A collaborative study of cancer incidence and mortality among vinyl chloride workers.

R. A collaborativestudy of cancer incidence and mortality among vinyl chloride workers. Scand J Work Environ Health 1991;17:159-69. A large European multicentric cohort study has been coordinated by the International Agency for Research on Cancer with the objectives of investigating the dose-response relationship between liver cancer and exposure to vinyl chloride and assessing cancer risk for sites other than the liver. A nearly threefold increase in liver cancer was detected on the basis of 24 observed deaths and 8.4 expected(standardized mortality ratio 286,95 % confidenceinterval 186-425). The excess from liver cancer wasclearly related to timesincefirst exposure, duration of employment, and estimatedranked and quantitativeexposures. Other cancer sites investigated on the basis of a priori hypotheses were either not in excess (lung) or apparently unrelated to the exposure variables (brain and lymphoma).

Vin yl chloride (VC) is an established carcinogen for humans (l , 2) , and it is also on e of the few substa nces for whi ch the experimental evidence of carcinogenicity was available (3) before the carcinogenic effects o n humans could be demonstrated.
The con cern about the haz ards from expo sure to this substance led to control measures which lowered th e levels of expo sure in excess o f 500 ppm in the ea rly 1950s to values gen erall y below I ppm in industri alized countries.
In recent years, the development of laboratory techniques in investigating pos sible effects at the molecu-Reprint requests to: Dr R Saracci, Unit of AnalyticalEpidemiology, International Agency for Research on Cancer, 150 Cours Albert-Thomas, F-69372 Lyon cedex 08, France. la r level (4) renewed the interest in this substance. In addition the ep idemiologic data have recently been updated (5-7) and reevaluated in a review (8).
Although occupational exposures ha ve decreased, impo rtant scientific and public health questions remain. In 1986 the occupational program of the International Agency for Research on Cancer (lARC) invited European ep idemiologists involved in research on the effec ts of VC to combine their efforts in a multicentric cohort study . The collaborative cohort study reported in this presentation was undertaken with the following three objectives: (i) to determine whether VC is associated with increased cancer risk at sites other than the liver, (ii) to in vesti gate the possible exposure-response relationship between VC and liver cancer , particularly angiosarcoma of the liver, and (iii) to construct a data ba se whi ch could be exploited in the future in relation to the ass essment of potential risk at low levels of exposure.
In thi s report , we present the methods and results of this collaborative study, f urt her details of which can be found in an IARC report (9).

