Cancer mortality and morbidity in employees of the United Kingdom Atomic Energy Authority, 1946-86.

In further analyses of a cohort of 39,718 United Kingdom Atomic Energy Authority employees after 7 more years follow-up, cancer mortality, based on 1,506 deaths in 1946-86, was 20% below the national average. Prostatic cancer mortality showed a statistically significant association with external radiation exposure, largely confined to men who were also monitored for internal contamination by radionuclides other than plutonium. Prostatic cancer mortality was highest in radiation workers at Winfrith. In women monitored for radiation exposure, mortality from cancer of the uterus (including the cervix uteri) was increased relative to other employees, and, showed a statistically significant association with external radiation exposure. While there were some other statistically significant results, as would be expected by chance alone when multiple comparisons are made, there were no other cancer sites with consistently exceptional findings. Point estimates for risk associated with increasing exposure to radiation suggest a decrease of four deaths per 10(4) person-years per Sv for leukaemia and an increase of 20 deaths for all cancers except leukaemia, but confidence intervals indicate that a wide range of values are compatible with the data. Cancer morbidity based on 1,699 registrations in 1971-84 was 12% below the national average. Findings from site-specific analyses largely replicated those of the mortality analyses.

Follow-up data for a further 7 years have accrued since we first described the mortality of employees of the United Kingdom Atomic Energy Authority (UKAEA),   . Cancer mortality then was 21% below the national average and prostatic cancer was the only malignancy with clearly increased mortality in relation to radiation exposure. However, the duration of follow-up to the end of 1979 was only 16 years on average, and small numbers of deaths from some cancers yielded imprecise estimates which were consistent with a wide range of effects. In order to increase precision and allow for longer latency, we have continued to collect follow-up data on the entire UKAEA study population, including employees still in service on 1 January 1980. Analyses of mortality data from 1946 to 1986, and cancer registration data from 1971 to 1984, are reported here with special reference to prostatic cancer, other genital tract cancers, and malignancies such as multiple myeloma which have been associated with radiation exposure in other studies of nuclear industry workers (Smith & Douglas, 1986;Gilbert et al., 1989).

Methods
The design and methods of data collection and validation in the UKAEA mortality study have been described previously  and a fully account of the methods used in this report will be published elsewhere (Fraser et al., in preparation).
Study population, personnel data andfollow-up The study population comprised all employees of the UKAEA establishments at Harwell, London, Culham, Dounreay and Winfrith who were ever employed between 1 January 1946 and 31 December 1979. Details of all employees were submitted for tracing to the National Health Service Central Registers (NHSCRs) in Southport and Edinburgh. For subjects recorded as having died, both the underlying and associated causes of death as stated on the death certificate were coded to the eighth (for deaths during 1946-78) and ninth (for deaths during 1979-86) revisions of the International Classification of Diseases (ICD) by the Office of Population Censuses and Surveys. Coded death certificates and notifications of emigrations and cancers registered since 1971 were obtained. When a subject could not be traced at the NHSCRs, identifying particulars were sent to the Department of Social Security's (DSS) Records Branch in Newcastle for ascertainment of vital status.
The UKAEA's records of deaths in service and among members of their pension scheme provided a cross-check on the completeness of notification of death and sometimes information of assistance in tracing deaths at the NHSCRs. Further checks on the completeness of notification of cancer deaths and random samples of non-cancer deaths were carried out at NHSCR in Southport. Deaths notified in the UKAEA study population were also cross-checked against deaths in radiation workers notified to the National Radiological Protection Board (NRPB) (Kendall. et al., 1992a,b). Finally, an extensive computerised check on the vital status of UKAEA study members using National Insurance numbers was carried out by the DSS' Information Technology Services Agency.
Radiation data Information on employees who had been monitored for exposure to radiation during 1946-85 was obtained from UKAEA records. Data abstracted for each employee included a yearly summary of external whole body radiation exposure (X-rays, gamma rays and neutrons) including exposures accumulated in previous employments when these had been notified to the UKAEA. Annual whole body exposures were cumulated adjusting exposures in 1946-79 for below-threshold measurements as described by Inskip et al. (1987). After 1979 no adjustments were made as this source of error became negligible. Adjustments were made throughout the study period for missing values due to lost or damaged films by incorporating an appropriate fraction of the worker's annual recorded exposure in that year. Information on internal exposure from radionuclides was generally limited to noting the years in which subjects were monitored for possible internal contamination by tritium, plutonium, or other unspecified radionuclides. Associated radiation exposures were not included in measures of whole body exposure except from tritium at Harwell from 1977.

