Investigation of cancer incidence and mortality at a Scottish semiconductor manufacturing facility

Occupational Medicine 2003;53:419 430 DOI: 10.1093/occmed/kqg111 Investigation of cancer incidence and mortality at a Scottish semiconductor manufacturing facility Damien M. McElvenny 1, Andrew J. Darnton 1, John T. Hodgson 1, Simon D. Clarke 1, Richard C. Elliott 2 and John Osman 1 Background We became aware of concern about cancer at a Scottish semiconductor manufacturing facility in 1998. Aim Method Results Conclusion Key words To compare cancer experience among current and former workers at the facility, with an appropriate comparison population, making use of any readily available exposure information. We obtained personnel and employment episode information from four sources within the company. Workers were flagged for death and cancer registrations at the National Health Service Central Register in Edinburgh. We constructed standardized registration and mortality ratios (SRRs and SMRs), using Scotland as the comparison, with and without an adjustment for deprivation. The main mortality analysis included 4388 workers, with a mean length of follow-up of 12.5 years. Overall mortality was substantially below that expected for men, and for women was slightly below expected. Total cancer registrations were close to expected levels for men and women. Four cancers produced noteworthy findings: malignant neoplasm of the trachea, bronchus and lung in women deprivation adjusted SRR [95% confidence interval (CI), number of cases] 273 (136 488, 11 cases); malignant neoplasm of the stomach in women adjusted SRR 438 (90 1281, three cases); and malignant neoplasm of the female breast adjusted SRR 134 (82 206, 20 cases). The unadjusted SMR for male brain cancer was 401 (83 1172, three cases) and there was an additional non-fatal case. These findings, particularly those relating to lung cancer, though inconclusive, point to the possibility of a work-related risk of cancer that justifies further investigation. Cancer; occupation; semiconductor. Received 17 December 2002 Revised 27 May 2003 Accepted 7 July 2003 Introduction Concern that there might be a cluster of cancers in the British semiconductor industry first came to our attention in late 1998. The concern was almost 1 Epidemiology and Medical Statistics Unit, Health & Safety Executive, Bootle, Merseyside, UK. 2 Medical Unit, Health & Safety Executive, Bootle, Merseyside, UK. Correspondence to: Dr John Osman, Epidemiology and Medical Statistics Unit, Health & Safety Executive, Room 236 Magdalen House, Stanley Precinct, Bootle, Merseyside L20 3QZ, UK. e-mail john.osman@hse.gsi.gov.uk exclusively centred on a single semiconductor manufacturing facility in Scotland, and was expressed in the form of workers support group actions, media coverage, and parliamentary activity. We identified only two epidemiological studies specific to cancer risk in semiconductor manufacturing, both conducted on the same cohort of workers at a semiconductor factory in the West Midlands region of England [1,2]. The initial investigation [1] provided evidence of a cluster of melanoma in women, which disappeared with further follow-up [2]. D. M. McElvenny, Health & Safety Executive. Crown Copyright 2003. Reproduced with the permission of the Controller of Her Majesty s Stationery Office. 419

420 OCCUPATIONAL MEDICINE The aims of this investigation were limited to establishing the level of cancer incidence and mortality in the workforce compared with an appropriate reference population and relating this incidence to any readily available information on individuals histories of work in the plant. The intention was to establish some basic facts relatively quickly to replace the existing anecdotal information, and decide whether more extensive investigation would be justified. The data collection was therefore approached in such a way that it would form a sound basis for further more detailed investigation, should that prove necessary. Methods A protocol was prepared for the investigation, and this was endorsed by an independent steering committee with expertise in statistics and epidemiology, and occupational and public health medicine. We compiled a list of known or suspected carcinogenic substances or agents that had been present or in use at the factory since the start of operations in 1970 (Table 1). We also utilized a list of cancers that were present on the (now defunct) worker support group s website (http://www.gla.ac.uk/acad/facsoc/mres/www98/ williamson/home.htm). These information sources enabled the inclusion of a focused set of disease groups in the study. In addition we looked at broad cancer groups and mortality from certain causes to help interpret the findings. An initial data set was compiled that consisted of surname, maiden name, forenames, sex, date of birth, National Insurance number, current or last known address and employment episodes. The data set included all men and women who were employed by the company at the facility on or before 30 April 1999. The data were checked for missing information and validated for Table 1. Known or suspected carcinogens present or in use at the factory Antimony trioxide Arsenic and arsenical compounds Arsine Asbestos (in buildings) Chromium trioxide Kaowool Highly refined mineral oil Arsine/arsenical compound contaminated oil Sulphuric acid mists Ionizing radiaton UV radiation Krypton 85 Carbon tetrachloride Chromic acid Trichloroethane Trichloroethylene internal consistency. Missing or inconsistent data items and potential duplicate records were queried with the company, and data amendments were made on the basis of their response. The completeness and accuracy of the personnel data on the master computer file were assessed by crosschecking with a sample of records from the original sources within the company (personnel, pension, payroll and occupational health). Initially, the aim was to sample ~10% of the source records in this way. However, the sample was extended to 100% for pension and payroll records after identifying a number of individuals from these sources missing from the master file. Altogether, an additional 425 workers were identified and added