Subjects and methods
Collaborators from four countries (Italy, Norway, Sweden, and the United Kingdom) participated in the cohort study and contributed a total of 14 351 subje cts to the combined data base. Both existing studies and newly collected cohorts were enrolled from 19 factories. Where existing study populations were included (10)(11)(12)(13)(14)(15) follo w-up was extended, and/or more factories were added. In the majority of factories in the cohort there was mixed VC monomer/polyvinyl chlo- ride (VCM/PVC) production (12 factories), two produced VCM only, four produced PVC only, and one was a PVC-processing plant. For the sake of homogeneity, we decided to include only subjects with at least one year of employment in the analysis, and therefore 1518 subjects (10.6 0J0 of the combined cohort) had to be excluded. An additional 127 subjects (0.9 C1Jo of the combined cohort) were excluded for the following reasons: female gender (N = 57), out of observation period (N = 48), member of more than one cohort (N = 21), and date of first exposure unknown (N = 1). After the total of 1645 exclusions (11.5 0J0 of the combined data base), 12706 subjects remained for the analysis.
The vital status of the subjects included in the mortality analysis is shown in table 1, and the completeness of follow-up at 97.7 0J0 can be considered satisfactory. For the 12706 subjects included in the analysis, the average length of follow-up was 17 (range 10-25) years, 36 0J0 of the cohort having a follow-up period of 20 years or more. The total number of person-years at risk was 222746, and the distribution of person-years according to duration and the number of years since first exposure is given in table 2.
National incidence or mortality rates (men only) specific for age and five-year calendar periods were used for reference. The observation period differed by factory, the majority of subjects having been followed from 1955 (first year for which reference rates were available) or from the start of the second year of employment, whichever came first, to 1986. Two of the four countries, Norway and Sweden, were also able to provide follow-up for incidence through nationbased cancer registries, while mortality rates were computed at IARC with the use of a data base belonging to the World Health Organization (WHO). As different revisions of the International Classification of Diseases (ICD) were used over the follow-up period, Table 1. Vital status of the cohort members.s a conversion table for causes of death was used and can be found in the detailed IARC report (9).
It should be noted that for this study liver cancer was defined as ICD 155-156 (seventh revision): liver, intrahepatic and extrahepatic bile ducts, and gallbladder specified as primary or secondary; ICD 155 (eighth revision): liver and intrahepatic bile ducts specified as primary; and ICD 155 (ninth revision): liver and intrahepatic bile ducts specified as primary and liver specified as secondary. Although ideally only primary liver cancer should be chosen as the definition, the seventh and ninth ICD revisions do not permit separation of primary and secondary (as this distinction is often difficult in reality). In the eighth and ninth revisions, secondary liver cancer is classified by a fourdigit code, but specification of the national mortality rates to four digits in the WHO data base is not available from all countries. Therefore, it was necessary to choose the aforementioned definition for liver cancer.
Angiosarcoma of the liver is distinguished from other types of liver cancer through histology . This type of information was sought from national investigators, and only cases histologically confirmed were included in the analyses.
For the analysis, person-years at risk were calculated with the person-years program using a modified lifetable approach (16). In calculating the person-years, no censoring at old age took place, and the date of entry for the tabulation of person-years started at the beginning of the observation period according to the availability of reference rates in 1955 or on day 1 of the second year of employment, whichever occurred later.
The standardized mortality ratio (SMR) or standardized incidence ratio (SIR) and the 95 0J0 confidence interval (95 0J0 CI) for the SMR or SIR were calculated on the assumption of a Poisson distribution. Prior to the analysis, it was decided that four causes of death suspected a priori (ie, liver cancer, lung cancer, brain cancer, and lymphosarcoma) would be examined in detail.
The mortality analysis was performed according to several temporal variables, specifically years since first exposure, calendar period at hire, calendar period at exit, age at hire, and age at exit. In the analysis all of these variables were based on individual information available for each subject. Exposure variables were job title as autoclave worker (ever/never), duration of employment, ranked level of exposure, and cumulative exposure in parts per million-years to VCM in the air.
For the ranked level of exposure and cumulative exposure indices, job histories were required, along with exposure estimates for specific jobs and calendar periods from job-exposure matrices. The job-exposure matrices specific for calendar period were provided by industrial hygienists for 13 of the 19 factories . These matrices were developed in various ways for the different factories. For most of them, job title was used as the basic unit with which exposure was assessed, and job histories were available for all factories except two. "Typical exposures" to VC in air were estimated as time-weighted averages by industrial hygienists using several sources of variable quality. In most factories, occasional measurements of VC provided the basis for past typical exposures, supplemented by knowledge of exposure conditions, processes, and technological changes over time. Systematic measurements taken since the mid-1970s provided the basis for more recent exposure assessments, and an indication of the variability of exposure levels between job titles.
In terms of agents to which the workers were exposed in VCM, PVC, and VCM/PVC production, VC was the main exposure, and virtually the only exposure in many of these factories. Use of butadiene was rare. In PVC processing (one factory in this study), however, additional exposures could have included PVC dust, asbestos, and other agents . All the job-exposure matrices referred only to VC exposure in air.
Each job-exposure matrix was checked and validated by two independent industrial hygienists, who were able to provide, prior to the statistical analysis, an index for the ranked level of exposure (low: < 50 ppm; intermediate: 50-449 ppm; and high:~500 ppm) in which the classification of the subjects was based on the highest level to which the workers were potentially exposed, specific to their jobs and the years in which they worked, according to levels of VC recorded in the job-exposure matrices.
The same information, that is, job histories and jobexposure matrices, was used to calculate cumulative exposure in parts per million-years (exposure level from the job-exposure matrices multiplied by duration of employment) for the subjects. In some of the analyses for cancer of the liver an estimated job-exposure matrix was used for the four factories in the United Kingdom which were unable to provide their own matrices . The estimated job-exposure matrix was based on the matrices provided for the other factories in the United Kingdom . This matrix was checked with the industrial hygienist from the United Kingdom. In the following text, it is clearly noted if the analyses under discussion include this estimated matrix.
For liver cancer only, the Poisson regression analysis was performed to assess the significance of several variables simultaneously. This analysis used observed deaths and the person-years distribution for crossclassified categories of temporal and exposure variables and performed an internal comparison with the base-line categories (ie, within the cohort only) (17,18).

