Cancer Mortality in Workers Exposed to 2,3,7,8-Tetrachlorodibenzo-P-Dioxin

Identification of Companies

In 1978 the National Institute for Occupational Safety and Health began an effort that would eventually identify the exposed workers at all U.S. chemical companies that had made TCDD-contaminated products between 1942 and 1984. TCDD was generated as a contaminant in the production of 2,4,5-trichlorophenol and was carried into subsequent production processes.²³ One derivative, 2,4,5-trichlorophenoxyacetic acid, was widely used in the United States to kill brush and was a constituent of defoliants such as Agent Orange. Other derivatives included the herbicides 2-(2,4,5-trichlorophenoxy)propionic acid (Silvex) and 2-(2,4,5-trichlorophenoxy)-ethyl 2,2-dichloropropionate (Erbon), the insecticide O^,O-dimethyl O-(2,4,5-trichlorophenyl)phosphorothioate (Ronnel), and the bactericide 2,2'-methylene-bis[3,4,6-trichlorophenol] (hexachlorophene).

Identification of Exposed Workers

Workers from 12 companies were included in the study cohort if a personnel or payroll record documented that they had been assigned to a production or maintenance job in a process involving TCDD contamination (n = 5000), or if they had been identified in a previously published study on the basis of exposure to TCDD (n = 172).²⁴ Personnel records for 202 workers did not reveal the duration of their assignment to processes involving TCDD contamination; they were therefore included in the analysis of overall mortality but excluded from analyses according to duration of exposure. Sixty-seven women are not included in this report; there were 10 deaths among them, including a single death from cancer (lung cancer).

At each plant, we made a thorough review of operating conditions, job duties, and records of TCDD levels in industrial-hygiene samples, intermediate reactants, products, and wastes. This review provided clear evidence of potential daily exposure to TCDD. The production of TCDD-contaminated substances at the various plants involved similar raw materials, processes, and job duties.²⁵ However, there were differences between jobs and between plants in the extent of TCDD exposures. Occupational exposure to substances contaminated with TCDD was confirmed by measuring serum TCDD levels, as adjusted for lipids, in 253 surviving members of the study cohort from two plants who were also participants in a related cross-sectional medical study.²⁶

Life-Table Analysis

Vital status was determined as of December 31, 1987, from records of the Social Security Administration or Internal Revenue Service, or from the National Death Index. All death certificates were independently classified by two nosologists according to the rules of the revision of the International Classification of Diseases (ICD) in effect at the date of death.²⁷

Life-table analysis was used to evaluate mortality in the cohort.²⁸ At each plant, the number of person-years at risk was calculated as the interval between the first systematically documented assignment to a process involving TCDD contamination and the date of death or December 31, 1987, whichever occurred first. Those whose vital status was unknown were assumed to be alive at the end of the study. Standardized mortality ratios (SMRs) were computed by dividing the observed number of deaths by the expected number and multiplying by 100, after stratification to adjust for the confounding effects of age, race, and year of death. Two-sided 95 percent confidence intervals were computed for each cause-specific SMR, with use of the Byar approximation for eight deaths or more and Fisher's exact method for fewer than eight deaths.²⁹ The U.S. population was used as the reference group, because the 12 plants were located in 11 states throughout the country.

Analyses According to Duration of Exposure and Employment

Figure 1. Figure 1. Serum Levels of TCDD, as Adjusted for Lipids, in 253 Workers, According to Years of Exposure.

Duration of exposure was defined as the number of years the worker was employed in processes involving TCDD contamination and was calculated with data from personnel records. We used duration of exposure as a surrogate for cumulative exposure to TCDD on the basis of the high correlation of the logarithm of serum TCDD levels with the logarithm of the number of years assigned to processes involving TCDD contamination in our sample of 253 workers (Pearson's product-moment coefficient r = 0.72) (Fig. 1), and on the assumption that the production processes were similar in the 12 plants.²⁵

Table 1. Table 1. Vital Status and Demographic and Employment Characteristics of the Study Cohort.

Because of the concentration of person-years in the short-duration categories, duration of exposure was stratified before analysis into categories of <1, 1 to <5, 5 to <15, and ≥15 years (Table 1). Mortality was also examined according to time since first exposure (latency) in periods of 0 to <10, 10 to <20, and ≥20 years since first exposure. To examine mortality in a subgroup with substantial exposure and adequate time for cancer to develop, we identified a group of workers who had 1 year or more of exposure to processes involving TCDD contamination and at least 20 years of latency. One year was chosen as a cutoff point for this high-exposure subcohort because in the sample of workers whose serum TCDD levels were measured, 100 percent of those exposed for more than one year had serum TCDD levels higher than the mean level in the unexposed reference group (7 pg per gram of lipid). For this sub-cohort, the number of person-years at risk was calculated from the date the person attained both 20 years of latency and 1 year of exposure.