to the master file. These could not be confirmed as UK nationals having worked for the company at the facility and were thus marked as unconfirmed workers. Where possible, inaccuracies in the data identified by the checking process were corrected. The master file was sent to the National Health Service Central Register (NHSCR) maintained by the General Register Officer for Scotland [GRO(S)] for tracing and notification of deaths and cancer registrations. Details of any workers initially untraced on the NHSCR were sent to the Department for Work and Pensions in Newcastle upon Tyne to enable the worker s National Insurance number to be used to identify fact and date of death information where applicable, or for confirming or correcting personnel information. Any records where new information was obtained via this method were then retraced at the NHSCR. In order to check that cancer registration data supplied by GRO(S) were complete, the data file was sent to the Scottish Health Service s Information and Statistics Division (ISD) and traced against the Scottish Cancer Registration Database. Any workers flagged as unconfirmed workers, after carrying out the check of completeness and accuracy of the personnel data, were excluded from the analysis data file. Although we did not attempt to collect detailed information on exposures at the individual level, since it was beyond the scope of this investigation, we nevertheless looked for a means to possibly distinguish different exposure subgroups within the study population. The only readily available source of data that could practically be used was assignment to work area as recorded by the occupational health department. Initially, this assignment was based on pre-employment medical check, but was updated opportunistically as workers came into contact with the company s occupational health department. The work areas were classified for analysis as fabrication (fab the areas in which the semiconductor chips are manufactured) or nonfabrication (non-fab). The dates when workers changed

D. M. MCELVENNY ET AL.: CANCER INCIDENCE, MORTALITY AND SEMICONDUCTOR MANUFACTURING 421 from working in fab to non-fab or vice versa were not usually available. Thus, workers were categorized as ever having worked in fab or not. These data were extracted by the study team and validated. For the mortality analysis, the end of follow-up was 31 December 2000, and for the cancer registration analysis it was 31 December 1998. At the time of the analysis, Scottish cancer registration data for 1998 from ISD were 96% complete and England and Wales data were 99% complete. Event data supplied by GRO(S) were coded to ICD8 [3] to 1978 and ICD9 [4] for 1979 onwards. Data supplied by ISD were coded to ICD10 [5]. All data were recoded by a member of the study team to ICD9 before analysis. The choice of the population on which to base the expected mortality or cancer incidence was a compromise between using national and local data. Local rates would best reflect the general health profile of the local population but were considered too unstable. Therefore our adjusted rates were Scottish rates but with an adjustment (different for males and females) based on the Carstairs index of deprivation [6], which was intended to take into account average health profile of the economically deprived area of Scotland in which the plant was based. The unadjusted rates were based on Scottish rates without the adjustment for deprivation. Because we did not have appropriate address information for all former employees, we could not stratify at the individual level, but rather chose to re-weight the Scottish deprivation specific rates in the standardized registration and mortality ratios (SRRs and SMRs) according to the geographical distribution of residence of current employees. The software used to calculate the SMRs and SRRs was OCMAP-PLUS v3.09 (Beta v1.0) [7], and all data manipulations and calculations were carried out using SPSS [8]. A single 10 year cut-off point was used for latency (time since first employment) analyses, since the relatively short average length of follow-up prohibited use of a longer one (e.g. 15 year cut-off point or >2 latency analysis categories). Analyses by date of start were run using two groups divided at 1982, the year which gave approximately equal expected numbers of registrations for all cancers in the two date of start subgroups. Similarly, analyses by age were divided at age 50, the five-year point giving approximately equal numbers of expected all malignant neoplasms in the two age subgroups. Appropriate ethical clearances were obtained for the investigation, and the data for the study were registered in accordance with current UK data protection legislation. As with all retrospective cohort studies, it was not practicable to seek the explicit consent of every individual included in the investigation. However, the study proposals were presented to all current workers, and current and former workers were sent a leaflet explaining the investigation and offering the opportunity to opt out of the investigation. We undertook to ensure that the management and workforce at the facility would hear the results of this investigation before they were made public, and our main means of doing this was to produce a results leaflet mailed to each current and former worker. In addition, we undertook to give presentations to all shifts of workers the same day as, or shortly after, receipt of the leaflet. We also undertook to provide a confidential freephone telephone line to deal with any concerns that individuals may have had about the findings of the investigation. Results Vital status Details of 4547 employees who were confirmed as working at the facility and who were not identified as non-uk nationals were submitted for tracing at the NHSCR, 159 (3.5%) of whom were untraced (64 males; 95 females). Of the remaining 4388 workers who were traced (2126 males, 2262 females), 4067 (93%) were traced in Scotland, 318 (7%) in England and Wales, and three (<1%) in Northern Ireland. The distribution of the vital status of the 4388 traced workers at the end of 2000, who were included in the mortality analyses is set out