® EPA
EPA/635/R-13/128b
www.epa.gov/iris
Evaluation of the Inhalation Carcinogenicity of
Ethylene Oxide
APPENDICES
(CASRN 75-21-8)
In Support of Summary Information on the
Integrated Risk Information System (IRIS)
July 2013
NOTICE
This document is a Revised External Peer Review draft. This information is distributed solely for
the purpose of pre-dissemination peer review under applicable information quality guidelines. It has
not been formally disseminated by EPA. It does not represent and should not be construed to
represent any Agency determination or policy. It is being circulated for review of its technical
accuracy and science policy implications.
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC
-------
DISCLAIMER
This document is a preliminary draft for review purposes only. This information is
distributed solely for the purpose of pre-dissemination peer review under applicable information
quality guidelines. It has not been formally disseminated by EPA. It does not represent and
should not be construed to represent any Agency determination or policy. Mention of trade
names or commercial products does not constitute endorsement or recommendation for use.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 ii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS
LIST OF TABLES vi
LIST OF FIGURES ix
LIST OF ABBREVIATIONS xi
APPENDIX A. CRITICAL REVIEW OF EPIDEMIOLOGIC EVIDENCE A-l
A.I. BACKGROUND A-l
A.2. INDIVIDUAL STUDIES A-2
A.2.1. Hogstedt (1988), Hogstedt et al. (1986) A-2
A.2.2. Gardner etal. (1989) A-5
A.2.3. Kiesselbach etal. (1990) A-6
A.2.4. Greenberg et al. (1990) A-7
A.2.5. Steenland etal. (1991) A-9
A.2.6. Teta etal. (1993) A-ll
A.2.7. Benson and Teta (1993) A-13
A.2.8. Stayneretal. (1993) A-14
A.2.9. Wong and Trent (1993) A-16
A.2.10. Bisanti etal. (1993) A-17
A.2.11. Hagmar et al. (1995) and Hagmar et al. (1991) A-18
A.2.12. Norman etal. (1995) A-20
A.2.13. Swaen etal. (1996) A-21
A.2.14. Olsen etal. (1997) A-22
A.2.15. Steenland et al. (2004) A-23
A.2.16. Steenland et al. (2003) A-25
A.2.17. Kardos et al. (2003) A-26
A.2.18. Tompa etal. (1999) A-27
A.2.19. Coggon et al. (2004) A-27
A.2.20. Swaen et al. (2009) A-28
A.3. SUMMARY A-35
A.4. CONCLUSIONS A-54
APPENDIX B. REFERENCES FOR FIGURE 3-3 B-l
APPENDIX C. GENOTOXICITY AND MUTAGENICITY OF ETHYLENE OXIDE C-l
C.I. DNAADDUCTS C-2
C.I.I. Detection of EtO Adducts in In Vitro and In Vivo Systems C-4
C.1.2. In Vitro DNA Binding Studies C-4
C.I.3. In Vivo Studies—Animal Experiments C-5
C.I.4. In Vivo Studies—Human Subjects C-9
C.I.5. DNA Adducts—Summary C-10
C.I.6. EtO-Hemoglobin Adducts C-10
C.2. GENE MUTATIONS C-ll
C.2.1. Bacterial Systems C-ll
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 iii DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
C.2.2. Mammalian Systems C-12
C.2.3. Gene Mutations—Summary C-20
C.3. CHROMOSOMAL ABERRATIONS C-20
C.4. MICRONUCLEUS FORMATION C-23
C.5. SISTER CHROMATID EXCHANGES (SCEs) C-24
C.6. OTHER ENDPOINTS (GENETIC POLYMORPHISM, SUSCEPTIBILITY) C-27
C.7. ENDOGENOUS PRODUCTION OF ETHYLENE AND EtO C-28
C.8. CONCLUSIONS C-33
APPENDIX D. RE ANALYSES AND INTERPRETATION OF ETHYLENE OXIDE
EXPOSURE-RESPONSE DATA D-l
D. 1. BREAST CANCER INCIDENCE BASED ON THE DATA WITH
INTERVIEWS D-4
D.2. BREAST CANCER MORTALITY D-29
D.3. LYMPHOID CANCER MORTALITY (SUBSET OF ALL
HEMATOPOIETIC CANCERS COMBINED) (N = 18,235) D-41
D.4. HEMATOPOIETIC CANCER MORTALITY (ALL HEMATOPOIETIC
CANCERS COMBINED) D-52
D.5. SUMMARY TABLE OF EC0iS FOR DIFFERENT OUTCOMES, USING 2-
PIECE SPLINE MODELS D-63
D.6. SENSITIVITY OF 2-PIECE SPLINE CURVES TO PLACEMENT OF
KNOT D-64
D.7. POSSIBLE INFLUENCE OF THE HEALTHY WORKER SURVIVOR
EFFECT D-65
D.8. POSSIBLE INFLUENCE OF EXPOSURE MIS-MEASUREMENT D-66
D.9. REFERENCES D-68
APPENDIX E. LIFE-TABLE ANALYSIS E-l
APPENDIX F. EQUATIONS USED FOR WEIGHTED LINEAR REGRESSIGN F-1
APPENDIX G. MODEL PARAMETERS IN THE ANALYSIS OF ANIMAL TUMOR
INCIDENCE G-l
APPENDIX H. SUMMARY OF 2007 EXTERNAL PEER REVIEW AND PUBLIC
COMMENTS AND DISPOSITION H-l
APPENDIX I. LIST OF REFERENCES ADDED AFTER THE 2006 EXTERNAL
REVIEW DRAFT 1-1
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 iv DRAFT—DO NOT CITE OR QUOTE
-------
CONTENTS (continued)
APPENDIX J. SUMMARY OF MAJOR NEW STUDIES SINCE THE LITERATURE
CUTOFF DATE J-l
J.I. SYSTEMATIC LITERATURE SEARCH J-l
J.2. REVIEWS OF MAJOR NEW STUDIES PUBLISHED SINCE THE
LITERATURE CUTOFF DATE J-4
J.2.1. Kiranetal. (2010) J-4
J.2.2. Mikoczyetal. (2011) J-6
J.3. REFERENCES J-12
APPENDIX K. DOCUMENTATION OF IMPLEMENTATIONS OF THE 2011
NATIONAL RESEARCH COUNCIL RECOMMENDATIONS K-l
REFERENCES R-l
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 v DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES
A- 1 . Estimated 8-hour time-weighted average ethyl ene oxide exposure, Plant 3 .................. A-3
A-2. Cox regression results for hematopoietic cancer mortality (15-year lag) in males ...... A-24
A-3. Cox regression results for lymphoid cell line tumors (15-year lag) in males ............... A-24
A-4. Exposure assessment matrix from Swaen et al. (2009) — 8-hour TWA exposures
in ppm [[[ A-29
A-5. Epidemiological studies of ethyl ene oxide and human cancer ..................................... A-36
C-l . Levels of endogenous (background) N7-HEG adducts in unexposed human and
experimental rodent tissues [[[ C-30
D-la. Distribution of cases in Cox regression for breast cancer morbidity analysis after
using a 15-year lag [[[ D-6
D-lb. Categorical analysis of breast cancer incidence by deciles (log RR model) .................. D-7
D-lc. Fit of 2-piece log-linear model to breast cancer incidence data, Cox regression ......... D-14
D-ld. Fit of log-linear model to breast cancer incidence data, Cox regression (RR = e(p x
exposure)^ D-15
D-le. Fit of the square root transformation log RR model to breast cancer incidence
data, Cox regression (RR = e(pxsqrt(exposure))) [[[ D-16
D-lf Fit of the log transform model to breast cancer incidence data, Cox regression
g(P x ln(exposure)K D- 1 7
D-lg. Change in -2 log likelihood for log RR models for breast cancer incidence, with
addition of exposure term(s) [[[ D-18
D-lh. Model fit statistics for linear RR models, breast cancer incidence ............................... D-20
D-li. Model coefficients for linear RR models, breast cancer incidence .............................. D-21
D-lj. Supplemental Results: Parameter estimates for exposure variables for categorical
(decile) linear RR model (RR= 1 + P), breast cancer incidence .................................. D-22
-------
LIST OF TABLES (continued)
D-2c. 2-piece log-linear spline, breast cancer mortality, 20-year lag, knot at 700 ppm-
days D-36
D-2d. Log-linear model, breast cancer mortality, 20-year lag D-37
D-2e. Log transform log RR model, breast cancer mortality, 20-year lag D-3 8
D-2f. 2-piece log-linear spline model, breast cancer mortality, 20-year lag, knot at
13,000ppm-days D-39
D-2g. Model results for breast cancer mortality, linear RR models D-40
D-3a. Exposure categories and case distribution for lymphoid cancer mortality D-42
D-3b. Lymphoid cancer mortality results by sex D-43
D-3c. Categorical results for lymphoid cancer mortality (log RR model), men and
women combined D-46
D-3d. Results of 2-piece log-linear spline model for lymphoid cancer mortality, men and
women combined, knot at 100 ppm-days D-47
D-3e. Results of the log transform log RR model for lymphoid cancer mortality, both
sexes combined D-48
D-3f Results of the log-linear model for lymphoid cancer mortality, both sexes
combined D-48
D-3g. Results of 2-piece log-linear spline model for lymphoid cancer mortality, men and
women combined, knot at 1600 ppm-days D-49
D-3h. Supplemental Results: Model fit statistics and coefficients for linear RR models,
lymphoid cancer mortality 50
D-4a. Exposure categories and case distribution for hematopoietic cancer mortality D-53
D-4b. All hematopoietic cancer mortality categorical results by sex (log RR model) D-54
D-4c. Categorical results for all hematopoietic cancer mortality (log RR model), men
and women combined, cumulative exposure with a 15-year lag D-57
D-4d. Results of 2-piece log-linear spline model for all hematopoietic cancer mortality,
men and women combined, cumulative exposure with a 15-year lag D-58
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 vii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF TABLES (continued)
D-4e. Results of log-transform log RR model for all hematopoietic cancer mortality,
men and women combined, cumulative exposure with a 15-year lag D-59
D-4f Results of log-linear model for all hematopoietic cancer morality, men and
women combined, cumulative exposure with a 15-year lag D-60
D-4g. Supplemental Results: Model fit statistics and coefficients for linear RR models,
hematopoietic cancer mortality D-61
D-5a. Summary of ECoi results (in ppm) in current analysis and previous EPA risk
assessment D-64
D-6a. Exposure-response coefficients and ECois based on selection of different knots,
using 2-piece log-linear models D-65
E-l. Extra risk calculation for environmental exposure to 0.0114 ppm (the LECoi for
lymphoid cancer incidence) using the weighted linear regression model based on
