Cancer morbidity in British military veterans included in chemical warfare agent experiments at Porton Down: cohort studyBMJ 2009; 338 doi: http://dx.doi.org/10.1136/bmj.b655 (Published 25 March 2009) Cite this as: BMJ 2009;338:b655
- L M Carpenter, reader in statistical epidemiology1,
- L Linsell, medical statistician1,
- C Brooks, data manager1,
- T J Keegan, research fellow1,
- T Langdon, assistant data manager1,
- P Doyle, professor of epidemiology2,
- N E S Maconochie, senior lecturer in epidemiology and medical statistics2,
- T Fletcher, senior lecturer in environmental epidemiology3,
- M J Nieuwenhuijsen, research professor in environmental epidemiology4, visiting professor5,
- V Beral, professor and director6,
- K M Venables, reader in occupational medicine1
- 1Department of Public Health, University of Oxford, Oxford OX3 7LF
- 2Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, London
- 3Department of Public Health and Policy, London School of Hygiene and Tropical Medicine
- 4Center for Research in Environmental Epidemiology, IMIM and CIBERESP, 08003 Barcelona, Spain
- 5Division of Epidemiology, Public Health and Primary Care, Imperial College, London
- 6Cancer Epidemiology Unit, University of Oxford
- Correspondence to: L M Carpenter ;
- K M Venables
- Accepted 5 December 2008
Objective To determine cancer morbidity in members of the armed forces who took part in tests of chemical warfare agents from 1941 to 1989.
Design Historical cohort study, with cohort members followed up to December 2004.
Data source Archive of UK government research facility at Porton Down, UK military personnel records, and national death and cancer records.
Participants All veterans included in the cohort study of mortality, excluding those known to have died or been lost to follow-up before 1 January 1971 when the UK cancer registration system commenced: 17 013 male members of the UK armed forces who took part in tests (Porton Down veterans) and a similar group of 16 520 men who did not (non-Porton Down veterans).
Main outcome measures Cancer morbidity in each group of veterans; rate ratios, with 95% confidence intervals, adjusted for age group and calendar period.
Results 3457 cancers were reported in the Porton Down veterans compared with 3380 cancers in the non-Porton Down veterans. While overall cancer morbidity was the same in both groups (rate ratio 1.00, 95% confidence interval 0.95 to 1.05), Porton Down veterans had higher rates of ill defined malignant neoplasms (1.12, 1.02 to 1.22), in situ neoplasms (1.45, 1.06 to 2.00), and those of uncertain or unknown behaviour (1.32, 1.01 to 1.73).
Conclusion Overall cancer morbidity in Porton Down veterans was no different from that in non-Porton Down veterans.
After the first world war, research conducted at the chemical defence establishment at Porton Down included a “human volunteer programme” and members of the armed forces took part in experiments studying the potential impact of chemical agents on military capability and the effectiveness of protective measures.1 2 Between 1941 and 1989, over 18 000 members of the armed forces were recorded as having taken part in this programme. Over 50% of these veterans were recorded as taking part in tests involving chemicals that are known or probable human carcinogens, the most common being dermal exposures to sulphur mustard.3 While their overall cancer mortality was similar to that of veterans who did not take part in tests at Porton Down,4 past exposure to these chemicals might have affected their risk of developing common cancers with relatively good survival such as certain skin cancers. We report here on cancer morbidity in Porton Down veterans by studying cancers reported for the period from 1971, when the UK cancer registration system commenced, to 2004.
Information provided here focuses on aspects relevant to cancer morbidity. More details of the methods used are provided elsewhere.4
We considered for inclusion all veterans included in the cohort study of mortality.4 Porton Down veterans were all male members of the British armed forces recorded as having participated in tests between 1 April 1941 and 31 December 1989, while non-Porton Down veterans were other similar members of the armed forces not recorded as having taken part in tests at Porton Down.4 We excluded veterans known to have died or been lost to follow-up before 1 January 1971.
We obtained death certificates, notifications of emigrations, and data on cancers registered since 1 January 1971 from the National Health Service central registers.4 Each condition listed on death certificates as causing or contributing to death was coded by the Office for National Statistics according to ICD-10 (international classification of diseases, 10th revision). Cancer registration data for all cancers registered on or after 1 January 1971 were provided electronically and included date of diagnosis and cancer site or type coded according to the ICD coding system at the time of registration (ICD-8, 9, or 10). The research team recoded ICD-8 and ICD-9 codes to ICD-10 codes.
