Mortality Patterns Among Industrial Workers Exposed to Chloroprene and Other Substances

Extended Follow-Up

Abstract

Objectives: 

To update the U.S. portion of an historical cohort mortality study of workers with potential exposure to chloroprene (CD) and vinyl chloride (VC) with focus on lung and liver cancer.

Methods: 

Subjects were 6864 workers from two sites with vital status determined through 2017 for 99% of subjects and cause of death for 97.2% of deaths. Historical exposures to CD and VC were estimated quantitatively. We performed external and internal mortality comparisons.

Results: 

External comparisons revealed mostly deficits in deaths; internal comparisons revealed no consistent evidence of exposure-response relationships with CD or VC.

Conclusions: 

Our update continues to support the conclusion that the risk of death from lung or liver cancer is unrelated to exposure to CD or VC at levels experienced by workers in the two U.S. sites.

Chloroprene (2-chlorobuta-1,3-diene) (CD) is a colorless volatile liquid used mainly as a monomer to produce the polymer polychloroprene, a type of synthetic rubber. Polychloroprene is also known by the trade name Neoprene.¹ Epidemiologic studies of CD exposure were prompted by a report of angiosarcoma of the liver among vinyl chloride (VC-a by-product of CD production) workers,² a case report of a liver angiosarcoma among a worker exposed to CD but not to VC³ and the similarity in chemical structural of CD and VC. Earlier CD studies found inconsistent evidence of liver, lung, and skin cancers.^4–11 The International Agency for Research on Cancer (IARC) first classified CD in 1979¹² as Group 3 (not classifiable as to its carcinogenicity to humans), and this Group 3 classification remained following the 1987 reevaluation.¹³ In 1999, CD was reclassified from Group 3 to Group 2B (possibly carcinogenic to humans)¹⁴ based on new bioassays in rats and mice that revealed sufficient evidence of carcinogenicity,^15,16 and two epidemiology studies^6,9 that did not reveal consistent evidence of cancer at any site. Those two epidemiological studies and those conducted earlier had inherent methodological limitations that cast doubt on their conclusions about the risk of human cancers.

In 2007, two of the authors (GM, JB) and colleagues published results of an international historical cohort study of workers potentially exposured to CD.^17,18 The cohort comprised 12,430 workers with some work experience at one of two U.S. industrial sites (Louisville, KY (Plant L, n = 5507) and Pontchartrain, LA (Plant P, n = 1357)) or two European sites (Maydown, Northern Ireland (Plant M, n = 4849) and Grenoble, France (Plant G, n = 717)), with earliest CD production dates ranging from 1942 (Plant L) to 1969 (Plant P). Two sites (Plants L and M) synthesized CD with the acetylene process that produced VC exposures. VC is a known risk factor for angiosarcoma of the liver, hepatocellular carcinoma and other types of cancer.¹³ The authors quantitatively estimated historical exposures for individual workers for CD and VC^19–21 and mortality follow-up was through 2000. This study found no evidence of elevated mortality risks from liver, lung, or any other cancer site or any non-cancer causes of death. The Marsh et al study has been considered as the most comprehensive and methodologically rigourous epidemiologic study of CD-exposed workers conducted to date.²²

In 2010, Environmental Protection Agency (EPA) published the Integrated Risk Information System “Toxicological Review of Chloroprene,” concluding that chloroprene was “likely to be carcinogenic to humans.²³” We believe that the EPA, in their overall causal evaluation of the epidemiological studies of CD, gave considerable weight to our original findings in Plant L of elevated mortality risks for liver cancer in relation to cumulative CD exposure,¹⁸ which they interpreted as evidence of exposure-response. These findings, which were based on internal cohort comparisons, were inflated by the inordinately low liver cancer mortality rates among workers in the lowest CD exposure category that served as the baseline for the internal comparisons.¹⁸ In light of EPA's 2010 evaluation of our original findings and to expand the epidemiological evidence base for CD, we sought to extend our original mortality follow-up of U.S. workers in Plants L and P from 2001 through 2017. By providing increased person-years at risk and observed numbers of deaths, our 17-year update enabled a more reliable and informative evaluation of total and cause-specific mortality patterns, including our original findings for liver cancer in Plant L in relation to CD exposure.

We report here our updated evaluation of general mortality patterns and of mortality in relation to CD and VC exposure among workers in Plants L and P with focus on the a priori cancer sites of interest (liver and lung).

METHODS

Study Sites and Subjects

Details of the original cohort enumeration can be found elsewhere,¹⁷ but in short, the cohort comprised all workers with possible CD exposure from the start of plant operations to the mortality follow-up end date (December 31, 2000). In the current update, we included only the two U.S. (Plants L and P) owned and operated by DuPont/Dow Elastomers LLC (DDE). We started follow-up for Plant L in 1949 to circumvent operational issues in converting cause of death codes from the earlier fifth revision of the International Classification of Diseases (ICD) to later revisions. For example, ICD6, published in 1949, represented a major operational revision to ICD5 converting from alpha-numeric to expanded, strictly numerical coding rules that yielded improved specificity and accuracy in cause of death coding. Also, in ICD6 the combined code section for injuries and their associated accidents was split into two, a chapter for injuries, and a chapter for their external causes.

Vital Status and Cause of Death Ascertainment

All workers not known to be deceased from the original study were traced for vital status using the National Death Index (NDI). For all persons identified as deceased through December 31, 2017, NDI-Plus provided the underlying cause of death code based on the ICD revision in effect at the time of death. Cohort members not found to be dead were assumed alive.

Exposure Assessment

In this mortality update, we did not update the work histories (or exposures) of study members who continued employment beyond the original follow-up end date of December 31, 2000. Details of the methods and results of the original worker job classification, exposure assessment and calculation of worker exposure metrics used in this update can be found elsewhere.^18–21 In short, historical exposures for individual workers were estimated quantitatively for CD and VC and then linked with detailed job histories to compute three metrics of CD and/or VC exposure (Duration of Exposure (Dur), Cumulative Exposure (Cum) and Average Intensity of Exposure (AIE)). These exposure metrics are described by the notation Agent_Metric. For example, CD_Cum indicates cumulative CD exposure. For Plants L and P, respectively, average intensity of CD exposures (summarized as the median value of exposed workers in ppm) were 5.23 and 0.028 and median cumulative exposures (ppm-years) were 18.35 and 0.133. For Plant L, VC_AIE (summarized as the median (25^th percentile, 75^th percentile) value of exposed workers in ppm) was 1.54 (0.27, 3.00) and median VC-Cum exposures (ppm-years) was 1.54 (0.33, 9.23).

We also computed other forms of the CD_Metrics and VC_Metrics using a 15-year lag period. Details of the lagged exposure analysis are provided by Marsh et al.¹⁸ A 15-year lag is commonly used in cohort analysis of solid tumors, such as the National Cancer Institute cohort study of formaldehyde-exposed workers.²⁴

Statistical Analysis

Details of the statistical methods used in our cohort analyses based on external and internal comparisons are provided in our original reports^17,18 and are summarized below. We conducted all statistical tests using a two-sided 0.05 significance level and no adjustments were made for multiple comparisons.

