Mortality among PVC-fabricating employees.

The results of a cross-sectional mortality study of 3847 deaths occurring among current and former (white) employees of 17 PVC fabricators during 1964-1973 are presented. Sex-race-cause-specific proportionate mortality ratios (PMR's) were computed by using two separate standards: one, the U.S. mortality in 1968; the second, U.S. mortality for the individual years 1964-1973. In addition, a case-control analysis, based upon 44 breast cancer deaths among white female employees, is presented. PMR's are significantly different from unity for all cancers, and for cancers of the digestive system among both white males and white females. Although observed deaths significantly exceeded expectations for cancer of the breast, a subsequent case-control analysis reveals no statistically significant relative risks for breast cancer.


Introduction
In March, 1974, Organization Resources Counselors (ORC) was requested by representatives of PVC producers who are members of ORC's Occupational Safety and Health Standards Group, to carry out a study of health risks to employees working in the PVC fabricating industry. After considering a number of alternative study designs, it was decided that a cross-sectional mortality study would best meet the need for providing information as rapidly as possible. Details regarding reasons for selecting this type of study, the actual study design and results of the inquiry have been reported previously (1).
In the original report, proportionate mortality ratios were calculated according to the method of Guralnick (2) and by using the sex-cause-age specific distribution of deaths among U.S. whites in 1968 as the basis for determining expectation (3). For the current report, we have analyzed the data using the Mantel-Haenszel procedure both for determination of PMR's and in order to test whether PMR's differ significantly from unity (4,5). In addition, PMR's are presented using two separate standards -as the basis for expectation. A follow-up to the cross-sectional mortality investigation, utilizing deaths among white women for whom the underlying cause of death was cancer of the breast, was undertaken (6), and results of the case-control analysis will be presented.

Materials and Methods
Although alternative study designs were considered, several factors led to the decision to carry out a cross-sectional mortality study. First, the primary study objective was to determine relatively quickly whether any angiosarcoma deaths could be identified among the study group and this was best accomplished by examining causes of death among relatively recent decedents. Second, it would not have been possible to identify clearly and completely (if at all) the cohort of workers necessary for a historical cohort mortality study within a reasonably short period of time. As a result, the study was based upon 4336 deaths which occurred during the years 1964-1973 among active or retired employees of 17 companies engaged in PVC fabrication (Table  1). This report will be restricted to the 3847 deaths among whites, since the number of deaths for most causes among nonwhites was quite small, making interpretation of cause specific mortality difficult at best.
Since the population at risk could not be deter-October 1981 The Mantel-Haenszel procedure for assessing the statistical significance of a difference between the observed and expected numbers of deaths for a specific cause requires the construction of a series of 2 x 2 contingency tables as shown in Table 2. Separate tables are constructed for each level of the factor(s) to be adjusted for. If, for example, we wish to compute a PMR for all decedents for cause X, adjusted for age (five age groups) and year of death (10 years: 1964-1973), i.e., 1964-1973 standard, we would have 50 2 x 2 tables: age < 35, year of death 1964; age < 35, year of death 1965; . . .; age < 65, year of death 1973. Observed deaths and expected deaths are summed over the 50 separate 2 x 2 tables to produce a PMR for cause X, adjusted for age and year of death. The resulting age-year of death-adjusted PMR compares the number of deaths from cause X observed in the study group to the number of deaths from cause X expected in the study group, if the proportion of total deaths The case-control analysis is based on 44 deaths among white females in the study with cancer of the breast as the underlying cause of death. Controls were drawn from deaths due to diseases of the circulatory system and accidents, and matched by age plus or minus five years and company, if possible. Forty of the cases were matched on both criteria and four on the basis of age alone. The basis for the case-control analysis, then, is the 44 breast cancer deaths and 134 matched controls distributed among eight of the 17 companies.

Proportionate Mortality Ratios
Cause specific PMR's adjusted for age are presented for white male and white female employees in Tables 3 and 4. For Table 3, expected numbers of deaths are based on the cause-sex-age specific distribution of deaths among U.S. whites applied to the study group. For Table 4, PMR's are adjusted for year of death as well as race, sex and age. Statistically significant departures from unity (i.e., PMR = 1) are evaluated by using the Mantel-Haenszel continuity corrected chi-square with one degree of freedom. Causes of death for which the calculated chi-square indicated a statistically significant excess or deficit at the a = 0.05 level of significance are indicated in the table. It is important to remember that when carrying out a large number of significance tests, some number will turn out to be significant on the basis of chance alone. When dealing with a large number of tests for PMR's, it may be more appropriate to consider the Mantel-Haenszel test as a screening device rather Environmental Health Perspectives than giving it a strict probabilistic interpretation. With these caveats, we see that based upon the 1968 standard, among white males there are statistically significant PMR's for all cancers, digestive cancer, respiratory cancer and other and unspecified cancers ( Table 3). Among white females, PMR's are significantly different from one for all cancers, digestive cancers, breast cancer, urinary cancer and other and unspecified cancers. In contrast to white males, the PMR for urinary cancer among white females is high and, although based upon only 11 deaths, is statistically significant. On the other hand, in contrast to the observation in white men, the PMR for respiratory cancer among white women is very close to unity. Similar to the observation in white men, mortality from cirrhosis of the liver is low, but the corresponding number of observed deaths is quite small. Among both white male and female employees, diseases of the circulatory system account for a large percentage of total deaths. In each case, observed numbers of deaths are close to expected,

