espmure (yearu

<10- n.a? 1.M2 (<0? 8.9) 10.20 -4 48 s -l . t h. 0.20) -l.15 (<0? 41) 20- 0. 4 t -0 44.1.16 6 14M (<0

• l t.1 )

g2 for homogeneits 1 60 (l do P = 0 21 14x (2 df) P = 0.48 Lags for Jme 4)sar4 0 4 21 d -4 4 019 1W (H.O. 5 2 j '

5 4 ! .% t -O 4. U2 f 2JIN (0.0. 5.7 r 1H 4 O' ( -4 .1. 0.11 ' 2. IN 4 d.1. 5.7 f 15 -d H4 (-0.5. 0.5 f 1.% (0.A. 8 M 20 H I4 4 -O 5. 0.9 f  ! 1.1 a 0.9.1.1.2 r

" Lower bound would lead to negause relat se risks in the low dose range (<300 mss t

$ot appiscabic; analyses for all cancers are lagged tS 10S cars.

' Score bened 90% con 6dence intersals.

dSimulated 90% con 6dence mtersals.

.T entirely the possibility of residual confounding by SEf, or results of two studies (44. 45) canied out. respectively, with-by a variety of lifestyle factors associated with cancer risk. in the Hanford and the UK AEA cohorts which showed lit-and for which the SES variable may be an imperfect meas- tie evidence for an association between smoking and occu-ure. The ERR for leukemia excluding CLL was little pational radiation dose.

changed by adjustment for SES. whereas that for ali can- A positive association between radiation dose and mor-cer excluding leukemia was decreased. The leukemia risk tality from circulatory disease was observed in the three estimate therefore appears to be less sensitive to con- cohorts where information on SES was least detailed founding by SES. (Rocky Flats. Sella 8 eld. AECL). It may the efore reflect As in most occupational cohort studies. information on residual confounding by lifestyle factors for which the SES lifestyle factors such as smoting bbits. diet and occupa- variable is an inadequate proxy. Alternatively, given the tional exposures could not be obtained retrospectively for large number of. associations tested. this could be a chance all members of the cohort.. There was little indirect evi- finding. It should be noted, however, that such an associa-dence, however, for an association between cumulative tion has also been seen in studies of atomic bomb survivors dose and mortality from smoking-related cancers. resptrato. (46) and U.S. radiologists (47).

ry diseases or liver cirrhosis: it is thus unlikely that smoking There was little evidence of an increase in the ERR for or alcohol consumption is strongly correlated with radiation all cancers excluding leukemia with attained age, as was dose and that adjustment for these factors would greatly reported in recent analyses of the U.S. data alone (31).

affect the conclusions of the study. This is supported. for tobacco, by the observation that the risk estimates for all Leukemur Risk cancers excluding leukemia and all cancers excluding both The combined analyses of the data for the workers leukemia and lung were nearly identical (34) and by the demonstrated a significant (P = 0.046) association between l

/of

i i

LOW. DOSE RADIATION AND CANCER:A TH REE. COL NTRY STI'DY 12-motwy from leukemia excluding CLL and radiation dose differeat weight given to doses received in different periods in a population receiving protracted low-dose rate espo- by the two models.The constant linear relatise nsk moJJ sures. The ERR for mortality from that disease was 2.18 used for these analyses gives equal weight to all dose 3 per Sv (90% CI: 0.1. 5.7). Out of 119 leukemia deaths received in the past. As discussed in greater detail in ref.

observed in the combined data set. however, six occurred in (34). the BEIR V model reduces the weieht by half for '

the 400 mSv and above category, a dose range comparable doses received 26-30 years in the past and to cloie to zero to the lowest dose range in which excesses were demon- for doses received 31 years or more previously. Four of the strated in the atomic bomb survivor population. When high-dose (400 mSv and above) leukemia deaths had their analyses were restricted to cumidative doses below 400 mSv cumulative dose reduced by this approach.