General Mortality—External Comparisons

Using the Occupational Cohort Mortality Program (OCMAP)²⁵ we calculated Standardized Mortality Ratios (SMRs) and their 95% confidence intervals (CI) to evaluate the total and cause-specific mortality of subjects from the two U.S. CD plants during their updated study periods (Table 1). We obtained U.S. and local county reference rates from the Mortality and Population Data System (MPDS) developed by UPitt.^26,27 The Appendix shows the cause of death categories and corresponding ICD codes used in our analysis. This listing includes the a priori cancer sites of interest to this study: malignant neoplasms of the “biliary passages and liver primary (liver cancer)”, “respiratory system” and “bronchus, trachea and lung”.

TABLE 1 - Key Features of Cohort Study Design

Characteristic	Plant L (Louisville, KY)	Plant P (Pontchartrain, LA)	Both Plants
Subjects	5,507	1,357	6,864
Person-years	245,218	50,602	295,820
Earliest hire date	1942	1962	1942
CD production dates	1942–72a	1969–date	—
CD production process	Acetylene	Butadiene	—
Observation period	1949–17b	1962–17	—
Maximum observation period through 2017	69 y	56 y	69 y

^aMonomer production ended in 1972, CD currently in use at plant.

^bDates chosen to avoid fifth revision of International Classification of Diseases (ICD).

Exposure-Response Based on Internal and External Comparisons

We fit multiplicative relative risk models of the form λ(t) = λ₀(t) exp{x(t)β} to internal cohort rates to evaluate the association between the CD and VC exposure metrics and risk of death from all cancers combined, respiratory system cancer (RSC) and liver cancer adjusting for the potential confounders (age, time period, sex and worker type).^28,29β was estimated using the conditional logistic regression programs in R^30,31 and SAS.³² To account for exposure heterogeneity between plants, all models were plant-specific and liver cancer models were confined to Plant L. To parallel the RR analyses, we calculated corresponding SMRs and their 95%CIs.

RESULTS

Tables 1 and 2 provide the key features of the two facilities and corresponding worker cohorts. The 2017 update added 47,299 and 19,942 person-years of observation and 1399 and 214 new deaths to the Plant L and Plant P cohorts, respectively. We identified 4118 deaths and underlying cause of death for 4004 or 97.2%. We ascertained cause of death for 97.0% of workers in Plant L and 99.7% in Plant P. No workers were untraced in Plant P and only 0.2% were untraced in Plant L.

TABLE 2 - Distribution of CD Cohort by Selected Study Factors

	Plant L (Louisville, KY)		Plant P (Pontchartrain, LA)		Both Plants
Characteristic	No.	%	No.	%	No.	%
Subjects	5,507	44.3	1,357	10.9	6,864	100.0
Race
White	3,425	62.1	698	51.4	4,123	60.7
Nonwhite	568	10.3	175	12.9	743	10.8
Unknown	1,514	27.6	484	35.7	1,998	29.1
Sex
Male	4,895	88.9	1,108	81.6	6,003	87.5
Female	612	11.1	249	18.4	861	12.5
Worker pay typea
Blue collar	5,317	96.6	947	69.8	6,264	91.3
White collar	190	3.4	410	30.2	600	8.7
Worker service type
Short-term (<5 y)	2,615	47.5	426	31.4	3,041	44.3
Long-term (5+ y)	2,892	52.5	931	68.6	3,823	55.7
Vital status (as of December 31, 2017)
Alive	1,696	30.8	1,041	76.7	2,737	39.9
Assumed	(1,696)	(100.0)	(837)	(100.0)	(2,737)	(100.0)
Dead	3,802	69.0	316	23.2	4,118	60.0
Cause of death known	(3,689)	(97.0)	(315)	(99.7)	(4,004)	(97.2)
Cause of death unknown	(113)	(3.0)	(1)	(0.3)	(114)	(2.8)
Untraceable	9	0.2	0		9	0.1
Working status (December 31, 2000b)
Active	380	6.9	418	30.8	798	11.6
Separated	5,127	93.1	912	67.2	6,039	88.0
Died while employed	0	0	27	2.0	27	0.4
Age at hire
<20	339	6.2	112	8.3	451	6.6
20–29	3,280	59.6	839	61.8	4,119	60.0
30+	1,888	34.3	406	29.9	2,294	33.4
Duration of employment (y)
<5	2,615	47.5	426	31.4	3,041	44.3
5–19	1,100	20.0	459	33.8	2,264	33.0
20+	1,792	32.5	472	34.8	1,559	22.7
Time since first employment (y)
<20	274	5.0	58	42.4	332	4.8
20–29	611	11.1	203	30.0	814	11.9
30+	4,622	83.9	1,096	35.7	5,718	83.3
CD exposure status
Unexposed	37	0.7	99	7.3	136	2.0
Exposed	5,470	99.3	1,258	92.7	6,728	98.0
VC exposure status
Unexposed	4,257	77.3	n/a		5,614	81.8
Exposed	1,250	22.7			1,250	18.2

n/a, not applicable.

^aPay type = blue collar if blue collar duration of employment > white collar duration of employment, else pay type = white collar. Used as time dependent variable in statistical models.

^bWork histories were not updated in current study and remain as in original study.

The CD cohort included mainly white, blue collar (wage earning) males who terminated work prior to the original follow-up end date (2000). Plant L included a larger percentage of short-term workers (<5 y). The age at hire for most workers was 20 to 29 years. About one-third of each cohort worked for more than 20 years or were traced for more than 30 years. Exposure to CD occurred in nearly all (>92%) workers at each plant. Only Plant L workers were exposed to VC (22.7% exposed). All VC-exposed workers in Plant L had exposure to CD.

General Mortality—External Comparisons

Tables 3 and 4 (Plants L and P, respectively) show for the corresponding total updated study period, observed deaths and SMRs based on both national and local county mortality rates. For Plant L (Table 3), U.S. rate-based SMRs are generally larger, indicating the generally higher local county rates. We observed statistically significant local county rate-based deficits in deaths for all causes of death (SMR = 0.74, 95%CI = 0.72 to 0.76) and all cancers (SMR = 0.75, 95%CI = 0.70 to 0.80). We also observed local rate based-deficits in deaths for almost all other malignant and nonmalignant categories, including a statistically significant 27% RSC deficit (358 deaths, SMR = 0.73, 95%CI = 0.65 to 0.81), a similar deficit for cancer of the bronchus, trachea or lung (340 or 95% of RSC, SMR = 0.72, 95%CI = 0.65 to 0.80) and liver cancer (31 deaths, SMR = 0.95, 95%CI = 0.65 to 1.35). Included among the 31 liver cancer deaths were: seven “liver primary” (ICD9 = 155.0); four “intrahepatic bile duct” (ICD9 = 155.1, ICD10 = C24.0); three “gall bladder” (ICD9 = 156.0); five “extrahepatic bile duct” (ICD9 = 156.1, ICD10 = C24.0); four “liver cell carcinoma” (ICD10 = C22.0); seven “hepatoblastoma” (ICD10 = C22.9); one “malignant neoplasm of overlapping sites of biliary tract” (ICD10 = C24.8) and one “biliary tract, unspecified” (ICD10 = C24.9). Only one statistically significant local rate-based mortality excess was observed in Plant L. We found a statistically significant 60% excess for breast cancer (27 observed deaths, SMR = 1.60, 95%CI = 1.05 to 2.33). Two of the cases were males and all 27 deaths were coded as “malignant neoplasm of breast, unspecified site”.