in Table 2. Twenty-seven (1.3%) out of 2126 male workers and 44 (1.9%) out of 2262 female workers had died. Eleven (<1%) males and seven (<1%) females had emigrated. Five workers were excluded from the cancer registration analyses due to their start of employment being after the cut-off for the cancer registration analysis (31 December 1998), or their being traced in Northern Ireland, for where cancer registration data were not available. After exclusion of cancer registrations that occurred prior to start of employment, or after the cut-off date for the analysis, there were 125 cohort members that had cancer registrations included in the analysis. Of these, two were associated with more than one primary cancer registration. Table 2. Vital status of traced workers at 31 December 2000 Male (%) Female (%) Total (%) Alive current worker 659 (31) 649 (29) 1308 (30) Alive former worker 1429 (67) 1562 (69) 2991 (68) Dead 27 (1) 44 (2) 71 (2) Emigrated from UK 11 (1) 7 (0) 18 (0) Total 2126 (100) 2262 (100) 4388 (100)

422 OCCUPATIONAL MEDICINE Descriptive statistics Two thousand, one hundred and fifty-two (49%) workers were aged under 25 years at hire, and 3688 (84%) were aged under 35 years at hire. There is some evidence that women on average tend to be slightly younger than men when hired. The overall mean length of follow-up for the mortality analysis was 12.5 years. Around 20% of men and women had <12 months employment, 37% of men had >1 and <5 years service, and the equivalent percentage for women was 33%. Twenty per cent of men had >10 years service, as did 25% of women. Thus, females tended to work slightly longer at the factory than males. From the company s occupational health records, 1081 out of 2126 (51%) males were recorded as working in fab areas whereas 1793 out of 2262 females (79%) were recorded as working in fab areas. The number of workers employed at the facility increased steadily from the mid-1970s to 1995, passing 500 in 1981 and 1000 in 1984, and reaching ~2000 in 1994. The number had decreased steadily from a peak of over 2000 in 1995 to ~1250 by April 1999. Mortality analyses The male mortality results by disease groups are presented in Table 3 and the equivalent results for females are presented in Table 4. There was a statistically significant deficit in male mortality, adjusted SMR = 40 [95% confidence interval (CI) = 27 59, 27 cases] (Table 3), providing evidence of a strong healthy worker effect in men, which persists when the analysis is restricted to a latency of 10 years. There was also a slight deficit of female mortality, adjusted SMR = 75 (95% CI = 54 101, 44 cases) (Table 4), thus providing weak evidence of a healthy worker effect in women. Due to the overlap between the mortality and cancer registration results, most of the results will concentrate on the cancer registration analyses. One mortality finding among men was noteworthy, however: the unadjusted SMR for three fatal brain cancers (a disease group not of specific interest at the outset of the investigation) was 401 (95% CI = 83 1172), and there was an additional non-fatal case. Three cases had a latency of <10 years with ages at diagnosis of 23, 35 and 56 years respectively. The non-fatal case was aged 32 years at diagnosis. Cancer registration analyses The male cancer registration results by disease groups are presented in Table 5 and the equivalent results for females are presented in Table 6. The adjusted SRR for all malignant neoplasms for males was 99 (95% CI = 64 147, 25 cases) (Table 5) and for females was 111 (95% CI = 83 145, 54 cases) (Table 6). A statistically significant excess of lung cancer and an excess of stomach cancer in women, which was of borderline statistical significance, were identified. There was also a nonsignificant excess of breast cancer [where there was an absolute excess of around five cases (34%, adjusted) above that expected]. Of the 11 female lung cancers, five had a latency of <10 years, with ages at diagnosis ranging from 46 to 55 years. Five female cases had a latency of 10 20 years with a range of ages at diagnosis of 44 61 years. One woman had a latency of >20 years and was aged 60 years at diagnosis. One of the female lung cancers was non-fatal. Note that one of the two male lung cancers was non-fatal, the other case died from myocardial infarction. Detailed lung cancer registration results for female lung cancers are set out in Table 7. There are three to four times as many lung cancer registrations as expected among women who have worked at the factory. This excess is reduced slightly, but persists after an adjustment is made for deprivation. The excess also remains, again with a slight reduction, after exclusion of workers employed for <12 months. The excess is more pronounced for a latency of <10 years than for 10 years. The excess is similar in observations on women aged 50 years compared with women aged under 50 years. The excess is more pronounced in women first employed in 1982 or later than those first employed before 1982. The excess seems marginally higher in women who work in fab areas. None of these internal contrasts was statistically significant. The cell types of the female lung cancers had the following distribution: adenocarcinoma not otherwise specified (NOS) (four); small cell carcinoma NOS (two); carcinoma NOS (two); squamous cell carcinoma NOS (one); large cell carcinoma (one); and oat cell carcinoma (one). The three women with stomach cancer were diagnosed between 5 and 10 years after (first) starting work at the factory, and all were aged <45 years at diagnosis. Two of the cancers were fatal. Detailed female stomach cancer results are set out in Table 8. There are four to five times more stomach cancers than expected. The excess was reduced slightly but persisted after adjustment for deprivation. The excess was increased slightly after exclusion of workers employed for <12 months. The excess was restricted to a latency of <10 years and observations on women aged <50. The excess was similar for women first employed before 1982 compared with those first employed after 1982 and also for those working in the fab compared with other women. These informal comparisons should be interpreted cautiously because of the small number of cases involved, and none was statistically significant. Four of the 20 breast cancers cases had died by the end of 2000. Ten of the cases had a latency of <10 years, with ages at diagnosis ranging from 31 to 60 years. For the eight cases with a latency between 10 and 20 years, the ages at diagnosis ranged from 35 to 57 years. For the two cases with latency of over 20 years, both were aged over

