Document RJ6vVoMaJ738BVVLVaMRKxwLv

MstteJ CL3B2; N\. or .es k. ELSEVIER Available online at www.sdencedirQctcom ScienceDirect Chemico-Bioiogical Interactions xxx (2006) xxx-xxx Chem ico-Bioiogical interaction/ w w w .clse vicr.co m /lo ca te /ch cm b io in t Mortality patterns among industrial workers exposed to chloroprene and other substances II. Mortality in relation to exposure Gary M. M arsha'*, Ada O. Youka, Jeanine M. Buchanich3, Michael Cunningham3, Nurtan A. Esm enb, Thomas A. H allc, Margaret L. Phillips c ' Department o f Biostatistics, Graduate School of Public Health, University o f Pittsburgh, 130 DeSoto Street, Pittsburgh, PA 15261, USA b Occupational and Environmental Health Sciences, School o f Public Health, University of Illinois at Chicago, 2121 West Taylor Street, Chicago, IL 60622, USA c University o f Oklahoma Health Sciences Center, 801 NE 13th Street, CHB 413, Oklahoma City, OK 73190, USA Abstract As part of an historical cohort study to investigate the mortality experience of industrial workers exposed to chloroprene (CD) and other substances, including vinyl chloride monomer (VC), we analyzed mortality from all cancers combined, respiratory system (RSC) and liver cancer in relation to CD and VC exposures. Subjects were 12,430 workers ever employed at one of two U.S. sites (Louisville, KY (ra= 5507) and Pontchartrain, LA (n = 1357)) or two European sites (Maydown, Northern Ireland (w= 4849) and Grenoble, France (n = 717)). Historical exposures for individual workers were estimated quantitatively for CD and VC. For sites L, M, P and G, respectively, average intensity of CD exposures (median value of exposed workers in ppm) were 5.23, 0.16, 0.028 and 0.149 and median cumulative exposures (ppm years) were 18.35, 0.084, 0.133 and 1.01. For sites L and M, respectively, average intensity of VC exposures (median value of exposed workers in ppm) was 1.54 and 0.03 and median cumulative exposures (ppm years) were 1.54 and 0.094. We performed relative risk (RR) regression modeling to investigate the dependence of the internal cohort rates for all cancers combined, RSC and liver cancer on combinations of the categorical CD or VC exposure measures with adj ustment for potential confounding factors. We categorized exposure measures into approximate quartiles based on the distribution of deaths from all cancers combined. We also considered 5- and 15-year lagged exposure measures and adjusted some RR models for worker pay type (white/blue collar) as a rough surrogate for lifetime smoking history. All modeling was site-specific to account for exposure heterogeneity. We also computed exposure category-specific standardized mortality ratios (SMRs) to assess absolute mortality rates. With the exception of a one statistically significant association with duration of exposure to CD and all cancers combined in plant M, we observed no evidence of a positive association with all cancers, RSC or liver cancer and exposure to CD and/or VC using both the unlagged and lagged exposure measures: duration, average intensity or cumulative exposure to CD or VC; time since first CD or VC exposure; and duration of CD exposure or time since first CD exposure in presence or absence of VC exposure. We observed elevated and statistically significantly elevated RRs for some analysis subgroups, but these were due to inordinately low death rates in the baseline categories. With the possible exception of all cancer mortality in plant G, our additional adjustment of RRs for pay type revealed no evidence of positive confounding by smoking. We conclude that exposures to CD or VC at the levels encountered in the four study sites do not elevate mortality risks from all cancers, RSC or liver cancer. This conclusion is corroborated by our analysis of general mortality patterns among the CD cohort * Corresponding author. Tel.: +1 412 624 3032; fax: +1 412 624 9969. E-mail address: gm arsh@ pitt.edu (G.M. Marsh). 0009-2797/$ - see front matter 2006 Elsevier Ireland Ltd. All rights reserved, doi: 10.1016/j .cbi.2006.08.012 Please cite this article os: Gary M. Marsh et aL Mortality patterns among industrial workers exposed to ehlomptene astd other substances. Chcrnieo-ilioiogicui Internelions (2006), doi: 10.1016/j .ehi.2006.08.(112 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00001  Mottet CBTBS: No. oi' I'ajses K> 2 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx reported in our companion paper [G. Marsh, A. Youk, J. Buchanich, M. Cunningham, N. Esmen, T. Hall, M. Phillips, Mortality patterns among industrial workers exposed to chloroprene and other substances. I. General mortality patterns, Chem.-Biol. Interact., submitted for publication ]. 2006 Elsevier Ireland Ltd. All rights reserved. Keywords: Chloroprene; Cohort study; Liver cancer; Lung cancer; M ortality; Vinyl chloride; Exposure-response 1. Introduction Chloroprene (2-chloro-l,3 butadiene) (CD) is a monomer used almost exclusively for the production of synthetic rubber and latexes [1], The chemical structure of CD is similar to that of vinyl chloride, a known human carcinogen [2]. CD is classified by the International Agency for Research on Cancer as a possible human carcinogen (group 2B) based on sufficient evidence of carcinogenicity in experimental animals [3]. Epidemio logical data on the carcinogenicity of CD are available from five cohort studies of chloroprene production work ers in the U.S. [4], China [5], Armenia [6] and France [7], and the study of shoe manufacturing workers in Russia [8]. An increased risk of liver [5,6,8,9] and lung cancer [7] has been suggested by some of these studies. The inherent methodological limitations in the previous epi demiology studies raise questions; however, about their significance regarding human cancer risks [10,11]. To provide more definitive and comprehensive epi demiological evidence regarding the long-term health effects of exposure to CD, a four-plant, multi-national epidemiologic study of workers with potential exposure to CD was commissioned in 1999 by the International Institute of Synthetic Rubber Producers (IISRP). The exposure assessment component of the study was con ducted at the University of Oklahoma (UOk) and the University of Illinois at Chicago (UIC); the epidemiol ogy and biostatistics component was conducted at the University of Pittsburgh (UPitt). Our analysis of gen eral mortality patterns among the CD cohort, reported in a companion paper [12], found no evidence of ele vated mortality risks from any of the causes of death examined, including all cancers combined and lung and liver cancer, the sites of a priori interest. We report here the results of our detailed analysis of mortality from all cancers combined, lung and liver cancer in relation to quantitative measures of CD and VC exposure. 2. Methods 2.1. Study sites and subjects The chloroprene (CD) cohort included all workers (.n = 12,430) with potential CD exposure at one of four CD production sites from plant start-up date through the end of 2000 (1999 for one site). The sites include two DuPont/Dow Elastomers LLC (DDE) plants in the U.S. (Louisville, KY and Pontchartrain, LA), one DDE plant in Maydown, Northern Ireland (NI) and one Enichem Elastomers France plant in Grenoble, France (FR) (called here plants L, P, M and G). CD production dates for each plant were: L (1942-1972), P(1969-date), M (1960-1998) and G (1966-date). In two plants (L and M), CD production included an acetylene-based process that produced vinyl chloride (VC) as a by-product. Plant L made CD only through the acetylene process that was phased out in between 1971 and 1976; plant M made CD by the acetylene process from 1960 to 1980 then only by the butadiene process from 1980 to 1998. Plants P and G used only the butadiene process to produce CD. The newer butadiene process did not involve VC exposures and resulted in lower CD exposures for jobs related to monomer production than those associated with the early production years of the older plants L and M. Details of the CD cohort and the history, processes and chemical exposures associated with each study plant are described elsewhere [12-16]. 