TABLE 3 - Observed (Obs) Deaths and SMRs for Selected Causes of Death Total Plant L (Louisville, KY) Cohort, U.S. and Local County Comparisons, 1949–2017^a

		U.S.		Local County
Cause of Death (Ninth Revision ICD Codes)	Obs	SMR	95%CI	( )b SMR	95%CI
All Causes of Death (001–999)	3802	0.85∗∗	0.83–0.88	(3756) 0.74∗∗	0.72–0.76
All Cancer (140–208)	974	0.89∗∗	0.84–0.95	0.75∗∗	0.70–0.80
Buccal Cavity and Pharynx (140–149)	19	0.85	0.51–1.32	0.61∗	0.37–0.96
Digestive Organs and Peritoneum (150–159)	241	0.89	0.78–1.01	0.80∗∗	0.70–0.90
Esophagus (150)	28	0.91	0.60–1.31	0.69∗	0.46–1.00
Stomach (151)	28	0.81	0.54–1.17	0.96	0.64–1.39
Large Intestine (153)	96	1.06	0.86–1.30	0.88	0.71–1.07
Rectum (154)	16	0.87	0.50–1.41	0.81	0.46–1.31
Biliary Passages & Liver Primary (155, 156)	31	1.06	0.72–1.51	0.95	0.65–1.35
Pancreas (157)	36	0.64∗∗	0.45–0.88	0.60∗∗	0.42–0.83
All Other Digestive (152, 158, 159)	6	0.60	0.22–1.30	0.53	0.20–1.16
Respiratory System (160–165)	358	1.01	0.90–1.11	0.73∗∗	0.65–0.81
Larynx (161)	13	1.10	0.59–1.88	0.76	0.41–1.30
Bronchus, Trachea, Lung (162)	340	1.0	0.89–1.11	0.72∗∗	0.65–0.80
All Other Respiratory (160, 163, 164, 165)	5	1.67	0.54–3.90	1.34	0.43–3.13
Breast (174, 175)	27	1.70∗	1.12–2.47	1.60∗	1.05–2.33
All Uterine (female only) (179, 180, 181, 182)	3	0.69	0.14–2.02	0.63	0.13–1.85
Prostate (Males Only) (185)	84	0.74∗∗	0.59–0.92	0.72∗∗	0.58–0.89
Kidney (189.0,189.1,189.2)	26	1.02	0.67–1.50	0.87	0.57–1.28
Bladder and Other Urinary Organs (188, 189.3, 189.4, 189.8, 189.9)	25	0.74	0.48–1.09	0.67∗	0.44–0.99
Malignant Melanoma of Skin (172)	8	0.54	0.23–1.06	0.55	0.24–1.09
Central Nervous System (191, 192)	22	0.94	0.59–1.42	0.86	0.54–1.30
Lymphatic-Hematopoietic Tissue (200–208)	84	0.80∗	0.64–0.99	0.74∗∗	0.59–0.92
Hodgkin's Disease (201)	4	0.81	0.22–2.08	0.74	0.20–1.89
Non-Hodgkin's Lymphoma (200, 202.0, 202.1, 202.8, 202.9)	31	0.81	0.55–1.15	0.77	0.52–1.09
Leukemia and Aleukemia (204–208)	35	0.84	0.59–1.17	0.78	0.54–1.08
All Other Lymphopoietic Tissue (202.2, 202.3, 202.4, 202.5, 202.6, 203)	14	0.67	0.37–1.13	0.62	0.34–1.03
All Other Malignant Neoplasms (171, 173, 195–199)	73	0.82	0.65–1.03	0.75∗	0.59–0.95
Diabetes (250)	84	0.83	0.66–1.03	(84) 0.71∗∗	0.57–0.88
Cerebrovascular Disease (430–438)	213	0.80∗∗	0.69–0.91	(212) 0.72∗∗	0.62–0.82
All Heart Disease (390–398, 402, 404, 410–429)	1219	0.80∗∗	0.76–0.85	(1211) 0.74∗∗	0.70–0.78
Nonmalignant Respiratory Disease (460–519)	326	0.77∗∗	0.69–0.86	(326) 0.63∗∗	0.56–0.70
Ulcer of Stomach and Duodenum (531–533)	11	0.87	0.44–1.56	(9) 0.78	0.36–1.48
Cirrhosis of Liver (571)	40	0.59∗∗	0.42–0.80	(40) 0.54∗∗	0.39–0.74
Nephritis and Nephrosis (580–589)	74	1.14	0.89–1.43	(72) 0.79∗	0.62–0.99
All External Causes of Death (E800–999)	170	0.71∗∗	0.61–0.83	(158) 0.63∗∗	0.54–0.74
Accidents (E800–949)	108	0.68∗∗	0.56–0.83	(102) 0.66∗∗	0.54–0.80
Suicides (E950–959)	42	0.82	0.59–1.11	(40) 0.68∗	0.49–0.93
Homicides and Other External (E960–978, E980–999)	20	0.67	0.41–1.03	(16) 0.43∗∗	0.25–0.70
Unknown Causes	113			(108) 113

^aObservation period is 1960–2017 for all causes combined and nonmalignant causes of death based on local comparisons.

^bObserved number of deaths during 1960–2017 study period.

^∗P < 0.05.

^∗∗P < 0.01.