D. M. MCELVENNY ET AL.: CANCER INCIDENCE, MORTALITY AND SEMICONDUCTOR MANUFACTURING 423 Table 3. Male mortality by disease group to 2000: numbers of deaths and SMRs with and without Carstairs adjustment Disease group (ICD9) Total No. of cases Unadjusted SMR Adjusted a SMR All causes (001 999) 27 46** (30 67) 40** (27 59) All malignant neoplasms (140 209) 6 51 (19 111) 47 (17 102) Malignant neoplasms of the lip, oral cavity and pharynx (140 149) 0 0 (0 880) 0 (0 682) Malignant neoplasms of the digestive organs and peritoneum (150 159) 2 60 (7 218) 57 (7 205) Malignant neoplasms of the stomach (151) 0 0 (0 608) 0 (0 550) Malignant neoplasms of respiratory and intrathoracic organs (160 165) 0 0 (0 106) 0* (0 90) Malignant neoplasm of the trachea, bronchus and lung (162) 0 0 (0 114) 0* (0 97) Malignant neoplasms of the pleura (163) 0 0 (0 4622) 0 (0 4464) Malignant neoplasms of the genitourinary organs (179 189) 0 0 (0 333) 0 (0 327) Malignant melanomas of the skin (172) 0 0 (0 1384) 0 (0 1488) Other malignant neoplasms of the skin (172) 0 0 (0 9862) 0 (0 8039) Malignant neoplasm of the thyroid gland (193) 0 0 (0 16 143) 0 (0 16 712) Malignant neoplasms of the lymphatic and haematopoietic tissue (200 208) 1 81 (2 453) 82 (2 459) Multiple myeloma (203) 0 0 (0 3576) 0 (0 3756) Leukaemia (204 208) 0 0 (0 824) 0 (0 812) Leukaemia excluding chronic lymphatic leukaemia (204.0, 204.2 208) 0 0 (0 919) 0 (0 900) Benign neoplasms (210 229) 0 0 (0 8474) 0 (0 8797) Neoplasms of uncertain behaviour or unspecified nature (235 239) 0 0 (0 3411 0 (0 3722) Diseases of the blood and blood-forming organs (280 289) 0 0 (0 3159) 0 (0 2828) Diseases of the circulatory system (390 459) 10 62 (30 114) 55 (26 101) Diseases of the respiratory system (460 519) 1 35 (1 196) 29 (1 160) Diseases of the genitourinary system (580 529) 0 0 (0 1248) 0 (0 1116) **Statistically significant at the 2-sided 1% level of significance. *Statistically significant at the 2-sided 5% level of significance. Table 4. Female mortality by disease group to 2000: numbers of deaths and SMRs with and without Carstairs adjustment Disease group (ICD9) Total No. of cases Unadjusted SMR Adjusted a SMR All causes (001 999) 44 92 (67 123) 75 (54 101) All malignant neoplasms (140 208) 23 120 (76 180) 110 (69 164) Malignant neoplasms of the lip, oral cavity and pharynx (140 149) 0 0 (0 1722) 0 (0 1677) Malignant neoplasms of the digestive organs and peritoneum (150 159) 3 92 (19 268) 88 (18 256) Malignant neoplasms of the stomach (151) 2 371 (45 1342) 327 (40 1181) Malignant neoplasms of respiratory and intrathoracic organs (160 165) 10 313** (150 575) 231* (111 425) Malignant neoplasm of the trachea, bronchus and lung (162) 10 325** (156 598) 241* (116 444) Malignant neoplasms of the pleura (163) 0 0 (0 17 453) 0 (0 20 449) Malignant neoplasms of the genitourinary organs (179 189) 2 60 (7 218) 54 (7 196) Malignant neoplasms of the uterus (179, 182) 0 0 (0 1659) 0 (0 1686) Malignant melanomas of the skin (172) 1 304 (8 1696) 354 (9 1973) Other malignant neoplasms of the skin (173) 0 0 (0 11 384) 0 (0 11 993) Malignant neoplasms of the female breast (174) 4 74 (20 191) 74 (20 190) Malignant neoplasm of the thyroid gland (193) 0 0 (0 11 172) 0 (0 8981) Malignant neoplasms of the lymphatic and haematopoietic tissue (200 208) 2 150 (18 543) 140 (17 506) Multiple myeloma (203) 0 0 (0 2647) 0 (0 2345) Leukaemia (204 208) 1 199 (5 1109) 172 (4 961) Leukaemia excluding chronic lymphatic leukaemia (204.0, 204.2 208) 1 207 (5 1156) 180 (5 1005) Benign neoplasms (210 229) 0 0 (0 6390) 0 (0 5312) Neoplasms of uncertain behaviour or unspecified nature (235 239) 0 0 (0 3211) 0 (0 3042) Diseases of the blood and blood-forming organs (280 289) 0 0 (0 2652) 0 (0 2325) Diseases of the circulatory system (390 459) 5 46 (15 108) 35** (11 81) Diseases of the respiratory system (460 519) 4 132 (36 337) 94 (26 241) Diseases of the genitourinary system (580 629) 1 237 (6 1321) 182 (5 1012) **Statistically significant at the 2-sided 1% level of significance. *Statistically significant at the 2-sided 5% level of significance.

424 OCCUPATIONAL MEDICINE Table 5. Male cancer registrations by disease group to 1998: numbers of cancer registrations and SRRs with and without Carstairs adjustment Disease group (ICD9) Total No. of cases Unadjusted SRR Adjusted a SRR All malignant neoplasms (140 208) 25 104 (67 153) 99 (64 147) Malignant neoplasms of the lip, oral cavity and pharynx (140 149) 0 0 (0 410) 0 (0 344) Malignant neoplasms of the digestive organs and peritoneum (150 159) 3 72 (15 211) 68 (14 198) Malignant neoplasms of the stomach (151) 0 0 (0 479) 0 (0 441) Malignant neoplasms of respiratory and intrathoracic organs (160 165) 3 83 (17 243) 71 (15 207) Malignant neoplasm of the trachea, bronchus and lung (162) 2 66 (8 238) 56 (7 202) Malignant neoplasms of the pleura (163) 0 0 (0 2666) 0 (0 2342) Malignant neoplasms of the genitourinary organs (179 189) 5 99 (32 231) 97 (31 225) Malignant melanomas of the skin (172) 2 165 (20 595) 186 (23 671) Other malignant neoplasms of the skin (173) 4 100 (27 256) 104 (28 265) Malignant neoplasm of the thyroid gland (193) 0 0 (0 2003) 0 (0 2255) Malignant neoplasms of the lymphatic and haematopoietic tissue (200 208) 3 104 (22 305) 106 (22 308) Multiple myeloma (203) 0 0 (0 2334) 0 (0 2401) Leukaemia (204 208) 0 0 (0 550) 0 (0 543) Leukaemia excluding chronic lymphatic leukaemia (204.0, 204.2 208) 0 0 (0 697) 0 (0 683) Carcinomas in situ (230 234) 0 0 (0 672) 0 (0 724) Table 6. Female cancer registrations by disease group to 1998: numbers of cancer registrations and SRRs with and without Carstairs adjustment Disease group (ICD9) Total No. of cases Unadjusted SRR Adjusted a SRR All malignant neoplasms (140 208) 54 111 (84 145) 111 (83 