Most of the 12 plants were large U.S. chemical manufacturing sites that produced thousands of chemicals. Complete documentation of each worker's exposures was impossible. A separate measure called "duration of employment," defined as the total time that each worker was employed at a study plant, was therefore used. Because of the long total employment at the plants, analyses according to duration of employment were stratified into periods of <5, 5 to <10, 10 to <15, 15 to <20, 20 to <25, 25 to <30, and ≥30 years (Table 1). For these analyses, latency was defined as time since first employment.

When the SMRs showed an apparent trend associated with duration of exposure or employment and when the observed numbers of deaths were sufficiently large, we conducted internal comparisons using directly standardized rate ratios and tests for trend.³⁰ For the standardized rate ratios, the cause-specific mortality rate in each of the categories of longer duration was compared with the rate in the category of shortest duration, after stratification of the rates for the potential confounding effects of age, race, and calendar time.

The cohort of 5172 male workers from 12 plants had 116,748 person-years of observation. Table 1 describes the vital status, race, latency, and duration of exposure and employment of the workers. Overall mortality for all causes of death was similar to national rates in the United States (1052 deaths; SMR, 99; 95 percent confidence interval, 93 to 105). Mortality from heart disease was also similar to national rates (393 deaths; SMR, 96; 95 percent confidence interval, 87 to 106). There were significant reductions in the mortality rates for diseases of the circulatory system (67 deaths; SMR, 77; 95 percent confidence interval, 60 to 98), primarily because of fewer deaths from stroke, and for diseases of the digestive system (38 deaths; SMR, 70; 95 percent confidence interval, 49 to 96), primarily because of fewer deaths from cirrhosis. There were also significantly fewer deaths from alcoholism and personality disorders (2 deaths; SMR, 23; 95 percent confidence interval, 3 to 87). The low mortality from circulatory disease may be a reflection of the "healthy worker" effect — cohorts of workers die at lower rates than the general population, particularly of causes other than cancer.³¹ The reduced number of deaths from cirrhosis and alcoholism implies that this cohort consumed less alcohol than the general population. Reduction may also have occurred simply by chance, since numerous comparisons were made between the cohort and the U.S. population. Fatal injuries were significantly more frequent in the cohort (106 deaths; SMR, 128; 95 percent confidence interval, 104 to 154), but they did not appear to be associated particularly with exposure to TCDD. Mortality from all cancers combined (265 deaths; SMR, 115; 95 percent confidence interval, 102 to 130) was significantly elevated in the cohort.

Cancers of a Priori Interest

Table 2. Table 2. Cancer Mortality in the Entire Cohort and in Workers with More Than 20 Years of Latency.

Table 3. Table 3. Deaths from Soft-Tissue Sarcoma among Workers in the Cohort.*

The term "soft-tissue sarcoma" describes the group of rare malignant neoplasms arising from supporting tissue other than bone.³² We restricted our analysis of mortality due to soft-tissue sarcoma to cases of soft-tissue sarcoma listed as the underlying cause of death on death certificates and assigned to the ICD category "malignant neoplasms of connective and other soft tissue." In the cohort, mortality from soft-tissue sarcoma was nonsignificantly higher than in the reference population (four deaths; SMR, 338; 95 percent confidence interval, 92 to 865) (Table 2). The deaths occurred at 2 of the 12 plants, with a significant increase at 1 plant (two deaths; SMR, 1512; 95 percent confidence interval, 183 to 5462). A review of tissue specimens from the four men whose deaths were attributed to soft-tissue sarcoma showed that only two were in fact soft-tissue sarcomas (Cases 1 and 4, Table 3).³³ Mortality from soft-tissue sarcomas was increased significantly in the subcohort of 1520 workers with 1 year or more of exposure and at least 20 years of latency (the high-exposure subcohort) (three deaths; SMR, 922; 95 percent confidence interval, 190 to 2695). Two other deaths in the cohort (Cases 5 and 6) were attributed to soft-tissue sarcoma according to hospital records, and one of them (Case 5) was confirmed by review of a tissue specimen. These two deaths did not contribute to mortality due to soft-tissue sarcoma in our life-table analysis, because the deaths were assigned other ICD codes. We are aware of a seventh death from soft-tissue sarcoma, which occurred in a group of 139 workers with chloracne who were excluded from the cohort because they did not meet the entry criteria.

In the cohort, the SMRs for the other cancers of a priori interest were nonsignificantly increased (Table 2). There were no deaths from nasal cancer, although approximately one was expected. In the high-exposure subcohort, the SMRs were nonsignificantly higher for Hodgkin's disease and stomach cancer and lower for non-Hodgkin's lymphoma and cancer of the liver, biliary passages, and gallbladder (Table 2).