Mortality results
Cause-specific mortality for the total cohort is presented in table 3. A statistically significant deficit for all-cause mortality was apparent (1438 deaths observed versus 1636.4 expected, SMR 88, 95 % CI 83-93). The following four main causes of death contributed to this deficit: (i) diseases of the circulatory system, (ii) diseases of the respiratory system, (iii) accidents, poisonings and violence, and (iv) other known causes.
In this report the results are not given by process. The statistically significant excess of liver cancer evident in the total cohort was mainly due to the excess in VCM/PVC production (19 deaths observed, SMR 311, 95 % CI 187-486).
Twenty-three sites were selected for analysis according to the years since first exposure, and the results for 10 of these sites are presented in table 4. Apart from liver cancer, which will be discussed in detail, there were no noteworthy patterns in risk according to this variable.
Four sites were investigated in detail for relationships with temporal and exposure variables. Of the two sites with excess risk for the total cohort, cancer of an unspecified site was not analyzed further due to the diversity of cancers found in this category; they are however discussed in a descriptive fashion in the text. Liver cancer, lung cancer, brain cancer, and lymphosarcoma were chosen a priori for further analysis. Excesses of bladder cancer in PVC production (in the United Kingdom) and of melanoma (in Norway) were not investigated further for the total cohort since they were confined to one country.

Liver cancer
No liver cancer deaths occurred before 15 years since first exposure, after which the SMR was 483 (95 % CI 208-951) for 15-19 years since first exposure, and it did not vary greatly from this level thereafter, the value always being statistically significant. In table 4, the pattern by years since first exposure is seen in lO-year groups. When only those with 15 years since first exposure or more were included in the analysis (l5-year latency), the overall SMR for liver cancer was 445 (95 % CI 285-663). Table 5 shows the SMR values for liver cancer according to the four exposure variables, without and with a 15-year latency analysis. According to job title, dichotomous as ever autoclave worker (suspected a priori as the highest risk job) versus never an autoclave worker, very high risk was experienced by those who were autoclave workers at some time (SMR 896, 95 0J0 CI 447-1603). In the analysis with a 15-year latency period, a statistically significant increased risk was also apparent for those classified as " never an autoclave worker," a group which however included workers with job "unspecified" also. Duration of employment was associated with an increasing mortality trend from liver cancer, which was statistically significant (X' 19.5, P < 0.001). With a 15-year latency period, the trend was not as strong (X' 5.70, P<0.025).
A very clear exposure-response relationship was seen for ranked level of exposure and liver cancer mortality. Although the ranked level of exposure was unknown for six deaths, increasing risk at progressively increasing levels of exposure was demonstrated (X' 7.99, P<O.Ol). With the use of a 15-yearlatency period and the estimated job-exposure matrices from factories 10 through 13 from the United Kingdom, the SMR for the intermediate category (50-499 ppm) was also statistically significant (7 deaths observed, SMR 551, 95 070 CI 222-1136). The risk clearly increased with increasing total cumulative exposure to VC in parts per million-years (Xl 2004, P < 0.001). An analysis with a five-year lag in cumulative exposure made virtually no difference in any of the results, and therefore the data 162 are not presented. The results of the multivariate analysis for liver cancer in an internal comparison are presented in table 6. Only two variable s, years since first exposure and cumulative exposure , had a statistically significant effect on the risk of liver cancer mortalit y.
Risk increased steadily with increasing exposure when years since first exposure was adjusted for. The tests for interactions were not statistically significant, and the addition of a quadratic term for cumulative exposure did not improve the fit of the model. The same procedure was followed when cumulative exposure from the estimated job-exposure matrix for the factories from the United Kingdom was included , and the regression results were similar. The relative risk estimates varied slightly from the previous model, probably because of the effect of misclassification from the job-exposure matrices of the four factories whose matrices were developed from those of other factories.