TABLE 4 - Observed (Obs) Deaths and SMRs for Selected Causes of Death, Plant P (Pontchartrain, LA) Cohort, U.S. and Local County Comparisons, 1962–2017

		U.S.		Local County
Cause of Death (Ninth Revision ICD Codes)	Obs	SMR	95%CI	SMR	95%CI
All Causes of Death (001–999)	316	0.63∗∗	0.56–0.71	0.58∗∗	0.51–0.65
All Cancer (140–208)	92	0.69∗∗	0.56–0.85	0.64∗∗	0.52–0.78
Digestive Organs and Peritoneum (150–159)	20	0.60∗	0.37–0.93	0.55∗∗	0.34–0.86
Esophagus (150)	2	0.45	0.05–1.61	0.46	0.06–1.66
Stomach (151)	2	0.64	0.08–2.31	0.63	0.08–2.28
Large Intestine (153)	10	1.03	0.49–1.90	0.93	0.45–1.71
Rectum (154)	2	1.04	0.13–3.77	1.10	0.13–3.96
Biliary Passages & Liver Primary (155, 156)	1	0.20	0.01–1.10	0.16∗	0.00–0.88
Pancreas (157)	3	0.39	0.08–1.15	0.35	0.07–1.03
Respiratory System (160–165)	32	0.73	0.50–1.03	0.64∗∗	0.44–0.90
Larynx (161)	1	0.73	0.02–4.05	0.59	0.02–3.31
Bronchus, Trachea, Lung (162)	30	0.71	0.48–1.02	0.62∗∗	0.42–0.89
All Other Respiratory (160, 163, 164, 165)	1	3.10	0.08–17.28	2.37	0.06–13.19
Prostate (Males only) (185)	4	0.43	0.12–1.10	0.46	0.13–1.17
Kidney (189.0, 189.1, 189.2)	3	0.83	0.17–2.43	0.70	0.14–2.04
Bladder and Other Urinary Organs (188, 189.3, 189.4, 189.8, 189.9)	1	0.27	0.01–1.48	0.28	0.01–1.58
Malignant Melanoma of Skin (172)	2	0.79	0.10–2.84	0.88	0.11–3.19
Central Nervous System (191, 192)	7	1.88	0.76–3.88	1.84	0.74–3.79
Lymphatic-Hematopoietic Tissue (200–208)	13	0.97	0.52–1.66	0.94	0.50–1.61
Non-Hodgkin's Lymphoma (200, 202.0, 202.1, 202.8, 202.9)	5	0.99	0.32–2.30	0.91	0.30–2.13
Leukemia and Aleukemia (204–208)	4	0.77	0.21–1.96	0.75	0.21–1.93
All Other Lymphopoietic Tissue (202.2, 202.3, 202.4, 202.5, 202.6, 203)	4	1.55	0.42–3.96	1.61	0.44–4.13
All Other Malignant Neoplasms (171, 173, 195–199)	9	0.78	0.36–1.48	0.67	0.31–1.28
Diabetes (250)	6	0.43∗	0.16–0.93	0.42∗	0.15–0.91
Cerebrovascular Disease (430–438)	15	0.70	0.39–1.15	0.63	0.35–1.04
All Heart Disease (390–398, 402, 404, 410–429)	78	0.57∗∗	0.45–0.71	0.50∗∗	0.40–0.62
Nonmalignant Respiratory Disease (460–519)	20	0.46∗∗	0.28–0.72	0.50∗∗	0.30–0.77
Cirrhosis of Liver (571)	3	0.28∗	0.06–0.82	0.34∗	0.07–0.99
Nephritis and Nephrosis (580–589)	6	0.79	0.29–1.72	0.48	0.18–1.05
All External Causes of Death (E800–999)	31	0.66∗	0.45–0.94	0.59∗∗	0.40–0.84
Accidents (E800–949)	21	0.74	0.46–1.13	0.67	0.42–1.02
Suicides (E950–959)	8	0.73	0.32–1.44	0.67	0.29–1.32
Homicides and Other External (E960–978 E980–999)	2	0.26∗	0.03–0.93	0.22∗	0.03–0.81
Unknown Causes (in All Causes category only)	1

^∗P < 0.05.

^∗∗P < 0.01.

For Plant P (Table 4), we observed a local rate-based statistically significant 42% deficit for all causes of death (SMR = 0.58, 95%CI = 0.51 to 0.65) and a statistically significant 36% deficit for all cancers (SMR = 0.64, 95%CI = 0.52 to 0.78). As in Plant L but by a smaller amount, most of the U.S.-based SMRs in Plant P are larger than those based on local parish rates. We also observed deficits in deaths for most of the malignant and nonmalignant cause of death categories examined, and many were statistically significant. Only four local rate-based mortality excesses were observed in Plant P and none was statistically significant. We observed in Plant P a statistically significant 36% deficit in RSC (32 deaths, SMR = 0.64, 95%CI = 0.44 to 0.90). Only one death from liver cancer and no cases of breast cancer were observed in Plant P compared with 2.7 expected deaths (data not shown).

Supplemental Tables S1 and S2, https://links.lww.com/JOM/A836 show SMRs for Plants L and P, respectively, for the original (1949 to 2000) and updated (1949 to 2017) follow-up periods using local county comparisons. For Plant L, the only notable difference between periods is an increase in the SMR for breast cancer from 0.99 to the statistically significant 60% excess noted in Table 3. A notable finding for Plant P is that the first three mortality excesses noted above in Table 4 were larger in the original study but none was statistically significant. Supplemental Table S3, https://links.lww.com/JOM/A836 compares the local county rate-based findings of the 2000 and 2017 periods for the combined U.S. cohort. Aside from 45% and 38% not statistically significant excesses for “all other respiratory system cancers” and breast cancer, respectively, we observed deficits in deaths for all other categories examined.

Supplemental Tables S4 to S6, https://links.lww.com/JOM/A836 show local county rate-based observed deaths and SMRs for all cancers, RSC and liver cancer, respectively, by selected study factors and plant (separate and combined). We observed deficits in deaths for most study variables and subcategories examined, many of which were statistically significant. A small number of SMRs in Tables S4 to S6 were elevated and these were slight not statistically significant. Overall, 1054 of 1066 cancer deaths in Plants L and P occurred among CD-exposed persons, yielding a statistically significant 26% deficit (SMR = 0.74, 95%CI = 0.69 to 0.78). Tables S4 to S6 provide no evidence that mortality from all cancers, RSC or liver cancer are related any of the surrogate CD or VC exposure metrics. Table S7, https://links.lww.com/JOM/A836 shows data for breast cancer in the format of Tables S4 to S6. The breast cancer SMR for Plant L females was of borderline statistical significance (SMR = 1.56, 95%CI = 1.00 to 2.32) and when combined with the Plant P, the total cohort reveals only a 38% excess in breast cancer that is not statistically significant. The breast cancer deaths in Plant L concentrated in whites, females, blue-collar workers, workers with 30 or more years since first employment and workers exposed to CD or unexposed to VC. Short-term (<5 y) and long-term (5+ y) workers had similar close to 60% excesses in breast cancer that were not statistically significant.

Exposure-Response Based on Internal and External Comparisons

Results of our CD exposure-response analyses for Plants L and P are shown in Tables 5 and 6, respectively. To evaluate further the statistically significant 1.6-fold mortality excess observed for breast cancer in Plant L (Table 3), we included breast cancer in Table 5. Shown for the internal comparisons are exposure category-specific numbers of observed deaths (cases) and corresponding non-cases. External comparisons include exposure category-specific numbers of person-years at risk.