A Posteriori Findings

A small but significant increase in mortality due to all cancers combined was observed in the entire cohort (SMR, 115; 95 percent confidence interval, 102 to 130). In the high-exposure subcohort the SMR was 146 (95 percent confidence interval, 121 to 176) (Table 2). At 9 of the 12 plants, mortality from all cancers combined was increased; at one of these plants the increase was statistically significant. Mortality was significantly higher than expected in the category of cancers of unspecified sites, which included those of rare sites not included in a category of the life-table analysis and those for which no primary site was listed on the death certificate. Hospital records, which were obtained for 96 percent of these cancers, revealed no particular clustering according to site.

The cohort had a nonsignificant increase in mortality from cancers of the trachea, bronchus, and lung (ICD code 162; SMR, 111; 95 percent confidence interval, 89 to 137). Mortality from cancers of the respiratory system (ICD codes 160 to 165) was significantly higher than expected in the high-exposure subcohort (SMR, 142; 95 percent confidence interval, 103 to 192) (Table 2). To estimate the effect of smoking on the increase in lung cancer, the expected number of lung cancers was adjusted according to the smoking prevalence found in lifetime histories obtained in 1987 by interviewing 223 workers from two plants.²⁵ This adjustment increased the expected number of lung cancers in the overall cohort by 5 percent and in the high-exposure subcohort by 1 percent, which reduced the SMR in the full cohort to 105 (95 percent confidence interval, 85 to 130) and in the high-exposure subcohort to 137 (95 percent confidence interval, 98 to 187).

Analyses According to Duration of Exposure and Employment

The study cohort worked a mean of 2.7 years in processes involving TCDD contamination and 12.6 years at the plants. The high-exposure subcohort worked a mean of 6.8 years in processes involving TCDD contamination and a mean of 19.2 years in total employment at the plants.

Table 4. Table 4. Mortality from All Cancers and from Cancers of the Trachea, Bronchus, and Lung, According to Latency Period and Duration of Exposure to Processes Involving TCDD Contamination.*

Table 5. Table 5. Mortality from All Cancers and from Cancers of the Trachea, Bronchus, and Lung, According to Latency Period and Duration of Employment at the Study Plants.*

The numbers of deaths due to the rare cancers of a priori interest were too small to permit meaningful analyses according to duration. For all cancers combined and for cancers of the trachea, bronchus, and lung, Table 4 shows the distribution of mortality with increasing duration of exposure to products contaminated with TCDD. The standardized rate ratios were increased in the strata of longer duration for both these categories, but significant linear trends were not found. Mortality increased with increasing latency for both these categories of cancer. Table 5 shows the distribution of mortality for the same categories with increasing duration of employment. Significant linear trends were not observed for either category with increasing length of employment, although standardized rate ratios were higher than expected in several strata of employment ≥20 years. Mortality increased with increasing latency for both categories of cancer.

Serum Levels of TCDD

The mean serum TCDD level, as adjusted for lipids, in the sample of 253 workers from two plants was 233 pg per gram of lipid (range, 2 to 3400) (Fig. 1). A mean level of 7 pg per gram was found in the comparison group of 79 unexposed persons, all of whose levels were under 20, a range found in other unexposed populations.³⁴ The mean for 119 workers with one year or more of exposure was 418 pg per gram. All the workers had received their last occupational exposures 15 to 37 years earlier.

TCDD, widely known as dioxin, has acquired the reputation of a potent carcinogen. Our study, although limited in its ability to detect increased numbers of rare cancers, found little increase in mortality from the cancers associated with TCDD in previous studies in humans. The exception was an increase in soft-tissue sarcoma. The difficulties of evaluating soft-tissue sarcomas in a cohort study of mortality have been described.³³ These include variability in pathological diagnosis and misclassification on death certificates. Consequently, the interpretation of the increased mortality from soft-tissue sarcoma in our study is limited by the small number of cases and the fact that the cause of death was sometimes misclassified on the death certificates of the workers (Table 3) and in the U.S. comparison population.³⁵

Several case–control studies have found significant fourfold increases in non-Hodgkin's lymphoma in persons reporting exposure to phenoxy herbicides or chlorophenols, some of which contained TCDD.^{6 , 8} The magnitude of the increase in mortality in the cohort described here (SMR, 137; 95 percent confidence interval, 66 to 254) suggests a smaller increase in this risk, or no increase at all. Mortality was not significantly higher than expected for other cancers of a priori interest — liver and stomach cancers and Hodgkin's disease. No deaths from nasal cancer were observed. The inconsistency between the results reported here and those of earlier epidemiologic studies is accentuated by the longer and probably greater exposure of this cohort to phenoxy herbicides and chlorophenols contaminated with TCDD.

