Leukemia risk associated with benzene exposure in the Pliofilm cohort.

A reanalysis of the Pliofilm cohort was conducted incorporating six additional years of follow-up information gathered by the National Institute of Occupational Safety and Health (NIOSH) and a new set of exposure estimates developed recently. The distribution of individual worker exposures calculated with the Paustenbach exposure estimates was compared to those derived using two earlier sets of job-, plant-, and year-specific exposure estimates. A traditional standardized mortality ratio analysis and the Cox proportional hazards model were used to investigate the impact of these exposure estimates and the NIOSH updated information on evaluation of benzene's leukemogenicity. There were no additional cases of multiple myeloma or any indication of increased incidences of solid tumors. The data added in the update did not greatly modify the estimated relative risk of all leukemias associated with benzene exposure but confirmed previous findings that occupational exposure only to very high concentrations had leukemogenic potential. Leukemia has not been observed in anyone who began employment in Pliofilm production after 1950. Neither the Paustenbach nor the Crump exposures gave dose-response estimates as steep as that resulting from the Rinsky exposures.


Introduction
The Pliofilm rubberworker cohort has been central to setting health-based standards for benzene, such as the U.S. Environmental Protection Agency's (U.S. EPA) cancer potency factor, the Occupational Safety and Health Administration's (OSHA) permissible exposure limit (PEL), and the American Conference of Governmental Industrial Hygienists' (ACGIH) proposed threshold limit values (TLVs). The reanalysis summarized here from more detailed accounts (1,2) was motivated by the availability of new information on two aspects of the cohort's experience. The National Institute of Occupational Safety and Health (NIOSH) had gathered six additional years of observation on the vital status of the more highly exposed portion of the cohort. The two existing sets of job-, plant-, and year-specific exposure estimates were supplemented by a third.
The detailed work histories of the individual workers composing the Pliofilm cohort represent a unique resource for estimating the dose response for leukemia that may follow occupational exposure to benzene. Detailed records of specific jobs performed at particular times within the Pliofilm plants were maintained for all workers. Although industrial hygiene data were gathered only intermittently, this information represents a better basis for estimation of individuals' exposures than is available for most industrial situations. Rinsky et al. (unpublished data) and Crump and Allen (unpublished data) derived sets of exposure estimates corresponding to specific jobs performed at the Akron and St. Marys, Ohio, plants over the years of Pliofilm production. Paustenbach et al. (3) developed a third and more refined set of exposure estimates by performing a detailed historical reconstruction of the industrial practices at the Pliofilm plants. The cumulative exposure estimates derived by applying the new exposure estimates to the individual workers' job histories are compared here with those obtained using previously reported exposure estimates (Rinsky et al., unpublished data; Crump and Allen, unpublished data).
The results of NIOSH's monitoring the vital status of workers from these Goodyear plants through 1981 were published in 1987 (4). NIOSH extended its follow-up of this cohort for an additional six years from 31 December 1981 to 31 December 1987. With the last "first exposure" in 1965 and follow-up through 1987, at least 22 years have elapsed since the most recent possible "first exposure" of a worker; a sufficient latency period has passed for assessment of leukemogenicity. The impact of this update on mortality risk, especially from leukemia, following occupational exposure to benzene is presented in the form of results of standardized mortality ratios (SMR) and dose-response analyses. The proportional hazards model was used in the dose-response analysis to extend the conditional logistic analysis by Rinsky et al. (4) of the then-identified nine leukemia cases each matched to ten controls.