and below 200 mSv. to assess the innuence of death in the The coef6cients of the BEIR V leukemia model were higher-dose categories on the dose-response relationship. esumated from analyses of data on atomic bomb survivors the assocue.1 was no longer statistically signincant but the and ankylosing spondylitis patients. The uncertainty in estimates of the slope parameter were compatible with that these is relatively large. For example, the BEIR V model based on the full data set (34). for leukemia predicts that the risk varies with time since Although positive associations were seen between radia- exposure. An analysis of risk by time since exposure (2-2.5 tion done and icukemia mortality in four of the six facilities 26-30.31+ yeers) using the data for the nuclear workers or groupings of facilities (they were signi6 cant in two). the provided no evidence of a reduction of risk after 25 or 30 risk estimate largely renects the association in the Sella 6 eld years (/ test for homogeneity: 0.84. 2 df. P = 0.66); the cohort. Activities at the Sellaneld plant included reprocess- power to test such an effect in the data for the nuclear ing of nuclear fuel; the possibility that chemical esposures workers was, however. extremely low, could have confounded the association between radiation When a factor of 1.2 was applied t3 the risk estimate. as exposure and leukemia risk cannot be excluded (32). suggested by the Dosimetry Subcommittee to adjust for the The observed association between radiation dose and probable overestimation of the dose to the bone marrow.

mortahty from leukemia excluding CLL appeared to be the ERR for leukemia excluding CLL became 2.6 per 3s restricted to myeloid leukemia. particularly CML. although and was compatible with risks up to 2.4 times that of one could not exclude the possibility that the risk of acute BEIR V and twice the linear estimate based on male atom-myeloid leukemia was of the same order as that of CML. ic bomb survivors exposed between the ages of 20 and 60.

This finding is consistent with the results of a recent reanalysis of the data from the National Registry of Radia- AllCancers Excluding Leukemia tion Workers (NRRW) in the UK (48); most of the NRRW The combined analyses of the data for the workers did cases, however. were included in the combined data set. not provide evidence for an association between all-cancer While the risk estimate for leukemia excluding CLL mortality and radiation dose (P = 0.51). The estimated derived from the data for the workers was less than ERR for all cancers excluding leukemia was lower than both the linear estimate derived at I ARC using data for both the estimate derived at IARC from the data for male atomic tiomb survivors exposed as adults and the atomic bomb survivors and the BEIR V estimate. The UNSCEAR estimate. the 90% confAence interval was con 6dence interval was wide, however, and the possibility wide, and the possibility of a lower risk or that of a risk up of a risk up to 1.7 times the linesi catimates obtained from to 1.5 times the linear estimates obtained from high-dose data for atomic bomb survivors. or up to 0.6 times the data (and up to four times the linear-quadratic estimate) BEIR V estimate could not be excluded (Taue VI). This could not be excluded.The risk for leukemia excluding apparent discrepancy in upper con 6dence limits arises l CLL was also estimated to be less than the BEIR V esti- mainly from the faut that the BEIR V models, based on ,

mate with a con 6dence interval ranging from less than zero observed patterns of risk over time in a number of studies l to two times that estimate. As the BEIR V model for of high-dose exposures, give different weights to doses i Irukemia includes both a linear and a quadratic term in received at different ages and in different time-since expo- )

dose, however, the BEIR V estimate of risk at low doses sure intervals (34). l and dose rates is about one half that which would have There is uncertainty concerning the appropriateness of l been obtained with a linear model. Hence the estimate for the BEIR V time and age-speci6c coef6cients for cancers the workers is compatible with risks of the order of the esti- other than leukemia:indeed. recent analyses of the data for mate which would be obtained using a BEIR V type linear the atomic bomb survivors have shown little effect of time model but not much higher. since exposure on the risk estimates for respiratory cancer The apparent discrepancy between the comparisons and little indication of differences in the temporal behavior based on the analyses of the data for atomic bomb survivors of risk between the BEIR V groupings for respirator).

and those based on the BEIR V modelis explained by the digestive and other cancers (41).The compansons with the J10

128 CARDIS ET 4L BEIR V Committee estimates for all cancers excluding malignancies. did not provide evidence of an association leukemia should therefore be interpreted with caution, with radiation dose in that population (5/).

Overall. the results of our analyses for all cancers exclud- Among nuclear industry workers. statistically sigrtificant ing leukemia provide evidence that the estimates obtained associations between multiple myeloma mortality and radi-fromi studies of high-dose exposures are unlikely to under- ation dose have been reported previously for workers at est mate the actual risk after protracted low doses substan. Hanford (2/) and Sella 6 eld (ll), the two facilitie> included tially. De estimates for the nuclear workers are compatible in the current combined analyses where reprocessing of with risks up to the order of the BEIR V estimate and twice nuclear fuel is carried out. In the most recent analyses of the estimates based on male atomic bomb survivors the Hanford data (27). however, this association was not exposed between the ages of 20 and 60. statistically signincant (P = 0.10).

