METHODS

Establishment of the PVC cohort

The original cohort consisted of 4096 workers from six PVC polymerisation plants. The participant name list was collected from the Bureau of Labor Insurance (BLI), where workers' details, including personal identification number (ID), date of birth, gender, and dates of new and change of employment, and retirement are stored. As every factory that employs more than five workers is required to join the National Labor Insurance (NLI) by law, inclusion of our cohort of PVC workers should be representative of PVC workers in general.

Eligibility criteria for analysis

As the number of female workers was small (n = 85), and all were office workers, they were excluded from the study. Furthermore, 80 workers whose ID numbers were not available from BLI were also excluded. To be eligible for this cohort, employees at six separate plants had to meet the following criteria: male workers exposed to VCM for a period of at least one year in the period 1950–92. Thus 468 workers who were employed before 1992 and had VCM exposure less than one year were excluded; 128 workers who were employed after 1992 were also excluded from analysis. As the personal ID number scheme in the National Mortality Registry (NMR) that was used as a linkage “key” has only been available since 1985, the study group was further restricted to men who were known to have been alive on 1 January 1985. Thus a further 42 workers were excluded because their vital status could not be established by 1985. A total of 3293 subjects were therefore included for estimation of the standardised mortality ratio (SMR).

Determination of individual vital status

For each cohort subject, through data linkage to Taiwan's National Health Insurance, information was collected to ascertain the cohort's vital status. Furthermore, the NMR was also contacted in order to identify the occurrence of any deaths of cohort members. The NMR has maintained a computerised file of all death certificates issued in Taiwan since 1971. The underlying cause of death has been systematically coded according to the ninth revision of the International Classification of Diseases (ICD9).

The period of observation of this study was within the period 1985–97. A total of 144 deaths of cohort members occurred during the follow up period, with a successful follow up rate of 99%.

Statistical analysis

The SMR was estimated by counting person-years at risk, which were accumulated into five year age groups beginning at the age of 15, and five year calendar periods commencing in 1985. Person-years calculation for a subject begins on 1 January 1985 or on the date when the subject accrued one year of exposure, whichever was later. Subsequently, the observed mortality number was divided by the expected number of deaths, computed from the general male population of Taiwan during the period 1985–97. The age and calendar period adjusted SMR were established and the corresponding 95% confidence intervals (95% CI) calculated using Poisson distribution.¹⁴ SMR analysis was also performed for the different classifications of exposure variables, including duration of VCM exposure, age and calendar year of first exposure, and duration of latency. Latency is the total length of time between the start of work and the end of follow up.

RESULTS

Descriptive statistics

A total of 3293 male workers met the enrolment criteria. By the end of the study, 3146 (95.5%) members of the cohort were still alive, 144 (4.4%) had died, and three (0.01%) had an unknown status. The total number of person-years at risk was 40 577 (table 1).

The majority of the cohort members were born between the period 1940–59, and only 12.5% were born after 1960. Over half (58.5%) of the cohort participants commenced employment before 1980. Most of the cohort (72.6%) were hired before the age of 30, although 6.3% were hired after the age of 40. Virtually half of the cohort (49.1%) had been employed for more than 10 years, and only 34.5% had remained employed for less than five years. The median duration of employment for the entire cohort was 9.0 years.

Cause specific mortality analysis

Table 2 summarises the observed and expected death by cause specific SMR for the whole cohort. The 144 observed deaths in the cohort resulted in an overall SMR of 0.78 (95% CI: 0.65 to 0.91) in comparison to normal Taiwanese male mortality. Mortality from cancer of all sites combined in PVC workers was higher than the expected number based on normally expected male Taiwanese mortality (57 observed v 43.7 expected). The excess deaths in our study group were caused mainly by liver cancer (25 observed v 14.1 expected), and lymphatic and haematopoietic cancer (7 observed v 2.6 expected). There were no excess deaths for any other category of cause of death. There was a significant reduction in mortality from all diseases of the circulatory system (12 observed v 35.0 expected), which were mainly from cerebrovascular disease. There were also significant decreases in deaths from all diseases of the digestive system (12 observed v 23.3 expected), caused mainly by chronic liver disease and cirrhosis (7 observed v 16.4 expected).

Analysis by duration of exposure, age, calendar year of first exposure, and latency for mortality

The data were further analysed by duration of exposure, age, and calendar year of first exposure, as well as for different latency periods (table 3). Longer duration of exposure did not seem to be a significant determinant of increased mortality for all causes and all cancers including liver cancer.