the categorical cumulative exposure results of Steenland et al. (2004), reanalyzed
by Steenland for both sexes combined with a 15-year lag, as described in
Section 4.1.1 E-2
G-l. Analysis of grouped data, NTP mice study (NTP, 1987); multistage model
parameters G-l
G-2. Analysis of grouped data, Lynch et al. (1984a); Lynch et al. (1984c) study of male
F344 rats; multistage model parameters G-2
G-3. Analysis of grouped data, Garman et al. (1985) and Snellings et al. (1984) reports
on F344 rats; multistage model parameters G-2
G-4. Time-to-tumor analysis of individual animal data, NTP mice study (NTP, 1987);
multistage-Weibull model parameters G-3
J-1. Disposition of 56 new references identified as potentially relevant J-2
J-2. New epidemiological studies of ethylene oxide and human cancer J-9
K-l. The EPA's implementation of the National Research Council's recommendations
in the ethylene oxide (EtO) carcinogenicity assessment K-2
K-2. National Research Council recommendations that the EPA is implementing in the
long term K-8
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 viii DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES
D-1 a. Likelihoods vs knots, 2-piece linear log RR model for breast cancer morbidity D-9
D-lb. Breast cancer incidence. Plot of the dose-response relationship for continuous
exposure generated using a 2-piece log-linear spline overlaid with a plot using
restricted cubic (logRR) splines D-10
D-lc. Breast cancer incidence. Plot of a log-linear dose-response relationship overlaid
with a dose-response relationship generated using restricted cubic log RR model
with continuous exposure D-ll
D-ld. Breast cancer incidence. Comparison of log-linear curve (log RR = P x cumexp)
with all the data and the log-linear curve (higher slope) after excluding those in
the top 5% of exposure (>27,500 ppm-days) D-ll
D-le. Breast cancer incidence. Plot of a logarithmic transformation log RR dose-
response model (log RR = P x log(cumexp)) overlaid with a dose-response
relationship generated using categorical log RR analyses (deciles) D-12
D-lf Breast cancer incidence. Plot of a square-root transformation log RR dose-
response model overlaid with a dose-response relationship generated using
categorical logRR analyses (deciles) D-13
D-lg. Breast cancer incidence exposure-response curves, linear RR models (units are
ppm-days, 15-year lag) D-19
D-lh. Knot location for Figure D-lg above, 2-piece linear spline model, breast cancer
incidence (units are ppm-days, 15-year lag) D-20
D-2a. Likelihoods vs knots for the 2-piece log-linear model, breast cancer morality D-31
D-2b. Likelihoods vs knots for the 2-piece log-linear model, breast cancer morality D-32
D-2c. Plot of the dose-response relationship of continuous exposure (lagged 20 years)
for breast cancer mortality, with the 2-piece linear spline, the categorical, and the
linear log RR models D-33
D-2d. Plot of the dose-response relationship of continuous exposure (lagged 20 years)
for breast cancer mortality, generated using a logarithmic transformation log RR
model D-34
D-2e. Linear RR models for breast cancer mortality D-40
D-3 a. Likelihoods vs knots for 2-piece log-linear model, lymphoid cancer mortality D-45
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 ix DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF FIGURES (continued)
D-3b. Plot of the exposure and lymphoid cancer mortality rate ratios generated using a
2-piece log-linear spline model overlaid with other log RR curves and categorical
log RR model points D-45
D-3c. Linear RR models for lymphoid cancer D-50
D-4a. Likelihood vs knots for 2-piece log-linear model, all hematopoietic cancer D-55
D-4b. Plot of exposure and rate ratios for all hematopoietic cancer generated using a 2-
piece log-linear spline model and log transform, linear, and categorical log RR
models D-56
D-4c. Linear RR models for hematopoietic cancer mortality D-61
H-l. Induction ofhprt mutations by EtO with data from ethylene (using estimated EtO
equivalents) shown H-l6
H- 2. Induction of recessive lethal mutations by EtO in Drosophila (wild-type) H-17
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 x DRAFT—DO NOT CITE OR QUOTE
-------
LIST OF ABBREVIATIONS
ADAF age-dependent adjustment factor
AIC Akaike information criterion
AIDS acquired immune deficiency syndrome
AML acute myeloid leukemia
AUC areas under the curve
BEIR Committee on the Biological Effects of Ionizing Radiation
CI confidence interval
DSB double-strand breaks
EC effective concentration
EOIC Ethylene Oxide Industry Council
EPA U.S. Environmental Protection Agency
EtO ethylene oxide
FRG Federal Republic of Germany
GST glutathione S-transferase
HAP hazardous air pollutants
N7-HEG N7-(2-hydroxyethyl)guanine
IARC International Agency for Research on Cancer
ICD International Classification of Diseases
IRIS Integrated Risk Information System
LEC lower confidence limit
MLE maximum likelihood estimate
NCEA National Center for Environmental Assessment
NHL non-Hodgkin lymphoma
NIOSH National Institute for Occupational Safety and Health
NTP National Toxicology Program
O6-HEG O6-hydroxyethylguanine
OBS observed number
OR odds ratios
PBPK physiologically based pharmacokinetic
POD point of departure
RR relative rate, i.e., rate ratio
SCE sister chromatid exchanges
SE standard error
SEER Surveillance, Epidemiology, and End Results
SIR standardized incidence ratio
SMR standard mortality ratios
TWA time-weighted average
UCC Union Carbide Corporation
UCL upper confidence limit
WHO World Health Organization
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
xi DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX A.
2 CRITICAL REVIEW OF EPIDEMIOLOGIC EVIDENCE
3 [EDITORIAL NOTE: Please note that in this assessment document the responses to the
4 2007 external peer review and public comments can be found in Appendix H.]
5
6 A.l. BACKGROUND
7 On the basis of studies indicating that EtO was a strong mutagen and that exposure to
8 EtO produced increased chromosomal aberrations in human lymphocytes (Ehrenberg and
9 Gustafsson, 1970; Ehrenberg and Hallstrom, 1967; Rapoport, 1948), Hogstedt and colleagues
10 studied three small, independent cohorts of workers from Sweden. Reports on two of these
11 cohorts (Hogstedt et al., 1984; Hogstedt et al., 1979b; Hogstedt et al., 1979a) were reviewed in
12 the earlier health assessment document (U.S. EPA, 1985). These two small cohorts plus a third
13 group of EtO-exposed workers from a third independent plant in Sweden were then combined
14 and studied as one cohort (Hogstedt, 1988; Hogstedt et al., 1986). A review of this reconstituted
15 cohort study and subsequent independent studies is presented in Section A.2.
16 Shortly after the third Hogstedt study was completed, another independent study of
17 EtO-exposed employees was completed (Gardner et al., 1989) on a cohort of workers from four
18 companies and eight hospitals in Great Britain, and it was followed by a third independent study
19 on a cohort of exposed workers in eight chemical plants from the Federal Republic of Germany
20 (Kiesselbach et al., 1990). A follow-up study of the Gardner et al. (1989) cohort was recently
21 conducted by Coggon et al. (2004).
22 Greenberg et al. (1990) was the first in a series of studies of workers exposed to EtO at
23 two chemical manufacturing facilities in the Kanawha Valley (South Charleston, WV). The
24 workers at these two facilities were studied later by Teta et al. (1993), Benson and Teta (1993),
25 Teta et al. (1999), and Swaen et al. (2009) and became the basis for several important
26 quantitative risk assessment analyses (Valdez-Flores et al., 2010; EOIC, 2001; Teta et al., 1999).
27 Another independent study of EtO-exposed workers in 14 sterilizing plants from across
28 the United States was completed by the National Institute for Occupational Safety and Health
29 (NIOSH) (Stayner et al., 1993; Steenland et al., 1991). The Stayner et al. (1993) paper presents
30 the exposure-response analysis performed by the NIOSH investigators. These same workers
31 were studied again from a different perspective by Wong and Trent (1993). The NIOSH
32 investigators recently completed a follow-up of the mortality study Steenland et al. (2004) and a
33 breast cancer incidence study based in the same cohort (Steenland et al., 2003). The results of
34 the Steenland et al. (2004) and Steenland et al. (2003) analyses are the basis for the quantitative
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-1 DRAFT—DO NOT CITE OR QUOTE
-------
1 assessment in this document, for reasons explained in the review and summary sections of this
2 appendix.
3 Several additional studies of lesser importance have been done on EtO-exposed cohorts
4 of workers in Sweden (Hagmar et al., 1995; Hagmar et al., 1991), Italy (Bisanti et al., 1993),
5 Belgium (Swaen et al., 1996) and western New York State (Norman et al., 1995), and other parts
6 of the United States (Olsen et al., 1997). These studies are discussed in the following review, but
7 they provide limited information to the overall discussion of whether EtO induces cancer in
8 humans.