2.2. Exposure estimation 2.2.1. Chemical process-based exposure reconstruction Historical individual worker exposure profiles were estimated using a chemical process-based exposure reconstruction approach detailed elsewhere by Esmen et al. [13-15]. In brief, the exposure reconstruction was based on mathematical models which utilized exposure models based on the physics and chemistry associated with a given chemical process as determined from pro cess documentation and task performance habits gleaned from interviews with knowledgeable plant personnel. The mathematical models used were based on the dis persion of the contaminant in the breathing zone of the worker performing the task associated with the exposure of interest. The simplest models scaled the contaminant vapor pressure by task execution time; more sophisti cated models considered the contamination generation and dispersion rates. To the extent possible, all mathe- Plcasc die this article as; Gary M. Marsh et al,, Mortality paneras among industrial workers exposed to chloroprene and other substances. CfoemictvBiological Interaetions (2006). dot: 0.i016/j .chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00002  Mixte! CB-SB: NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx 3 Table 1 Chloroprene (CD) and vinyl chloride (VC) exposure level classes Level CD (ppm) VC (ppm) Range Nominal Range Nominal 0 N< 0.0005 0 V cO .O l 0 1 0.0005-0.005 0.0016 0.01-0.1 0.03 2 0.005-0.05 0.016 0.1-1 0.3 3 0.05-0.5 0.16 1-10 3 4 0.5-5 1.6 10+ 16 5 5-50 16 6 50-100 l 7 100+ 160 N: negligible exposure. matical models were validated with existing air moni toring data available for the later years (1975-1992). In comparison to the available exposure measurement data, the estimated exposure levels could have been obtained for many job categories using only the existing exposure measures leading to the same results obtained through modehng. This suggested that the exposure assignment in job categories with sparse or missing exposure mea surement data was satisfactory. 2.2.2. Estimated exposures to chloroprene and vinyl chloride In each plant, CD and VC exposures were modeled for all unique job title classes using six exposure classes for CD and four exposure classes for VC (Table 1). The width of the exposure classes (one order of magnitude) was calculated to minimize exposure misclassification on the basis of the specificity available in the job dic tionary [13-15]. The nominal values (geometric mean of the associated class limits) of each agent were used as the daily average intensity value for a specific job with the associated exposure level. Exposures were esti mated for the entire period of CD production in each plant. Although the four plants varied considerably with respect to the mix of production methods, CD expo sures were remarkably similar in both estimated and measured values. CD exposures were found to be much more dependent on the improvement of the production methods, rather than deliberate reduction in exposures for occupational hygiene considerations. CD exposures were generally lower than the contemporaneous expo sure limits or guidehnes. Specifically, average CD expo sures were less than 20 ppm in the pre-1960 era, less than 10 ppm in the 1960-1980 era, less than 1ppm in the 1980-1990 era and less than 0.5 ppm thereafter. VC exposures, which occurred only in monomer production, were estimated to be relatively high in the pre-1960 era of CD production, but the highest exposures for the CD monomer operator job class did not exceed 2 ppm. For the same job class, VC exposures in the 1960-1970 era were less than 0.5 ppm. 2.2.3. Construction o f worker summary exposure measures For each plant, the job title classes and correspond ing time-specific exposure class estimates for CD and VC were linked to the detailed subject work histories held by UPitt to enable the construction of working life time exposure profiles for each subject. These individual subject exposure profiles were used to compute three summary measures of CD and/or VC exposure for each subject as follows: duration of exposure (Dur) = the sum of the days spent in jobs with nonzero exposure to CD or VC (in years); cumulative exposure (Cum) = the prod uct of the number of days in each job and the estimated average daily exposure to CD or VC (in ppm years); and average intensity of exposure (AIE) = the ratio of Cum to Dur (in ppm). The notation used to describe these sum mary measures is Agent-Measure, for example, CDvME refers to the average intensity of CD exposure. We also considered a latent time-related exposure measure, time since first exposure to CD (or VC), expressed as Agent_TSFExp and constructed two com posite summary measures, CD exposure in the presence of VC exposure (CDwVC Measure) and CD exposure in the absence of VC exposure (CDwoVC_Measure). With the former composite measure, CD exposures are com puted only for those jobs with concomitant VC exposure. Alternative characterizations of the CD and VC exposure measures described above were also computed using a lag period as described in detail by Youk et al. [17]. Sim ply put, the lag period refers to the fixed period of time before the time of observation during which exposures are not counted. Thus, with a 5-year lag period, expo sures received up to five years before a given observation time are given zero weight, and exposures received five or more years before observation time are given full weight. Fagging attempts to characterize only the most etiologically relevant exposures and is particularly rele vant for examining diseases, such as cancer, associated with long latent periods. We considered both a 5 and 15 years lag in our analysis of respiratory system cancer and hver cancer. 2.2.4. Worker exposures to chloroprene and vinyl chloride Table 2 shows for each plant the distribution of sub jects exposed to CD and/or VC. More than 92% of the workers at each plant were exposed to CD, with Please die this article as: Gary M. Marsh et aL Mortality patterns among industrio] workers exposed to ehlaroprenc and other! substances. CfoemictvBiological Itrieraetkms (2006k dot: 10.10 16/j.chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00003  MixfcJ C-3332; No. oi' 'ayes 16 4 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx Table 2 Distribution of workers exposed to chloroprene and vinyl chloride Plant Number exposed (%) CD VC Louisville Maydown Pontchartrain Grenoble All plants 5468 (99.3) 4474 (92.3) 1258 (92.7) 717(100.0) 11919 (95.9) 1250 (22.7) 265 (5.5) 0(0) 0(0) 1515 (12.2) CD and VC 1250 (22.7) 265 (5.5) 0(0) 0(0) 1515 (12.2) Total subjects 5507 4849 1357 717 12430 Table 3 Summary statistics for chloroprene (CD) and vinyl chloride (VC) exposure measures by plant, exposed workers only Exposure indicator Louisville Pontchartrain Maydown Chloroprene Person-years (total) Person-years exposed Person-years unexposeda Duration (CD_Dur, years) 25 percentile Median 75 percentile Maximum Mean Standard deviation Coefficient of variation 197919 197034 885 0.997 5.78 23.2 45.2 12.15 12.35 101.6 30660 26842 3818 3.08 13.34 23.91 30.998 14.12 10.72 75.91 127036 117640 9396 0.003 3.32 10.96 42.5 7.11 8.87 124.8 Average intensity (CD_A1E, ppm) 25 percentile Median 75 percentile Maximum Mean Standard deviation Coefficient of variation 0.871 5.23 16.00 71.00 8.42 10.40 123.5 0.0016 0.0283 0.552 12.37 0.269 0.552 205.7 0.0016 0.160 1.60 16.00 1.43 3.27 228.7 Cumulative (CD_Cum, ppm-years) 25 percentile Median 75 percentile Maximum Mean Standard deviation Coefficient of variation 1.52 18.35 106.3 1351.5 80.35 134.9 167.9 0.0088 0.133 13.13 110.9 5.64 10.33 183.1 0.0022 0.0837 7.33 357.3 11.1 32.79 295.4 Vinyl chloride Person-years (total) 197919 n/a Person-years exposed 50401 Person-years unexposeda 147518 127036 8792 118244 Duration (VC_Dur, years) 25 percentile Median 75 percentile Maximum Mean Standard deviation Coefficient of variation 0.392 n/a 2.55 12.39 29.42 6.795 7.97 117.3 0.296 2.24 5.57 27.23 3.94 4.75 120.5 Grenoble 17057 17057 0 5.71 15.5 24.21 33.996 15.51 9.88 63.74 0.0160 0.149 1.39 16.00 2.19 4.54 207.3 0.0690 1.005 19.38 458.7 37.04 85.93 231.96 n/a n/a Please die this article as: <3ary M Marsh el al.. Mortality paneras amen industrial workers exposed to chtoioprcnc and other substances. Chemieo-Bioiogca11nferactions (2006). dot: 10.101 Wj.eb.2()()6.