TABLE 5 - Exposure-Response Analysis for Chloroprene and Selected Cancer Sites by Exposure Metric, Plant L (Louisville, KY) Cohort, Relative Risks (RR) and Standardized Mortality Ratios (SMR)

	Internal Rate Analysis				External Rate Analysis
Metrica	Deaths	Noncasesb	RRc (95%CI)	P value	Pyrsd	SMR-L (95%CI)
All cancers combined
CD_Dur (y)
<10	518	89,393	1.00	Global = 0.42	162585	0.71∗∗ (0.65–0.77)
10–19	93	12,593	1.09 (0.87–1.39)	Trend = 0.21	35877	0.71∗∗ (0.57–0.87)
20+	363	49,880	1.09 (0.95–1.26)		46755	0.83∗∗ (0.75–0.92)
CD_AIE (ppm)
<3.6216	302	46,572	1.00	Global = 0.08	96080	0.73∗∗ (0.65–0.82)
3.6216–8.1245	223	27,806	1.26∗ (1.04–1.53)	Trend = 0.79	32769	0.91 (0.80–1.04)
8.1246–15	142	20,306	1.05 (0.85–1.31)		31995	0.73∗∗ (0.61–0.85)
16+	307	51,182	1.04 (0.87–1.24)		84375	0.71∗∗ (0.64–0.80)
CD_Cum (ppm-years)
<4.747	282	45,265	1.00	Global = 0.10	90636	0.73∗∗ (0.65–0.82)
4.747–55.918	253	39,971	1.03 (0.86–1.23)	Trend = 0.40	71287	0.70∗∗ (0.62–0.80)
55.919–164.052	229	30,779	1.23∗ (1.02–1.49)		47344	0.86∗ (0.76–0.98)
164.053+	210	29,851	1.02 (0.84–1.24)		35950	0.73∗∗ (0.64–0.84)
Respiratory system cancer
CD_Dur (y)
<10	196	33,042	1.00	Global = 0.98	162585	0.72∗∗ (0.62–0.83)
10–19	30	4,689	0.99 (0.66–1.48)	Trend = 0.85	35877	0.63∗∗ (0.43–0.90)
20+	132	20,474	0.98 (0.78–1.23)		46755	0.77∗∗ (0.64–0.91)
CD_AIE (ppm)
<3.6216	96	17,883	1.00	Global = 0.31	96080	0.62∗∗ (0.50–0.75)
3.6216–8.1245	83	11,348	1.28 (0.93–1.76)	Trend = 0.29	32769	0.87 (0.69–1.07)
8.1246–15	49	8,179	1.04 (0.72–1.50)		31995	0.67∗∗ (0.49–0.88)
16+	130	20,795	1.24 (0.93–1.66)		84375	0.80∗∗ (0.67–0.95)
CD_Cum (ppm-years)
<4.747	95	17,565	1.00	Global = 0.04	90636	0.66∗∗ (0.53–0.80)
4.747–55.918	97	15,899	1.08 (0.81–1.45)	Trend = 0.89	71287	0.71∗∗ (0.57–0.86)
55.919–164.052	96	12,368	1.37∗ (1.02–1.85)		47344	0.94 (0.76–1.15)
164.053+	70	12,373	0.88 (0.63–1.23)		35950	0.65∗∗ (0.51–0.82)
Liver cancera
CD_Dur (y)
<10	15	2,402	1.00	Global = 0.19	162585	0.83 (0.46–1.37)
10–19	6	357	2.44 (0.75–6.92)	Trend = 0.55	35877	1.77 (0.65–3.84)
20+	10	1375	1.21 (0.48–2.96)		46755	0.91 (0.44–1.67)
CD_AIE (ppm)
<3.6216	9	1,552	1.00	Global = 0.34	96080	0.76 (0.35–1.45)
3.6216–8.1245	8	750	2.45 (0.76–7.75)	Trend = 0.43	32769	1.34 (0.58–2.63)
8.1246–15	5	526	1.84 (0.44–6.80)		31995	1.09 (0.35–2.55)
16+	9	1,306	1.69 (0.53–5.42)		84375	0.90 (0.41–1.71)
CD_Cum (ppm-years)
<4.747	9	1,400	1.00	Global = 0.18	90636	0.87 (0.40–1.66)
4.747–55.918	6	1,203	0.84 (0.24–2.70)	Trend = 0.18	71287	0.65 (0.24–1.42)
55.919–164.052	10	849	2.39 (0.82–7.09)		47344	1.54 (0.74–2.83)
164.053+	6	682	1.57 (0.41–5.69)		35950	0.92 (0.34–2.01)
Breast cancera
CD_Dur (y)
<10	16	539	1.00	Global = 0.22	162585	1.34 (0.76–2.17)
10–19	4	72	2.01 (0.46–6.74)	Trend = 0.14	35877	1.77 (0.48–4.53)
20+	7	233	1.96 (0.63–5.54)		46755	2.66∗ (1.07–5.47)
CD_AIE (ppm)
<0.602	13	332	1.00	Global = 0.16	533456	1.29 (0.69–2.20)
0.603–7.12	7	177	2.81 (0.86–8.52)	Trend = 0.18	70050	2.92∗ (1.18–6.03)
7.12+	7	316	1.70 (0.50–5.43)		121822	1.60 (0.64–3.30)
CD_Cum (ppm-years)
<0.00541	7	172	1.00	Global = 0.32	23142	1.48 (0.59–3.04)
0.00541–2.40	6	173	1.07 (0.28–3.98)	Trend = 0.18	49351	1.14 (0.42–2.49)
2.41–36.6	7	178	2.73 (0.70–10.88)		74704	2.63∗ (1.06–5.41)
36.7+	7	302	1.79 (0.41–7.88)		98021	1.68 (0.67–3.46)

^aAnalyzed using proc logistic exact statement.

^bThe number of persons in decedent's risk set used in calculation of RR.

^cAlso adjusted for sex and worker pay type.

^dThe number of person-years used in calculation of SMR.

^∗P < 0.05.

^∗∗P < 0.01.

TABLE 6 - Exposure-Response Analysis for Chloroprene and Selected Cancer Sites by Exposure Metric, Plant P (Pontchartrain, LA) Cohort, Relative Risks (RR) and Standardized Mortality Ratios (SMR)