Statistical analysis
Person-years at risk were calculated from each worker's first day of employment by the UKAEA (or from 1 January 1946 for workers recruited before that date) to 31 December 1986 or the date of emigration, death, or the last date traced if any of these preceded 1 January 1987. Person-years-at-risk and deaths were stratified by sex, age in 15 groups, calendar year both in single years and in nine groups, establishment in three groups (Harwell with Culham and London, Dounreay and Winfrith), and social class in six groups. Radiation exposure was treated as a time-dependent variable in all analyses. In order to allow for the delayed effects of radiation exposure, the exposures were lagged in some analyses by 2 years for leukaemia and 10 years for other causes of death using the same method as before . Some analyses were repeated with a 5-year lag. All mortality analyses were based on the underlying cause of death. Deaths for which an underlying cause could not be ascertained were included in analyses of deaths from all causes but not in cause-specific analyses. Age, sex and singleyear specific death rates for England and Wales were used to calculated expected deaths and standardised mortality ratios (SMR). Exact 95% confidence intervals (CI) and statistical significance levels for SMRs were obtained using the Poisson distribution (Bailer & Ederer, 1964) unless the numbers of observed deaths was 200 or more, when approximate significance levels were derived from the standard chi-squared test (Breslow & Day, 1987). Rate ratios (RR) adjusted for age, sex, calendar period, establishment and social class, and approximate 95% CIs, were estimated by the method of maximum likelihood using the GLIM computer package (Baker & Nelder, 1982). Statistical significance of the RRs was tested using the likelihood ratio statistic and checked using the score statistic (Breslow & Day, 1987). When the total number of deaths contributing to the RR was less than 20, exact confidence intervals and significance levels were generated from the stratum-specific deaths and person-years using the likelihood for binomial data.
For workers with a radiation record, the relation between level of external whole body radiation exposure and mortality was examined without reference to national rates. An overall chi-squared statistic was obtained to test for a linear trend across five levels of exposure stratified by age, sex, calendar period, establishment and social class (Breslow & Day, 1987). When the resultant test statistic was based on a total number of deaths of 20 or less, probability values were checked using 10,000 simulations as described elsewhere . Changes in cancer risk associated with increasing exposure to radiation were estimated using an additive relative risk model. Excess relative risks and absolute risks (and their 95% CIs) were estimated from this model using maximum likelihood methods described by Gilbert (1989).
The cancer registration data were analysed using methods similar to those employed for mortality. Cancers first identified through a death certificate as the underlying or an associated cause of death were included in all internal (within workforce) comparisons, the date of death being used as a surrogate for the registration date. Where an ill-defined, secondary or unspecified neoplasm was registered but a primary malignant neoplasm was specified on the death certificate, the date of registration was retained but the primary site was substituted in all internal analyses. For comparability with national cancer registration rates in England and Wales, cancers first identified through a death certificate in 1971-73 were not included in calculations of standardised registration ratios (SRR).
P-values obtained from tests of statistical significance are quoted as two-sided throughout. Separate analyses were performed for many individual anatomical sites -29 for cancer mortality and 14 for cancer registrations. Some statistically significance results may therefore be expected to occur on the basis of chance alone. In interpreting the findings, those for which similar results have not been described in other studies are considered as being likely to be due to chance, except when the P-value is less than 0.01. Where similar results have been described before, attention is drawn to findings with a P-value of less than 0.05 in the expected direction. Confidence intervals for SMRs and RRs are quoted throughout the text, to give an indication of the range of possible values for the true effect, and to examine its consistency with that of other studies.

Results
The total study population numbered 39,869. Of these, 151 lacked essential information such as sex, date of birth, or dates of employment. These individuals could neither be traced nor included in the analyses; only one had a radiation record but no recorded dose. The analyses presented here are therefore based on 39,718 subjects -172 more than the 39,546 subjects analysed previously . Of these, 29,085 (73%) were men and 21,545 (54%) had a radiation record, of whom 19,760 (92%) were men. The average length of follow-up was 22 years. Subjects with a radiation record had been followed-up for an average of 23 years and those without for an average of 21 years.