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00004  Model CBi-SB; NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx Table 3 (Continued ) Exposure indicator Louisville Pontchartrain Average intensity (VC_AIE, ppm) 23 percentile 0.265 n/a Median 1.54 75 percentile 3.00 Maximum 3.00 Mean 1.50 Standard deviation 1.26 Coefficient of variation 83.7 Cumulative (VC_Cum, ppm-years) 25 percentile 0.331 n/a Median 1.54 75 percentile 9.25 Maximum 58.04 Mean 8.66 Standard deviation 13.55 Coefficient of variation 156.4 Includes unexposed portion of person-years among subjects ultimately exposed. Maydown 0.030 0.030 0.030 0.300 0.0648 0.0787 121.4 0.0089 0.0941 0.212 5.34 0.335 0.672 200.8 5 Grenoble n/a n/a 99% of the Louisville workers exposed and all Greno ble workers exposed. Exposure to VC occurred only in plants L and M with 22.7 and 5.5% of subjects exposed, respectively. By nature of the production process involved, all workers exposed to VC were also exposed to CD. Table 3 shows selected summary statistics for CD and VC exposures in each study plant. For each plant, the bulk of the accumulated person-years occurred during periods of exposure to CD. The median CD_Dur computed across individual workers ranged from 3.32 years at plant M to 15.5 years at plant G; the median CD_AIE ranged from 0.028 ppm (plant P) to 5.23 ppm (plant L) and the median CD Cum ranged from 0.084 ppm years (plant M) to 18.35 ppm years (plant L). The median VC_Dur, VC-AIE and VC_Cum values for plants L and M were, respectively, 2.55 and 2.24 years, 1.54 and 0.03 ppm and 1.54 and 0.094 ppm years. For both CD and VC, the corresponding means of the three exposure measures were considerably greater than the medians, reflecting the positive skewness of the subjects' exposure distributions. Figs. 1 and 2 provide plant-specific dot plots of the individual worker values for CD_AIE and CD.Cum; Figs. 3 and 4 provide corresponding dot plots for VC. The clustering of points at various levels of CD_AIE and VC_AIE corresponds with the nominal exposure values shown in Table 1. The figures show that plant L had, by far, the largest exposures to both CD and VC, due to its earlier period of operation and use of the acety lene process. Figs. 5 and 6 show for plants L and M, respectivel y, scattergrams of individual worker values of Grenoble il Louisville i l l i u t i L Maydown Pontchartrain * * * * -- -*------ .----------.--------- .--------- .--------- .-------- -- 0 10 20 30 40 50 60 70 CD_AIE (ppm) Exposed workers only, each dot represents up to 62 subjects. Fig. 1. Average intensity of chloroprene exposure by plant. CD_AIE by VC_AIE. These are relevant to the compos ite exposure measures CDwVC_AIE and CDwoVC_AIE described above. The figures show that the CD_AIE and VCVvIE values were essentially uncorrelated. G re n o b le --L Louisville - i l l . Maydown - Pontchartrain 0 200 400 600 800 1000 1200 1400 CD_Cum (ppm-years) Exposed workers only, each dot represents up to 84 subjects. Fig. 2. Cumulative chloroprene exposure by plant. Please cite this article as; Gary M. Marsh et aL Mortality patterns among industrial workers exposed to ehlmuprenc and otberi substances. Chimico- Biological Itrieraetkms (2006). dot: 0.10 6/j.chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00005  Model C:i-3332; No. oi' !':i<tes K> 6 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx I 2.3. Statistical analyses Louisville -- Maydown -- 0.0 -- ,----------- ,------------- ,------------ ,------------,------------ -- 0.5 1.0 1.5 2.0 2.5 3.0 VC_AIE (ppm) Exposed subjects only, each dot represents up to 11 subjects. Fig. 3. Average intensity of vinyl chloride exposure by plant. ! * 1. ! Maydown 0 --------- T--------- ,----------,--------- T--------- ,---------,-- 8 16 24 32 40 48 56 VC__Cum (ppm-years) Exposed workers only, each dot represents up to 16 subjects. Fig. 4. Cumulative vinyl chloride exposure by plant. CD-exposed workers only Fig. 5. Louisville plant. VC_AIF1 by CD_AIE. CD-exposed workes only Fig. 6. Maydown plant, VC_AIE by CD_AIE. 2.3.1. Exposure-response based on internal comparisons We used relative risk (RR) regression modeling to investigate the dependence of the internal cohort rates (modeled as time to death) for all cancers combined, respiratory system cancer (RSC) and liver cancer (cate gorized as "cancer of the biliary passages and liver" [12]) on combinations of the various CD and VC exposure measures, with adjustment for potential confounding factors. In our analysis of all cancers combined (each plant) and respiratory system cancer (plant L only), we also included adjustment for worker pay type, which was dichotomized as blue collar or white collar and analyzed as a time-dependent variable. The worker pay type variable was constructed by two of the authors (NE and TH) from detailed work history data for use in our exposure-response analysis for respiratory system can cer as a rough surrogate of education and socioeconomic status, which are highly correlated with smoking preva lence in both the U.S. and Europe [18]. In our analysis of general mortality patterns among the CD cohort reported elsewhere [12], we observed SMRs for all cancers com bined and RSC were generally higher among blue collar workers, giving this variable the potential to confound exposure-response associations. We categorized all exposure measures a priori into approximate quartiles based on the distribution of deaths from all cancer. The equal subgroup sizes approximately balance the precision of the risk estimates across sub groups; the use of the all cancer category standardizes comparability across cause of death categories and pro duces nearly equal subgroup sizes for many of the cause of death categories examined. We analyzed no alter native categorizations. For each cause of death, risk sets were explicitly constructed from the cohort data file with age as the primary time dimension, using the RISKSET program module in OCMAP-PLUS [19]. Risk sets were matched further on year of birth to control for cohort effects, and time-dependent exposures and the time-dependent variable, worker pay type, were evalu ated for each individual at each event time they were at risk. Multiplicative relative risk models of the form k(t) = ko(t) exp{x(i)f3} were fit to the internal cohort rates [20,21], and the stratified conditional logistic regression programs in Stata [22] were used to estimate ft from the explicitly constructed risk sets. For plants with less than 20 deaths for the causes of interest, the stratified exact conditional logisti c regression program in LogXact was used to estimate f3 [23]. To parallel the descrip tive SMR analysis of mortality in relation to exposure ['lease die this article as; Gary M. Marsh et al,. Moriulliy patients among industrial workers exposed to chlomprenc and other! substances. Chemico- Biological I(iterations (2006). dob 10.1016/j .chi.200ft.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00006  Mode} CB-SB: NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx 7 (described below), categorized forms of the covariates were considered. Demographic and exposure variables were first considered univariately as categorical vari ables to identify patterns of univariate associations with outcome variables and possible sparse data problems. Possible exposure-response associations were then eval uated with a forward stepwise approach to adjust for possible confounders. Effect modification was assessed if warranted by the main effects. We assessed the statistical significance of each main effect (expressed as a global p-value) with a likelihood ratio statistic. For the exposure variables that exhib ited a monotonic increasing or decreasing pattern in the parameter estimates, we conducted a test for hnear trend (expressed as a trend p-value). All tests on RRs were done at the 0.05 significance level and no adjustment was made for multiple comparisons. All modeling was plant-specific to account for the marked heterogeneity in CD and VC exposures and other large disparities (e.g., vital records systems, time periods, culture, ethnicity, etc.) across the two U.S. and two European study plants. Models for fiver cancer were limited by small numbers of deaths. 