	Internal Rate Analysis				External Rate Analysis
Metrica	Deaths	Noncasesb	RRa (95%CI)	P value	Pyrsc	SMR-L (95%CI)
All cancers combined
CD_Dur (y)
<10	33	1,339	1.00	Global = 0.10	28761	0.67∗ (0.46–0.94)
10–19	23	997	1.06 (0.67–1.67)	Trend = 0.63	11697	0.56∗∗ (0.36–0.84)
20+	36	1,403	0.64 (0.35–1.15)		10145	0.66∗∗ (0.46–0.91)
CD_AIE (ppm)
<0.0017	40	1,776	1.00	Global = 0.07	25819	0.52∗∗ (0.37–0.70)
0.0017–0.1329	9	272	2.85∗ (1.25–6.53)	Trend = 0.12	4364	1.12 (0.51–2.13)
0.1330–0.8174	30	1241	1.83 (0.98–3.44)		11671	0.71 (0.48–1.02)
0.8175+	13	450	1.48 (0.71–3.09)		8748	0.74 (0.40–1.27)
CD_Cum (ppm-years)
<0.0193	29	1,148	1.00	Global = 0.25	21072	0.63∗∗ (0.42–0.90)
0.0193–1.8944	21	1,014	0.74 (0.41–1.35)	Trend = 0.20	12231	0.51∗∗ (0.31–0.78)
1.8945–16.1918	11	493	1.16 (0.54–2.50)		8720	0.63 (0.31–1.13)
16.1919+	31	1,084	1.46 (0.76–2.82)		8579	0.69∗ (0.47–0.98)
Respiratory system cancer
CD_Dur (y)
<10	4	363	1.00	Global = 0.09	28761	0.25∗∗ (0.07–0.63)
10–19	12	333	2.51 (0.74–8.48)	Trend = 0.03∗	11697	0.84 (0.43–1.46)
20+	16	623	3.72 (0.98–14.10)		10145	0.81 (0.46–1.31)
CD_AIE (ppm)
<0.0017	10	564	1.00	Global = 0.09	25819	0.38∗∗ (0.18–0.69)
0.0017–0.1329	2	53	5.15 (0.95–27.97)	Trend = 0.05∗	4364	0.79 (0.1–2.85)
0.1330–0.8174	14	516	3.62∗ (1.10–11.99)		11671	0.93 (0.51–1.56)
0.8175+	6	186	2.90 (0.82–10.26)		8748	0.96 (0.35–2.08)
CD_Cum (ppm-years)
<0.0193	6	336	1.00	Global = 0.23	21072	0.39∗ (0.14–0.84)
0.0193–1.8944	7	299	1.42 (0.44–4.59)	Trend = 0.05∗	12231	0.49 (0.20–1.02)
1.8945–16.1918	3	170	1.55 (0.32–7.42)		8720	0.49 (0.10–1.43)
16.1919+	16	514	3.08 (0.92–10.36)		8579	0.96 (0.55–1.56)
Respiratory system cancer (revised)
CD_Dur (y)
<15.3	8	500	1.00	Global = 0.04∗	34351	0.37∗∗ (0.16–0.72)
15.3–24.5	13	321	3.62∗ (1.18–11.10)	Trend = 0.03∗	10109	0.81 (0.43–1.39)
24.6+	11	498	3.43 (0.91–12.93)		6142	0.87 (0.44–1.56)
CD_AIE (ppm)
<0.0642	11	596	1.00	Global = 0.13	28178	0.39∗∗ (0.20–0.70)
0.0642–0.632	10	324	4.18∗(1.28–13.62)	Trend = 0.05∗	9891	1.07 (0.51–1.96)
0.632+	11	399	2.54 (0.84–7.69)		12532	0.85 (0.42–1.51)
CD_Cum (ppm-years)
<1.26	11	625	1.00	Global = 0.05∗	31969	0.38∗∗ (0.19–0.67)
1.26–17.8	10	403	2.70 (0.88–8.26)	Trend = 0.005∗∗	14543	0.77 (0.37–1.41)
17.8+	11	291	5.27∗∗ (1.62–17.18)		4090	1.40 (0.70–2.51)

^aAlso adjusted for sex and worker pay type.

^bThe number of persons in decedent's risk set used in calculation of RR.

^cThe number of person-years used in calculation of SMR.

^∗P < 0.05.

^∗∗P < 0.01.

For Plant L (Table 5), neither mortality comparison revealed evidence of a positive association between CD exposure and mortality from all cancers or RSC as indicated by the corresponding not statistically significant RR trend tests. While we observed statistically significantly elevated RRs in a few exposure categories for both cause of death categories, these were offset by deficits in deaths or only slight excesses in the higher exposure categories of that exposure metric. For liver cancer, we observed elevated RRs in most non-baseline exposure categories, however, none was was statistically significant.

Table 5 also shows that corresponding SMRs for the three cause of death categories were consistently below 1.00 and mostly statistically significant. Of particular note is that the elevated RRs for these categories reflect the exceedingly low SMRs associated with the corresponding baseline categories (ie, for any non-baseline exposure category, the RR is roughly equal to the non-baseline SMR divided by the corresponding baseline SMR). Moreover, almost all the observed elevations in RRs were based on comparisons of a non-baseline exposure category associated with a small deficit in deaths to a baseline category associated with a larger deficit in deaths as indicated by the corresponding SMRs.

For breast cancer, the post hoc cause of death category of interest, Table 5 shows elevated but not statistically significant RRs for all the non-baseline categories of the metrics considered with no evidence of a positive exposure-response relationship (based on all breast cancer deaths, we categorized CD_Cum into approximate quartiles, CD_AIE into approximate tertiles). Unlike the a priori cause of death categories in Table 5, SMRs are elevated in all baseline and non-baseline categories but with no evidence of positive exposure-response relationship.

For Plant P (Table 6), neither mortality comparison revealed evidence of a positive association between CD exposure and mortality from all cancers combined as indicated by the not statistically significant RR trend tests. As in Plant L, the external comparisons revealed deficits in deaths for all categories of the CD exposure metrics considered. For RSC, the original exposure categorizations yielded an unbalanced distribution of observed deaths, thus, we recategorized each of the CD exposure metrics to form nearly balanced tertiles and show results based on both categorizations in Table 6.

The internal comparisons for RSC revealed large, often statistically significant RRs for all non-baseline categories of CD exposure using either categorization scheme, and in some cases evidence of a positive exposure-response relationship indicated by the statistically significant trend test. However, in stark contrast to these findings, the external comparisons revealed mostly deficits in RSC deaths for all categories of the exposure metrics considered under both categorization schemes, and with the possible exception of CD_Cum under the new categorization, no evidence of a positive exposure-response relationship. Of particular note in Table 6 are the very large, statistically significant deficits in deaths in the baseline categories of all CD exposure metrics under both categorization schemes, ranging from 61% ((CD_Cum-original categorization (SMR = 0.39, 95%CI = 0.14 to 0.84) and CD_AIE-new categorization (SMR = 0.39, 95%CI = 0.20 to 0.70) to 75% (CD_Dur-original categorization (SMR = 0.25, 95%CI = 0.07 to 0.63)). As noted and discussed above for the parallel analyses in Plant L, these large and statistically significant deficits in RSC deaths among Plant P workers in the baseline exposure categories contributed heavily to the elevated RRs in the internal comparisons rendering a meaningful interpretation of their absolute values and associated CI, global and trend tests difficult.

Table 7 provides results of our analyses of VC exposure-response for Plant L. Because of the much higher percentage of Plant L workers unexposed to VC, we were able to create an “unexposed” category as the baseline for the internal comparisons. For all cancers and RSC, RRs and related SMRs are all below 1.0 for each category of the VC exposure metrics considered and there is no evidence of any positive exposure-response relationships. In fact, in each analysis we observed statistically significant inverse trend tests. Aside from a not statistically significant elevated RR of 1.28 and SMR of 1.47 for the lower exposure category of VC_AIE, we observed similar findings for liver cancer in Table 7, however trend tests were not statistically significant. For breast cancer, 24 of the 27 deaths occurred among unexposed workers yielding a not statistically significant 54% excess in deaths based on the external comparison. The three deaths among VC-exposed workers led to an RR of 2.04 and SMR of 2.37, both not statistically significant. While not shown, the findings from our 15-year lagged exposure-response analyses of RSC, liver cancer and breast cancer mortality for CD and VC (Plant L only) did not differ substantially from those in the unlagged analyses.