145) Malignant neoplasms of the lip, oral cavity and pharynx (140 149) 1 160 (4 890) 143 (4 799) Malignant neoplasms of the digestive organs and peritoneum (150 159) 4 88 (24 225) 86 (23 220) Malignant neoplasms of the stomach (151) 3 491* (101 1435) 438 (90 1281) Malignant neoplasms of respiratory and intrathoracic organs (160 165) 11 339** (169 606) 245* (122 438) Malignant neoplasm of the trachea, bronchus and lung (162) 11 373** (186 668) 273** (136 488) Malignant neoplasms of the pleura (163) 0 0 (0 10097) 0 (0 6809) Malignant neoplasms of the genitourinary organs (179 189) 7 72 (29 149) 66 (27 136) Malignant neoplasms of the uterus (179, 182) 1 76 (2 426) 74 (2 410) Malignant melanomas of the skin (172) 2 70 (9 253) 88 (11 319) Other malignant neoplasms of the skin (173) 6 108 (40 235) 125 (46 272) Malignant neoplasms of the female breast (174) 20 125 (76 193) 134 (82 206) Malignant neoplasm of the thyroid gland (193) 0 0 (0 447) 0 (0 577) Malignant neoplasms of the lymphatic and haematopoietic tissue (200 208) 3 107 (22 314) 110 (23 322) Multiple myeloma (203) 0 0 (0 1872) 0 (0 1795) Leukaemia (204 208) 1 139 (4 776) 145 (4 806) Leukaemia excluding chronic lymphatic leukaemia (204.0, 204.2 208) 1 161 (4 899) 167 (4 933) Carcinomas in situ (230 234) 47 86 (63 114) 94 (69 125) **Statistically significant at the 2-sided 1% level of significance. *Statistically significant at the 2-sided 5% level of significance. 50 years at diagnosis. Detailed female breast cancer results are set out in Table 9. Although the breast cancer result was not as striking as those for lung and stomach cancers, it represents an ~30% excess of observed over expected cases, and was based on 20 cases. The excess was increased slightly when an adjustment was made for deprivation. However, the excess was substantially reduced when the analysis was restricted to women employed for 12 months. The excess was slightly higher for latency <10 years compared with a latency of 10 years. The excess was higher for observations on women aged 50 years than for those aged <50 years. The excess was markedly higher for those women first employed in 1982 or later compared with those first employed before 1982. Within those women first employed after 1982, there seems little difference between the analysis based on

D. M. MCELVENNY ET AL.: CANCER INCIDENCE, MORTALITY AND SEMICONDUCTOR MANUFACTURING 425 Table 7. Female lung cancer (ICD9 162) registrations to 1998: numbers and SRRs by latency for selected subgroups Analysis type Total Latency < 10 years Latency 10 years Total females Unadjusted SRR 373** (186 668) 544** (177 1270) 296* (109 644) No. of cases 11 5 6 Adjusted a SRR 273** (136 488) 390* (127 911) 218 (80 475) Females employed for <12 months excluded Unadjusted SRR 325** (148 616) 381 (79 1113) 302* (111 658) No. of cases 9 3 6 Adjusted SRR 238* (109 451) 273 (56 798) 223 (82 485) Females aged <50 years Unadjusted SRR 370* (101 946) 500* (103 1462) 207 (5 1154) No. of cases 4 3 1 Adjusted SRR 262 (71 671) 355 (73 1038) 146 (4 816) Females aged 50 years Unadjusted SRR 376** (151 774) 627 (76 2264) 324* (105 755) No. of cases 7 2 5 Adjusted SRR 279 *( 112 576) 458 (55 1654) 242 (79 564) Females first employed before 1982 Unadjusted SRR 299* (110 650) 0 (0 963) 369* (135 803) No. of cases 6 0 6 Adjusted SRR 220 (81 479) 0 (0 688) 274* (101 596) Females first employed 1982 or later Unadjusted SRR 533** (173 1244) 933** (303 2177) 0 (0 918) No. of cases 5 5 0 Adjusted SRR 383* (124 894) 671** (218 1567) 0 (0 658) Females working in fab areas Unadjusted SRR 436** (199 827) 743** (241 1733) 287 (78 735) No. of cases 9 5 4 Adjusted SRR 317** (145 602) 532** (173 1242) 211 (57 540) Females working in non-fab areas Unadjusted SRR 227 (28 821) 0 (0 1502) 315 (38 1139) No. of cases 2 0 2 Adjusted SRR 167 (20 605) 0 (0 1080) 235 (28 847) **Statistically significant at the 2-sided 1% level of significance. *Statistically significant at the 2-sided 5% level of significance. latency <10 years compared with that based on a latency of 10 years. There was a non-significant excess in women recorded as working in fab areas, and a non-significant deficit for women recorded as working outside the fab. For women, apart from the cancers discussed in the previous paragraphs, there were no findings of note. For men, none of the cancer groupings selected for analysis produced statistically significant results. The highest SRR in men was produced by two cases of malignant melanoma, nearly twice the expected level (on an adjusted basis), but not statistically significant. Our approach It had been argued from the outset that the study should have looked at a wider range of health outcomes and encompassed the whole of the industry in Great Britain [9,10]. In reply, we argued that the pre-existing evidence for concern did not justify the commitment of resources implied by such a study [11, a further response was not published]. We believe that despite its moderate size and these limited aims, the present study represents a substantial addition to the available evidence on cancer risks in the semiconductor manufacturing industry. Discussion To our knowledge, this is only the second formal study of cancer in a semiconductor plant. Its aim was limited to establishing cancer incidence in the workforce population and to assess the need for further investigation. Completeness and accuracy of data Considerable efforts were made to ensure that the cohort was as complete as possible. Of the 425 workers who could not be confirmed as UK employees who worked at the factory, only 36% were traced at the NHSCR. This low trace rate is consistent with most of them being