We analysed the cause specific SMR stratified by calendar year of first exposure. The SMR for liver cancer was 4.82 (11 observed v 2.3 expected, 95% CI: 2.41 to 8.63) for those first employed before 1970, 1.92 (8 observed v 4.2 expected, 95% CI: 0.83 to 3.79) for those employed between 1970 and 1979, and 0.78 (6 observed v 7.7 expected, 95% CI: 0.28 to 1.69) for those employed after 1980.

Subsequently, we analysed the data by stratifying the latency period into three groupings: less than 15 years, 15–24 years, and more than 25 years. The results also reflected an increased SMR value for liver cancer corresponding to the largest duration category (SMR 3.13; 10 observed v 3.2 expected, 95% CI: 1.50 to 5.75).

Analysis by age at first exposure in the categories less than 30 years of age, 30–39 years of age, and greater than 40 years of age was conducted. Mortality from liver cancer was inversely associated with age at first exposure. The SMR value for these three groups was 2.24 (10 observed v 4.5 expected, 95% CI: 1.07 to 4.12), 1.78 (6 observed v 3.4 expected, 95% CI: 0.65 to 3.88) and 1.43 (9 observed v 6.3 expected, 95% CI: 0.65 to 2.71), respectively. Further analysis was performed on SMR for liver cancer by latency and age at first exposure. In workers with latency greater than 25 years, SMR for liver cancer in those who had first exposure before 30 years of age was greater than that in those who had first exposure after 30 years of age (SMR 4.05, 95% CI: 1.48 to 8.82 v SMR 2.33, 95% CI: 0.63 to 5.95). In workers with latency less than 25 years, a similar relation was not observed.

We successfully obtained medical records in 18 of 25 cases of cohort members having died from liver cancer. Among 18 subjects who had medical records (as summarised in table 4), five workers were confirmed to have HCC by histology. An additional five workers with extremely high serum concentrations of α fetoprotein (AFP >1000 μg/l) and at least one positive image from angiography, sonography, liver scan, and/or a computed tomography (CT) scan, were regarded as having HCC. Previous clinical observations indicated that an AFP concentration exceeding 400 μg/l with at least one positive image suggesting liver tumour, confirms the presence of HCC.^{15, 16} Most liver cancers among VCM exposed workers in this study were therefore HCC. However, we found no deaths caused by ASL.

DISCUSSION

In this study, we found that workers from PVC manufacturing plants, compared to the general male population in Taiwan, showed a higher mortality from digestive system cancers, especially liver cancer. The risk increase as a result of VCM exposure was almost five times greater than that for the general Taiwanese male population, if the first exposure occurred before 1970. We also observed that the risk of liver cancer increased in workers who were first exposed to VCM before age of 30.

The most compelling evidence for the carcinogenic potential of VCM in humans comes from the cluster of reports of a greater than expected incidence of angiosarcoma of the liver in workers occupationally exposed to VCM.^2–6 Angiosarcoma of the liver is considered to be a very rare type of cancer, and it has become apparent that workers exposed to VCM exhibit an exceptionally high incidence of this type of tumour.^2–6 In this study, we found no deaths caused by ASL. As tissue proof was relatively uncommon for liver cancer in Taiwan,¹⁷ a diagnosis of ASL may not be convincingly made. In previous studies, ASL accounted for more than half of the liver cancers.^{5, 6} The lack of ASL in our cohort was probably a result of the much lower levels of VCM exposure in our study subjects compared to those in Western workers. The PVC industry began in 1958 in Taiwan, two to three decades later than that in Western countries. It has been estimated that in the early years, workers in Western countries might have been exposed to concentrations as high as 1000–2000 ppm of VCM before 1960.¹⁸ A job exposure matrix model developed by Du and colleagues¹⁹ in 2001 revealed that VCM exposure for Taiwanese workers was about 500 ppm in 1960s. Furthermore, p53 mutation in liver cancer cells and serum oncoprotein p53 were observed in VCM exposed workers.^{20, 21} The prevalence of serum oncoprotein p53 in Taiwanese PVC workers was much lower than that in Western workers.²² This also suggests that the levels of VCM exposure in our workers were not as high as those in Western workers.