The collective external radiation exposure in the UKAEA study population from 1946-85 was 862 Sv giving an average exposure of 40 mSv per monitored worker. The distribution was highly skewed with half (10,806) of the 21,545 workers with a radiation record having a final cumulative whole body exposure of 10 mSv or more, 10% (2,211) a final exposure exceeding 100 mSv, and less than 1% (150) a final exposure exceeding 500 mSv. Exposures varied between establishments, average exposures at Winfrith (56 mSv) and Dounreay (49 mSv) being higher than at Harwell with Culham and London (32 mSv). Among workers who had a radiation record, 7,480 (35%) were monitored for internal contamination by radionuclides; 1,702 (8%) were monitored for tritium, 3,564 (17%) for plutonium, and 6,412 (30%) for other unspecified radionuclides. Fifty-one percent (3,834) of radionuclide-monitored workers fell into more than one of these categories. A total of 5,509 deaths (14% of the study population) had occurred by 31 December 1986. The cause of death had been confirmed by necropsy in 781 subjects with radiation records (25.9%) and 655 subjects without (26.3%). There were 1,506 deaths from cancer and 45 deaths for which the underlying cause could not be ascertained. A total of 1,706 ex-employees (4% of the study population) were reported to have emigrated by 31 December 1986. Of the 111 (0.3%) ex-employees who were lost to follow-up after leaving the UKAEA, only ten had a radiation record.
Among the 28,594 study members (72% of the study population) whose records were identified through National Insurance numbers, DSS reported belatedly 82 deaths in the UK during 1970-86 which had not been notified by the NHSCRs, 50 of them in monitored workers. Three more deaths in Dounreay radiation workers came to light in the cross-check against deaths notified to NRPB. The distribution of all 53 missed deaths in radiation workers across dose categories was not dissimilar to the distribution of the 3,021 notified deaths in the study population so their omission from the analyses is unlikely therefore to have resulted in serious bias. An examination of deaths by year suggested a deficit in the numbers of deaths notified by the NHSCRs in 1985, and possibly in 1986. The proportions of missed deaths in 1985 and 1986, 3.5% and 2.9% respectively, whilst above the annual average were not extreme.
For employees without a radiation record, no SMR was significantly increased. Significant deficits were apparent however for all malignant neoplasms, cancers of the bronchus and lung, uterus and bladder.
Mortality of employees with radiation records compared with other employees: rate ratios When death rates for employees with a radiation record were compared with rates for other employees using rate ratios there were few significant differences (Table I). When no lag was assumed, rate ratios were significantly low for all causes of death combined (RR = 0.94, 95% CI 0.88-0.99) and breast cancer (RR = 0.32, 95% CI 0.14-0.77). A 4-fold excess of cancer of the uterus (including the cervix uteri) in women with a radiation record was observed (RR = 4.55, 95% CI 1.20-16.66, P = 0.008). Confidence intervals for all other rate ratios embraced 1.00, including those for prostatic cancer (RR = 0.81, 95% CI 0.48-1.35) and testicular cancer (RR = 0.92, 95% CI 0.34-2.07).
Rate ratios were recalculated lagging exposure by 2 years for leukaemia and 10 years for other causes of death (Table  I). The 4-fold excess of cancer of the uterus in women with a radiation record persisted (RR = 4.28, 95% CI 1.03-19.97, P = 0.02). The deficit of breast cancer was also apparent in the lagged analyses, giving a rate ratio of 0.33 (95% CI 0.13-0.84, P= 0.008) for both sexes combined and, for women alone, a rate ratio of 0.36 (95% CI 0.14-0.91, P= 0.02). The lagged analyses revealed a 2-fold excess of cancers of the bladder and urinary organs (excluding kidney) or borderline significance (RR = 1.98, 95% CI 0.98-3.97, P = 0.05) in employees with a radiation record. The rate ratio for prostatic cancer was 0.84 (95% CI 0.50-1.41) and 0.92 (95% CI 0.55-1.54) with 5 and 10-year lags respectively.