Case-Control Analysis
Comparison of cases and controls on a variety of variables where information was available, including length of employment, ever married versus never married, and child-bearing history, reveals no statistically significant differences (p > 0.05) between cases and controls for any of these variables.
Review of employment histories for cases and controls revealed a wide variety of jobs ranging from office and clerical work to production jobs such as bench inspector, press operator, trimmer, assembler and sweeper. The wide range of job content and location made it impossible to determine precisely whether there was PVC exposure in every case, or the precise length of that exposure. Therefore, a subjective ranking system was developed to classify exposures. Work histories were reviewed with plant personnel and PVC exposure potential was categorized into five classes-no exposure, improbable exposure, possible exposure, definite exposure and unknown exposure. This classification scheme enabled some definitive exposure statement in 80% of the cases and 92% of the controls (Table  5).
After matching by age and company, 38 matched sets of cases and controls were developed from the 44 cases and 134 controls. There are fewer matched sets than cases because, in six instances, it was October 1981 necessary to combine cases of similar age within the same set in order to have at least one control per matched set. The Mantel-Haenszel procedure was used to derive a summary estimate of relative risk and to test for significant departures from unity (4,7). In this procedure, each of the 38 sets can be viewed as a 2 x 2 contingency table. Thus, the ith set can be represented as in Table 6.
Relative risk is defined as the ratio of the probability of dying from cancer of the breast among women exposed to PVC to the probability for women not exposed. Estimates of these individual probabilities are not available from a case-control study. However, a measure of estimated relative risk from case-control studies as suggested by Mantel and Haenszel has been calculated as in Eq. (2). To assess whether the departure from unity of an observed relative risk is too great to have occurred by chance alone, a summary chi-square test corrected for continuity was performed using the Mantel-Haenszel procedure. The calculated chisquare must be 3.84 or larger in order to conclude with 95% assurance that the observed relative risk did not differ from unity by chance alone. R = Y,i (AiDiTi)li (BiCilTi) (2) Combining the five exposure categories into two may be accomplished in a variety of ways, resulting in several possible relative risk measures as shown in Table 7.
None of the calculated relative risks, including the second shown which treats anything other than no exposure as definitely exposed, are statistically T, significant; i.e., they may have occurred by chance alone. Similar analyses were carried out on a company-by-company basis. None of the relative risks so calculated is significantly different from unity.

Discussion and Summary
The cross-sectional mortality study was designed with two objectives. The first was to determine if any angiosarcoma deaths had occurred among employees of the PVC fabricators under study. Since no angiosarcoma deaths were found among the employees studied, the first question has an unequivocal answer. A secondary objective was that of examining the distribution of deaths by cause among the employees under study. Implicit in that objective is the question of whether or not that distribution is, in some sense, unusual. There is no unequivocal answer to the latter question. Whether or not an observed distribution of causes of death is unusual clearly relates to the standard or comparison population as well as the analytic methodology (8). This is illustrated by the observation that there are PMR's which are significantly different from unity on the 1968 standard and not on the [1964][1965][1966][1967][1968][1969][1970][1971][1972][1973] standard. However, it seems much more important to focus on the large area of agreement between the two standards rather than on the few areas of disagreement.
On the basis of the PMR analyses, there are statistically significant excesses in total cancer mortality among both white males and white females when compared to the distribution of deaths for the total United States specific for color and sex and adjusted for age. Excesses in cancer mortality appear concentrated in cancers of the digestive system and, in particular, in cancers of the intestine for both men and women. In addition, there is a suggestion that mortality from cancer of the breast and urinary organs among white women employees is higher than that for the total U.S. There are, however, several reasons why definitive interpretation is difficult. Factors meant to suggest that proportionate mortality analyses must be interpreted cautiously have been reviewed previously (9)(10)(11)(12)(13). However, while these results must be interpreted with caution, they appear to be consistent with previously studied workers and suggest the need for some continuing investigation (1).
One such follow-up investigation has been presented here in the form of a case-control study involving the 44 breast cancer deaths. On the basis of a case-control analysis, estimates of relative risk were derived but none of these relative risk estimates is significantly different from one, although such results must be interpreted with caution.
Absence of a statistically significant relative risk does not demonstrate that there is not an excess risk of death from breast cancer among women employees with PVC exposure. In fact, when no statistically significant relative risks are found, it is pertinent to ask what the chances are of detecting an increase of a given magnitude from the available data. Using the method described by Walter (14), we have subjected each of the relative risk estimates (Table 7) to a least significant relative risk analysis under the conditions that we desire 95% assurance that a risk of such magnitude, if observed, did not occur by chance alone and 80% probability of detecting the least significant relative risk if it exists.
Even in the case where all but no exposure are counted as exposed, the smallest relative risk which could be detected from these data is nearly 3:1. Nearly 200 cases and 200 controls would have been necessary to detect a true doubling of the risk under the specified conditions. Given the sample size in this study and the percentage of controls exposed (a percentage which was unknown at the aNone of the relative risk estimates are statistically significantly different from unity (p > 0.05). bThe true relative risk would have to be at least this large to have an 80% assurance of detection (i.e., power = 0.80) with a type I error of 0.05 (i.e., a = 0.05).
Environmental Health Perspectives 142 start of the study), it would be possible to detect only very large increases in risk. There seems reasonable assurance, therefore, that such very large increases in the risk for breast cancer do not exist among these PVC fabricators, but no such statement can be made for possibly smaller increases in risk.