Classification of chemical exposures
We used contemporaneous experimental records of tests carried out as part of the “human volunteer programme” in the Porton Down historical archive to retrospectively assess exposure.3 5 Data abstraction involved linking, when possible, each test to an individual veteran and the classification of that test as involving a chemical or not. The type of chemical was recorded as being a vesicant (blistering agent), a nerve agent, or other chemical. There were eight specific chemicals for which there were records of at least 1000 Porton Down veterans having been tested: three vesicants (sulphur mustard, Lewisite, and nitrogen mustard), one nerve agent (sarin), two lachrymators (CS and CR), and two anti-nerve agent pharmaceutical chemicals (pralidoxime and atropine). Two of these chemicals have been classified by the International Agency for Research on Cancer (IARC) as either a known (sulphur mustard and cancer of the upper and lower respiratory tract) or probable (nitrogen mustard and squamous cell carcinoma of the skin) human carcinogen.6 We also considered benzene, a chemical classified by IARC as a leukaemogen6 and used as a diluent in tests in 994 veterans.3
Of the veterans recorded as being exposed to sulphur mustard, 74% of those in the analyses of cancer morbidity could be classified according to cumulative exposure. Veterans exposed to a total of ≥10.63 mg of sulphur mustard were classified as having “high” exposures.3 Equivalent classifications were possible for 66% of veterans exposed to Lewisite and 75% of those exposed to nitrogen mustard, the corresponding “high” exposure levels being ≥13.69 mg and ≥23.73 mg, respectively.3 For 93% of tests involving sulphur mustard, we could classify veterans according to the biological effect recorded, equivalent figures being 98% for Lewisite and 83% for nitrogen mustard. Veterans with a least one dermal vesicle or a necrosed area recorded after tests involving these vesicants were classified as having had an exposure with at least one “high” biological effect. The derivation of these groups, and of similar groups for nerve agent exposures, is described elsewhere.3 For chemicals other than vesicants or nerve agents, we collected information on the number of tests recorded.
All veterans included in the previously reported mortality analysis as alive and under follow-up on or after 1 January 1971 contributed to the analysis. The primary method involved comparing registration rates in Porton Down veterans with those of non-Porton Down veterans by calculating rate ratios. For each Porton Down veteran, we counted person years of follow-up from 1 January 1971 or the earliest subsequent date that they were first recorded as being included in a test. For non-Porton Down veterans, we counted person years from the date derived by adding to their enlistment date the time between enlistment and first attendance at Porton Down for the veteran for whom they had been selected. For each specific cancer site or type, person years stopped at the earliest of date of registration of the first cancer, death, loss to follow-up, or 31 December 2004. For veterans lost to follow-up, person years stopped at the last date they were known to be alive (such as date of emigration or discharge from the services). When we compared rates in Porton Down and non-Porton Down veterans we included cancers identified only from the death certificate. For cancers identified in this way, we used the date of death as the date of cancer registration.
Rate ratios were estimated for all neoplasms combined and according to type (malignant, in situ, benign, uncertain or unknown behaviour). For malignant neoplasms, we also estimated rate ratios separately for 16 predetermined cancer sites or types. Veterans with more than one cancer of the same site or type had only the first cancer counted while those with cancers registered for two or more different types (such as primary cancers of skin and bladder) contributed a cancer to each. We calculated person years separately for each cancer site or type stratified according to 15 five year age groups (15-19, 20-25, etc, to ≥85) and five year calendar periods (1971-4, 1975-9, etc, to 2000-4). All rate ratios were adjusted for age group and calendar period and estimated with 95% confidence intervals, and P values calculated, with the Mantel-Haenszel method.7 When the number of cancers expected in either group of veterans was less than 10, we estimated adjusted rate ratios and tested for significance by fitting exact conditional Poisson regression models to the stratified data.8
We also compared rates of cancer in each group of veterans with rates in England and Wales, using only those cancers notified from registrations. We counted person years from the same date as that used for rate ratio calculations but stopped at the date when the veteran was last known to be alive in the UK. We stratified person years of follow-up according to 15 five year age groups (15-19, 20-24, etc, to ≥85) and by single calendar year (from 1941 to 2004). Expected registrations were calculated by multiplying the person years for each five year age group and single calendar year by the corresponding registration rates for England and Wales. Standardised registration ratios were estimated from the ratio of observed to expected registrations. We obtained 95% confidence intervals and P values using the normal approximation to the Poisson distribution.8
Analyses reported here relate to 17 013 Porton Down veterans, and 16 520 non-Porton Down veterans, all of whom were included in mortality analyses.4 Updating of databases after the mortality analysis removed two Porton Down veterans and four non-Porton Down veterans from the analysis but added three non-Porton Down veterans. All analyses were performed with Stata 10 (www.stata.com). All tests of significance were two sided.