Mortality and level of cumulative external radiation exposure Table II shows mortality from selected causes of death among workers with a radiation record according to cumulative external whole body radiation exposure. In the unlagged mortality analysis in men, prostatic cancer alone showed a positive trend in mortality with increasing exposure (X2 for trend = 6.12, P = 0.01), the trend being most evident for employees at Harwell (with Culham and London) (X2 for trend = 5.11 ,P = 0.02). The trends persisted when exposures were lagged by 5 years (X2 for trend = 4.39, P = 0.04 overall and 6.45, P = 0.01 at Harwell) but were less apparent with a 10 year lag (X2 for trend = 1.65, P = 0.20 overall and 3.70, P = 0.05 at Harwell). Contrary to the findings in the previous analysis , there was no significant trend in prostatic cancer mortality at Winfrith (X2 for trend = 0.04, P=0.8).
The unlagged analysis of mortality in women showed a trend for all causes of death (X2 for trend= 4.52, P= 0.03) and all malignant neoplasms (X2 for trend= 5.69, P= 0.02), the main components of which were cancer of the uterus (X2 for trend = 5.11, P = 0.02) and cancer of the bronchus and lung(X2 for trend = 4.14, P = 0.04). These site-specific trends were based on very few deathseight cancers of the uterus (two cervix, five body, one part unspecified) and five cancers of the bronchus and lung. When P-values for these sites were checked by simulation, the values obtained were similar (P = 0.04 and P = 0.01 respectively). Though the significant trend for all malignant neoplasms was lost in the lagged analysis(X2 for trend = 2.66, P = 0.1), the trends for the two specific cancer sites persisted when exposures were lagged by 10 years (X2 for trend = 5.10, P= 0.02 for uterus and x2 = 8.80, P = 0.003 for bronchus and lung). Again the simulated P-values for these two cancers were similar (P = 0.04 and P = 0.003 respectively). There were no other statistically significant trends in site-specific cancer mortality with increasing exposure to radiation from either lagged or unlagged analyses.
Mortality and monitoring for internal exposure to radionuclides Mortality in workers monitored for possible internal exposure to specific radionuclides was examined by calculating SMRs for comparison with national rates and RRs for comparison with other workers who had a radiation record but were not monitored for exposure to that particular radionuclide (Table III). When no lag was assumed, mortality from all malignant neoplasms in workers monitored for exposure to any radionuclide was below national rates (SMR = 80, 95% CI 70-91) but simlar to that of workers with a radiation record who were not monitored for radionuclide exposure (RR = 1.05, 95% CI 0.90-1.24).
Mortality from prostatic cancer was raised nearly 3-fold in men monitored for exposure to tritium (SMR = 282, 95% CI 113-580, P=0.03; RR= 2.85, 95% CI 1.17-6.95). At Winfrith, the SMR of 689 (95% CI 188-1763) though based on only four deaths from prostatic cancer in tritium-monitored workers was highly significant (P = 0.006) as was the SMR of 345 (95% CI 139-710) associated with other unspecified radionuclides (P = 0.01 based on seven deaths). The rate ratio for prostatic cancer in tritium-monitored workers remained elevated when exposures were lagged by 5 years (RR = 3.39, 95% CI 1.40-8.24) but not with a 10-year lag (RR= 1.75, 95% CI 0.50-6.10). Mortality from cancer of the uterus was raised in women monitored for exposure to any radionuclide by comparison with other radiation workers (RR = 17.99, 95% CI 2.14-290.14) but based on only two deaths. There were no deaths from cancers of the bronchus or lung in women monitored for radionuclide exposure. A 2-fold excess mortality from cancers of ill-defined and secondary sites was observed in radiation workers monitored for radionuclide exposure compared to other radiation workers (RR= 1.95, 95% CI 1.01-3.77). This was largely confined to workers monitored for radionuclides other than tritium or plutonium (RR = 2.39, ).