Standardized Mortality Ratios in Updated Cohort
The Pliofilm cohort consists of 1868 nonsalaried individuals employed by Goodyear in the production of Pliofilm (Table 1 Table 2 delineates the changes in SMRs between the two data-gathering periods for various causes of death. The SMR for overall mortality was approximately 1 in both periods. Previously, benzene exposure was associated with an increase in nonmalignant blood disorders in this cohort; such end points are not likely to have lengthy latencies, and no new cases were reported in the 1987 update. For all types of cancer, the SMR remained nonsignificant. There were no new deaths from multiple myeloma in the 1987 update, and Wong (6) has determined that the updated SMR for multiple myeloma of 2.91, with a 95% confidence interval (CI) of 0.79 to 7.45, is no longer significant. The addition of five new deaths from leukemia and one from lymphoma left the SMR for all lymphatic and hematopoietic cancers in males little altered and still significant. For other types of cancer combined, the SMR remained less than 1. Animal studies (7) have suggested that benzene might produce solid tumors in humans, but no individual type of solid tumor showed a significant increase.
Descriptions of the old and new cases of lymphatic and hematopoietic cancer are presented in Table 3, including the first female (case 21) in the Pliofilm cohort with this form of cancer (specifically, an acute myeloid leukemia). As before, the new leukemia cases represented a variety of histologic types; unfortunately, the exact type was not specified for two of the new leukemias and this information is not readily retrievable. Note that all of the cases had held jobs exposing them to benzene by 1950. Table 4 compares the SMRs for leukemia from the old and new updates by plant location. For the analysis through 1981, the SMR for the overall cohort was significantly elevated, but the SMR for St. Marys alone was not. After the update through 1987, the addition of five new male leukemia cases increased the SMR for both locations combined, from 3.15 to 3.60. The addition of two more cases in Akron had little impact on the SMR, but the addition of three cases in St. Marys increased the SMR to 3.36, making it statistically significant (p < 0.05). Table 5, which gives the SMRs according to year of first Pliofilm job, shows no leukemia cases among the 299 workers who started in 1951 or later. The leukemia risk was elevated, but nonsignificant for those starting Pliofilm work between 1946 and 1950. Leukemia risk was significantly elevated only for those who started work prior to the end of World War II. In that benzene exposures in the Pliofilm process were markedly and progressively reduced after World War II, these observations are consistent with a threshold mechanism for leukemogenesis induced by benzene but might also be attributed to the low statistical power of the epidemiology data for this exposure group.

Individual Worker Exposures from Three Sets of Exposure Estimates
For each worker, estimates of cumulative exposure in ppm-years were generated using three sets of exposure estimates corresponding to the occupational codes in the individual work histories. The original matrix of exposure estimates (the Rinsky exposure estimates) was presented in a 1985 draft (Rinsky et al., unpublished data) and used again but not presented in 1987 (4). Crump and Allen (unpublished data) developed another set of estimates (the Crump exposure estimates) based on the concept that the benzene levels in the workplace probably improved in parallel with progressively more restrictive standards for occupational exposure. Using more detailed information about the monitoring devices used at the plants, the varying     Figure 1 presents the cumulative distribution functions of individual cumulative exposure for all male workers at St. Marys and Akron under each of the three sets of exposure estimates. The further to the right a curve lies relative to the other curves in Figure 1, the higher the overall estimated exposure. More than 50% of the individuals have estimated cumulative exposures of less than 50 ppm-years using all three exposure matrices. The upper tails of these highly skewed distributions extend orders of magnitude above their medians, resulting in arithmetic averages ( Figure 2) that are about ten times higher than their medians.
The estimated values for the leukemia cases are superimposed upon each curve in Figure 1. For St. Marys, the estimates for the six leukemia cases are spread quite evenly over the range of cumulative exposure estimates, but for Akron, the estimates for almost all the eight leukemia cases are clustered at the upper extreme of the exposure distributions. As indicated under "cause of death" in Table 3  ._ m at risk in the categories: 0 to 5, 5 to 50, 50 Cases Controls to 500, and more than 500 ppm-years Akron ( Table 6). These findings suggest a strong dose-response relationship of risk increastive exposures of ing with cumulative exposure no matter which set of exposure estimates is used. For none of the three sets of exposure estimates, iology of the however, is there a statistically significant ys.
increase in the SMRs for cumulative expocumulative sures less than 50 ppm-years. This is rn noted by consistent with the hypothesis that expoe job-specific sure in excess of some threshold value (here or Akron are suggested to be greater than 50 ppm-years) or St. Marys; is necessary for leukemogenesis. This es are higher threshold hypothesis is further supported mates; and c) by work at the mechanistic level (8-11). s both exceed [In a similar analysis of cumulative expoee cumulative sure partitioning these leukemia cases into cular individacute myelocytic or acute monocytic lities a and b. leukemias (AMMLs) versus non-AMMLs, imp exposure Crump (12) showed that the dosevalue for the response relationship is driven by AMMLs )ach exposure with a suggested threshold in excess of 400 ver, the mean ppm-years and that the frequency of the other leukemias is not associated with benzene exposure.]