2.3.2. Exposure-response based on external comparisons Mortality excesses and deficits in relation to CD and VC exposure levels were also determined via external mortality comparisons expressed as standardized mortal ity ratios (SMRs) along with their 95% confidence inter vals (Cl). The methods used to compute SMRs for the CD cohort study are described in detail elsewhere [12]. SMRs were computed for the categories of the CD and VC exposure measures used in the RR analysis described above. Person-year counts in the unexposed baseline cat egories included the observation time of workers prior to their first exposure. Statistically significant deviations of the SMRs below and above 1.00 were identified using Poisson probabilities [24]. All tests were done at the 0.05 significance level and no adjustment was made for mul tiple comparisons. 3. Results Tables 4-8 show for plants L, M, P and G, respec tively, the results of our exposure-response analyses based on internal and external comparisons. Results for the internal comparisons include for each category of the exposure measures considered, the number of observed deaths (cases) and associated non-cases summed across individual risk sets. The external comparisons include the number of person-years accumulated in each expo sure category. For plant L (Table 4), none of the expo sure measures was positively associated with mortality from all cancers combined or RSC using either inter nal or external comparisons. RRs fluctuate sporadically around 1.00 and the corresponding SMRs are consis tently less than 1.00 and almost always statistically sig nificant. For fiver cancer, we observed elevated RRs in all non-baseline categories of each exposure measure. However, none of the RRs was statistically significant and there was no evidence of a positive association with any exposure measure. The elevated RRs result mainly from the exceedingly low death rates associated with the baseline categories of each measure, as reflected by the correspondingly low SMRs (i.e., the RR for a given non-baseline category is roughly related to the ratio of the corresponding SMR for that category to the SMR for the basehne category). For plant M (Table 5), we observed a statistically significant positive association with CDJDur and all can cers combined based on RRs and the associated non baseline RRs were statistically significant (RRs= 1.53 (95% Cl =1.00-2.34) and 1.78 (95% Cl = 1.11-2.84) for CD Dur 10-19 and 20+ years, respectively, trend p = 0.007). However, as noted for liver cancer in Table 4, the elevated RRs and positive association with CDJDur appear to be due mainly to an inordinately low death rate associated with the baseline category, as reflected by corresponding statistically significant SMR of 0.53 (95% Cl = 0.41-0.67). In fact, both elevated RRs for CDJDur arise as the ratio of death rates that are less than those of the corresponding external standard pop ulation (i.e., SMR = 0.85 and 0.97 for CDJDur 10-19 and 20+ years, respectively). There is no evidence in Table 5 of a positive association with all cancers com bined and the other exposure measures considered, and the corresponding SMRs are consistently less than 1.00. For RSC, we observed some limited evidence of a posi tive association with CD AIE and CD Cum along with a marginally statistically significant (0.05 <p < 0.10) trend test for CD,,AIE, however, this appears again to be driven by inordinately low baseline death rates for both expo sure measures as reflected by the statistically significant baseline SMRs (SMR for baseline CD AIE = 0.47 (95% Cl = 0.23-0.83) and SMR for baseline CD_Cum=0.54 (95% Cl = 0.29-0.90)). For plants P and G (Tables 6 and 7), the evalua tion of exposure-response was made more difficult by the smaller numbers of observed deaths, particularly for RSC. In plant P (Table 6), none of the exposure measures was positively associated with mortality from all can cers combined or RSC using either internal or external comparisons. We observed one statistically significant Please die this article as; Gary M. Marsh et aL Mortality patterns among industrial workers exposed to ehlamprene and other substances. Cheatico-Biological Iuterrartons i 2096k dot: 11).i0 16/j.chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00007 CBKW2: Hq<q( 8 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx--xxx Table 4 Exposure-response analysis for chloroprene and selected cancer sites by exposure metric, Louisville plant, relative risks (RR) and standardized mortality ratios (SMR) Metric Deatlis Internal rate analysis External rate analysis0 Noncases RRd (95% Cl) //-Value Person-years SMR (95% Cl) Ail cancer combined CD_Dur <10 326 10-19 64 20+ 262 CD_AIE <3.6216 163 3.6216-8.1245 163 8.1246-15.99 97 16.0+ 229 CD ..Cum <4.747 163 4.747-55.918 163 55.919-164.052 163 164.053+ 163 60363 9559 39856 29840 22373 16147 40418 30338 29222 24222 24996 1.00 1.06 (0.80-1.41) 1.07 (0.90-1.27) 1.00 1.19(0.94-1.50) 0.93 (0.71-1.21) 1.07 (0.86-1.32) 1.00 0.98 (0.78-1.23) 1.14 (0.91-1.43) 0.93 (0.73-1.17) Global = 0.71 Trend = 0.42 131276 30404 36239 Global = 0.27 Trend = 0.97 69274 27933 28689 72023 Global = 0.35 Trend = 0.83 68918 56737 39840 32424 0.70** (0.63-0.78) 0.68" (0.53-0.87) 0.82** (0.72-0.93) 0.73*' (0.62-0.83) 0.88(0.75-1.02) 0.65" (0.53-0.79) 0.72** (0.63-0.82) 0.75" (0.64-0.87) 0.71** (0.60-0.82) 0.79** (0.68-0.92) 0.70** (0.60-0.82) Respiratory system cancer CD_Dur <10 137 10-19 23 20+ 106 CD_AIE <3.6216 56 3.6216-8.1245 70 8.1246-15.99 33 16.0+ 107 CD ..Cum <4.747 62 4.747-55.918 67 55.919-164.052 77 164.053+ 60 25995 3806 17174 12642 9812 6950 17571 12961 12656 10471 10887 1.00 0.98 (0.62-1.57) 0.97 (0.75-1.27) 1.00 1.34(0.93-1.95) 0.88 (0.56-1.38) 1.36 (0.97-1.91) 1.00 1.00 (0.71-1.43) 1.32 (0.94-1.88) 0.85 (0.58-1.23) Global = 0.98 Trend = 0.84 131276 30404 36239 Global = 0.06 Trend = 0.20 69274 27933 28689 72023 Global = 0.07 Trend = 0.71 68918 56737 39840 32424 0.74** (0.62-0.87) 0.66** (0.42-0.99) 0.79* (0.65-0.96) 0.63*' (0.48-0.82) 0.90 (0.70-1.14) 0.56" (0.38-0.78) 0.83 (0.68-1.00) 0.71" (0.55-0.91) 0.71" (0.55-0.90) 0.92 (0.73-1.15) 0.65" (0.50-0.84) Liver cancerf CD_Dur <10 6 10-19 4 20+ 7 CD-AIE <3.6216 3 3.6216-8.1245 7 8.1246-15.99 3 16.0+ 4 CD.Cum <4.747 2 4.747-55.918 3 55.919-164.052 7 164.053+ 5 1500 216 965 714 568 388 1011 744 725 653 559 1.00 3.85(0.76-17.09) 1.75 (0.49-6.44) 1.00 3.81 (0.77-25.76) 1.84(0.22-15.74) 1.31 (0.20-10.07) 1.00 1.90 (0.21-23.81) 5.10 (0.88-54.64) 3.33 (0.48 39.26) Global = 0.24 Trend = 0.36 131276 30404 36239 Global = 0.22 Trend = 0.84 69274 27933 28689 72023 Global = 0.17 Trend = 0.09 68918 56737 39840 32424 0.61 (0.22-1.32) 2.08 (0.57-5.33) 0.99 (0.40-2.04) 0.62 (0.13-1.80) 1.73 (0.70-3.56) 0.94(0.19-2.74) 0.59 (0.16-1.52) 0.43(0.05-1.55) 0.59 (0.12-1.74) 1.62 (0.65-3.33) 1.00 (0.33-2.34) a Categoiies based on approximate quartiles of all cancer deaths; deci mal places of cutpoints reflect precision needed for computational purposes only and not precision of exposure assessment. b Local county rates. c The number of persons in decedent's risk set used in calculation of RR. d Also adjusted for sex. e The number of person-years used in calculation of SMR. f Analyzed using LogXact. * p<0.05. ** pcO.Ol. Please die ihis article as: Cary MLMarsh el al.< Mortality pnflcms among indtfsldaJ workers exposed to chtowprcnc and other substances. Chemieo~Bioiogical Infemotions (2006). dot: 10.101 Wj.ebL2()()6.