In order to investigate further the effects of internal and external radiation exposure, the trends in prostatic cancer mortality were re-examined after stratification for radionuclide exposure (Table IV). Rate ratios for prostatic cancer mortlaity increased with increasing cumulative whole body exposure in men monitored for exposure to any radionuclide "Tz E (x2 for trend = 6.46, P = 0.01) but not in radiation workers who were not monitored for radionuclide exposure (X2 for trend = 0.01, P = 0.9). Prostatic cancer mortality increased with increasing external exposure in men monitored for exposure to tritium (x2 for trend = 4.24, P = 0.04) but not in radiation workers who were not monitored for this radionu-'0 clide (X2 for trend = 0.31, P = 0.6). oX also increased with increasing whole body exposure in men monitored for exposure to radionuclides (unspecified) other o z > " " " " than tritium or plutonium (x2 for trend = 5.97, P = 0.02) but + ++ + + + + not irradiation workers who were not monitored for other 'IT |> E-°0 < |o -| ;^|> t es radionuclides (X2 for trend = 0.22, P = 0.6). These trends however are not independent as more than half of the E radionuclide-monitored workers were monitored for exposure to more than one radionuclide and thus contribute to more 2^than one category in Table IV. >Z o1 _ _ I-, There was no association between prostatic cancer mortal-°< D CA WI ity and monitoring for plutonium exposure. The significant -0 _ ___, o trend in radiation workers who were not monitored for plutonium is due to multiple monitoring; seven of the eight 0 men in the highest exposure category, whilst not monitored for plutonium, were monitored for exposure to tritium and/ Q0 0 G;-°o°£°or other radionuclides. There were no significant trends when e 0 w, _ N these analyses were repeated incorporating a 10-year lag en | (Table IV). previously (Inskip et al., 1987  Registrations for employees with radiation records compared with other employees: rate ratios Fatal cancers dominante the cancer registrations for most cancer sites and analyses of fatal cancers have been reported Table V Risk estimatesa (with 95% confidence limits) for leukaemia and all malignant neoplasms except leukaemia obtained from additive relative risk models using lag periods of 2 years for leukaemia and 10 years for all malignant neoplasms except leukaemia Absolute excess risk Excess relative risk (per 10i person years (per Sv) per Sv)b Leukaemia (ICD8 204-208) Unlagged -4.1 (-5.7, +2.9) -3.5 (-5.4, + 1.8) Lagged -4.2 (-5.7, +2.6) -3.9 (-5.9, + 1.7) All malignant neoplasms except leukaemia (ICD8 140 -203, 209) Unlagged +1.2 (-0.4, +3.1) +19.9(-6.4, +48.0) Lagged +0.8 (-1.0, +3.1) +20.3(-26.0, +71.1) aAdjusted for age, sex, calendar period, establishment and social class. b Equivalent to 106 person-years per 10 mSv used in previous analyses (Inskip et al., 1987). in detail in Tables I to IV. The registration analyses reported here were restricted therefore to cancer sites where previous analyses have suggested a relationship between mortality or cancer incidence and radiation exposure, and sites which carry a better prognosis where an examination of cancer registrations might be expected to provide additional information. The extent to which fatal cancers dominated the cancer registrations even for these selected sites is shown in Table VI; only 429 (27%) individuals with a malignant neoplasm registered in 1971-84 were alive at the end of the study period in December 1986.
When registration rates for employees with a radiation record were compared with rates for other employees there were no significantly raised ratios in either lagged or unlagged analyses (Table VI). In the latter, the rate ratio of 0.38 (95% CI 0.15-0.98) for testicular cancer was significantly low (P = 0.05). There was no evidence of excess registrations Discussion Cancer mortality This report describes mortality from all causes, and cancer in particular, among 39,718 employees of the UKAEA over a 41-year period from 1946 to 1986. The analyses are based on a total of 5,509 deaths, representing an increase of 63% over the number of deaths included in our first report which covered 34 years from 1946 to 1979 . This substantial increase in material permitted more detailed analyses to be carried out for site-specific cancer sites than previously. Comparisons have been made with mortality in the general population of England and Wales, between employees with a radiation record and other employees, and between groups of radiation workers accumulating different levels of external radiation exposure. Mortality has also been examined in workers monitored for internal exposure to tritium, plutonium and other unspecified radionuclides. The of non-fatal breast cancer among women with a radiation record (not shown) as had been the case in the analysis reported in our previous paper (Inskip et al., 1987).
Cancer registrations and level of cumulative external radiation exposure In unlagged but not lagged analyses there was a significant trend in prostatic cancer registrations with increasing exposure overall (x2 for trend = 6.34, P = 0.01) and at Winfrith alone (X2 for trend = 5.52, P = 0.02). These trends are based on 74 and 23 registrations respectively. In women the unlagged analyses showed trends for all malignant neoplasms (X2 for trend = 5.43, P = 0.02 based on 61 registrations) and for invasive cancer of the uterus (X2 for trend = 4.83, P = 0.03 based on eight registrations) which persisted in the lagged analyses. The uterine cancers were all fatal and thus the findings largely replicate those of the mortality analysis.