TABLE 7 - Exposure-Response Analysis for Vinyl Chloride Monomer and Selected Cancer Sites by Exposure Metric, Plant L (Louisville, KY) Cohort, Relative Risks (RR) and Standardized Mortality Ratios (SMR)

	Internal Rate Analysis				External Rate Analysis
Metrica	Deaths	Noncasesb	RRc (95%CI)	P-value	Pyrsd	SMR-L (95%CI)
All cancers combined
VCM_Dur (y)
Unexposed	743	102,820	1.00	Global = 0.03∗	184744	0.78∗∗ (0.72–0.83)
>0–5	134	24,578	0.82∗ (0.68–0.99)	Trend = 0.007∗∗	39210	0.68∗∗ (0.57–0.81)
5–9	21	3,625	0.78 (0.50–1.21)		7161	0.65 (0.41–1.00)
10+	76	14,843	0.77∗ (0.61–0.98)		14104	0.67∗∗ (0.53–0.84)
VCM_AIE (ppm)
Unexposed	743	102,820	1.00	Global = 0.009∗∗	184744	0.78∗∗ (0.72–0.83)
>0–0.27	63	9,452	0.95 (0.73–1.23)	Trend = 0.01∗	14974	0.81 (0.62–1.04)
0.28–1.75	56	12,864	0.64∗∗ (0.49–0.84)		17256	0.52∗∗ (0.39–0.68)
1.751–2	29	7,124	0.77 (0.55–1.07)		6400	0.66∗ (0.44–0.94)
3+	83	13,606	0.87 (0.68–1.12)		21844	0.58∗∗ (0.46–0.72)
VCM_Cum (ppm-years)
Unexposed	743	102,820	1.00	Global = 0.04∗	184744	0.78∗∗ (0.72–0.83)
>0–0.4476	60	10,299	0.83 (0.63–1.08)	Trend = 0.003∗∗	18890	0.70∗∗ (0.53–0.90)
0.4477–1.9482	57	9,538	0.91 (0.69–1.20)		13803	0.66∗∗ (0.50–0.85)
1.9483–14.5832	60	13,142	0.71∗ (0.54–0.93)		16691	0.54∗∗ (0.41–0.70)
14.5833+	54	10,067	0.79 (0.59–1.04)		11091	0.67∗∗ (0.50–0.87)
Respiratory system cancer
VC_Dur (y)
Unexposed	295	40,762	1.00	Global <0.0001∗∗	184744	0.82∗∗ (0.73–0.92)
>0–5	41	9,947	0.57 (0.21–1.07)	Trend <0.0001∗∗	39210	0.52∗∗ (0.38–0.71)
5+	22	7,496	0.41∗∗ (0.27–0.64)		21264	0.38∗∗ (0.24–0.57)
VC_AIE (ppm)
Unexposed	295	40,762	1.00	Global <0.0001∗∗	184744	0.84∗∗ (0.74–0.94)
>0–1.751	30	8,867	0.46∗∗ (0.31–0.67)	Trend <0.0001∗∗	32230	0.42∗∗ (0.29–0.61)
1.751+	33	8576	0.55∗∗ (0.38–0.80)		28244	0.52∗∗ (0.36–0.74)
VC_Cum (ppm-years)
Unexposed	295	40,762	1.00	Global <0.0001∗∗	184744	0.83∗∗ (0.73–0.93)
>0–0.4476	20	4,035	0.66 (0.42–1.05)	Trend <0.0001∗∗	18890	0.59∗ (0.36–0.91)
0.4477–1.9482	15	3,856	0.55∗ (0.33–0.93)		13803	0.44∗∗ (0.25–0.73)
1.9483–14.5832	17	5,324	0.46∗∗ (0.28–0.76)		16691	0.39∗∗ (0.23–0.63)
14.5833+	11	4,228	0.36∗∗ (0.20–0.66)		11091	0.35∗∗ (0.17–0.62)
Liver cancera
VC_Dur (years)
Unexposed	24	3,007	1.00	Global = 1.0	184744	0.99 (0.64–1.48)
>0–5	4	654	0.86 (0.21–2.62)	Trend = 0.89	39210	0.81 (0.22–2.08)
5+	3	473	0.87 (0.16–2.98)		21264	0.85 (0.18–2.48)
VC_AIE (ppm)
Unexposed	24	3,007	1.00	Global = 0.65	184744	0.99 (0.64–1.48)
>0–0.27	3	304	1.28 (0.23–4.57)	Trend = 0.72	14974	1.47 (0.3–4.29)
0.28+	4	823	0.70 (0.17–2.10)		45500	0.57 (0.15–1.45)
VC_Cum (ppm-years)
Unexposed	24	3,007	1.00	Global = 1.0	184744	0.99 (0.64–1.48)
>0–0.19842	4	592	0.90 (0.22–2.78)	Trend = 0.89	32693	0.97 (0.26–2.49)
1.9483+	3	535	0.82 (0.15–2.82)		27781	0.66 (0.14–1.94)
Breast cancera
VC
Unexposed	24	686	1.00	Global = 0.38	184744	1.54 (0.84–1.95)
Exposed	3	139	2.04 (0.32–9.60)		60474	2.37 (0.49–6.93)

^aAnalyzed using proc logistic exact statement.

^bThe number of persons in decedent's risk set used in calculation of RR.

^cAlso adjusted for sex and worker pay type.

^dThe number of person-years used in calculation of SMR.

^∗P < 0.05.

^∗∗P < 0.01.

DISCUSSION

As in our original study, we observed generally consistent total and cause-specific mortality patterns across both U.S. plants including statistically significantly reduced mortality risks for all causes combined, all cancer sites combined, and many other malignant and nonmalignant disease categories, including the two sites of a priori interest (lung and liver). These reduced mortality risks, especially for the chronic diseases examined, were likely impacted by the “healthy worker effect” and the “health worker survivor effect”, although we expected that the extended follow-up period would have diminished the impact of these commonly observed effects.