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00008  MixfcJ C-3B2; NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx 9 Table 5 Exposure-response analysis for chloroprene and selected cancer sites by exposure metric, Maydown plant, relative risks (RR) and standardized mortality ratios (SMR) M etric" Deaths Internal rate analysis External rate analy;sisb Noncasesc RRd (95% Cl) p-Value Person-years6 SMR (95% Cl) All cancers combined C D _Dur <10 66 10-19 35 20+ 27 C D .A IE <0.1538 43 0.1538-1.269 28 1.270-1.69 36 1.70+ 21 9674 2965 2349 5660 2931 3973 2424 1.00 1.53* (1.00-2.34) 1.78* (1.11-2.84) 1.00 1.02 (0.60-1.72) 1.07 (0.67-1.71) 0.95 (0.54-1.65) Global = 0.03 Trend-0,007 Global = 0.98 Trend = 0,97 102413 17257 7366 57453 22489 32973 14121 0.53** (0.41-0.67) 0.85 (0.59-1.18) 0.97 (0.64-1.41) 0.54** (0.39-0.73) 0.83 (0.55-1.20) 0.70* (0.49-0.96) 0.70 (0.43-1.07) CD_Cum <0.0387 43 0.0387-6.7310 28 6.7311-24.50 29 24.51+ 28 6062 3266 3065 2595 1.00 1.12(0.67-1.89) 0.94(0.53-1.66) 0.95 (0.53-1.70) Global = 0.92 Trend = 0.75 63130 32527 19539 11840 0.50** (0.36-0.67) 0.74 (0.49-1.07) 0.79 (0.53-1.13) 0.85 (0.56-1.22) Respiratory system cancer CDJDur <10 28 10-19 12 20+ 8 3649 1143 992 1.00 0.81 (0.36-1.79) 1.17 (0.23-5.92) Global = 0.82 Trend = 0.84 102413 17257 7366 0.73 (0.48-1.05) 0.86 (0.44-1.50) 0.83 (0.36-1.64) C D .A IE <0.1538 11 0.1538-1.269 12 1.270-1.69 16 1.70+ 9 2180 1133 1522 949 1.00 2.83* (1.09-7.38) 2.63* (1.11-6.23) 2.23 (0.83-5.97) Global = 0.08 Trend = 0.09 57453 22489 32973 14121 0.47** (0.23-0.83) 1.08 (0.56-1.89) 0.93 (0.53-1.51) 0.87 (0.40-1.65) CD_Cum <0.0387 14 0.0387-6.7310 9 6.7311-24.50 12 24.51+ 13 2300 1263 1181 1040 1.00 1.65 (0.66-4.15) 1.89 (0.72-4.96) 2.28 (0.86-6.01) Global = 0.39 Trend = 0.10 63130 32527 19539 11840 0.54* (0.29-0.90) 0.74 (0.34-1.40) 0.97 (0.50-1.69) 1.13 (0.60-1.92) a Categories based on approximate quartiles of all cancer deaths; decimal places of cutpoints reflect precision needed for computational purposes only and not precision of exposure assessment. b National rates. c The num ber of persons in decedent's risk set used in calculation of RR. d Also adjusted for worker service type and duration of employment. e The number of person-years used in calculation of SMR. p < 0.05. ** p < 0 .0 1 . RR for all cancers combined in the second category of CD.AIE (RR = 4.76; 95% Cl = 1.39-6.27); however, this appears to be an isolated finding. RRs for RSC were also elevated in all baseline categories of each exposure measure, again driven by the inordinately low baseline death rates (i.e., baseline SMRs for CD Dur, CD_AIE and CD_Cum = 0.28 (95%CI = 0.03-1.00), 0.33 (95% Cl = 0.09-0.85) and 0.40 (95% Cl = 0.08-1.18), respectively. In plant G (Table 7), none of the exposure measures was positively associated with mortality from all cancers combined using either internal or external comparisons. There is some limited evidence of a pos itive association with CD_AIE and CD.Cum and RSC; however, the linear trends and the exposure categoryspecific RRs were not statistically significant. While the death rates for RSC associated with the baseline cat egories of CD_AIE and CD.Cum were not as low as those in the other plants, they were still about 25% less Please die this article as; Gary M. Marsh et a L Morialliy patterns among desina! workers exposed to chloroprene and other! substances. Chamico- Biological rieraetkms i 2006k dot: 10.10 16/j .chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00009  MixfcJ C-3332; No. oi' i':t<tes K> 10 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx Table 6 Exposure-response analysis for chloroprene and selected cancer sites by exposure metric, Pontchartrain plant, relative risks (RR) and standardized mortality ratios (SMR) M etric" Deaths Internal rate analysis External rate analysis'3 Noncasesc RR (95% Cl) //-Value Person-yearsd SMR (95% Cl) All cancers combined CD _Dur <10 15 10-19 12 20+ 7 2039 2095 1473 1.00 0.71 (0.32-1.58) 0.83 (0.27-2.53) G lobal = 0.71 Trend = 0.56 19067 7668 3926 0.75 (0.42-1.24) 0.59 (0.31-1.03) 0.57 (0.23-1.18) C D .A IE <0.0017 17 0.0017-0.1329 4 0.1330-0.8174 7 0.8175+ 6 2735 298 1409 1165 1.00 4.76* (1.39-16.27) 1.58 (0.57-4.37) 1.34 (0.49-3.65) Global = 0.17 Trend = 0.44 15858 1574 5522 7707 0.55" (0.32-0.88) 2.76 (0.75-7.06) 0.76 (0.31-1.57) 0.69 (0.26-1.51) C D .C um <0.0193 15 0.0193-1.8944 6 1.8945-16.1918 6 16.1919+ 7 1794 1434 766 1613 1.00 0.52 (0.20-1.37) 1.50 (0.55-4.11) 0.80 (0.29-2.16) G lobal = 0.31 Trend = 0.91 15354 6363 6027 4916 0.75 (0.42-1.24) 0.41* (0.15-0.90) 1.07 (0.39-2.32) 0.61 (0.24-1.25) Respiratory system cancere CD .Dur <10 2 653 1.00 Global = 0.33 19067 10-19 7 799 3.08(0.62-15.31) Trend = 0.32 7668 20+ 3 747 2.09 (0.26-16.85) 3926 0.28 (0.03-1.00) 0.85 (0.34-1.75) 0.60 (0.12-1.76) C D -A IE <0.0017 4 932 1.00 Global = 0.25 15858 0.0017-0.1329 1 102 7.28 (0.09-167.13) Trend = 0.14 1574 0.1330-0.8174 4 646 5.03 (0.59-58.02) 5522 0.8175+ 3 519 3.50 (0.37-33.64) 7707 0.33* (0.09-0.85) 2.05 (0.05-11,44) 1.11 (0.30-2.85) 0.90 (0.19-2.62) CD_Cum <0.0193 3 600 1.00 Global = 0.70 15354 0.0193-1.8944 3 468 1.60 (0.20-12.77) Trend = 0.34 6363 1.8945-16.1918 2 322 2.90 (0.20-34.11) 6027 16.1919+ 4 809 2.32 (0.30-21.83) 4916 0.40 (0.08-1.18) 0.52 (0.11-1,53) 0.96 (0.12-3.48) 0.85 (0.23-2.18) a Categories based on approximate quartiies of all cancer deaths; decimal places of cutpoints reflect precision needed for computational purposes only and not precision of exposure assessment. b Local county rates. c The num ber of persons in decedent's risk set used in calculation of RR. d The number of person-years used in calculation of SMR. e Analyzed using LogXact. p < 0.05. ** p < 0 .0 1 . than those in the external comparison populations (i.e., baseline SMRs for CD AIE and CD Cum were 0.76 and 0.72, respectively). As noted for the other plants, the low baseline rates at least partly explain the elevated RRs for many of the non-baseline categories. Table 8 shows our exposure-response analyses for VC that was limited to plant L. For all cancers combined and RSC, deficits in deaths based on RRs and SMRs were observed in all exposure categories; many were statistically significant. Fifteen of the 17 hver cancer deaths in plant L occurred among unexposed workers; RRs and SMRs in the non-baseline categories were unremarkable. While not shown, our analysis of mortality among plant L and M workers in relation to the four composite ;xposure measures, CDwVC_AIE, CDwVC.Cum, CDwoVC_AIE and CDwoVC_Cum, produced risk estimates similar to those based on the marginal CD exposure measures (i.e., exposure to CD regardless of VC exposure) and none of the composite measures revealed evidence of increasing cancer risks with increasing expo- Please die this article as; Gary M. Marsh et al.. Mortality patterns among Industrial workers exposed to chloroprene and other substances. ChemictvBiological Interactions (2(106), dot: 10.1016/j.chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00010  MixfcJ CB-SB; NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx 11 Table 7 Exposure-response analysis for chloroprene and selected cancer sites by exposure metric, Grenoble plant, relative risks (RR) and standardized mortality ratios (SMR) M etric" Deaths Internal rate analysis External rate analysi sb Noncasesc RR (95% Cl) /7-Value Person-yearsd SMR (95% Cl) All cancers combined C D _Dur <10 9 934 10-19 5 819 20+ 6 585 1.00 0.60 (0.20-1.80) 1.32 (0.43-4.08) Global = 0.43 Trend = 0.82 9813 4900 2344 0.63 (0.29-1.20) 0.40* (0.13-0.94) 0.83 (0.31-1.82) C D ..AIE <0.0051 5 551 0.0051-0.0880 5 477 0.0881-1.2246 5 616 1.2247+ 5 694 1.00 1.21 (0.35-4.22) 1.21 (0.34-4.40) 1.04 (0.29-3.76) Global = 0.99 Trend = 0.95 3393 4694 3189 5781 0.60 (0.20-1.41) 0.55 (0.18-1.28) 0.67 (0.22-1.56) 0.56 (0.18-1.31) CD_Cum <0.0497 5 584 0.0497-1.4149 5 532 1.4150-23.9306 5 683 23.9307+ 5 539 1.00 1.16 (0.33-4.08) 1.07 (0.30-3.84) 1.54 (0.43-5.60) Global = 0.92 Trend = 0.57 4267 4749 4619 3422 0.56 (0.18-1.31) 0.53 (0.17-1.23) 0.55 (0.18-1.28) 0.79 (0.26-1.85) Respiratory system cancere CDJDur <10 3 500 10-19 20+ 5 448 2 272 1.00 1.84(0.44-7.77) 1.46 (0.22-9.61) Global = 0.70 Trend = 0.58 9813 4900 2344 0.64(0.13-1.87) 1.16 (0.38-2.71) 0.71 (0.09-2.58) C D .A IE <0.0051 2 294 0.0051-0.0880 1 260 0.0881-1.2246 3 325 1.2247+ 4 341 1.00 0.63 (0.06-6.96) 2.29 (0.22-34.16) 2.99 (0.36-41.87) Global = 0.45 Trend = 0.19 3393 4694 3189 5781 0.76 (0.09-2.80) 0.32 (0.01-1.76) 1.06 (0.22-3.09) 1.25 (0.34-3.19) CD_Cum <0.0497 2 312 0.0497-1.4149 1 288 1.4150-23.9306 4 356 23.9307+ 3 264 1.00 0.61 (0.05-6.76) 2.87 (0.35-39.70) 3.14(0.30-47.99) Global = 0.40 Trend = 0.17 4267 4749 4619 3422 0.72 (0.09-2.61) 0.30 (0.01-1.69) 1.19 (0.32-3.04) 1.28 (0.26-3.73) a Categories based on approximate quartiies of all cancer deaths; decimal places of cutpoints reflect precision needed for computational purposes only and not precision of exposure assessment. b National rates. c The num ber of persons