In our previous analysis of non-fatal cancers in relation to cumulative whole body exposure (Inskip et al., 1987) the trends for skin cancer and bladder cancer were suggestive of an association (X2 = 3.65, P = 0.06 and x2 = 3.57, P = 0.06 respectively). These findings were not replicated when the trends for these two sites were re-examined in subjects registered in 1971-84 who were still alive at the end of the study period in December 1986. The chi-squared trend statistic for skin cancer was 0.12 (P = 0.7) in unlagged analyses and 0.03 (P = 0.9) in lagged analyses based on 71 registrations. The corresponding statistics for bladder cancer were 1.92 (P = 0.2) and 0.97 (P = 0.3) based on 25 registrations. Non-fatal cancers of the prostate were also examined in relation to cumulative whole body exposure but there was no evidence of an association (X2 (unlagged) = 1.91, P = 0.3 and x2 (lagged)= 0.17, P = 0.7, based on 21 registrations).  EMPLOYEES, 1946-86 623 results will be discussed here within the context of the previous findings, which have been described fully elsewhere Inskip et al., 1987;Carpenter et al., 1988;Carpenter et al., 1990).
As previously, overall mortality in the UKAEA workforce was lower than the national average in England and Wales as a consequence of health-related selection and other differences between employed people and the general population (Carpenter et al., 1990). Cancer mortality remained about 20% below national rates and rate ratios for all malignant neoplasms indicated no overall difference in cancer mortality between workers with a radiation record and other employees. As before, mortality from testicular cancer, thyroid cancer, leukaemia and non-Hogkin's lymphoma was above average in the workforce as a whole but only that for testicular cancer approached statistical significance. Attention had been drawn previously to increased mortality from testicular cancer, particularly at Harwell , but there was no evidence in these new analyses that this finding was associated with exposure to radiation.
As in previous analyses, prostatic cancer mortality showed a clear relation with radiation exposure. There was a statistically significant association with increasing cumulative whole body exposure although the trend diminished in strength with longer lags. The pattern of prostatic cancer mortality across age groups suggests, as before, that the highest risk (relative to national rates) is in men aged 45-54. Mortality was significantly raised in men monitored for exposure to tritium, the rates being increased nearly 3-fold in comparison with national rates and other radiation workers who were not monitored for tritium exposure. At Winfrith, where prostatic cancer mortality was significantly increased in employees with a radiation record, mortality was also raised in men monitored for exposure to tritium, and to radionuclides other than tritium and plutonium. Further examination of the dose-response relationships after stratification for radionuclide exposure indicated that the association between prostatic cancer mortality and cumulative whole body exposure was largely confined to radiation workers who had also been monitored for radionuclide exposure, notably tritium and unspecified radionuclides. The highest risks were in radionuclide-monitored workers who had accumulated at least 100 mSv external exposure suggesting as before that men with these dual exposures are most at risk. The findings for prostatic cancer are now being investigated further in a case-control study based on UKAEA records to determine if prostatic cancer can be linked to any particular aspect of the employment history of the individuals concerned. Particular attention is being given to radionuclide exposure and the potential for contamination of the work environment by radionuclides, chemicals and other substances.
Mortality from cancer of the uterus (including the cervix uteri) in women with a radiation record was increased 4-fold by comparison with other employees. As in the previous report , cancers of the body of the uterus made the major contribution to the excess mortality. The persistent trend in these analyses, whilst based on only eight deaths, is supportive of an association between mortality from cancer of the uterus and external exposure to radiation and warrants further study. Contrary to the findings in the previous analysis, there was no suggestion of increased mortality from ovarian cancer.
Although the increase in the average duration of follow-up from 16 to 22 years allowed for longer latency and greater statistical power, cancer of the prostate, and possible the uterus, are the only malignancies which show an association with radiation exposure in the UKAEA workforce. It is noteworthy that mortality from multiple myeloma which has been associated with radiation exposure in the Sellafield (Smith & Douglas, 1986) and Hanford (Gilbert et al., 1989) workforces was not associated with radiation exposure in UKAEA employees. Indeed, the SMR of 36 (95% CI 10-92) based on four deaths from multiple myeloma in employees with a radiation record was significantly low by comparison with national rates (P = 0.03).