9 The more important studies, which are discussed in detail in the summary, are those at
10 two facilities in the Kanawha Valley in West Virginia (Valdez-Flores et al., 2010; Swaen et al.,
11 2009; Teta et al., 1999; Benson and Teta, 1993; Teta et al., 1993; Greenberg et al., 1990) and at
12 14 sterilizing plants around the country (Steenland et al., 2004; Steenland et al., 2003; Stayner et
13 al., 1993; Steenland et al., 1991). These studies have sufficient follow-up to analyze latent
14 effects, and the cohorts appear to be large enough to test for small differences. In addition,
15 exposure estimates were derived for both cohorts, and attempts were made to assess
16 dose-response relationships.
17
18 A.2. INDIVIDUAL STUDIES
19 A.2.1. Hogstedt (1988), Hogstedt et al. (1986)
20 Hogstedt et al. (1986) combined workers from several cohorts for a total of 733 workers,
21 including 378 workers from two separate and independent occupational cohort mortality studies
22 by Hogstedt et al. (1979b) and Hogstedt et al. (1979a) and 355 employees from a third EtO
23 production plant who had not been previously examined. The combined cohort was followed
24 until the end of 1982. The first cohort comprised employees from a small technical factory in
25 Sweden where hospital equipment was sterilized with EtO. The second was from a production
26 facility where EtO was produced by the chlorohydrin method from 1940 to 1963. The third was
27 from a production facility where EtO was made by the direct oxidation method from 1963 to
28 1982.
29 In the update of the 1986 occupational mortality report (Hogstedt, 1988), the cohort
30 inexplicably was reduced to 709 employees (539 men; 170 women). Follow-up for mortality
31 was extended to the end of 1985. The author reported that 33 deaths from cancer had occurred,
32 whereas only 20 were expected in the combined cohort. The excesses that are significant are due
33 mainly to an increased risk of stomach cancer at one plant and an excess of blood and lymphatic
34 malignancies at all three. Seven deaths from leukemia occurred, whereas only 0.8 were expected
35 (standard mortality ratio [SMR] = 9.2). Ten deaths due to stomach cancer occurred versus only
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
1.8 expected (SMR = 5.46). The results tend to agree with those from clastogenic and short-term
tests on EtO (Ehrenberg and Gustafsson, 1970). The authors believe that the large number of
positive cytogenetic studies demonstrating increased numbers of chromosomal aberrations and
sister chromatid exchanges at low-level exposure to EtO indicate that the lymphatic and
hematopoietic systems are particularly sensitive to the genotoxic effects of EtO. They concluded
that the induction of malignancies even at low-level and intermittent exposures to EtO should be
"seriously considered by industry and regulating authorities."
The average air EtO concentrations in the three plants were as follows: In Plant 1
(Hogstedt et al., 1979a) in 1977, levels ranged from 2 to 70 ppm in the storage hall. The average
8-hour time-weighted average (TWA) concentration in the breathing zone of the employees was
calculated as 20 ppm ±10 ppm. Measured concentrations were 150 ppm on the floor outside of
the sterilized boxes and 1,500 ppm inside.
In Plant 2 (Hogstedt et al., 1979b), EtO was produced through the chlorohydrin process.
Between 1941 and 1947, levels probably averaged about 14 ppm, with occasional exposures up
to 715 ppm. Between 1948 and 1963, levels were in the range of 6 ppm to 28 ppm. After 1963,
when production of EtO came to an end, levels ranged from less than 1 ppm to as much as
6 ppm.
In Plant 3 (Hogstedt et al., 1986), the 355 employees were divided into subgroups.
Subgroup A had almost pure exposure to EtO. Subgroup B had principal exposure to EtO but
also exposure to propylene oxide, amines, sodium nitrate, formaldehyde, and 1,2-butene oxide.
Workers in the remaining subgroup C were maintenance and technical service personnel, who
had multiple exposures, including EtO. Concentration levels in Plant 3 are shown in Table A-l.
Table A-l. Estimated 8-hour time-weighted average ethylene oxide
exposure, Plant 3
Group
A (n = 128)
B (n = 69)
C (» = 158)
1963-1976
5-8 ppm
3 ppm
1-3 ppm
1977-1982
1-2 ppm
1 ppm
0.4-1.6 ppm
Source: Hogstedt etal. (1986).
In the earlier studies (Hogstedt et al., 1979b) and (Hogstedt et al., 1979a) of two of the
plants that contributed workers to this cohort, the authors allude to the fact that there was
exposure to benzene, ethylene dichloride, ethylene chlorohydrin, ethylene, and small amounts of
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 bis-(2-chloroethyl) ether, as well as other chemicals in the respective plants. Although
2 170 women were present in the workforce, no gender differences in risk were analyzed
3 separately by the investigators. Of 16 patients with tumors in the two exposed cohorts, there
4 were three cases of leukemia (0.2 expected), six cases of alimentary tract cancer, and four cases
5 of urogenital cancer. Of the 11 cancer cases in the full-time exposed cohort, 5.9 were expected
6 (p< 0.05). This study was criticized by Divine and Amanollahi (1986) for several reasons.
7 First, they believed that the study's strongest evidence in support of a carcinogenic claim for EtO
8 was only a "single case of leukemia" in subgroup C of Plant 3, where the workers had multiple
9 chemical exposures; however, there were no cases in subgroups A or B of Plant 3. Hogstedt et
10 al. (1986) countered that the expectation of leukemia in these two subgroups were 0.04 and 0.02,
11 respectively, and that the appearance of a case could only happen if EtO had "outstanding
12 carcinogenic properties at low levels." Divine and Amanollahi also pointed out that a study
13 (Morgan et al., 1981) of a cohort similar to that of Plant 3 found no leukemia cases or evidence
14 of excessive mortality. Hogstedt et al. (1986) replied that Morgan et al. (1981) stated in their
15 paper that the statistical power of their study to detect an increased risk of leukemia was not
16 strong.
17 Divine and Amanollahi (1986) also stated that the exposures to EtO were higher in
18 Plants 1 and 2 than in Plant 3; therefore, combinations would "normally preclude comparisons
19 between the plants for similar causes of adverse health." This potential problem could be
20 resolved by structuring exposure gradients to analyze risk. Furthermore, they noted Plant 1 was
21 a nonproduction facility involved in sterilization of equipment. Plant 2 used the chlorohydrin
22 process for making EtO, and Plant 3 used the direct oxygenation process. Although these
23 conditions are obviously different, they "are grouped together as analogous." This criticism
24 would, in most instances, be valid only because the methods for producing EtO differ and there
25 were differing exposures to multiple chemicals.
26 However, these concerns are not supported by the evidence. In all three plants the
27 leukemia risk was elevated, even if only slightly in Plant 3. This suggests that there may have
28 been a common exposure, possibly to EtO, endemic to all three plants that was responsible for
29 the measured excesses. Noteworthy is the elevated risk of leukemia seen in Plant 1 (3 observed
30 vs. 0.14 expected), where the exposures were almost exclusively to EtO in the sterilization of
31 equipment. The argument that Plant 1 leukemias form a "chance cluster," as Shore et al. (1993)
32 claim, and as such should be excluded from any analysis does not preclude the possibility that
33 these cases are in reality the result of exposure to EtO. Hogstedt (1988) argues that earlier
34 remarks by Ehrenberg and Gustafsson (1970) that EtO "constituted a potential cancer hazard" on
35 the basis of a considerable amount of evidence other than epidemiologic should have served as a
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 warning that the increased risk seen in Plant 1 was not necessarily a "chance cluster," and
2 because the chlorohydrin process was not in use in Plant 1, it cannot be due to exposure to a
3 chemical in the chlorohydrin process.
4
5 A.2.2. Gardner et al. (1989)
6 Gardner et al. (1989) completed a cohort study of 2,876 men and women who had
7 potential exposure to EtO. The cohort was identified from employment records at four
8 companies that had produced or used EtO since the 1950s and from eight hospitals that have had
9 EtO clinical sterilizing units since the 1960s, and it was followed to December 31, 1987. All but
10 1 of the 1,012 women and 394 of the men in the cohort worked at one of the hospitals. The
11 remaining woman and 1,470 men made up the portion of the cohort from the four companies.
12 By the end of the follow-up, 226 members (8% of the total cohort) had died versus
13 258.8 expected. Eighty-five cancer deaths were observed versus 76.64 expected.
14 No clear excess risk of leukemia (3 observed vs. 2.09 expected), stomach cancer
15 (5 observed vs. 5.95 expected), or breast cancer (4 observed vs. 5.91 expected) was present as of
16 the cutoff date. "Slight excesses" of deaths due to esophageal cancer (5 observed vs.
17 2.2 expected), lung cancer (29 observed vs. 24.55 expected), bladder cancer (4 observed vs.
18 2.04 expected), and non-Hodgkin lymphoma (NHL) (4 observed vs. 1.63 expected) were noted,
19 although an adjustment made to reflect local "variations in mortality" reduced the overall cancer
20 excess from 8 to only 3. According to the authors' published tabulations, all three leukemias
21 identified in this study fell into the longest latent category (20 years or longer), where only
22 0.35 were expected. All three were in the chemical plants. This finding initially would seem to
23 be consistent with experimental animal evidence demonstrating excess risks of hematopoietic
24 cancer in animals exposed to EtO. But the authors note that since other known leukemogens
25 were present in the workplace, the excess could have been due to a confounding effect.
26 The hospitals began using EtO during or after 1962, whereas all of the chemical
27 companies had handled EtO from or before 1960. In the hospitals there was occasional exposure
28 to formaldehyde and carbon tetrachloride but few other confounding agents. On the other hand,
29 the chemical workers were exposed to a wide range of compounds including chlorohydrin,
30 propylene oxide, styrene, and benzene. The earliest industrial hygiene surveys in 1977 indicated
31 that the TWA average exposures were less than 5 ppm in almost all jobs and less than 1 ppm in
32 many. No industrial hygiene data were available for any of the facilities prior to 1977, although
33 it is stated that peaks of exposure up to several hundred ppm occurred as a result of operating
34 difficulties in the chemical plants and during loading and unloading of sterilizers in the hospitals.