in decedent's risk set used in calculation of RR. d The number of person-years used in calculation of SMR. e Analyzed using LogXact. p < 0.05. sure. Also not shown, our analyses of RSC and liver cancer mortality in relation to 5 and 15 year lagged measures of CDJDur, CD_AIE and CD_Cum did not materially alter the findings from the unlagged analy ses. Likewise, for RSC and liver cancer, we observed no evidence of an association with time since first exposure to CD or to VC (not shown). We attempted to roughly additionally adjust RRs for RSC for potential confounding by smoking, via the surrogate variable worker pay type (blue/white collar). Because of the small number of white-collar RSC deaths among the white-collar workers in each plant (3, 0, 2 and 0 white collar RSC deaths for plants L, M, P and G, respectively), the adjusted analysis was limited to plant L. For all cancers combined, additional adjustment for potential confounding by worker pay type had lit tle effect on the all cancer RRs for either CD AIE or CD_Cum. For plant P, additionally adjusted RRs were higher for all exposure categories of both measures, sug gesting negative confounding by smoking; for plant G, Please die this anide as; Gary M. Marsh er aL Morialliy paneras among industrial workers exposed to chloroprene and orher substances. Chemico-Biological interactions (2(K)6k dot: iO.IOih/j.chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00011  MixfcJ C-3332; No. oi' i'ayes 16 12 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx Table 8 Exposure-response analysis for vinyl chloride monomer and selected cancer sites by exposure metric, Louisville plant, relative risks (RR) and standardized mortality ratios (SMR) M etric" D eaths Internal rate analysis External rate analysis11 Noncasesc RRd (95% Cl) p-Value Person-yearse SMR (95% Cl) All cancers combined VC_Dur Unexposed 524 >0-5 71 5-9 9 10+ 48 VCM_AIE Unexposed 524 >0-0.27 32 0.28-1.75 34 1.751-2.99 15 3.0+ 47 V CM X um Unexposed 524 >0-0.4476 32 0.4477-1.9482 32 1.9483-14.5832 32 14.5833+ 32 77268 17509 2444 11557 77268 6335 9312 4087 11776 77268 7057 6747 9578 8128 1.00 0.66* (0.52-0.84) 0.78(0.55-1.11) 0.50* (0.29-0.85) 1.00 0.80 (0.56-1.15) 0.59* (0.41-0.83) 0.56* (0.35-0.87) 0.70* (0.51-0.97) 1.00 0.72 (0.50-1.03) 0.82 (0.57-1.18) 0.57* (0.40-0.82) 0.60* (0.42-0.86) Global = 0.0002 Trend = 0.0001 Global = 0.0004 Trend = 0.0001 Global = 0.0004 Trend <0.0001 147518 31911 6122 12369 147518 10880 14543 5768 19210 147518 14506 11583 14267 10045 0.80*' (0.73-0.87) 0.58*' (0.45-0.73) 0.47* (0.21-0.88) 0.60** (0.18-0.61) 0.80*' (0.73-0.87) 0.71 (0.49-1.00) 0.52*' (0.36-0.73) 0.46** (0.26-0.76) 0.47*" (0.35-0.63) 0.80*' (0.73-0.87) 0.63*' (0.43-0.89) 0.58** (0.39-0.81) 0.44** (0.30-0.62) 0.53** (0.36-0.75) Respiratory system cancer VCJDur Unexposed 232 >0-5 20 5-9 2 10+ 12 VCM_AIE Unexposed 232 >0-0.27 12 0.28-1.75 6 1.751-2.99 3 3.0+ 13 VCM X um Unexposed 232 >0-0.4476 13 0.4477-1.9482 8 1.9483-14.5832 6 14.5833+ 7 33132 8732 3313 1798 33132 2743 4035 2443 4622 33132 3009 2964 4232 3638 1.00 0.38* (0.24-0.59) 0.48* (0.25-0.90) 0.15* (0.04-0.60) 1.00 0.64 (0.35-1.15) 0.22* (0.10-0.50) 0.23* (0.09-0.62) 0.41* (0.23-0.73) 1.00 0.64 (0.36-1.12) 0.42* (0.21-0.86) 0.22* (0.10-0.49) 0.27* (0.13-0.58) G lobal< 0.0001 Trend <0.0001 G lobal< 0.0001 Trend <0.0001 G lobal< 0.0001 Trend <0.0001 147518 31911 6122 12369 147518 10880 14543 5768 19210 147518 14506 11583 14267 10045 0.89 (0.78-1.02) 0.38** (0.23-0.59) 0.25* (0.03-0.89) 0.35*' (0.18-0.61) 0.89 (0.78-1.02) 0.62 (0.31-1.08) 0.22** (0.08-0.47) 0.22*' (0.05-0.65) 0.31*' (0.16-0.53) 0.89 (0.78-1.02) 0.60 (0.32-1.02) 0.34** (0.15-0.67) 0.19*' (0.07-0.42) 0.27*' (0.11-0.57) Liver cancerf VC_Dur Unexposed >0-5 5-9 10+ VCM_AIE Unexposed >0-0.27 0.28-1.75 1.751+ V CM X um Unexposed 15 1952 1.00 Global = 0.24 147518 1 407 2.49s (0.41--oc) Trend = 0.23 31911 1 61 0.69s (0.02-oc) 6122 0 261 3.96s (0.10-oc) 12369 15 1952 1.00 Global = 0.46 147518 1 139 1.04 (0.02-7.04) Trend = 0.20 10880 0 223 0.49s (--co, 2.98) 14543 1 367 0.43 (0.01-2.92) 24978 15 1952 1.00 Global = 0.54 147518 1.07 (0.60-1.77) 0.37 (0.01-2.08) 2.38 (0.06-13.29) -(0 -2 .0 5 ) 1.07 (0.60-1.77) 0.98 (0.03-5.48) -(0 -2 .5 9 ) 0.37 (0.01-2.04) 1.07 (0.60-1.77) Please die this article as: <3arv MLMarsh el al.< Mortality patterns among indtfsldaJ workers exposed to chtowprcnc and miter substances. Chernieo-Bioiogical Inferections (2006). dot: 10.101 Wj.ebl2006,08.(112 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00012  Mette} CB-SB; NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx 13 Table 8 (Continued ) M etric3 Deaths Interna] rate analysis External rate analysis*3 Noncasesc RRd (95% Cl) //-Value Person-years6 SMR (95% Cl) >0-0.4476 0 164 0.4477-1.9482 1 168 1.9483+ 1 397 0.66 (-oo,4.00) 0.97 (0.02-6.67) 0.38 (0.01-2.58) Trend = 0.31 14506 11583 24312 -(0 -3 .2 5 ) 0.86 (0.02-4.79) 0.36 (0.01-1.99) a Categories based on approximate quartiles of all cancer deaths; decimal places of cutpoints reflect precision needed for computational purposes only and not precision of exposure assessment. b Local county rates. c The num ber of persons in decedent's risk set used in calculation of RR. d Also adjusted for sex. c The number of person-years used in calculation of SMR. f Analyzed using LogXact. g Median-unbiased estimate. * p < 0.05. ** p < 0.01. the opposite pattern emerged, suggesting positive con founding by smoking. In neither case; however, did the additional adjustment for worker pay type materially alter the unadjusted findings of no association with either exposure measure. In plant L, additional adjustment for pay type had little effect on the RSC RRs for CD_AIE or CD_Cum (data not shown). 4. Discussion and conclusions As described in detail in our analysis of general mor tality patterns [12], our historical cohort study of workers from four CD production sites in the U.S. and Europe represents the largest and the most comprehensive and rigorous investigation of the long-term health effects of exposure to CD conducted to date. It overcomes most of the shortcomings and uncertainties noted by Rice and Bofetta [10] and Acquavella and Leonard [11] that have limited the interpretation of findings from the five avail able cohort studies [4-8]. A particular strength of our study was the rigorous, chemical process-based expo sure reconstruction for chloroprene and vinyl chloride conducted by Esmen et al. [13-15] and Hall et al. [16] that enable us to examine mortality from all cancers com bined and from the apriori sites of interest, lung and liver cancer, in relation to several quantitative measures of CD and/or VC exposure. Another strength of our exposure-response analy ses was the use of national and local county mortality comparisons and robust statistical modeling of internal cohort rates. The strengths of the internal study group comparison are that it will usually reduce the healthy worker effect [25], and it allows direct comparison of relative risk across strata. However, internal compar isons can be unstable when the study population is small and/or the disease under study is rare (producing wider confidence limits), and may be misleading if workers included in the baseline category (i.e., least exposed) have different underlying cancer risks than workers in the exposed groups. On the other hand, external com parisons based on regional rates have the strengths of being able to adjust for geographic variability in social, cultural and economic factors related to disease [26] and are generally very stable. The disadvantages of the external comparison group are an inability to adjust for the healthy worker effect and a difficulty in comparing standardized mortality ratios between groups when their confounder distributions differ [27]. When we used external comparisons of the surround ing county populations of each study plant, we observed many deficits in death from all cancers combined, RSC and liver cancer that were often largest among the unex posed workers, but still present among workers in the non-baseline exposure categories. This pattern of find ings by exposure category in the external populationbased SMRs led to elevated relative rates (RRs) of dis ease when rates for