426 OCCUPATIONAL MEDICINE Table 8. Female stomach cancer (ICD9 151) registrations to 1998: numbers and SRRs by latency for selected subgroups Analysis type Total Latency < 10 years Latency 10 years Total females Unadjusted SRR 491* (101 1435) 1213**( 250 3545) 0 (0 1015) No. of cases 3 3 0 Adjusted a SRR 438 (90 1281) 1093**( 225 3193) 0 (0 900) Females employed for <12 months excluded Unadjusted SRR 528* (109 1543) 1402**( 289 4096) 0 (0 1041) No. of cases 3 3 9 Adjusted SRR 469 (97 1371) 1251**( 258 3657) 0 (0 923) Females aged <50 years Unadjusted SRR 957** (198 2797) 1560**( 322 4560) 0 (0 3043) No. of cases 3 3 0 Adjusted SRR 884* (182 2582) 1416**( 292 4139) 0 (0 2889) Females aged 50 years Unadjusted SRR 0 (0 1241) 0 (0 6698) 0 (0 1523) No. of cases 0 0 0 Adjusted SRR 0 (0 1070) 0 (0 5879) 0 (0 1308) Females first employed before 1982 Unadjusted SRR 532 (64 1921) 2009**( 243 7258) 0 (0 1334) No. of cases 2 2 0 Adjusted SRR 459 (56 1657) 1671* (202 6038) 0 (0 1166) Females first employed 1982 or later Unadjusted SRR 426 (11 2375) 677 (17 3770) 0 (0 4246) No. of cases 1 1 0 Adjusted SRR 403 (10 2244) 646 (16 3597) 0 (0 3949) Females working in fab areas Unadjusted SRR 457 (55 1649) 1076* (130 3888) 0 (0 1462) No. of cases 2 2 0 Adjusted SRR 410 (50 1483) 977* (118 3530) 0 (0 1305) Females working in non-fab areas Unadjusted SRR 579 (15 3226) 1626 (41 9060) 0 (0 3317) No. of cases 1 1 0 Adjusted SRR 508 (13 2828) 1431 (36 7973) 0 (0 2901) **Statistically significant at the 2-sided 1% level of significance. *Statistically significant at the 2-sided 5% level of significance. non-uk nationals, and it was thus appropriate to exclude them from our analysis. (It was known that a number of non-uk nationals worked at the plant in its early years of operation.) Further, it was not judged appropriate to include them via a sensitivity analysis. In contrast, 96.5% of the confirmed workers were traced at the NHSCR. Staff at the factory had the option of utilizing private health care. We were advised that this would be unlikely to affect completeness of cancer registration in the cohort (C. R. Gillis and D. J. Hole, personal communications). There are doubts about the completeness of cancer registration in the local area compared with the rest of Scotland in the early years of operation of the factory (C. R. Gillis and D. J. Hole, personal communications). However, most of the cancers in this workforce would be expected after the 1970s and, although a potential problem, this was unlikely to mask any excesses that would otherwise have been found. At the time of the analyses cancer registrations in England and Wales were complete to the end of 1998 (M. Quinn, personal communication). Cohort size and follow-up The investigation cohort is limited in size, and since the factory only began operations in 1970 and did not recruit large numbers until the early 1980s, the average length of follow-up of 12.5 years is relatively short for a statistical analysis of the type undertaken here. However, the limitations of cohort size and follow-up were unavoidable in the context of this investigation. Also, at the outset of this investigation, there was insufficient justification for an industry-wide study to be conducted, and the study reported here was designed so that data could be pooled with any data from such a future study. Thus an urgent assessment was required of the evidence for or against

D. M. MCELVENNY ET AL.: CANCER INCIDENCE, MORTALITY AND SEMICONDUCTOR MANUFACTURING 427 Table 9. Female breast cancer (ICD9 174) registrations to 1998: numbers and SRRs by latency for selected subgroups Analysis type Total Latency < 10 years Latency 10 years Total females Unadjusted SRR 125 (76 193) 137 (65 251) 116 (55 212) No. of cases 20 10 10 Adjusted a SRR 134 (82 206) 145 (70 267) 124 (59 228) Females employed for <12 months excluded Unadjusted SRR 109 (63 178) 96 (35 210) 119 (57 219) No. of cases 16 6 10 Adjusted SRR 117 (67 190) 102 (38 223) 128 (61 235) Females aged <50 years Unadjusted SRR 93 (45 171) 112 (45 230) 67 (14 196) No. of cases 10 7 3 Adjusted SRR 99 (48 182) 118 (48 244) 72 (15 210) Females aged 50 years Unadjusted SRR 191 (92 351) 286 (59 836) 167 (67 344) No. of cases 10 3 7 Adjusted SRR 204 (98 375) 304 (63 888) 179 (72 368) Females first employed before 1982 Unadjusted SRR 61 (20 143) 43 (1 239) 68 (19 175) No. of cases 5 1 4 Adjusted SRR 65 (21 152) 45 (1 251) 73 (20 188) Females first employed 1982 or later Unadjusted SRR 192* (108 317) 180 (82 342) 213 (78 465) No. of cases 15 9 6 Adjusted SRR 206* (115 339) 192 (88 365) 229 (84 498) Females working in fab areas Unadjusted SRR 141 (82 226) 139 (60 275) 142 (65 270) No. of cases 17 8 9 Adjusted SRR 150 (88 241) 148 (64 292) 153 (70 289) Females working in non-fab areas Unadjusted SRR 77 (16 224) 126 (15 455) 43 (1 239) No. of cases 3 2 1 Adjusted SRR 82 (17 238) 134 (16 482) 46 (1 256) **Statistically significant at the 2-sided 1% level of significance. *Statistically significant at the 2-sided 5% level of significance. the claim of excess cancer incidence in this workforce. This assessment would then inform consideration of the need for further more detailed and industry-wide investigations. Adjustment for deprivation The expected numbers in the present analyses have been calculated on two different bases. First, simply using national Scottish rates, and secondly using rates based on the national Scottish data but adjusted to reflect the socio-economic profile of the local area (strictly, the areas of residence of current employees). A priori, the adjusted data should provide a better basis for assessing the presence of workplace risks, though any evidence of risk has to be interpreted with the possibility of further distortion from the healthy worker effect. Also there remains a possible issue in relation to how appropriate the adjustments made in this investigation are, since it is not clear whether the background rate of cancer in this workforce should be thought of as being the same as in the local population, due to the potential influence from influx of workers not indigenous to the area. Furthermore, the extent of this could be different for men and women. In any case the adjustment for deprivation does not materially change our conclusions. Overall mortality experience Mortality from all causes was substantially less than expected for men and slightly below expected levels for women. The pattern of these deficits by latency is consistent with a healthy worker effect, being rather stronger in observations with a latency of <10 years than in observations beyond 10 years latency. This healthy worker effect is seen even more clearly in the results for circulatory disease and respiratory disease (Tables 3 and 4).