Angiosarcoma of the liver
Major characteristics of the 24 liver cancer deaths certified in the mortality data as ICD code 155 (eighth and ninth revision, as no death was coded according to the seventh revision) can be seen in table 7. When  a The values in parenth eses were determ ined in analyses includi ng the est imated [ob-exposure matrices. b In Norway, the longest-held job was used. In Sweden job rotati on was prac tic ed, and no one was cl assifi ed as an autoc lave wor ker. histology had been performed and this information was available, a liver cancer death could be classified definitely as an angiosarcoma or not. When histology had not been performed or the information was not available, it was classified " unknown" as to whether or not it was an angiosarcoma. Deaths that occurred before 1974 when angiosarcoma of the liver was first reported in relation to VC (19) may not have been investigated as thoroughly as they would have been following the 1974 report. Of the 17 deaths for which histology (pathology) information was available, 16 were histologically confirmed angiosarcomas of the liver and one was a primary liver cancer. For the remaining seven, it is unkno wn whether or not they were angiosarcomas of the liver . A regression analysis was performed to assess the risk of angiosarcoma of the liver. In total , 22 angiosarcomas were included, with 16 coming from the liver cancer death s coded to leD 155 (as seen in table 7) and six additional angiosarcomas coming ' from other deaths, as seen in table 8. This table shows additional liver cancer deaths , not coded as 155 in the mortality data.
In total, there were six angiosarcomas, one primary liver cancer, and five " unknown" liver cancers among the 12 additional death s. The small numb ers necessitated combining the lowest two categories of cumulative exposur e « 2000 ppm-years) for stability of the base-line category . The results were similar to those obtained for the 24 liver cancer deaths in that the final model included years since first exposure and cumulative exposure. The major difference is seen in table 9, where the relative risks for angiosarcoma are higher at each levelof cumulative exposure than those for liver cancer.
The absolute risk of angiosarcoma is shown in table 10, on the basis of the results of the regression analysis shown in table 9, when cumulative exposure and years since first exposure were both included in the model. At~25 years since first exposur e and 10 000 ppm-years, the absolute risk was 280 per 100000.
In summary, the results from the regression analyses indicated that, while cumulative exposure and years since first exposure had had a detectable effect on the risk of liver cancer mortality and angiosarcoma of the    A detailed ana lysis of the seven deaths fro m lympho sarcoma showed no pattern in the SMR values according to year s since first exposure. All seven deaths occurred in VCM/P VC production , and for this pro cess alone th ere was no excess apparent by calendar period of hire or exit or age at hire or exit. liver, age at first exposur e and calendar period of exposure did not. Very clear exposure-respo nse relat ionships were evident between the cumulative exposure to VC a nd the risk of liver cancer and an giosa rcoma of the liver . Fina lly, an effect of misclassification was dem onstrated when the estim ated job-exposure mat rix was included , although the effect was minimal for the angiosa rcoma results .
L ung cancer, brain cancer and lymphosarcoma Th e SMR fo r trachea, bro nchu s, and lung cance r was 97 (95 070 CI 82-114) for the total cohort, on the basis of 144 observed deaths. Th e SMR values did not show any remarkable associati on with a particular process, and a lthough no patt ern was evident fo r years since first expos ure, ther e was a statistically significa nt increase at 25-29 years since first exposure on the basis of 33 observe d deat hs (SMR 147, 95 % CI 101-207), ma inly fro m an excess in VCM production in th is time period (5 deaths observed , SMR 486, 95 % CI 158-1134). Calendar period at exit and at hire did not reveal any con sistent pat tern. Fourteen death s from brain cancer occ urre d in the cohort, and, alt hough the overa ll SMR was no t increased (SMR 107, 95 % C I 59-1 80), the re was a statistically signific ant excess at~30 years since fir st exposure on the basis of fo ur observ ed deaths (SMR 407,95 % CI 111-1041) (ta ble 4). The excess was confined to the calenda r period of hire of 1945-1954 and was the most evident for VCM/PVC production. Analyses by calenda r period of exit, age at hire, an d age at exit did not reveal any pattern s of risk for brain cancer mo rtality.
In table 11, the SMR values for lung cancer, and in table 12 those for brain cancer and lymphosarcoma , are shown accord ing to the exposure variables. The analyses by job title, duration of employment, and cumulative exposure showed no relation ship with any of the three sites. Neither was lung cancer associated with the ranked level of exposur e, while bra in cancer was slightly in excess in the high category of exposure . Lymphosarcoma showed a slight increasing pattern of mortality with increasing category of ran ked level of exposure, but the entire analysis was based on four deaths only. a The values in parentheses were determ ined in analyses inc lUding the est imated job-exposure matrices. D In Norway, the longes t·held job was used. In Sweden job rotati on was pract iced, and no one was classif ied as an autoclave worker. a In Norway , the longest-held job was used. In Sweden job rotation was practi ced, and no one was classified as an autoclave wor ker. D All six deaths in the periods 1-4 years.