non-baseline categories were com pared to the baseline (unexposed) rates. For example, for RSC by CD_AIE in plant P (Table 6), an RR of 3.50 (95%Cl = 0.37-33.64) for the highest exposure category (0.8175+ppm), or an apparent 3.5-fold excess, results because a small 10% deficit in deaths in the highest exposure category (SMR = 0.90; 95% Cl = 0.19-2.62) is essentially being compared to a exceedingly large, sta tistically significant 67% deficit in the baseline category (SMR = 0.33; 95%CI = 0.09-0.85). Thus, the question arises as to whether the ratio of small to large deficits in deaths (essentially, but not exactly, what is expressed Please die this article as; Gary M. Marsh et aL Merialiiy patterns among industrio] workers exposed to ehiaroprenc and other substances. CfoemictvBiological Itrieraettosis i 2006k dot: 11).i0 16/j.chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00013  Mottet CBI-52: No. oi' i'ajtes K> 14 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx via RRs) should be interpreted as a meaningful "excess" in deaths. This enigmatic feature of exposure-response analyses created by inordinately low basehne rates has been observed in other major occupational cohort stud ies, such as the cohort studies of formaldehyde [28-30] and acrylonitrile [31] workers conducted by the National Cancer Institute, and has stimulated reanalyses and rein terpretation of the NCI cohort data [32-34], Although RRs for the cancer sites and exposure measures con sidered were elevated in many non-baseline categories due to the low basehne rates, we observed no consis tent evidence that RRs were positively associated with increasing exposure in any of the study plants. There are at least two possible explanations for the large differences in the cancer relative risks in the CD cohort when internal or external comparison rates are used. The first is that internal comparisons pro duce more valid results because selection bias stem ming from the "healthy worker effect" can reduce the putative effect of high exposure to CD (or VC) when external comparison rates are used. The healthy worker effect is evident in this population by the low relative risks for all causes of death for CD-exposed (SMR = 0.71; 95% Cl = 0.69-0.73) and CD-unexposed workers (SMR = 0.88; 95% Cl = 0.69-1.10). However, the selection for workers who are healthy at time of hire is usually more relevant for cardiovascular and nonmalignant respiratory diseases than lung cancer, which has a relatively sudden onset, short survival time and high case-fatality rate [35]. A second explanation is that the external comparisons produce more valid results because the unexposed group has a different underlying cancer risk than the exposed group. As shown above, the risk in the highest exposure category when internal comparisons are used may simply be the result of an unusually low lung cancer death rate among workers in the unexposed basehne category. In fact, had the death rates for all cancer, RSC or liver cancer among the unexposed workers been closer to or equal to those of the general regional populations from which the four plant workforces were drawn, the internal RRs calculated for quartiles of CD (or VC) exposure across the total cohort would probably have been uniformly near or less than 1.00. The very low SMRs for all cancer, lung and fiver cancer, especially among unexposed workers, are puz zling given that we used regional standard population rates. Although a small percentage of deaths (estimated at about 5%) may have been missed among transferees in plant P and among subjects who emigrated in plants M and G [12], under-ascertainment of deaths is an unlikely explanation for these low SMRs. Also, because regional rates can help adjust for the social, cultural and economic factors related to diseases such as lung cancer, and even help to adjust for geographic variability in tobacco use [26], it is difficult to postulate what non-occupational factors may have had such a profound influence on the cancer mortality experience of this cohort. It was hoped that our additional model adjustment for worker pay type, a correlate of education/socioeconomic status, and thus, smoking history, might help to explain the inor dinately low and often statistically significant baseline SMRs for all cancers combined and RSC found for each study plant in the baseline categories of each exposure measure. For example, if subjects at risk in the baseline exposure categories were lighter smokers than subjects at risk in the non-baseline categories, this would neg atively confound baseline SMRs for RSC relative to non-baseline SMRs and positively confound the corre sponding non-baseline RRs. To a lesser extent, the same pattern could occur for all cancers combined. However, with the possible exception of plant G, where pay typeadjusted RRs for all cancers combined were uniformly less, suggesting positive confounding by smoking, the additional adjustment for worker pay type did not mate rially alter the pattern of RRs for all cancer and RSC found in the unadjusted models. With the possible exception of liver cancer in plant L (based on small numbers of death), chance alone does not appear to be an explanation for the cancer deficits observed among unexposed workers in this study. Our U.S. and regional rate-based SMRs (and RRs) for all cancers and RSC in all categories of the CD expo sures examined were based on sufficiently large num bers of observed deaths to provide stable risk estimates, and deficits were generally consistent across the CD exposure categories considered. Also, the general qual ity of the follow-up and cause of death ascertainment in this study rule out under-ascertainment of cancer deaths as a reason for the deficits. Given the absence of a viable explanation derived from the available study data, what remains is the possibility that some hereto fore unknown selection factors for low cancer inci dence or mortality were operating on the unexposed subjects in this cohort, or that some type of protective effect for lung cancer arose from a particular exposure or combination of exposures encountered at the study plants. Without further formal investigation of this phe nomenon in the CD cohort, the reason(s) for the marked deficits in cancer in unexposed workers will remain unknown. While the possible occurrence of the rare VC-related cancer, angiosarcoma of the fiver, was of interest in this study, methodological limitations precluded a full ['lease die this article as; Gary M. Marsh et al. Merialiiy patients among industrial workers exposed to chlomprenc and other! substances. Chernico-Biological | ntenactions (2006). dot: I),i016/j .chi.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00014  Moder CBi-SB2: NV or r*:.s!es k. G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx-xxx 15 evaluation. A full account of our evaluation of angiosar coma was provided in our companion paper [12]. In brief, because angiosarcoma of the liver does not have a specific ICD code until the 10th revision (1999+), it can only be roughly identified in earlier revisions by manually reviewing text fields of death certificates. A comprehensive death certificate review was not possible in this study as we obtained death certificates for the two U.S. plants only for deaths that occurred before the National Death Index (before 1979) and in some cases cause of death for pre-1979 deaths was obtained as an ICD code from the DuPont mortality registry. For plant G we obtained ICD codes only from our French collabo rators and in plant M we obtained only a limited number of death certificates. These limitations notwithstanding, "angiosarcoma of the liver" was not mentioned on any death certificates available for our review. In summary, our analysis of the cancer mortality experience of the CD cohort provides no evidence that exposure to CD or VC, at the levels encountered in the four study plants, increases the risk of death from all cancers or the sites of a priori interest, lung (included within the broader category respiratory system cancer) and hver (categorized as "cancer of the biliary passages and liver"). Our findings based on external comparisons using regional rates produced exposure