The follow-up of the UK AWE study (16) and recent in the combined data set the association was statistically updated analyses of the ORNL study (24) have provided signi6 cant (P = 0.037) largely re6ecting the associations in excess relative risk estimates for all cancers which were sev- the Hanford and Se'.la6 eld cohorts reported previously.

eral times greater than that estimated in the study of the Although there was no evidence of inconsistency between atomic bomb survivors. In our analyses, the ERR for all facilities, tests of consistency have very limited power since cancers excluding leukemia in Oak Ridge workers was 1.66 multiple myeloma is a relatively rare type of cancer. A causal per Sv. higher than for workers in other facilities [but lower association between multiple myeloma and radiation dose is than that reported by Wing and colleagues (24) for reasons not inconsistent with the evidence from other studies. Since a explained by Gilbert and collaborators in ref. (2/)j. large number of associations was studied in this report. how. .

although the formal test of consistency provided no evi- ever. a P value of 0.037 is not particularly unusual and I dence that the ORNL estimate fell outside the espected chance is a possible explanation for the observed association.

• range of random sariation. As the data for the AWE sup. The ERR for multiple myeloma was 4.2 per Sv (90% CI:

plied to IARC had been combined with the data from 0.3 to l4.4) in the combmed data set. larger than the estimate AEA facilities. it was not possible to calculate a separate for leukemia and much larger than the estimate for all can-estimate for that cohort. Overall. however, there was no cers excluding leukemia. This cancer type was. howeser.

evidence of non-homogeneity of nsk across facilities. Selected for risk estimation because it was the only one (apart from leukemia excluding CLL) which was significantly Sing /c Cancer Types associated with radiation dose. The risk estimate. like all sim-Most of the 36 cancer types or groupings of cancers stud- ilarly chosen estimates. may therefore be biased upward.

ied showed little or no association with radiation exposure. Since multiple myeloma is a disease with a late age at This could, however, have resulted from lack of power of onset, there is concern that it is underdiagnosed and thus the combined analyses for detecting such risks. As dis- unrecorded on death certi6 cates. Further independent stud-ct'ssed above, the estimate of risk for all cancers excluding ies. and in particalar studies of cancer incidence rather than leukemia was consistent with risks larger than those based mortality, with histological review of hematological malig-on high-dose data. Although risk estimates and con 6dence nancies. are needed to clarify the association between radi-intervals are not presented for most individual cancer types. ation and risk of multiple myeloma.

in most cases, these confidence intervals were much wider .

than those for all cancers excluding leukemia and were 2. Prostare Cancer therefore compatible not only with no risk but also with Two studies in the UK (data from which are included in frirly large positive risks. South England facilities) have reported statistical associa-tions between radiation dose and mortality from cancer of

1. Multiple Myeloma the prostate (9.16). Updated analyses of mortality in the UK t pan from leukemia excluding CLL multiple myeloma AEA cohort founa that the association between cancer of i

c '.he only type of cancer to exhibit a statistically the prostate and external radiation dose was largely confined significar.t association with radiation dose in the combined to workers who had also been monitored for radionuclide data set (one sided P value: 0.037). Evidence exists for radi- exposure (26). These results were con 6rmed by a subsequent ation-induced multiple myeloma from other studies, yet it is case control study of twostate cancer in the AEA workforce not consistent (l. 49). In panicular, although a dose-related (52). in which the authors concluded that tne association with increase in multiple myeloma mortality hans been observed external dose was largely a result of the correlation between systematically among survivors of the atomic bombings in external doses and radionuclide contammation. That popula-Hiroshima and Nagasaki since the late 1960s (46. 50), tions receivmg high doses of external radiations have shown ecent analyses of data for the incidence of multiple myelo- no evidenge of an increased risk of prostate cancer H6) pro-na, after an extensive review of cases of hematological vides further support for this conclusion.

IH

- . . - - - _.- - - - _. . - - - ~ - - _ . _ _ _ - _ - ~ _ ~ - - - .. ~ . - - -. ----

l

) -

i h

4 LOW. DOSE RADIATION AND CANCER A THREE. col:NTRY STL D) IN i

In the present combined analyses, no association was risk. possibly by an order of magnitude or more (53). Thew

, seen for cancer of the prostate in the combined data set or analyses indicate that if there has been underestimation. it i in any single facility, although the AEA could not be exam- is unlikely to have been by more than a factor of about two.

izad separately. There is therefore, little evidence for an Another question ofimportance in radiation risk assess.