Despite the additional 7 years of follow-up yielding a substantially increased number of cancer deaths, uncertainty associated with the risk estimates remains large (Table V). For leukaemia, the data are compatible with decreasing risks or increases of up to two deaths per 104 person-years per Sv (or an excess RR of 2.6 per Sv). These risk estimates are very similar to those obtained for US nuclear industry workers (Gilbert et al., 1989) but contrast with the positive values recently provided for adult males from A-bomb survivor data of 5.0 deaths per 104 person-years per Sv and an excess RR of 3.7 per Sv (UNSCEAR, 1988). Comparing these latter estimates with those relating to nuclear industry workers is questionable because they are likely to be affected by many factors thought to modify the risk of radiation-induced leukaemia. These include the type and duration of exposure and other characteristics of the populations studied. Recent analysis of a much larger population of UK nuclear industry workers (Kendall et al., 1992a,b) (of which the current UKAEA cohorts forms a part) are more appropriate for comparison. This provided an excess RR estimate for leukaemia of 4.3 per Sv (90% CI 0.4 to 13.6).
In contrast with data on US workers, the central risk estimate for all cancers except leukaemia obtained from our data was positive (excess RR per Sv (lagged) = 0.8, 95% CI -1.0 to 3.1). Data from the larger population of UK workers are also suggestive of a generally positive association (excess RR for all malignant neoplasms = 0.5 per Sv, 90% CI -0.1 to 1.2) (Kendall et al., 1992a,b). Estimates of absolute risk from our data ranged from a decrease of 26 deaths per i04 person-years per Sv to an increase of about 70. Whilst these include the value recently recommended by the International Commission on Radiological Protection (ICRP) for working populations of approximately 14 deaths per 104 person-years per Sv (ICRP, 1990) (assuming a 25-year expression for lifetime risks), our estimates are not sufficiently precise to rule out estimates of up to five times larger than this. To lessen this uncertainty, we have combined the UKAEA, Atomic Weapons Establishment  and Sellafield (Smith & Douglas, 1986) study populations in a further analysis (the Nuclear Industry Combined Epidemiological Analysis). The combined study population of 75,000 nuclear industry workers will permit the relationship between exposure to low-level ionising radiation and cancer mortality to be estimated with greater precision than was possible in any of the three studies individually.
Cancer morbidity Because of high case fatality for most cancer sites, the findings for cancer registrations largely replicated those of the mortality analyses. In particular, trends in registrations of prostatic cancer and cancer of the uterus with increasing exposure were apparent. The dose-response relationships reported before for non-fatal skin and bladder cancers (Inskip et al., 1987) were not replicated here. The previous findings were based only on cancer registrations in ex-employees who may not have been typical of the workforce as a whole. These new results based on cancer registrations in all employees suggest that the previous findings may have been biased.

Conclusion
This analysis of a much larger body of material than that reported in our first analysis of the UKAEA workforce generally confirms the robustness of the previous findings for most cancer sites. The association between prostatic cancer and both internal and external exposure to radiation is still evident though the dose-response relationship is diminished in strength. Prostatic cancer is under investigation in a casecontrol study within the UKAEA workforce. The association between cancer of the uterus and external radiation exposure which has strengthened in this analysis also warrants further study. There are a number of other statistically significant results, as would be expected by chance alone when such a large number of comparisons are made, but no other cancer sites with consistently exceptional findings. Uncertainty still surrounds estimates of the increase in cancer risk per unit dose. Further combined analyses will provide more precise estimates.
Members of the Epidemiological Monitoring Unit were funded by the Medical Research Council through contracts held with the United Kingdom Atomic Energy Authority. We thank members of these organisations for their support and co-operation. We are particularly grateful to Len Salmon, Dallas Law and Jean Rose for their help. We also thank staff of the National Health Service Central Registers at Southport and Edinburgh, and the Department of Social Security Records Branch and Information Technology Services Agency in Newcastle for providing follow-up information. We are grateful to Evelyn Middleton and Juliet Jain for secretarial assistance.