Of the veterans included in the mortality analysis, 93% (17 013/18 276) of Porton Down veterans and 94% (16 520/17 600) of non-Porton Down veterans contributed data to analyses on cancer morbidity. Of those excluded from analyses, in over 60% of Porton Down (796/1263) and non-Porton Down veterans (703/1,080) this was because they had died between 1941 and 1970. Service at enlistment and other characteristics were similar to that of non-Porton Down veterans (table 1)⇓. As in the mortality study, Porton Down veterans had a longer total duration of military service. The distribution of Porton Down veterans according to frequency, calendar period, and type of chemical test was also similar to that reported previously (table 2)⇓.
The number of men with one or more cancer registrations was 3029 in Porton Down veterans and 3015 in non-Porton Down veterans. We identified a further 530 veterans with cancer from death certificates (259 and 271). The total number of cancers recorded in each group was 3457 and 3380, respectively. The percentage with two or more registrations was slightly larger in Porton Down than non-Porton Down veterans: 12.3% (374/3029) v 10.6% (320/3015).
Overall, rates for all neoplasms were the same in both groups of veterans (rate ratio 1.00, 95% confidence interval 0.95 to 1.05) as were those for all malignant neoplasms (0.99, 0.94 to 1.04) (table 3)⇓. Porton Down veterans had higher rates of ill defined, secondary, or unspecified malignant neoplasms (1.12, 1.02 to 1.22), in situ neoplasms (1.45, 1.06 to 2.00), and those of uncertain or unknown behaviour (1.32, 1.01 to 1.73) and lower rates of skin cancer other than melanoma (0.87, 0.77 to 0.99). Some neoplasms classified as in situ or of uncertain or unknown behaviour (16 in Porton Down and 10 in non-Porton Down veterans) were non-melanoma skin cancers. When we combined these with malignant skin cancers other than melanoma, the rate ratio for this cancer site increased only slightly (0.89, 0.78 to 1.01).
Rate ratios for all neoplasms and all malignant neoplasms were just below unity for veterans exposed to tests involving any vesicant, sulphur mustard, or nitrogen mustard and just above unity for Lewisite (fig 1)⇓. Veterans exposed to Lewisite had rates of cancer of the trachea, bronchus, and lung 25% higher than non-Porton Down veterans (1.25, 1.05 to 1.48, P=0.01). Porton Down veterans exposed to tests involving any nerve agent, sarin, or any chemical other than nerve agents or vesicants had rate ratios for all neoplasms and all malignant neoplasms just below unity (figs 2 and 3)⇓ ⇓. The rate of cancer of the oesophagus was raised in veterans exposed to CS (2.17, 1.04 to 4.52, P=0.03).
Table 4⇓ shows results for veterans classified as having had high levels of exposure or biological effect either for chemicals with previous evidence of carcinogenicity (such as sulphur mustard for cancers of the trachea, bronchus, and lung) or where rate ratios were raised in Porton Down veterans (such as for CS and oesophageal cancer). When we compared the resultant estimates of rate ratios to the results for all exposed Porton Down veterans, in nearly all cases the estimate obtained was lower or similar to that obtained for all veterans. The only instance where the estimate for the high exposure group was above the upper limit of the 95% confidence interval for all veterans was for skin cancers other than melanoma in relation to high dermal exposure to nitrogen mustard (1.59, 0.97 to 2.61, v 1.13, 0.82 to 1.57). For veterans with a dermal vesicle recorded (high biological effect), however, the rate ratio estimate was lower than that of all exposed veterans (1.05, 0.59 to 1.87).
When compared with cancer registration rates in England and Wales, rates in Porton Down veterans were 10% lower for all neoplasms (standardised registration ratio 0.90, 0.87 to 0.93) and 6% lower for all malignant neoplasms (0.94, 0.91 to 0.98). Corresponding results for non-Porton Down veterans were 0.87 (0.85 to 0.90) and 0.94 (0.90 to 0.97), respectively.
Overall occurrence of cancer in veterans who took part in tests at Porton Down was similar to that of other veterans and lower than in the general population. These findings are in accord with their cancer mortality.4 We also found no evidence of an excess of any specific, clearly defined, cancer type or site in all Porton Down veterans combined. As expected, the similarity to our findings for mortality was particularly notable for highly fatal cancers such as those of the trachea, bronchus, and lung: the 9% higher rates of lung cancer morbidity in Porton Down veterans is identical to the excess reported for mortality.4 In the current study, we were particularly interested in cancers with a relatively good survival. There was no evidence to suggest rates of one of the commonest of these—malignant skin cancers other than melanoma—were higher in Porton Down veterans, even after we reclassified neoplasms previously coded as in situ or of uncertain or unknown behaviour. Porton Down veterans did, however, experience increased rates of ill defined, secondary or unspecified malignant neoplasms, in situ neoplasms, and neoplasms of uncertain or unknown behaviour. Such findings are difficult to interpret in the absence of information on smoking habits and other risk factors for cancer.