Cancer of unspecified site
Twenty-four deaths in the cohort were classified as malignant neoplasms of unspecified sites (ICD 199), giving a statistically significant excess (24 deaths observed, SMR 187,95070 CI 120-278). Additional information, such as cancer incidence data, was available for 20 of the 24 deaths. Three subjects had liver cancer and were therefore included in table 8. It was unknown, however, whether they had angiosarcoma of the liver since no histological information was available. No other clear excess of a particular cancer among those classified as cancer deaths of an unspecified site was apparent.

Cancer incidence results
For the 2643 subjects from the four factories in Norway and Sweden included in the cancer incidence analysis, the total number of cancers observed was 127(SIR 107, 95 % CI 89-127), and their distribution by site is shown in Although the lung cancer increase was not statistically significant, it was investigated in more detail as other studies have suggested increased risk for lung cancer. There was no excess according to process type or category of years since first exposure . A slight excess was suggested for < 15 years of employment, while the SIR values were close to 100 for 15-19 and~20 years of employment. For the ranked level of exposure, the risks in the high and low categories were virtually identical. According to cumulative exposure, no exposure-response was apparent (results not presented in tabular form). Without statistical significance and with little apparent relationship to VC exposure, some indication of increased lung cancer risk remained for one PVC-processing plant and one Norwegian VCMI PVC production plant. The national investigator for the PVC-processing plant attributed the excess to exposure to asbestos, which was utilized in the process (7), while the excess remained unexplained in Norway.

Discussion
This collaborative study was carried out with the main purpose of analyzing exposure-response relationships between exposure to VC and liver cancer and investigating whether exposure to VC could increase cancer risk for sites other than the liver.
The results confirmed the association between exposure to VC and liver cancer. The excess ofliver cancer mortality was associated with duration of employment, and a clear association with ranked level of exposure was found. The results were strengthened by the regression analyses, which indicated that the risk of liver cancer depended on cumulative exposure and years since first exposure.
Twenty-two subjects had histologically confirmed angiosarcoma of the liver, and the regression analyses demonstrated that the risk was mostly influenced by cumulative exposure to VC. The relative risks were higher at each level of cumulative exposure than those for all liver cancer deaths, but it must be remembered that the same 16 angiosarcoma deaths were included in both analyses. The approximate incidence rate of angiosarcoma of the liver in Norway , for example, based on 1953-1988 data, was 1 in 10 million per year (personal communication from A Andersen, 1989). Others have estimated the annual incidence at 1 to 2 in 10 million (20,21).
Given 222746 person-years at risk accumulated by this cohort, with an annual incidence of angiosarcoma of the liver of 2 per 10 million in the general population , the overall expected figure for the cohort would be 0.045. The rarity of this tumor supports the use of internal comparisons to assess the significance of exposure variables.