Risks Estimated by Proportional Hazards Dose-Response Model
Statistical analysis with the proportional hazards dose-response model (13) relates exposure to the incidence of leukemia while controlling for demographic, socioeconomic, and other variables that might confound interpretation of the data. With this same objective, Rinsky et al. (4) had used a conditional logistic regression analysis on a subset of the Pliofilm data consisting of the nine leukemia cases (all white males), each of whom was matched on the basis of sex, race, date of birth, and date of first employment in a Pliofilm job to ten controls who were alive at the time their corresponding case died. Because it accepts all controls falling within a defined stratum (Table 7), the proportional hazards analysis includes more workers than would the conditional logistic's matching process. It does not require the arbitrary assignment of each control to one of several cases sharing similar profiles or the utilization of very poorly matched controls when the fixed number of appropriate individuals is not available, as does the conditional logistic analysis.
The matching criteria of Rinsky et al. (4) were used to define the strata in the proportional hazards model (i.e., sex, race, date of birth, and date of first Pliofilm job). The possibility that the plants in Akron and St. Marys might have differed in the leukemogenic response they induced because of factors other than benzene exposure levels was controlled for by using location as an additional matching criterion. Although the vital status of the dryside workers had been followed up only through 1981, they Environmental Health Perspectives -Vol 104, Supplement 6 * December 1996    (4) had been applied to the 15 leukemia cases found through 1987.
The estimates of the slope parameter in this dose-response model before and after the 1987 update under the three sets of exposure estimates are presented in Table 8. For either update period, the slope of dose-response relationship was estimated to be about 20 to 40% as steep using the Crump or the Paustenbach estimates of exposure as when the Rinsky exposure estimates were used. Updating the cohort's leukemia status through 1987 resulted in a lowering of the estimated slope parameters under all three sets of exposure estimates.
The predictions of additional leukemia deaths and associated 95% confidence intervals are plotted in Figure 3. Risks for occupational exposure are given for working lifetimes (8 hr/day, 5 days/week, 50 weeks/year for 45 years) for continuous exposure at 0.1, 0.3, 1, 5, or 10 ppm. In the current analysis, using the preferred exposure estimates of either Crump or Paustenbach predicts about 0.3 to 0.5 additional leukemia deaths per thousand workers in the concentration range around 1 ppm. Overall, the results of this update and reanalysis suggest that the previous estimate of 5.1 additional leukemia deaths for that exposure scenario [(16), based on the approach of Rinsky et al. (4)], is about an order of magnitude too high. More detailed investigation of the dose-response relationships in the Pliofilm cohort using all the workers and focusing on the most pertinent end point, AMLs only, may be found in the sensitivity analysis of Crump (12) summarized at this conference.

Conclusion
The additional information assembled in updating the Pliofilm cohort strengthens but does not greatly alter the thrust of the data through 1981 concerning leukemia risk in this occupational setting (4,16). The absence of any additional cases of multiple myeloma in the update through 1987 weakens to nonsignificance the previous statistical association of this end point with benzene exposure. There continues to be no evidence of any increase in the incidence of any type of solid tumors. The more rigorously defined exposure estimates derived by Paustenbach et al. (3) are consistent with those of Crump and Allen (unpublished data) in giving estimates of the slope of the leukemogenic dose response that are not as steep as the slope resulting from the exposure estimates of Rinsky et al.
(unpublished data). The newly gathered information continues to be consistent with a threshold model for leukemogenesis by benzene. The leukemia deaths in the entire cohort occurred exclusively among individuals who commenced their work in Pliofilm production in 1950 or earlier. The simplest explanation would be that industrial hygiene improved over the years at both Akron and St. Marys and achieved a critical level of reduced benzene exposure in the early 1950s, so that workers entering the workplace after that time were no longer at risk for developing leukemia.