35 An odor threshold of 700 ppm was reported by both manufacturers and hospitals, according to
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-5 DRAFT—DO NOT CITE OR QUOTE
-------
1 the authors. The authors assumed that past exposures were somewhat higher without knowing
2 precisely what they were. An attempt was made to classify exposures into a finite number of
3 subjectively derived categories (definite, possible, continual, intermittent, and unknown). This
4 exercise produced no discernable trends in risk of exposure to EtO. However, the exposure
5 status classification scheme was so vague as to be useless for determining risk by gradient of
6 exposure to EtO.
7 It is of interest that all three of the leukemia deaths entailed exposure to EtO, with very
8 little or no exposure to benzene, according to the authors. The findings are not inconsistent with
9 those of Hogstedt (1988) and Hogstedt et al. (1986). The possibility of a confounding effect
10 other than benzene in these chemical workers cannot entirely be ruled out. Other cancers were
11 slightly in excess, but overall there was little increased mortality from cancer in this cohort. It is
12 possible that if very low levels of exposure to EtO had prevailed throughout the history of these
13 hospitals and plants, the periods of observation necessary to observe an effect may not have been
14 long enough.
15 A follow-up study of this cohort conducted by Coggon et al. (2004) is discussed below.
16
17 A.2.3. Kiesselbach et al. (1990)
18 Kiesselbach et al. (1990) carried out an occupational cohort mortality study of 2,658 men
19 from eight chemical plants in the Federal Republic of Germany (FRG) that were involved in the
20 production of EtO. The method of production is not stated. At least some of the plants that were
21 part of an earlier study by Thiess et al. (1981) were included. Each subject had to have been
22 exposed to EtO for at least 1 year sometime between 1928 and 1981 before person-years at risk
23 could start to accumulate. Most exposures occurred after 1950. By December 31, 1982, the
24 closing date of the study, 268 men had died (about 10% of the total cohort), 68 from malignant
25 neoplasms. The overall SMR for all causes was 0.87, and for total cancer the SMR was 0.97,
26 based on FRG rates. The authors reported that this deficit in total mortality indicates a
27 healthy-worker effect.
28 The only remarkable findings here are slightly increased risks of death from stomach
29 cancer (14 observed vs. 10.15 expected, SMR = 1.4), cancer of the esophagus (3 observed vs.
30 1.5 expected, SMR = 2), and cancer of the lung (23 observed vs. 19.86 expected, SMR = 1.2).
31 Although the authors claimed that they looked at latency, only stomach cancer and total
32 mortality has a latency analysis included. This was accomplished by not counting the first
33 10 years of follow-up in the parameter "years since first exposure." This study is limited by the
34 lack of further latency analyses at other cancer sites. The risk of stomach cancer shows only a
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 slight nonsignificant trend upward with increasing latency. Only two leukemias were recorded
2 versus 2.35 expected.
3 This is a largely unremarkable study, with few findings of any significance. No actual
4 exposure estimates are available. The categories of exposure that the authors constructed are
5 "weak," "medium," and "strong." It is not known whether any of these categories is based on
6 actual measurements. No explanation of how they were derived is provided except that the
7 authors claim that the information is available on 67.2% of the members of the cohort. If the
8 information was based on job categories, it should be kept in mind that exposures in jobs that
9 were classified the same from one plant to the next may have produced entirely different
10 exposures to EtO. The tabular data regarding these exposure categories shows that only 2.4% of
11 all members of the cohort were considered "strongly" exposed to EtO. Although 71.6% were
12 classified as "weak," the remaining 26% were considered as having "medium" exposure to EtO.
13 This is largely a study in progress, and further follow-up will be needed before any
14 definite trends or conclusions can be drawn. The authors reported that only a median 15.5 years
15 of follow-up had passed by the end of the cutoff date, whereas the median length of exposure
16 was 9.6 years. Before any conclusions can be made from this study several additional years of
17 follow-up would be needed with better characterization of exposure.
18
19 A.2.4. Greenberg et al. (1990)
20 Greenberg et al. (1990) retrospectively studied the mortality experience of 2,174 men
21 who were assigned to operations that used or produced EtO in either of two Union Carbide
22 Corporation (UCC) chemical plants in West Virginia. In 1970 and 1971, EtO production at the
23 two plants was phased out, but EtO was still used in the plants for the production of other
24 chemicals. SMRs were calculated in comparison with the general U.S. population and the
25 regional population. Results based on regional population death rates were found to be similar to
26 those based on the U.S. general population. Follow-up began either on January 1, 1940, if
27 exposure to EtO began sooner, or on the date when exposure began, if it occurred after January
28 1, 1940. Follow-up ended on December 31, 1978. Note that this cohort is thus a mixture of a
29 prevalent cohort and an incident cohort, and the prevalent part of the cohort may be especially
30 vulnerable to bias from the healthy worker survivor effect. The healthy worker survivor effect
31 might have occurred if workers who were employed before 1940 and who were of greater
32 susceptibility preferentially developed a disease of interest prior to 1940 and were no longer
33 employed when cohort enumeration began. It appears that the chemical facilities began
34 operating in 1925, so the maximum latency for the development of a disease of interest between
35 the time of first exposure and cohort enumeration was 15 years; however, these early (pre-1940)
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 hires would also have had the highest EtO exposures (Swaen et al., 2009) and may thus have had
2 short latency periods as well. The healthy worker survivor effect bias can also dampen
3 exposure-response relationships (Applebaum et al., 2007). According to Greenberg et al. (1990),
4 slightly over 10% of the cohort was comprised of prevalent hires (223 of 2,174). This is not a
5 large proportion, but, as noted above, these early hires would also have had the highest exposures
6 (Swaen et al., 2009). It is unknown how many workers employed before 1940 were no longer
7 employed when cohort enumeration began. Two years of pre-1940 exposure were reportedly
8 taken into account when categorizing the cohort into groups with >2 years exposure in the
9 different potential exposure categories (see below); however, it is unclear how pre-1940 years of
10 exposure were treated in other analyses, e.g., the analyses based on duration of exposure
11 (although presumably they were taken into account for those analyses as well).
12 Total deaths equaled 297 versus 375.9 expected (SMR = 0.79,p< 0.05). Only 60 total
13 cancer deaths were observed versus 74.6 expected (SMR = 0.81). These deficits in mortality
14 suggest a manifestation of the healthy-worker effect. In spite of this, nonsignificant elevated
15 risks of cancer of the liver, unspecified and primary, (3 observed vs. 1.8 expected, SMR = 1.7),
16 pancreas (7 observed vs. 4.1 expected, SMR = 1.7), and leukemia and aleukemia (7 observed vs.
17 3.0 expected, SMR = 2.3) were noted.
18 The authors also reported that in 1976, 3 years prior to the end of follow-up, an industrial
19 hygiene survey found that 8-hour TWA EtO levels averaged less than 1 ppm, although levels as
20 high as 66 ppm 8-hour TWA had been observed. In maintenance workers, levels averaged
21 between 1 and 5 ppm 8-hour TWA. Because of the lack of information about exposures before
22 1976 (e.g., when EtO was in production), the authors developed a qualitative exposure
23 categorization scheme with three categories of exposure (low, intermediate, and high) on the
24 basis of the potential for exposure in each department. The number of workers in each exposure
25 category was not reported; however, it appears from Teta et al. (1993) (see below) that only
26 425 workers were assigned to EtO production departments, which were apparently the only
27 departments with high potential exposure. No significant findings of a dose-response
28 relationship were discernable.
29 Except for two cases of leukemia, all the victims of pancreatic cancer and leukemia
30 began their work—and hence exposure to EtO—many years prior to their deaths. The leukemia
31 and pancreatic cancer deaths were concentrated in the chlorohydrin production department. Four
32 of the seven leukemia victims had been assigned to the chlorohydrin department; only 0.8 deaths
33 (SMR = 5.0) would have been expected in this department of only 278 workers. Six pancreatic
34 cancer victims were assigned to the chlorohydrin department, whereas only 0.98 deaths would
35 have been expected to occur (SMR = 6.1). All seven leukemia victims, including the four in the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-8 DRAFT—DO NOT CITE OR QUOTE
-------
1 chlorohydrin department, were listed by the authors as having only low potential exposure to
2 EtO. In contrast, among workers ever assigned to a department in the high exposure category,
3 no leukemia deaths and only one pancreatic cancer death occurred.
4 The authors hypothesized that the excesses in leukemia and pancreatic cancers were
5 associated with production of ethylene chlorohydrin or propylene chlorohydrin or both in the
6 chlorohydrin department. Some later follow-up studies (described below) were done of the
7 cohort excluding the chlorohydrin production workers (Teta et al., 1993) and of the chlorohydrin
8 production workers alone (Benson and Teta, 1993) to further examine this hypothesis.
9
10 A.2.5. Steenland et al. (1991)
11 In an industry-wide analysis by NIOSH, Steenland et al. (1991) studied EtO exposure in
12 18,254 workers (55% female) identified from personnel files of 14 plants that had used EtO for
13 sterilization of medical equipment, treating spices, or testing sterilizers. Each of the 14 plants
14 (from 75 facilities surveyed) that were considered eligible for inclusion in the study had at least
15 400 person-years at risk prior to 1978. Within each eligible facility, at least 3 months of
16 exposure to EtO qualified an employee for inclusion in the cohort. Employees, including all
17 salaried workers, who were "judged never to have been exposed to EtO" on the basis of
18 industrial hygiene surveys were excluded. Follow-up ended December 31, 1987. The cohort
19 averaged 16 years of latency. Approximately 86% achieved the 9-year latent point, but only 8%
20 reached the 20-year latency category. The average year of first exposure was 1970, and the
21 average length of exposure was 4.9 years. The workers' average age at entry was not provided,
22 nor was an age breakdown. Nearly 55% of the cohort were women.