167
The exposure estimates used for the ranked level of exposure and cumulative exposure indices were based on the reconstruction of past exposures, and they appeared to be an efficient tool for investigating exposure-response relationships. Although the job-exposure matrices utilized for the analysis were often based on rough estimates and thus resulted in a certain degree of imprecision, the results demonstrate exposure-response relationships for the carcinogenicity of VC. A recent study also estimated cumulative exposure to VCM, PVC, and butadiene and found that only cumulative exposure to VCM had any effect on liver cancer risk (6).
No increase was evident for lung cancer mortality, nor was there any association with the exposure variables, including ranked and cumulative exposure indices. It should be noted that the power of the study would allow detection of a statistically significant (at the 5 % level, one-sided) SMR for lung cancer of 114 with 80 0,70 probability.
A slight increase in lung cancer was suggested by the incidence data, which corresponded to about 10 % of the mortality data. There was no apparent relation between lung cancer incidence and VC exposure. In one factory of the four included in the incidence analysis, a case of pleural mesothelioma was reported, and exposure to asbestos has been documented for the PVCprocessing plant (7).
Two other sites investigated for excess risk were brain cancer and lymphosarcoma according to the a priori hypotheses. The results from this collaborative study did not suggest an effect of exposure to VC on mortality from brain cancer, although the power of the study only permitted detection of an SMR of 182 or more, which could be labeled as statistically significant with 80 % probability. The excess at~30 years since first employment was, however, an indication that an effect of exposure cannot be fully dismissed. For lymphosarcoma, although an excess was suggested, the small numbers and missing information on exposure variables for some of the subjects prohibited interpretation in relation to exposure to vC.
None of the other causes of death was in excess; instead, some statistically significant deficits were apparent. The deficit in total mortality was probably due to the healthy worker effect, both in the hiring of healthy workers compared with the general population and in a survival effect within the cohort due to the criteria for inclusion of employment for one year or more.
These results are very similar to those reported recently in the United States (6). In that study only liver cancer was in excess for the cohort of VCM workers (SMR 333, 95 % CI 202-521), with no statistically significant excess of lung cancer (SMR 115, 95 % CI 95-139), brain cancer (SMR 145, 95 0,70 CI 79-248) or hematopoietic cancers (SMR 78, 95 % CI 48-121). In the nested case-referent analysis, liver cancer risk increased with increasing cumulative exposure (esti-168 mated as duration times categorized exposure level). Of the 19 liver cancers, 12 were angiosarcomas, and for this subgroup only, unlike our results, was the positive dose-response evident.

Concluding remarks
The results of this multicentric collaborative study on workers in the VC industry indicate that exposure to VC is associated with an increase in liver cancer. An exposure-response relationship was observed for both ranked and estimated cumulative exposure. The relationship was even more evident when only liver angiosarcoma was analyzed.
No significant excess of mortality was found for the other sites suspected a priori to be affected by exposure to VC. Although the incidence of lung cancer was slightly increased, neither it nor lung cancer mortality appeared to be associated with any of the exposure variables. Brain cancer and lymphosarcoma mortality, although showing slight increases, did not appear to be consistently associated with exposure, although the small numbers prohibited firm conclusions.
An increased risk of bladder cancer and melanoma of the skin was detected which did not appear to be related to exposure in that the association with employment in the VC industry was confined to one country only.
No increased mortality was observed for the other main causes of death.