! assooation between protracted external low doses of radia- ment is the choice of model to extrapolate risk estimates d

tion and increased mortality from this disease. The possibil- across populations with different background incidence and ity that an association exists between radinnuclide exposure mortality rates of cancer. Comparisons of risk for specific and mortality from cancer of the prostate could not. howev- cancer types (particularly lung and stomach the incidence l l

er, be tested in this study. of which varies greatly between Japan and North America '

and Europe) could provide some information about this Implicationsfor Radiation Protection issue. At present, however, site specinc estimates of risk A primary objective of studies of cancer risk among among nuclear industry workers are too uncertain for such  ;

nuclear industry workers is the assessment of the adequacy comparisons to be meaningful. In general. however. the of existing protection standards. These are based on risk problem of extrapolating risks across populations would estimates derived from analyses of the mortality of atomic appear to be greater for specinc cancer types than for all bomb survivors and studies of other high dose exposures. cancers combined, for which the baseline mortality rates in its most recent recommendations. the ICRP (4) varv less between industrialized countries.

states that. based on consideration of experimental data. The problem of predicting absolute risk has not been a dose and dose-rate effectiveness factor (DDREF) of addressed directly in these analyses. One of the most two should be applied to estimates from these studies to important steps in obtaining such estimates, namely extrap-set protection standards for low-dose protracted expo- olatica from high to low doses and dose rates. has been dis-sures. BEIR V used a linear-quadratic model for estimat- cussed brieliy above. as has that ot extrapolating from one -

ing leukemia risk and also made a recommendation that population to another. Additional factors to be taken into risks should be reduced to account for lower-dose rate account include projection of risk over time and modifying exposures (1). There is therefore particular interest in effects of sex and age at exposure. Even for exposure using data for populations receistng such exposures. in received at high doses and dose rates, there is uncertaint) particular nuclear industry workers. to assess the appro- concerning the appropriate methods for handling these priateness of this DDREF. The risk estimates presented issues. The data for the workers do not provide information in this paper for leukemia excluding CLL and all cancers regarding these issues and, at present, no information is excluding leukemia are compatible with a range of risks. asailable to judge whether the modifying effects of tim'e ranging from close to zero to a risk approximately twice since exposure. age at exposure and sex are similar after tha linear estimates from analyses of atomic bomb sur. Iow dose protracted exposures to those observed in studies vivors. If we assumed that the difference between the risk of high dose exposures  !

estimates derived from the nuclear workers and the stud-its of the atomic bomb survivors was entirely attributable CONC 1.US10NS to the effect of dose and dose rate, we could infer (by dividing the estimate for the atomic bomb survivors in Combining data from seven cohorts in three countries Table VI by the estimate for the workers and its has provided the opportunity to obtain the most compre-con 6dence limits) that the DDREF for leukemia exclud- hensive and precise direct estimates to date of the carcino-ing CLL is of the order of 1.7 with a lower limit of 0.6 and genic effect of low-LET radiation at low doses and low dose an upper limit of 28. There may be othet differences, rates. Overall, the estimates resulting from these analyses however, including differences in dose and outcome were consistent across studies, as well as with those derived assessment, as well as in the distribution of host factors from high-dose, high-dose-rate studies. A significant and environmental exposures, which could modify the increase in leukemia (in particular myeloid leukemia) risk association between radiation dose and leukemia risk. was demonstrated by the combined analyses at relatively Furthermore, the connoence intervals presented describe low dose levels. The study has also provided the opportuni.

only part of the uncertainty of the risk estimates. These ty to examine some of the previously reported associations are therefore not suf6ciently precise to test the need for a between low doses of ionizing radiation and mortality from DDREF or for estimating its magnitude. speci6c cancer types: we have found a dose-related increase The upper con 6dence bounds presented in this paper are in mortality from multiple myeloma that largely reflected of particular interest because it has been said that the the experience, at two facilities. Additional follow up of extrapolation process used to assess cancer risk after low. these cohorts, as well as studies of additional groups of dose protracted exposure may seriously underestimate this workers, will be useful to reduce the uncertairey further.

l hA