23 Some 1,137 workers (6.4%) were found to be deceased at the end of the study period,
24 upon which the underlying cause of death was determined for all but 450. If a member was
25 determined to be alive as of January 1, 1979, but not after and no death record was found in the
26 National Death Index through December 31, 1987, then that member was assumed to be alive for
27 the purposes of the life-table analysis and person-years were accumulated until the cutoff date.
28 Altogether, 4.5% of the cohort fell into this category. This procedure would tend to increase the
29 expected deaths and, as a consequence, potentially bias the risk ratio downward if a sizable
30 number of deaths to such persons during this period remained undiscovered to the researchers.
31 In the total cohort no significantly increased risks of death from any site-specific cancer
32 were noted. Analyses by job categories and by duration of exposure indicated no excess risks of
33 cancer when compared with the rate in the general population. However, there was an increased
34 trend in the risk of hematopoietic cancers, all sites, with increasing lengths of time since first
35 exposure. After 20 years latency, the SMR was 1.76, based on 13 cases. The test for trend was
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-9 DRAFT—DO NOT CITE OR QUOTE
-------
1 significant atp = 0.03. For men (45%), without regard for latency, the SMR for hematopoietic
2 cancer was a significant 1.55 (p < 0.05), based on 27 cases. Among men with long latency
3 (greater than 20 years) and the longest duration of exposure (greater than 7 years) the SMR for
4 hematopoietic cancers was 2.63, based on 7 deaths (p < 0.05).
5 The authors pointed out that the SMR for leukemia among men was 3.45, based on
6 5 deaths (p < 0.05), for deaths in the latter period of 1985 to 1987. For kidney cancer, the SMR
7 was 3.27, based on six deaths (p < 0.05), after 20 years latency. The authors also reported on a
8 significant excess risk (p < 0.05) of lymphosarcoma-reticulosarcoma in men (SMR = 2.6), based
9 on seven deaths. Women had a lower nonsignificant rate. The risk of breast cancer was also
10 nonsignificant (SMR = 0.85 based on 42 cases). The authors hypothesized that men were more
11 heavily exposed to EtO than were women because "men have historically predominated in jobs
12 with higher levels of exposure." However, the lack of an association between EtO exposure and
13 lymphohematopoietic cancer in females was also observed in the exposure-response analyses of
14 this cohort, including in the highest exposure category, performed by Stayner et al. (1993) and
15 discussed below.
16 Industrial hygiene surveys indicated that sterilizer operators were exposed to an average
17 personal 8-hour TWA EtO level of 4.3 ppm, whereas all other workers averaged only 2 ppm,
18 based on 8-hour samples during the period 1976 to 1985. These latter employees primarily
19 worked in production and maintenance, in the warehouse, and in the laboratory. This was during
20 a time when engineering controls were being installed to reduce worker's exposure to EtO;
21 earlier exposures may have been somewhat higher. The authors reported that no evidence of
22 confounding exposure to other occupational carcinogens was documented.
23 The authors concluded that there was a trend toward an increased risk of death from
24 hematopoietic cancer with increasing lengths of time since the first exposure to EtO. This trend
25 might have been enhanced if the authors had added additional potential deaths identified from
26 the 820 (4.5%) "untraceable" members of the cohort from 1979 to 1987. The authors felt that
27 their results were not conclusive for the relatively rare cancers of a priori interest, based on the
28 limited number of cases and the short follow-up. The cohort averaged 16 years of latency and
29 86% had at least 9 years but only 8% reached the 20-year latent category.
30 Exposure-response analyses were conducted by Stayner et al. (1993) and are discussed
31 below. More recently, a follow-up mortality study (Steenland et al., 2004) and a breast cancer
32 incidence study (Steenland et al., 2003) of this cohort were conducted; these are also discussed
33 below.
34
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-10 DRAFT—DO NOT CITE OR QUOTE
-------
1 A.2.6. Teta et al. (1993)
2 In a follow-up analysis of the cohort of 2,174 male UCC workers studied by Greenberg et
3 al. (1990), Teta and her colleagues excluded the 278 workers in the chlorohydrin unit in which
4 Greenberg and colleagues found a high risk of leukemia and pancreatic cancer, thereby removing
5 the potential confounding of the chlorohydrin production process. The 1,896 men in the
6 remaining cohort were followed for an additional 10 years, through all of 1988. (Among the
7 278 men who were excluded because they had worked in the chlorohydrin unit, 49 had also been
8 assigned to EtO production departments, which were considered high potential ETO exposure
9 departments, according to Greenberg et al. (1990). Data were reportedly examined with and
10 without the inclusion of these 49 workers with overlapping assignments; however, the results of
11 these analyses are not fully presented). According to Benson and Teta (1993), 112 of the
12 278 excluded workers were employed before 1940, reducing the prevalent part of the remaining
13 cohort to 111 of 1,896 workers, or just under 6%. (It is unclear how pre-1940 years of exposure
14 were treated in the analyses based on duration of exposure, although presumably they were taken
15 into account.) The update did not include additional work histories for the study subjects. Teta
16 et al. (1993) note that duration of assignment to an EtO production unit was not affected by the
17 update because EtO was no longer in production at the two plants; however, assignment to
18 EtO-using departments might have been affected, and according to Greenberg et al. (1990), some
19 of these departments had medium EtO exposure potential.
20 Teta et al. (1993) reported that the average duration of exposure was more than 5 years
21 and the average follow-up was 27 years. Furthermore, at least 10 years had elapsed since first
22 exposure for all the workers. The reanalysis demonstrated no increased risk of overall cancer, or
23 of leukemia, NHL, or cancers of the brain, pancreas, or stomach. The SMR for total deaths,
24 based on comparison with mortality from the general population, was 0.79 (p < 0.01;
25 observed = 431). The SMR for total cancer was 0.86 (observed = 110). No site-specific cancers
26 were significantly elevated. Although the authors concluded that this study did not indicate any
27 significant trends of increasing site-specific cancer risk with increasing duration of potential
28 exposure to EtO, there appeared to be a nonsignificant increasing trend for leukemia and
29 aleukemia (p = 0.28, based on five cases) as well as stomach cancer (p = 0.13; eight cases).
30 According to Greenberg et al. (1990), 8-hour TWA EtO levels averaged less than 1 ppm,
31 based on the 1976 monitoring (after EtO production at the plants had ceased), although levels as
32 high as 66 ppm 8-hour TWA were reported. Teta et al. (1993) estimated that in the 1960s,
33 exposure in the units producing EtO by direct oxidation ranged from 3 to 20 ppm 8-hour TWA,
34 with peaks of several hundred ppm. These estimates were based on an industrial hygiene survey
35 conducted at another UCC facility in Texas that used the same direct oxidation process as the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-11 DRAFT—DO NOT CITE OR QUOTE
-------
1 two plants in West Virginia from which the UCC EtO cohort was taken. Ethylene oxide was
2 also produced via the chlorohydrin process in a closed building during the years 1925 to 1957.
3 Levels of exposure to EtO would have been higher than in the direct oxidation production
4 process because of start-up difficulties, fewer engineering controls, less complex equipment, and
5 the enclosed building. Employee nausea, dizziness, and vomiting were documented in the
6 medical department in 1949. These acute effects occur in humans at exposures of several
7 hundred ppm, according to the authors.
8 During the time periods under investigation, the estimated exposure ranges for
9 departments using or producing EtO were >14 ppm from 1925 to 1939; 14 ppm from 1940 to
10 1956; 5-10 ppm from 1957 to 1973; and <1 ppm from 1974 to 1988, with frequent peaks of
11 several hundred ppm in the earliest period and some peaks of similar intensity in the 1940s to
12 mid-1950s. In the absence of monitoring data prior to 1976, these estimates cannot be
13 confirmed. Furthermore, workers were eliminated from the analysis if they had worked in the
14 chlorohydrin unit because of the assumption that the increased risks of leukemia and pancreatic
15 cancer were possibly due to exposure to something in the chlorohydrin process, as conjectured
16 by Greenberg et al. (1990). However, even when the potential confounding influence of the
17 chlorohydrin process is removed, there remains the suggestion of a trend of an increasing risk of
18 leukemia and aleukemia with increasing duration of exposure to EtO in the remaining cohort
19 members (p = 0.28, based on 5 cases).
20 The authors indicated that their findings do not confirm the findings in experimental
21 animal studies and are not consistent with the earliest results reported among EtO workers. They
22 also noted that they did not observe any significant trend of increasing risks of stomach cancer
23 (n = 8), leukemia (n = 5) or cancers of the pancreas or brain and nervous system with increasing
24 duration of exposure. No lagged exposure or latency analyses were conducted in this study.
25 In a later analysis, Teta et al. (1999) fitted Poisson regression dose-response models to
26 the UCC data (Teta et al., 1993) and to the NIOSH data (Steenland et al., 1991). They reported
27 that latency and lagging of dose did not appreciably affect the fitted models. Because Teta et al.
28 (1999) did not present risk ratios for the categories used to model the dose-response
29 relationships, the only comparison that could be made between the UCC and NIOSH data is
30 based on the fitted models. These models are almost identical for leukemia, but, for the
31 lymphoid category, the risk according to the fitted model for the UCC data decreased as a
32 function of dose, whereas the risk for the modeled NIOSH data increased as a function of dose.
33 However, the models are based on small numbers of cases (16 [5 UCC, 11 NIOSH] for
34 leukemia; 22 [3 UCC, 19 NIOSH] for lymphoid cancers), and no statistics are provided to assess
35 model goodness of fit or to compare across models. This analysis is superseded by the more
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-12 DRAFT—DO NOT CITE OR QUOTE
-------
1 recent analysis by the same authors (Valdez-Flores et al., 2010) of the results of more recent
2 follow-up studies of these two cohorts [see discussion of the Swaen et al. (2009) study below].