Of particular note is that we observed no overall excess mortality risk for liver cancer, given that VC exposures occurred in Plant L. Excess liver cancer risks have also been associated with CD exposures in an experimental animal studies¹⁶ and in occupational epidemiology studies in China,⁷ Armenia,⁸ and Russia.⁹ These earlier epidemiology studies had major methodological limitations that cast doubt on their conclusions about human cancer risks.^33,34 Likewise, for liver cancer we found no evidence of a positive exposure-response relationship with any of the CD or VC exposure metrics considered. As noted in the original study, the continued lack of elevated cancer risks among workers exposed to VC is likely the result of the low historical VC exposures in Plant L.^17–21 The higest tertile of cumulative VC exposure in this study was greater than 1.9483 ppm-years (Table 7, liver cancer), considerably lower than the levels found in the Mundt et al evaluation of angiosarcoma of the liver.³⁵ Mundt et al found a statistically significant association between angiosarcoma of the liver (SMR = 2.87) and cumulative VC exposure of 865 ppm-years.

The absence of an excess mortality risk for RSC (and lung cancer) in Plants L or P is also noteable in view of the elevated lung cancer risk reported in the earlier study of French CD production workers.¹¹ As noted in the original study, the absence of an elevated risk for RSC is consistent with experimental animal studies that showed considerable inter-species differences in sensitivity to CD-induced lung tumorigenicity and how human lung cancer risk can be estimated from these findings.^15,36–37

Two features of our results from exposure-response findings for CD and RSC, render meaningful interpretation difficult. First, the evidence for exposure-response in Plant L is considerably weaker than for the Plant P cohort, despite the higher CD exposures in Plant L. Second, particularly in Plant P, we observed major inconsistencies between the results of the internal and external exposure-response analyses. Specifically, our external comparisons for Plant P revealed large deficits in RSC deaths in most all categories of the CD exposure metrics considered, and the largest deficits occurred among the least exposed workers that served as the baseline category for the RR calcuations. In this situation, a small deficit in deaths was essentially compared to a larger deficit in deaths resulting in an elevated RR that belies interpretation as a meaningful “excess” risk (ie, can the ratio of two deficits be interpreted as an excess?). As an example of the phenomenon, our analysis of RSC by CD_AIE in Plant P (Table 6), revealed a statistically significant 3.62-fold excess (RR = 3.62, 95%CI = 1.10 to 11.19) for the exposure category (0.1330 to 0.8174 ppm). The basis of this exposure category-specific “excess” was the comparison of a slight 7% deficit in deaths (SMR = 0.93, 95%CI = 0.51 to 1.56) to an inordinately large, statistically significant 62% deficit in the baseline category (SMR = 0.38, 95%CI = 0.18 to 0.69).

We observed this low baseline mortality rate-based phenomenon in our original CD study¹⁸ and in other cohort studies, including the National Cancer Institure (NCI) studies of formaldehyde^38–41 and acrylonitrile⁴² workers. This phenomenon has also prompted reanalyses and reinterpretation of the NCI cohort data.^43–45 A full and more detailed discussion of the possible explanations for these inconsistences and for the low baseline RSC rates in our CD cohort is provided in our original paper.¹⁷

Although we observed elevated RRs in many exposure categories, we found no compelling evidence of a positive exposure-response relationship in either study plant. The one possible exception with the recategorized CD_Cum in Plant P (Table 6) occurred because the unremarkable pattern of deficits in deaths indicated by exposure category-specific SMRs (<1.26 ppm-years (baseline): SMR=0.38 (p < 0.05); 1.26-17.8 ppm-years: SMR=0.77; 17.8+ ppm-years: SMR=1.40) = 0.39 (P < 0.05); 1.26 to 17.8 ppm-years = 0.77; 17.8 ppm-years = 1.40) led to a similar but highly elevated pattern of RRs due the inordinately low and statistically significant baseline SMR. Because of the inflated RRs stemming from low baseline RSC rates we do not believe that this pattern in RRs reflects a real positive exposure-response relationship.

We observed a new finding in Plant L only of a statistically significant 60% excess in breast cancer based on 27 deaths. In the original 2000 follow-up period, the local county-based SMR for Plant L was 0.99 based on 11 deaths. Breast cancer mortality was elevated in nearly all Plant L subgroups examined (eg, males and females, short-term and long-term workers, and workers exposed and unexposed to VC) and was concentrated in workers with 30 or more years of time since first employment (23/27 deaths, SMR = 1.94, 95%CI = 1.23 to 2.91). However, we did not discern any clear exposure-response patterns in our findings. Additionally, breast cancer mortality was not elevated in the second U.S. plant (Plant P); in fact, we observed no deaths from breast cancer in Plant P in the original or updated follow-up period. Therefore, interpreting the meaning and importance, if any, of our finding for breast cancer is difficult. Although the 60% excess was statistically significant, it is possible that at least part of this excess is due to our inability to adjust for other potential confounding factors, including genetic susceptibility and gene-environment interactions. Breast cancer has strong genetic, environmental, and personal risk factors (eg, smoking, alcohol consumption, obesity, nulliparity, age at first birth, and postmenopausal hormone usage) but is not generally considered to be related to occupational chemical exposures.^46–52 Exposure to endocrine-disrupting chemicals is associated with increased risk of developing breast cancer,⁵³ but there were no known endocrine disrupters used at the facilities. The lack of well established occupational risk factors for breast cancer and the absence of any clear patterns or trends in breast cancer mortality among the Plant L cohort suggest that our finding may be due to factors external to employment at Plant L.

Our cohort study of CD production workers has many underlying strengths that are discussed in detail in our original reports.^17,18 Our update of the U.S. cohort of 6867 workers contributed 295,820 person-years, the majority (160,852 or 54%) of which occurred among workers with 20 or more years of follow-up. Through 2017, we observed 4072 deaths, including 1066 from all cancers combined, increases of 1607 and 374 from the original 2000 follow-up. We also observed an additional 111 and 16 deaths from RSC and liver cancer. With the increased number of observed deaths, our updated study overall had excellent statistical power to detect an overall two-fold or greater excess in RSC or liver cancer, especially in the larger and older Plant L which included 31 of 32 total deaths from liver cancer.

Although we did not update work histories of study members active at the original follow-up date of December 31, 2000, this likely had neglible impact on our updated findings, especially for the larger Plant L, as 97% and 70% of workers from Plant L and Plant P had terminated employment before 2001, respectively. Also, the 15-year lagged exposure analyses effectively discounted most of the exposure that would have occurred during the 17-year updated follow-up period (2001 to 2017). Potential selection bias from the 191 Plant P workers who were known to have transferred to other facilities but for whom detailed work histories were unavailable¹⁷ may have impacted our results, but only slightly considering the small number of workers. As our statistical tests were not adjusted for multiple comparisons, some statistically significant results may be simply chance occurences.

CONCLUSIONS

Our 2017 mortality update of the U.S. portion of our CD cohort study has many strengths and provides new findings from the most definitive study of the human carcinogenic potential of exposure to CD conducted to date. Our update continues to support the conclusion that the risk of death from all cancers or from the sites of a priori interest (lung and liver cancer) is unrelated to exposure to CD or VC at levels experienced by workers in the two U.S. sites. The post hoc elevated risk for breast cancer was confined to Plant L and does not appear to be associated with workplace factors. Additional evaluations of the CD cohort may help to elucidate explanations for the differential mortality risk estimates based on internal and external mortality comparisons.