We identified almost 500 different chemicals in the records at Porton Down for 1941-89.3 We focused on nine specific chemicals to which the largest numbers of veterans were exposed. Given the large number of outcomes and exposure groups, some of the associations found might be due to chance. Of the nine chemicals, sulphur mustard and benzene are known human carcinogens, while nitrogen mustard is a probable carcinogen.6 Of these, the chemical most often used in tests was sulphur mustard, a carcinogen known to affect the upper and lower respiratory tract. Over half of Porton Down veterans were included in tests involving this chemical but they experienced no increased cancer morbidity, either overall or for any specific cancer. This probably reflects the low cumulative exposures received, especially compared with those of manufacturing workers in whom excesses of upper and lower respiratory cancer have been reported.9 10 In addition, use of physical protective barriers (such as masks and clothing), which were recorded as used in over 30% of tests involving vesicants,3 will have reduced the actual exposure of veterans to these chemicals.
Despite it being a known human carcinogen, there was no evidence of an overall excess of cancer in veterans exposed to benzene. Previous epidemiological evidence indicates an association between occupational exposure to benzene and leukaemia.6 Although the rate ratio for leukaemia was above unity, the confidence interval was wide and the rate ratio did not increase when we restricted analysis to veterans with two or more tests. The increased strength of association with skin cancer other than melanoma in veterans recorded to have high exposure to nitrogen mustard is noteworthy, given previous evidence that this chemical is a human carcinogen.6 This needs to be qualified, however, by the absence of an increase in those in whom the biological effect was high.
Although arsenic and arsenic compounds are human carcinogens,6 and Lewisite is an organic arsenic compound, there have been no previous epidemiological studies of the carcinogenicity of this specific chemical.11 While there was no evidence of an overall excess of cancer in Porton Down veterans exposed to Lewisite, there was an excess of lung cancer, which was also present in our previous mortality analyses.4 The lack of association with increasing exposure or effect levels for this chemical, together with the lack of data on a known key confounding factor (smoking), make it difficult to attribute this excess to tests at Porton Down and it might be a chance finding.
For the other chemicals examined, there were no clear associations between any specific defined cancer type or site with sarin, pralidoxime, or atropine. For CS, while the overall cancer occurrence was lower than in non-Porton Down veterans, it was raised for cancer of the oesophagus, particularly in those who had two or more tests. A similar excess was seen in veterans exposed to CR, the other lachrymator included in our analyses.
In summary, the overall rates of cancer morbidity in Porton Down veterans were not raised relative to other veterans or the general population. The excesses of ill defined, secondary or unspecified malignant neoplasms, in situ neoplasms, and neoplasms of uncertain or unknown behaviour were seen across several of the chemical exposure groups analysed and are difficult to interpret.
What is already known on this topic
Many tests in the Porton Down research programme into chemical warfare agents involved known carcinogens
What this study adds
Overall cancer morbidity was no higher in military personnel who had been included in tests at Porton Down than in those who had not been included or in the general population
Cite this as: BMJ 2009;338:b655
We thank the many people and organisations who contributed to this research project and full acknowledgements are listed elsewhere.4 In particular, we thank the MRC Liaison Group established for this study, the Porton Down Veterans Support Group, the MoD Veterans Policy Unit, the Defence Science and Technology Laboratory at Porton Down, the National Health Service Central Register, the General Register Office for Scotland, the Office for National Statistics, and the individual cancer registries in England, Wales, Scotland, and Northern Ireland. We especially thank Bianca De Stavola for her work on exact analytical methods, David Cox and Michael Hills for their valuable guidance. The figures were originally plotted by Adrian Goodill, Cancer Epidemiology Unit, University of Oxford.
Contributors: KMV and LMC jointly led the study and are its guarantors. All authors participated in the study design, plan of analysis, interpretation of the results, and drafting of the paper. LL carried out the statistical analyses, with contributions from CB.
Funding: This study was funded by the MRC with funds provided to it by the MoD. Additional funds were provided by the Department of Public Health and Nuffield College, University of Oxford. The MRC set up a liaison group to oversee the progress of the study. The MoD had no role in study design, analysis, and interpretation of data. The MoD facilitated access to its archives but had no role in deciding which data were abstracted. The MoD was sent a copy of the paper for factual comment but had no role in writing the paper or in the decision to submit for publication. The Universities of Oxford and London are independent of both the MRC and the MoD.
Competing interests: None declared.
Ethical approval: The study was approved by the south east multicentre research ethics committee, the Defence Medical Services clinical research committee, and the Patient Information Advisory Group.
This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.