3
4 A.2.7. Benson and Teta (1993)
5 In a companion mortality study (Benson and Teta, 1993), the remaining 278 employees
6 who were identified by Greenberg et al. (1990) as having worked at some time in the
7 chlorohydrin unit and who were not included in the cohort of Teta et al. (1993) were followed to
8 the end of 1988. Note that the prevalent part (i.e., those workers first employed before the cohort
9 enumeration date of 1 January 1940) of this reduced cohort is 112 of the 278 workers, or 40%,
10 and, therefore, the potential for bias from a healthy worker survivor effect, as discussed for the
11 Greenberg et al. (1990) study above (see Section A.2.4), may be more pronounced in this study
12 of the chlorohydrin unit workers. It is unknown how many chlorohydrin unit workers employed
13 before 1940 were no longer employed when cohort enumeration began.
14 Altogether, 40 cancer deaths occurred versus 30.8 expected (SMR = 1.3) in the subcohort
15 of chlorohydrin workers. In Greenberg et al. (1990), significant elevated risks of pancreatic
16 cancer and leukemia and aleukemia occurred in only those workers assigned to the chlorohydrin
17 process. Benson and Teta (1993) noted a significantly increased risk of pancreatic cancer
18 (SMR = 4.9, eight observed deaths,/? < 0.05) in the same group and a significantly increased risk
19 of cancer in the enlarged category of lymphohematopoietic cancer (SMR = 2.9, eight observed
20 deaths,/* < 0.05), which included leukemia and aleukemia, after an additional 10 years of
21 follow-up.
22 The authors concluded that these cancers were likely work-related and some exposure in
23 the chlorohydrin unit, possibly to the chemical ethylene dichloride, was probably the cause.
24 They pointed out that Greenberg et al. (1990) found that the chlorohydrin unit was likely to be a
25 low-EtO exposure area in the West Virginia plants. The other possibility was bis-chloroethyl
26 ether, which the authors pointed out is rated by the International Agency for Research on Cancer
27 (IARC) as a group 3 ("not classifiable as to its carcinogenicity to humans") chemical.
28 Circumstantial evidence seems to support the authors' contention that ethylene dichloride is the
29 cause: IARC designated ethylene dichloride as a group 2B chemical ("possibly carcinogenic to
30 humans"), exposure was likely heavier throughout the history of the facility, and plant medical
31 records documented many accidental overexposures occurring to the pancreatic cancer victims
32 prior to diagnosis. However, this conclusion is disputed by Olsen et al. (1997) whose analysis is
33 discussed later.
34
35
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-13 DRAFT—DO NOT CITE OR QUOTE
-------
1 A.2.8. Stayner et al. (1993)
2 Stayner et al. (1993) provide an exposure-response analysis for the cohort study of EtO
3 workers described by Steenland et al. (1991). Nothing was modified concerning the follow-up,
4 cohort size, vital status, or cutoff date of the study. The exposure assessment and verification
5 procedures were presented in Greife et al. (1988) and Hornung et al. (1994). In brief, a
6 regression model was developed, allowing the estimation of exposure levels for time periods,
7 facilities, and operations for which industrial hygiene data were unavailable. The data for the
8 model consisted of 2,700 individual time-weighted exposure values for workers' personal
9 breathing zones, acquired from 18 facilities between 1976 and 1985. These data were divided
10 into two sets, one for developing the regression model and the second (from six randomly
11 selected plants) for testing it. Job titles were grouped into eight categories with similar potential
12 for EtO exposure, and arithmetic mean exposure levels by facility, year, and exposure category
13 were calculated from the data used for model development. The arithmetic means were
14 logarithmically transformed, and weighted linear regression models were fitted. Seven out of
15 23 independent variables tested for inclusion in the model were found to be significant predictors
16 (p < 0.10) of EtO exposure and were included in the final model. This model predicted 85% of
17 the variation in average EtO exposure levels in the test data. The model was also evaluated
18 against estimates for the test data derived by a panel of 11 industrial hygienists familiar with EtO
19 levels in the sterilization industry and provided with the values for the independent variables
20 used in the model corresponding to the arithmetic means from the test data. The overall mean of
21 the modeled estimates was not highly biased nor biased in one direction when compared to the
22 overall mean exposure estimates of the individual industrial hygiene experts. Using the test data
23 as the standard, the model estimates showed less bias (average difference) than 9 of the
24 11 industrial hygienists and more precision (standard deviation of the differences) than all 11.
25 Similarly, the model outperformed the panel in terms of both bias and precision when the panel
26 results were averaged.
27 Average exposure levels, including early historical exposure levels, for the exposure
28 categories in the study plants were estimated using this industrial hygiene-based regression
29 model. Then, the cumulative exposure for each worker was estimated by calculating the product
30 of the average exposure in each job the worker held by the time spent in that job and then
31 summing these over all the jobs held by that worker. This value became the cumulative
32 exposure index for that employee and reflected the working lifetime total exposure to EtO.
33 Stayner et al. (1993) generated SMRs based on standard life-table analysis. The three
34 categories of cumulative exposure were less than 1,200 ppm-days, 1,200 to 8,500 ppm-days, and
35 greater than 8,500 ppm-days. Additionally, the Cox proportional hazards model was used to
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-14 DRAFT—DO NOT CITE OR QUOTE
-------
1 model the exposure-response relationship between EtO and various cancer types, using
2 cumulative exposure as a continuous variable.
3 Stayner and colleagues noted a marginally significant increase in the risk of
4 hematopoietic cancers, with an increase in cumulative exposure by both the life-table analysis as
5 well as the Cox model, although the magnitude of the increased risk was not substantial. At the
6 highest level—greater than 8,500 ppm-days of exposure—the SMR was a nonsignificant 1.24,
7 based on 13 cases. However, 12 of these cases were in males, whereas only 6.12 were expected.
8 Thus, in this highest exposure category, a statistically significant (p < 0.05) SMR of 1.96 in
9 males was produced. This dichotomy produced a deficit in females (1 observed vs. 4.5 expected,
10 /?<0.05).
11 The Cox analysis produced a significantly positive trend with respect to lymphoid cell
12 tumors (combination of lymphocytic leukemia and NHL) when EtO exposures were lagged
13 5 years. The authors stated that these data provide some support for the hypothesis that exposure
14 to EtO increases the risk of mortality from lymphatic and hematopoietic neoplasms. They
15 pointed out, however, that their data do not provide evidence for a positive association between
16 exposure to EtO and cancer of the stomach, brain, pancreas, or kidney or leukemia as a group.
17 Breast cancer was not analyzed in this report.
18 This cohort was not updated with vital status information on the "untraceables" (4.5%),
19 and cause of death information was not provided on deaths with unknown causes; thus, it lacks a
20 complete follow-up and, therefore, the risk estimates may be understated. Another potential
21 limiting factor is the information regarding industrial hygiene measurements of EtO that were
22 completed in the plants. According to the authors, the median length of exposure to EtO of the
23 cohort was 2.2 years and the median exposure was 3.2 ppm. It may be unreasonable to expect
24 any findings of increased significant risks because follow-up was too short to allow the
25 accumulation of mortality experience (average follow-up =16 years; only 8% of cohort had
26 >20 years follow-up).
27 The authors also remind us that there is a lack of evidence for an exposure-response
28 relationship among females or for a sex-specific carcinogenic effect of EtO in either laboratory
29 animals or humans. In fact, the mortality rate from hematopoietic cancers among the women in
30 this cohort was lower than that of the general U.S. population. Therefore the contrast seen here
31 is unusual.
32 The positive findings are somewhat affected by the presence in the cohort of one heavily
33 exposed case (although the authors saw no reason to exclude it from the analysis), and there is a
34 lack of definite evidence for an effect on leukemia as a group. Despite these limitations, the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-15 DRAFT—DO NOT CITE OR QUOTE
-------
1 authors believe that their data provide support for the hypothesis that exposure to EtO increases
2 the risk of mortality from hematopoietic neoplasms.
O
4 A.2.9. Wong and Trent (1993)
5 This study is a reanalysis of the same cohort that was studied by Stayner et al. (1993) and
6 Steenland et al. (1991), with some differences. The cohort was incremented without explanation
7 by 474 to a total of 18,728 employees and followed one more year, to the end of December 1988.
8 This change in the cohort resulted in the addition of 176 observed deaths and 392.2 expected
9 deaths. The finding of more than twice as many expected deaths as observed deaths is baffling.
10 A reduced total mortality of this magnitude suggests that many deaths may have been
11 overlooked. This resulted in a further reduction of the overall SMR to a significant deficit of
12 0.73. Sixty additional cancer deaths were added versus 65.9 expected, for an SMR = 0.9, based
13 on 403 total cancer deaths observed versus 446.2 expected.
14 The authors reported no significant increase in mortality at the cancer sites found to be of
15 most interest in previous studies, that is, stomach, leukemia, pancreas, brain, and breast. They
16 also reported the lack of a dose-response relationship and correlation with duration of
17 employment or latency. They did report a statistically significant increased risk of NHL among
18 men (SMR = 2.47; observed = 16, expected = 6.47; p < 0.05) that was not dose-related and a
19 nonsignificant deficit of NHL among women (SMR = 0.32; observed = 2, expected = 6.27). The
20 authors concluded that the increase in men was not related to exposure to EtO but could in fact
21 have been related to the presence of acquired immune deficiency syndrome (ADDS) in the male
22 population. When this explanation was offered in a letter to the editor (Wong, 1991) regarding
23 the excess of NHL reported in Steenland et al. (1991), it was dismissed by Steenland and Stayner
24 (1993) as pure speculation. Steenland and Stayner (1993) responded that most of the NHL
25 deaths occurred prior to the ADDS epidemic, which began in the early 1980s. They also
26 indicated that there was no reason to suspect that these working populations would be at a higher
27 risk for AIDS than was the general population, the comparison group.