category-specific risks very different than those based on internal rates due largely to inordinately low death rates among workers in the unexposed categories. We conclude that chance or selection bias in the form of the healthy worker effect were unlikely explanations for these differences. Fur ther investigation of the CD cohort may help to explain the reasons for the differences in risk estimates based on internal and external rates. Acknowledgements The International Institute of Synthetic Rubber Pro ducers (IISRP) sponsored this research, but the design, conduct, analysis and conclusions are those of the authors. Sponsoring companies were DuPont Dow Elas tomers LLC and Enichem Elastomers France. We would like to acknowledge the cooperation and support of the representatives and consultants of IISRP and its member companies, in particular, Sheila lones, Robin Leonard, Mike Lynch, Stuart Pollard and Paul Poullet. Our special thanks to Dr. Marc Colonna of the Registre du Can cer de l'Isre who coordinated the cohort enumeration and vital status tracing of the Grenoble, France cohort and provided us with a copy of the data file. In addi tion, we acknowledge the computer programming work of Stephen Sefcik. The research proposal was approved by the Institutional Review Boards (IRB) of the Univer sity of Pittsburgh, the University of Oklahoma and the University of Illinois at Chicago. Portions of the data were presented at the 2005 annual meeting of the British Occupational Hygiene Society, April 19, 2005, Manch ester, U.K. and the 2005 annual meeting of the American Industrial Hygiene Association, May 25, 2005, Ana heim, CA. References [1] J. Lynch, BD m onom er and elastom er production processes, Chem.-Biol. Interact. 135/136 (2001) 147-153. [2] IARC Monographs, Suppl. 7, Vinyl Chloride, Lyon, France, 1987. [3] IARC M onographs, vol. 71, Chloroprene: Re-evaluation of Some Organic Chemicals, Hydrazine and Hydrogen Peroxide, Lyon, France, 1999. [4] S. Pell, Mortality of workers exposed to chloroprene, J. Occup. Med. 20 (1978) 21-29. [5] L. Shouqi, D. Quinan, L. Yuqing, L. Yingfei, Epidem iologic study of cancer mortality among chloroprene workers, Biomed. Envi ron. Sci. 2(1989) 141-149. [6] M. Bulbulyan, A. M argaryan, S. Ilychova, S. Astashevsky, S. Uloyan, V. Cogan, D. Colin, P. Boffetta, D. Zaridze, Cancer inci dence and mortality in a cohort of chloroprene workers from Armenia, Int. J. Cancer 81 (1999) 31-33. [7] M. Colonna, G. Laydevant, A cohort study of workers exposed to chloroprene in the departm ent of Isere, France, Chem.-Biol. Interact. 135/136 (2001) 505-514. [8] M. Bulbulyan, O. Changuina, D. Zaridze, S. Astashevsky, D. Colin, P. Boffetta, Cancer mortality am ong Moscow shoe w ork ers exposed to chloroprene, Cancer Causes Contr. 9 (4) (1998) 381-387. [9] D. Zaridze, M. Bulbulyan, O. Changuina, A. M argaryan, P. Bof fetta, Cohort studies of chloroprene-exposed workers in Russia, Chem.-Biol. Interact. 135/136 (2001) 487-503. [10] J. Rice, P. Boffetta, 1,3-Butadiene, isoprene and chloroprene: reviews by the IARC monographs programme, outstanding issues, and research p riorities in epidem iology, Chem .-B iol. Inter act. 135/136 (2001) 11-26. [11] J. Acquavella, R. Leonard, A review of the epidem iology of 1,3butadiene and chloroprene, Chem.-Biol. Interact. 135/136 (2001) 43-52. [12] G. Marsh, A. Youk, J. Buchanich, M. Cunningham, N. Esmen, T. Hall, M. Phillips, M ortality patterns am ong industrial workers exposed to chloroprene and other substances. I. General mortality patterns, Chem.-Biol. Interact., doi: 10.1016/j.cbi.2006.08.011, in press. [13] N. Esm en, K. Kennedy, T. Hall, M. Phillips, Classification of worker exposures, Chem.-Biol. Interact., doi: 10.1016/j.cbi.2006. 08.008, in press. [14] N. Esmen, T. Hall, M. Phillips, G. Marsh, Chemical process based reconstruction of exposures for an epidem iological study. I. Theoretical and m ethodological issues, Chem.-Biol. Interact., doi: 10.1016/j.cbi.2006.08.009, in press. [15] N. Esmen, T. Hall, M. Phillips, E. Jones, H. Basara, G. Marsh, J. Buchanich, Chemical process based reconstruction of exposures for an epidem iological study. II. Estim ated expo sures to chloroprene and vinyl chloride, Chem.-Biol. Interact., doi: 10.1016/j.cbi.2006.08.010, in press. Please die (his article as: Gary M. Marsh et aL Mortality patterns among incHrsfrkt.1 workers exposed to ehlaroprcnc anti otbcrj substances. Chimico- Biological Itriemettons (2006k dot: 10.10 16/j.chi.200ft.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00015 16 G.M. Marsh et al. / Chemico-Biological Interactions xxx (2006) xxx--xxx [16] T. Hall, N. Esm en, E. Jones, H. Basara, M. Phillips, G. Marsh, A. Youk, J. Buchanich, R. Leonard, Chemical process based reconstruction of exposures for an epidem iological study. III. Analysis of industrial hygiene samples, Chem.-Bioi. Interact., doi:10.1016/j.cbi.2006.09.004, in press. [17] A. Youk, G. Marsh, R. Stone, J. Buchanich, T. Smith, H istorical cohort study of US man-made vitreous fiber production workers. III. Analysis of exposure-weighted measures of respirable fibers and formaldehyde in the nested case-control study of respiratory system cancer, J. Occup. Environ. Med. 43 (2001) 767-778. [18] NIOSH The Work-Related Lung Disease Surveillance Report, NIOSH (2002), Publication No. 2003-111; CDC. [19] G. Marsh, A. Youk, R. Stone, S. Sefcik, C. Alcorn, OCM APPLUS, a new program for the comprehensive analysis of occupa tional cohort data, J. Occup. Environ. Med. 40 (1998) 351-362. [20] D.R. Cox, Regression m odels and life tables, J. R. Stat. Soc. 34 (1972) 187-220. [21] D.R. Cox, Partial likelihood, Biometrika 62 (1975) 269-276. [22] StataCorp., Stata Statistical Software: Release 8.0, Stata Corpo ration, College Station, TX, 2004. [23] Cytel Software: LogXact-5, Software for Exact Conditional Logistic Regression, Cambridge, MA, 2003. [24] N. Breslow, N. Day. The D esign and Analysis of Cohort Studies. Statistical Methods in Cancer Research, vol. II, Oxford University Press, Lyon, 1987. [25] N. Pearce, H. Checkoway, C. Shy, Time-related factors as poten tial confounders and effect modifiers in studies based on an occupational cohort, Scand. J. W ork Environ. Health 12 (1986) 97-107. [26] R. Doll, Occupational cancer: a hazard for epidemiologi sts, Int. J. Epidemiol. 14 (1985) 22-31. [27] H. Checkoway, N. Pearce, D. Crawford-Brown, Research Meth ods in Occupational Epidemiology, Oxford University Press, New York, 1989. [28] A. Blair, P. Stewart, M. O 'Berg, W. Gaffey, J. Walrath, J. Ward, R. Bales, S. Kaplan, D. Cubit, Mortality among industrial work ers exposed to form aldehyde, J. Natl. Cancer Inst. 76 (1986) 1071-1084. [29] M. Hauptm ann, J. Lubin, P. Stewart, R. Hayes, A. Blair, M or tality from lymphohematopoietic malignancies among work ers in form aldehyde industries, J. Natl. Cancer Inst. 95 (2003) 1615-1623. [30] M. Hauptm ann, J. Lubin, P. Stewart, R. Hayes, A. Blair, M ortality from solid cancers among workers in formaldehyde industries, Am. J. Epidemiol. 159 (2004) 1117-1130. [31] A. Blair, P. Stewart, D. Zaebst, L. Pottern, J. Zey, T. Bloom, B. Miller, E. Ward, J. Lubin, M ortality study of industrial workers exposed to acrylonitrile, Scand. J. Work H ealth Environ. 24 (suppl 2) (1998) 25-41. [32] G. Marsh, A. Youk, J. Collins, Rvaluation of lung cancer risk in the acrylonitrile cohort study of the National Cancer Institute and the National Institute for Occupational Safety and Health, Scand. J. W ork Health Environ. 27 (2001) 5-13. [33] G. Marsh, A. Youk, Rvaluation of mortality risks from leukemia in the form aldehyde cohort study of the National Cancer Institute, Regul. Toxicol. Pharm. 40 (2004) 113-124. [34] G. Marsh, A. Youk, Rvaluation of mortality risks from nasopharyngeal cancer in the formaldehyde cohort study of the National Cancer Institute, Regul. Toxicol. Pharm. 42 (2005) 275283. [35] P. Enterline, Pitfalls in epidem iological research, J. Occup. Med. 18(1976) 150-156. Please die ihis article as: Cary M, Marsh el al.< Mortality paneras anion industrial workers exposed to chtotoprcnc and other substances. Chemieo-Bioiogca11nferaettons (2006). dot: 10.1Ot Wj.eb.2006.08.012 Sierra Club v. EPA 18cv3472 NDCA Tier 3/4 ED 002061 00082580-00016