428 OCCUPATIONAL MEDICINE Overall cancer registration experience Since cancer registration gives a more complete picture of cancer incidence than cancer mortality, the rest of this discussion will concentrate on the Carstairs adjusted cancer registration data. For men and women, the total number of cancers was close to the expected number. For two of the four cancers of public concern (leukaemia and uterine), and for five of the eight cancer groups examined because a relevant carcinogen was known to have been present in the workplace [pleura, skin (excluding melanoma), thyroid, multiple myeloma and leukaemia (excluding chronic lymphatic)], there were no findings of note. Four cancers produced findings worthy of more detailed description, which is set out in the following sections. Female lung cancer The excess of lung cancer in women in the cohort is the most striking finding from this investigation. The 11 cases observed are 2.7 times as many as expected, a statistically significant finding. In contrast to the raised risk in women, there were only two cases in men, around half as many as expected. To the extent that men and women have distinct occupational profiles in the plant, this difference could indicate a risk arising from work largely occupied by women. We know that ~80% of the women work in fabrication areas, which might suggest a source of risk present in these areas. On the other hand, a substantial proportion of men (around half) also work in the fabrication areas, and if this was the source of exposure to a lung carcinogen that generated the excess in the women, one might expect to see at least some sign of an excess in men. We have not assessed the extent to which exposures for males and females in fabrication might differ, since it was beyond the scope of this initial investigation. The female lung cancer excess was higher in those cases with <10 years latency than for a latency of 10 years, the reverse of the pattern normally expected in the presence of an occupational carcinogen. However, the factory has only been in existence since 1970 and the workforce has expanded considerably from the small numbers present in the earlier years, meaning that there is limited potential for observations with a long latency period, particularly in the more recently employed. Furthermore, if a potent carcinogen had been recently introduced into this working environment, an effect could only be evident in short latency observations. It is also worth noting that within the group of women who were first employed before 1982 there are no cases of lung cancer with <10 years latency, and an excess which is just significant for longer latency observations. Other factors complicate the interpretation of these findings. For example, the relatively high age at hire of the lung cancer cases (data not shown) raises the possibility of relevant exposures prior to employment at the facility. We have not collected information on smoking habit, although smoking behaviour in the cohort would have to be substantially higher than the comparison population to account for a relative risk of the size observed [12], particularly given that the adjustment for socio-economic factors will have adjusted for the effects of smoking to some extent. The distribution of cell types for the cases is not unusual, as the distribution varies historically and geographically [13]. The interpretation of the lung cancer findings is therefore complex. After careful consideration, we have concluded that we cannot exclude the possibility of some of the observed excess being related to work at the plant. Female stomach cancer There is an excess of stomach cancer in this cohort. The three cases observed are 4.4 times as many as expected, a finding of borderline statistical significance. All three cases occurred in women aged <50 years and were associated with a latency of between 5 and 10 years. There were no stomach cancer cases in men (0.8 cases were expected). Similar lines of argument based on the contrast between the findings in men and women, and the observation of higher risk at shorter latencies apply to these stomach cancer findings as to the lung cancer findings. Stomach cancer was selected for study at the outset of the investigation because it was of concern to the local worker support group. In this situation where cases have been identified prior to formal investigation, statistical significance will tend to be overstated and findings must be interpreted more cautiously. There remains the possibility that some of this excess may be related to work at the plant. Female breast cancer There is a moderate (34%) excess of breast cancer in the cohort as a whole. The only statistically significant finding in relation to breast cancer is the observation of approximately twice as many cases as expected among women first employed in 1982 or later. Breast cancer was selected for study at the outset of the investigation because of its association with ionizing radiation, but also because it was of concern to the local worker support group. The finding was not statistically significant, and to some extent the reservations about statistical significance as for stomach cancer still apply, although to a lesser extent. Breast cancer risk is strongly modified by hormonal factors and by reproductive history. The influence of such factors on the breast cancer findings in this investigation are unknown. Several recent papers have suggested a role for shiftwork in breast cancer [14 16]. The proposed mechanism is the suppression of melatonin levels produced by exposure to light at night.