28 Wong and Trent (1993) also reported a slightly increased risk of cancer in other
29 lymphatic tissue (14 observed vs. 11.39 expected). In men, the risk was nonsignificantly higher
30 (11 observed vs. 5.78 expected). Forty-three lymphopoietic cancers were observed versus
31 42 expected. In men, the risk was higher (32 observed vs. 22.22 expected). Fourteen leukemia
32 deaths were noted versus 16.2 expected. The authors did not derive individual exposure
33 estimates for exposure-response analysis, such as in Stayner et al. (1993). Rather, they used
34 duration of employment as a surrogate for exposure.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-16 DRAFT—DO NOT CITE OR QUOTE
-------
1 This study has many of the same limitations as the Stayner et al. (1993) study. The
2 authors assumed that those individuals with an unknown vital status as of the cutoff date were
3 alive for the purposes of the analysis, and they were unable to obtain cause-of-death information
4 on 5% of the known deaths.
5 The differences between this cohort study and that of Stayner et al. (1993) are in the
6 methods of analysis. Stayner etal. (1993) used the 9* revision of the International Classification
7 of Diseases (ICD) to develop their site-specific cancer categories for comparison with expected
8 cancer mortality, whereas Wong and Trent (1993) used the 8th revision. This could account for
9 some of the differences in the observed numbers of site-specific cancers, because minor
10 differences in the coding of underlying cause of death could lead to a shifting of some unique
11 causes from one site-specific category to another. Furthermore, Wong and Trent (1993) did not
12 analyze separately the category "lymphoid" neoplasms, which includes lymphocytic leukemia
13 and NHL, whereas Stayner et al. (1993) did. Stayner et al. (1993) further developed cumulative
14 exposure information using exposure estimates, whereas Wong and Trent (1993) used length of
15 employment as their surrogate for exposure but did not code detailed employment histories.
16 Because Wong and Trent (1993) made no effort to quantify the exposures, as was the
17 case in Stayner et al. (1993), this study is less useful in determining a exposure-response
18 relationship. Furthermore, the assumption that a member of the cohort should be considered
19 alive if a death indication could not be found will potentially tend to bias risk ratios downward if,
20 in fact, a large portion of this group is deceased. In this study all untraceable persons were
21 considered alive at the end of the follow-up; therefore, the impact of the additional person-years
22 of risk cannot be gauged.
23
24 A.2.10. Bisanti et al. (1993)
25 These authors reported on a cohort mortality study of 1,971 male chemical workers
26 licensed to handle EtO by the Italian government, whom they followed retrospectively from
27 1940 to 1984. Altogether, 76 deaths had occurred in this group by the end of the study period,
28 whereas 98.8 were expected. Of those, 43 were due to cancer versus 33 expected. The cause of
29 one death remained unknown, and 16 workers were lost to follow-up. A group of
30 637 individuals from this cohort was licensed to handle only EtO; the remaining 1,334 had
31 licenses valid for handling other toxic gases as well. Date of licensing for handling EtO became
32 the initiating point of exposure to EtO, although it is likely that some of these workers had been
33 exposed previously to EtO. The regional population of Lombardia was used as the reference
34 group from which comparison death rates were obtained.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-17 DRAFT—DO NOT CITE OR QUOTE
-------
1 Although there were excess risks from almost all cancers, one of the greatest SMRs was
2 in the category known as "all hematopoietic cancers," where 6 observed deaths occurred when
3 only 2.4 were expected (SMR = 2.5). In the subgroup "lymphosarcoma, reticulosarcoma" there
4 were 4 observed deaths whereas only 0.6 were expected (SMR = 6.7, p < 0.05); the remaining
5 2 were leukemias. The authors note that five hematopoietic cancers occurred in the subgroup of
6 workers who were licensed to handle only EtO but no other chemicals versus only
7 0.7 hematopoietic cancers expected (SMR = 7.l,p< 0.05). These deaths occurred within
8 10 years from date of licensing (latent period), which is consistent with the shorter latent period
9 anticipated for this kind of cancer. According to the authors, all workers began their
10 employment in this industry when the levels of EtO were high, although no actual measurements
11 were available. The fact that this subgroup of workers was licensed only for handling EtO
12 reduces the likelihood of a confounding chemical influence.
13 The authors concluded that the excess risk of cancer of the lymphatic and hematopoietic
14 tissues in these particular EtO cohort members support the suggested hypothesis of a higher risk
15 of cancer found in earlier studies, but they added that the lack of exposure information on the
16 other industrial chemicals in the group that had a license for handling other toxic chemicals made
17 their findings inconclusive.
18 This study was of a healthy young cohort, and most person-years were contributed in the
19 latter years of observation. Many years of follow-up may be necessary in order to fully verify
20 any trend of excess risks for the site-specific cancers of interest and to measure latent effects.
21 Furthermore, the unusual deficit of total deaths versus expected contrasted with an excess of
22 cancer deaths versus expected raises a question about the potential for selection bias when the
23 members of this cohort were chosen for inclusion. Also, one of the study's major limitations is
24 the lack of exposure data.
25
26 A.2.11. Hagmar et al. (1995) and Hagmar et al. (1991)
27 Cancer incidence was studied in a cohort of 2,170 EtO-exposed workers from two plants
28 in Sweden that produced disposable medical equipment. To fit the definition for inclusion, the
29 subjects, 1,309 women and 861 men, had to have been employed for a minimum of 12 months
30 and some part of that employment had to have been during the period 1970-1985 in the case of
31 one plant and 1965-1985 in the case of the other. The risk ratios were not dichotomized by
32 gender. No records of anyone who left employment or died before January 1, 1972 in one plant
33 and January 1, 1975 in the other were included. Expected incidence rates were generated from
34 the Southern Swedish Regional Tumor Registries.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-18 DRAFT—DO NOT CITE OR QUOTE
-------
1 Because of a short follow-up period and the relative young age of the cohort, little
2 morbidity had occurred by the end of the cutoff date of December 31, 1990. Altogether,
3 40 cancers occurred, compared with 46.3 expected. After 10 years latency, 22 cases of cancers
4 were diagnosed versus 22.6 expected. However, 6 lymphohematopoietic cancers were observed
5 versus 3.37 expected, and when latency is considered, this figure falls to 3 versus 1.51 expected.
6 The authors pointed out that for leukemia the standard incidence ratio (SIR) is a nonsignificant
7 7.14, based on 2 cases in 930 subjects having at least 0.14 ppm-years of cumulative exposure to
8 EtO and a minimum of 10 years latency. The authors believed that the results provided some
9 minor evidence to support an association between exposure to EtO and an increased risk of
10 leukemia. However, for breast cancer, no increase in the risk was apparent for the total cohort
11 (SIR = 0.46, OBS = 5). Even in the 10-years or more latency period, the risk was less than
12 expected (SIR = 0.36, OBS = 2).
13 The authors made a reasonably good attempt to determine exposure levels during the
14 periods of employment in both plants for six job categories. Sterilizers in the years 1970-1972
15 were exposed to an average 40 ppm in both plants. These levels gradually dropped to 0.75 ppm
16 by 1985-1986. Packers and developmental engineers were the next highest exposed employees,
17 with levels in 1970-1972 of 20 to 35 ppm and by 1985-1986 of less than 0.2 ppm. During the
18 period 1964-1966 in the older plant, EtO levels averaged 75 ppm in sterilizers and 50 ppm in
19 packers. Peak exposures were estimated to have ranged from 500 to 1,000 ppm during the
20 unloading of autoclaves up to 1973. The levels gradually dropped to less than 0.2 ppm in both
21 plants by 1985-1986 in all job categories (developmental engineers, laboratory technicians,
22 repair men, store workers, controllers, foremen, and others) except sterilizers.
23 These exposure estimates were verified by measurement of hydroxy ethyl adducts to
24 N-terminal valine in hemoglobin in a sample of subjects from both plants. The adduct levels
25 reflect the average exposure during the few months prior to the measurement of EtO. The results
26 of this comparison were close except for sterilizers, whose air monitoring measurements were
27 2 to 3 times higher.
28 The authors pointed out two limitations in their study: a minority of subjects had a high
29 exposure to EtO, and the follow-up (median 11.8 years) resulted in relatively few person-years at
30 risk and was insufficient to assess the influence of a biologically relevant induction latency
31 period. Although this study has good exposure information and the authors used this information
32 to develop an exposure index per employee, they did not evaluate dose-response relationships
33 that might have been present, nor did they follow the cohort long enough to evaluate morbidity.
34 The strength of this study is the development of the cumulative exposure index as well as the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-19 DRAFT—DO NOT CITE OR QUOTE
-------
1 absence of any potential confounding produced by the chlorohydrin process, which was a
2 problem in workers who produced and manufactured EtO in other similar studies.
O
4 A.2.12. Norman et al. (1995)
5 These authors conducted a mortality/incidence study in a cohort of 1,132 workers, mainly
6 women (82%), who were exposed to EtO at some time during the period July 1, 1974, through
7 September 30, 1980. Follow-up was until December 31, 1987. Ethylene oxide was used at the
8 study plant to sterilize medical equipment and supplies that were assembled and packaged there.
9 This plant was selected for the study because in an earlier small study at this plant (Stolley et al.,
10 1984) there was an indication that in a sample of workers the average number of sister chromatid
11 exchanges was elevated over that of a control group selected from the nearby community.
12 Cancer morbidity was measured by comparing cancers occurring in this cohort with those
13 predicted from the National Cancer Institute's Surveillance, Epidemiology, and End Results
14 (SEER) Program for the period 1981-1985 and with average annual cancer incidence rates for
15 western New York for 1979-1984. Observed cancers were compared to expected cancers using
16 this method.
17 Only 28 cancer diagnoses were reported in the cohort; 12 were for breast cancers. Breast
18 cancer was the only cancer site in this study where the risk was significantly elevated, based on
19 the SEER rates (SIR = 2.55,/> < 0.05). No significant excesses were seen at other cancer sites of
20 interest: leukemia (1 observed, 0.54 expected), brain (0 observed, 0.49 expected), pancreas
21 (2 observed, 0.51 expected) and stomach (0 observed, 0.42 expected). The authors offered no
22 explanation except chance as to why the risk of breast cancer was elevated in these workers.