Epidemiological studies suggest that hepatitis B virus (HBV) infection is the most important cause of liver cancer in Taiwan, with more than 80% of liver cancer patients carrying hepatitis B virus surface antigen (HbsAg),^{23, 24} and 30–50% being anti-HCV positive.^{25, 26} This has suggested that the increased risk of liver cancer might be a result of confounding by hepatitis virus infection. Although we had no data pertaining to HCV infection prior to 1992, our recent cross sectional study has revealed the prevalence of 19.9% for HBsAg and 2.4% for anti-HCV in this cohort.²⁷ This level was comparable to that for the general Taiwanese population (HBsAg 15–20%,²³ anti-HCV 1–4%,²⁸ respectively). In this study, we observed that PVC workers exhibited almost a twofold greater risk of developing liver cancer than the general male Taiwanese population. The magnitude of such a risk could not be explained completely solely by hepatitis B or C virus infection. Whether there is a synergistic interaction between VCM exposure and hepatitis virus infection in the development of liver cancer, however, needs to be further investigated.

In addition to HBV and HCV infection, aflatoxin exposure, arsenic exposure, and alcohol consumption may also play important roles in the development of liver cancer. We carried out a data linkage between our cohort and another cohort in Taiwan, the blackfoot disease (BFD) cohort, which investigated the association of arsenic exposure and cancer.^{29, 30} Only one subject from our cohort was present in the BFD cohort list; that person remained healthy at the end of follow up. Moreover, our data also showed that only two cases of liver cancer death reflected a history of excessive alcohol consumption. Thus, arsenic exposure and alcohol drinking could not explain the twofold increase in the risk of SMR shown in this study.

Some previous studies have reported that PVC workers experienced a greater risk of developing cancer of the brain¹⁰ and/or the lymphatic/haematopoietic systems.^{10, 13} In our study of PVC workers, we also observed a significantly higher risk of developing a cancer in the lymphatic and haematopoietic systems, although the number of deaths from lymphatic and haematopoietic system cancer was too small to draw a conclusion. Further follow up may shed light on this relation. By contrast, we observed no increased mortality from brain tumours, although this result was similar to previous studies from the USA and Europe.^{5, 6}

A reduced overall mortality risk compared to the general population was also observed in our study, which was generally recognised as the so called “healthy workers” effect. Such a phenomenon could be explained by significant deficits in mortality arising from diseases of the circulatory system and digestive system among cohort members, although as cancer mortality was less likely to be influenced by the initial and subsequent selection of the healthiest individuals for work purposes, the effect might be less great.³¹ Even if such an effect existed in this Taiwanese cohort, it might lead to a null result. In addition, we also observed a reduced mortality risk for chronic liver disease and cirrhosis (ICD9 = 571) in this study. This was caused either by the healthy workers effect or the fact that these PVC workers had better medical care provided by NLI. These workers thus had a better chance to survive chronic liver disease or cirrhosis compared to the general population. It was also likely that VCM caused p53 mutation, which might promote neoplastic transformation of liver cirrhosis by means of increased proliferation of liver cells and reduced DNA damage.

Longer duration of exposure did not seem to be a significant determinant for increased cancer mortality. A possible explanation is the higher turnover for workers involved in high exposure activities compared to other workers who were not directly exposed to VCM. The other explanation is that those with diseases tended to leave the job earlier, which was consistent with the observation that SMR increased with shorter duration of exposure for all causes and all cancers. In this study, the available historical exposure data were deemed to be sparse for quantitative estimation of (historical) cumulative exposures. Thus, the cumulative dose of VCM exposure was not used in the analysis. However, it was interesting to observe that workers with VCM exposure prior to 1970 had higher mortality for liver cancer. PVC workers tended to be exposed to higher levels of VCM before 1981 when worksite VCM permissible exposure levels had not been set. Further analysis indicated that workers who had a latency greater than 25 years also had a higher risk of liver cancer. It may take more than two decades for liver cancer to develop from the beginning of exposure to death. Because the 10 deaths caused by liver cancer with latency greater than 25 years included nine deaths with first exposure prior to 1970, it is difficult to separate the effect of exposure prior to 1970 on liver cancer from latency. We also found that mortality from liver cancer increased inversely with age at first exposure after controlling for the effect of latency. In animal studies, induction of preneoplastic hepatocellular lesions in rats by vinyl chloride was restricted to young animals.^{32, 33} Thus the early period of life may be a sensitive period for developing tumours related to VCM exposure.

Another limitation of this study was that some of our cohort members might have died before 1985. Unfortunately, we were unable to carry out the data linkage on our computerised NMR prior to 1984 because of an absence of personal ID numbers in this file. We are exploring the possibility of linking the NMR with other government databases to increase the completeness of available data; until we can do this, an underestimation of occupationally influenced mortality among PVC workers appears likely.