D. M. MCELVENNY ET AL.: CANCER INCIDENCE, MORTALITY AND SEMICONDUCTOR MANUFACTURING 429 This hypothesis has not been fully evaluated, but has some biological plausibility, and epidemiological support. Since shiftwork is widespread in this workforce, it is possible that this is relevant to the observed excess. More detailed examination of the cases work histories is required before more definite conclusions can be drawn. Brain cancer Brain cancer was not one of the cancer categories selected for study in this investigation, and as such we were not able to calculate SRRs. However, three of the four cases identified were fatal, giving rise to an unadjusted SMR of 401 (95% CI = 83 1172). The cases had latencies of 2, 3, 6 and 11 years, and there were no cases among women. In the context of a moderately large number of different end points examined in this investigation, and given the short latencies of three of the four cases, it seems reasonable to see this result as more likely to be due to chance than a consequence of work at the plant, although the possibility of a work-related explanation cannot be entirely ruled out. Occupational factors In the current investigation, only data readily available on the whole population were used. The only exposure indicator available on this basis was a marker identifying individuals who worked in the fabrication areas. The marker did not, of course, give specific information about exposure, but does mark those with potential exposure to the wide range of agents characteristic of semiconductor fabrication. It would also include those with greater potential for exposure to physical agents such as ultraviolet or ionizing radiations. Conversely, the non-fab group would include all the office or electronic laboratory based staff with little or no potential for chemical exposures. It is possible that if misclassification of exposure exists between fab and non-fab workers then this would tend to mask an adverse health effect associated with hazardous exposure. However, only limited conclusions can be drawn from analyses based on this marker. Non-occupational factors As a result of the purposely limited aims of the investigation, no data on other possible risk factors were obtained, e.g. smoking (lung cancer) or diet (stomach cancer). This limits what can be concluded on the present data alone. The Carstairs index based adjustment does not directly adjust for variation in potential confounding factors such as smoking, diet or reproductive history. However, health outcomes correlate strongly with the index, and though the reasons for this are not fully known, established risk factors such as smoking and diet are thought to play an important role in the health of a population. In the original analyses of Carstairs and Morris [6] the disease group with the strongest correlation with the index was that for smoking related diseases. It is therefore reasonable to consider the adjustment as controlling, in some part, for the important socio-economic and lifestyle factors, including smoking. Conclusion These findings, particularly those relating to lung cancer, raise the possibility of a work-related risk of cancer, and more detailed studies will be needed to clarify this. We will continue case ascertainment within this cohort, to review the pattern of cancers over time. We are already planning a nested case control study to follow up the findings in relation to lung, stomach, breast and brain cancer, which will address both work and other relevant factors. The Health & Safety Executive has called for a replication of this study in the rest of the British semiconductor industry. Acknowledgements We thank the management and employees at the semiconductor manufacturing facility, without whom this investigation could not have been completed. We also acknowledge the advice of Raymond Agius, Freda Alexander and Lewis Reay, who acted as an independent scientific steering committee for this study, and would like to thank Roger Black, David Brewster, Charles Gillis, Veronica Harris, David Hole and Diane Stockton, for information and advice on Scottish cancer registration data; Sandra Ferguson, for information on cancer registration and cancer mortality rates in the locality of the semiconductor facility; staff at the National Health Service Central Registers in Edinburgh and Southport; Bernie Good, Jan Harding, Phil Irwin, Valerie Clayton and Andrea Grainger, for administrative support; Suzi Curtis, for her independent audit of study method and its implementation; and other staff at the Health & Safety Executive, too numerous to individually mention, for advice and support. An anonymous reviewer made very helpful comments on the paper. A copy of the full report of this investigation is available from the HSE website at http://www.hse.gov.uk/statistics/nsukrept.pdf. References 1. Sorahan T, Waterhouse JAH, McKiernen MJ, Aston RHR. Cancer incidence and cancer mortality in a cohort of semiconductor workers. Br J Ind Med 1985;42:546 550. 2. Sorahan T, Pope DJ, McKiernen MJ. Cancer incidence and cancer mortality in a cohort of semiconductor workers: an update. Br J Ind Med 1992;49:215 216. 3. WHO. Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death, 8th revision. Geneva: World Health Organization, 1967.

430 OCCUPATIONAL MEDICINE 4. WHO. Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death, 9th revision. Geneva: World Health Organization, 1977. 5. WHO. Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death, 10th revision. Geneva: World Health Organization, 1992. 6. Carstairs V, Morris R. Deprivation and Health in Scotland. Aberdeen: Aberdeen University Press, 1991. 7. Marsh GM, Youk AO, Sefcik ST, Alcorn CW. OCMAP-PLUS v3.09 (Beta v1.0) technical notes, draft, July 6. Pittsburgh, PA: University of Pittsburgh, 1999. 8. SPSS. SPSS Base 10.0 User s Guide. Chicago, IL: SPSS Inc., 1999. 9. Bailar JC III, Bobak M, Fowler B, et al. Reply to: rebuttal from the Health and Safety Executive. Int J Occup Environ Health 2000;6:262 [letter]. 10. Bailar JC III, Bobak M, Fowler B, et al. Open letter to the Health and Safety Executive. Int J Occup Environ Health 2000;6:71 72 [editorial]. 11. McElvenny DM, Osman J, Hodgson JT, Elliott RC. Rebuttal from the Health and Safety Executive. Int J Occup Environ Health 2000;6:261 262 [letter]. 12. Axelson O. Confounding from smoking in occupational epidemiology. Br J Ind Med 1989;46:505 507 [editorial]. 13. Modan B. Population distribution of histological types of lung cancer. Epidemiologic aspects in Israel and review of the literature. Israel J Med Sci 1978;14:771 784. 14. Davis S, Mirch DK, Stevens RG. Shift work, light at night and risk of breast cancer. J Natl Cancer Inst 2001;93:1557 1562. 15. Hansen J. Increased breast cancer risk among women who work predominantly at night. Epidemiology 2001;12:74 77. 16. Schernhammer ES, Laden F, Sperzer FE, et al. Rotating night shifts and risk of breast cancer in women participating in the Nurses Health Study. J Natl Cancer Inst 2001;93:1563 1568.