Mortality from cancers of the trachea, bronchus, and lung was nonsignificantly higher in the cohort. Among the workers with 20 years or more of latency, mortality from respiratory cancer was significantly increased in the high-exposure subcohort, which had 1 year or more of exposure (SMR, 142; 95 percent confidence interval, 103 to 192) but not in the subcohort with less than 1 year of exposure (SMR, 103; 95 percent confidence interval, 62 to 161) (Table 2). SMRs for lung cancer are known to be somewhat higher in blue-collar groups than in the general U.S. population because of more cigarette smoking in the blue-collar groups.³⁶ However, the increased number of lung cancers in the high-exposure subcohort was probably not due to confounding by smoking, for several reasons. First, other diseases related to smoking were not more common than expected in this subcohort; mortality from nonmalignant respiratory disease (ICD codes 470 to 478 and 490 to 519), which is often associated with smoking, was lower than expected (15 deaths; SMR, 96; 95 percent confidence interval, 54 to 158). Second, in the exposed population with 20 years of latency, whose members presumably shared similar smoking habits, the increase was confined to the high-exposure subcohort. Third, on the basis of empirical evidence from other studies, Siemiatycki et al.³⁶ have shown that between a blue-collar population and the general U.S. population, confounding by smoking is unlikely to account for an excess risk of more than 10 to 20 percent. Finally, a limited adjustment in the risk of lung cancer,^{37 , 38} based on the smoking prevalence of surviving workers at only two plants, did not substantially change our results.²⁵ Although confounding by smoking is unlikely to explain the higher rate of respiratory cancer in the high-exposure subcohort, it remains possible that the increase was due to confounding by occupational exposures other than TCDD. For example, asbestos may have contributed to mortality from lung cancer in the cohort, since two deaths were due to mesotheliomas.

An unexpected finding was the small but significant increase in mortality from all cancers combined. The observed increase is consistent with a carcinogenic effect of TCDD. For all cancers combined, mortality was significantly higher than expected in the entire cohort, more pronounced in the high-exposure subcohort, and increased at 9 of 12 plants. With mortality from cancers of the trachea, bronchus, and lung excluded, mortality from all remaining cancers combined was still higher than expected in the overall cohort (SMR, 117; 95 percent confidence interval, 100 to 136) and in the high-exposure subcohort (SMR, 150; 95 percent confidence interval, 118 to 189). Consequently, the increased risk for all cancers combined is not explained by smoking or by increased mortality due to cancer of the trachea, bronchus, and lung. The generation of tumors in a number of organs in animals exposed to TCDD^{12 , 13} and the demonstration that TCDD promoted tumors in two organs^{21 , 22} make it biologically plausible that TCDD may produce tumors in more than one organ in humans. Moreover, a significantly increased SMR for all cancers combined is unusual in occupational studies of chemical workers. Results similar to ours were observed in a study of German workers exposed to TCDD after a 2,4,5-trichlorophenol reactor accident in 1953. A subgroup of workers with chloracne (used as a surrogate for exposure) and at least 20 years of latency had an SMR of 201 (90 percent confidence interval, 122 to 315) for all cancers combined, based on 14 deaths.³⁹ This is the only other industrial cohort with both substantial exposure to TCDD and a long period of latency during which mortality was examined. Workers from U.S. production cohorts described in previous studies were included in the current study if they met our entry criteria.^{40 41 42}

Two observations argue against a carcinogenic effect of TCDD. First, there was not a significant linear trend of increasing mortality with increasing duration of exposure to products contaminated with TCDD (Table 4). However, our use of duration of exposure may have misclassified the cumulative dose of some workers. In addition, a dose–response relation is generally viewed as strong evidence for an association when it is present, but as fairly weak evidence against an association when it is absent.⁴³ Second, our study did not directly assess the effect of exposure to TCDD alone. The workers were exposed concurrently to the chlorophenols and phenoxy herbicides that were contaminated with TCDD. In addition, they may have been exposed to numerous other chemicals while employed at the plants.

Because the exposure of our cohort was substantially higher than that of most nonoccupational populations, the estimates of effect in this study may provide an upper level of risk to be anticipated in humans. For several types of cancer previously associated with TCDD, we found no increases above expected levels. Soft-tissue sarcoma was an exception; a ninefold increase was found among workers who were exposed for 1 year or more and who had at least 20 years of latency. Interpretation of the increased SMR is limited, however, by the small number of cases and because this cause of death was sometimes misclassified on the death certificates of the workers and in the national comparison population. Continued surveillance of the cohort may provide a firmer estimate of risk.

Mortality from all cancers combined was 15 percent higher than expected in the overall cohort. The subcohort with 1 year or more of exposure and 20 years or more of latency had a 46 percent increase in all cancers combined and a 42 percent increase in cancers of the respiratory tract. Although the study could not completely exclude the possible contribution of other occupational carcinogens or smoking, the increased mortality, especially in the subcohort with one year or more of exposure, is consistent with the status of TCDD as a carcinogen.