23 In 1980, three 2-hour samples from the plant provided 8-hour TWA exposures to
24 sterilizer operators that ranged from 50 to 200 ppm. Corrective action reduced the levels to 5 to
25 20 ppm.
26 This study has little power to detect any significant risk of cancer at other sites because
27 morbidity was small, chiefly as a consequence of the short follow-up period. The mean number
28 of years from the beginning of follow-up to the end of the study was 11.4 years. In fact, the
29 authors stated that breast cancer was the only cancer site for which there was adequate power to
30 detect an increased relative risk. Additional weaknesses in this study include no historic
31 exposure information and too short a period of employment in some cases (<1 month) to result in
32 breast cancer. The authors maintained that their study was inconclusive.
33
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-20 DRAFT—DO NOT CITE OR QUOTE
-------
1 A.2.13. Swaen et al. (1996)
2 A significant cluster of 10 Hodgkin lymphoma cases in the active white male workforce
3 of an unidentified large chemical manufacturing plant in Belgium led to a nested case-control
4 study by Swaen et al. (1996) to determine which, if any, chemical agents within the plant may
5 have led to the increase. By comparison with regional cancer incidence rates, the SIR for this
6 disease was 4.97 (95% CI = 2.38-9.15) over a 23-year period, from 1966 to 1992. This
7 suggested that an occupational exposure may have produced the significant excess risk of
8 Hodgkin lymphoma seen in these workers.
9 The investigators randomly selected 200 individuals from a computerized sampling frame
10 of all men ever employed at the facility. From this list of 200, workers who were actively
11 employed at the time of diagnosis of each case were chosen as controls. No age matching was
12 done because the authors stated that age-specific incidence rates for Hodgkin lymphoma in the
13 United States were relatively flat for men between ages 18 and 65. The investigators felt that a
14 control could serve for more than one case.
15 Verification of the 10 cases revealed that 1 case was, in reality, a large-cell anaplastic
16 lymphoma. Two others could not be confirmed as Hodgkin lymphoma due to the lack of tissue.
17 The remaining seven were confirmed as Hodgkin lymphoma. In the ensuing case-control
18 analysis, significant odds ratios (ORs) for Hodgkin lymphoma were observed for five chemicals,
19 ammonia (6 cases, OR =5.6), benzene (5 cases, OR = 11), EtO (3 cases, OR =8.5), NaOH
20 (5 cases, OR = 8) and oleum (3 cases, OR = 6.9), based on the number of cases and controls
21 known to be exposed to the chemicals in question. This does not mean they were exposed only
22 to the chemical in question.
23 The availability of exposure information made it possible to calculate cumulative
24 exposure to the cases and controls of two chemicals, benzene and EtO. The cumulative exposure
25 for benzene-exposed cases was 397.4 ppm-months versus an expected 99.7 ppm-months for the
26 matched controls. The authors stated that one heavily exposed case was chiefly responsible for
27 the high cumulative total for all the benzene-exposed cases; however, it was not statistically
28 significant. Only a few studies have suggested that exposure to benzene could possibly be
29 related to an increase in the risk of Hodgkin lymphoma. The cumulative total exposure to EtO
30 for the cases was 500.2 ppm-months versus 60.2 for the matched controls, which was statistically
31 significant, the significance being due to one extreme case.
32 This study is limited because the authors enumerated only cases among active employees
33 of the workforce; therefore, the distinct possibility exists that they could have missed potential
34 cases in the inactive workers. It is possible that latent Hodgkin lymphoma cases could have been
35 identified in the controls after the controls left active employment. However, given that there
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-21 DRAFT—DO NOT CITE OR QUOTE
-------
1 were many different possible exposures to the chemicals produced in the workplaces of these
2 employees, it is not likely that EtO or benzene could be considered solely responsible for the
3 excess risk of Hodgkin lymphoma in this working group.
4
5 A.2.14. Olsen et al. (1997)
6 Olsen et al. (1997) studied 1,361 male employees of four plants in Texas, Michigan, and
7 Louisiana who were employed a minimum of 1 month sometime during the period 1940 through
8 1992 in the ethylene chlorohydrin and propylene chlorohydrin process areas. These areas were
9 located within the EtO and propylene oxide production plants. Some 300 deaths had occurred by
10 December 31, 1992.
11 Plant A in Texas produced EtO beginning in 1941 and ceased production in 1967.
12 Bis-chloroethyl ether, a byproduct of EtO continued to be produced at this plant until 1973. The
13 plant was demolished in 1974. Plant B, which was nearby, manufactured EtO from 1951 to 1971
14 and then again from 1975 until 1980. This plant continues to produce propylene oxide. The
15 Louisiana plant produced EtO and propylene oxide through the propylene chlorohydrin process
16 from 1959 until 1970, when it was converted to propylene oxide production. The Michigan plant
17 produced ethylene chlorohydrin and subsequently EtO beginning in 1936 and continuing into the
18 1950s. This plant produced propylene chlorohydrin and propylene oxide up to 1974.
19 The authors suggested that exposure to EtO was possible at the plants studied in this
20 report but that exposure was unlikely in the 278 chlorohydrin unit workers who were excluded
21 from the cohort studied by Teta et al. (1993). Unfortunately, no actual airborne measurements
22 were reported by Olsen et al. (1997), and thus only length of employment could be used as a
23 surrogate for exposure.
24 The SMR for all causes was 0.89 (300 observed). For total cancer the SMR was 0.94
25 (75 observed, 79.7 expected). There were 10 lymphohematopoietic cancers versus 7.7 expected
26 (SMR = 1.3). No significantly increased risks of any examined site-specific cancer (pancreatic,
27 lymphopoietic, hematopoietic, and leukemia) were noted even after a 25-year induction latency
28 period, although the SMR increased to 1.44 for lymphopoietic and hematopoietic cancer. When
29 only the ethylene chlorohydrin process was examined after 25 years latency, the SMR increased
30 to 1.94, based on six observed deaths. The data to support the latter observation by the authors
31 were not presented in tabular form.
32 The authors concluded that there was a weak, nonsignificant, positive association with
33 duration of employment for lymphopoietic and hematopoietic cancer with Poisson regression
34 modeling. They stated that the results of their study provide some assurance that their cohort has
35 not experienced a significant increased risk for pancreatic cancer and lymphopoietic and
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-22 DRAFT—DO NOT CITE OR QUOTE
-------
1 hematopoietic cancer in ethylene chlorohydrin and propylene chlorohydrin workers. They
2 believed that this study contradicted the conclusions of Benson and Teta (1993) that ethylene
3 dichloride, perhaps in combination with chlorinated hydrocarbons, appeared to be the causal
4 agent in the increased risk of pancreatic cancer and hematopoietic cancer seen in their study.
5 They pointed out that ethylene dichloride is readily metabolized and rapidly eliminated from the
6 body after gavage or inhalation administration; therefore, they questioned whether experimental
7 gavage studies (NCI, 1978) are appropriate for studying the effects of ethylene dichloride in
8 humans. One study (Maltoni et al., 1980) found no evidence of tumor production in rats and
9 mice chronically exposed to ethylene dichloride vapor concentrations up to 150 ppm for 7 hours
10 a day. Also, because this chemical is a precursor in the production of vinyl chloride monomer,
11 the authors wondered why an increase in these two site-specific cancers had not shown up in
12 studies of vinyl chloride workers. However, they believe that an additional 5 to 10 years of
13 follow-up of this cohort would be necessary to confirm the lack of risk for the two types of
14 cancer described above.
15 Another maj or weakness of this study is the lack of any actual airborne measurements of
16 EtO and the chlorohydrin chemicals.
17
18 A.2.15. Steenland et al. (2004)
19 In an update of the earlier mortality studies of the same NIOSH cohort of workers
20 exposed to EtO described by Steenland et al. (1991) and Stayner et al. (1993), an additional
21 11 years of follow-up were added. This increased the number of deceased to 2,852. Work
22 history data were originally gathered in the mid-1980s. Approximately 25% of the cohort
23 continued working into the 1990s. Work histories on these individuals were extended to the last
24 date employed. It was assumed that these employees continued in the job they last held in the
25 1980s. Little difference was noted when cumulative exposure was calculated with and without
26 the extended work histories, chiefly because the exposure levels after the mid-1980s were very
27 low (see Section A.2.8 for a discussion of the NIOSH exposure assessment). Again, no excess
28 risk of hematopoietic cancer was noted based on external rates. However, as in the earlier paper,
29 exposure-response analyses reported positive trends for hematopoietic cancers limited to males
30 (p = 0.02 for the log of cumulative exposure with a 15-year lag) using internal comparisons and
31 Cox regression analysis.l (See Table A-2 for the categorical exposure results.)
1 Valdez-Flores et al. (2010) suggest that Steenland et al. (2004) incorrectly used one degree of freedom in their
evaluation of statistical significance and that a second degree of freedom should have been included for estimating
the lag. However, Steenland et al. (2004) did not estimate the lag using the likelihood; rather, lagged exposure was
treated as an alternate exposure metric.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
The excess of these tumors was chiefly lymphoid (NHL, myeloma, lymphocytic
leukemia) (see Table A-3), as in the earlier paper. A positive trend was also observed for
Hodgkin lymphoma in males, although this was based on small numbers.
Table A-2. Cox regression results for hematopoietic cancer mortality
(15-year lag) in males
Cumulative exposure (ppm-days)
0
>0-1,199
1,200-3,679
3,680-13,499
13,500+
Odds
ratio (95% CI)
1
1.23 (0.32-4.73)
2.52 (0.69-9.22)
3.13 (0.95-10.37)
3.42 (1.09-10.73)
Source: Steenland et al. (2004)
Table A-3. Cox regression results for lymphoid cell line tumors (15-year lag)
in males
Cumulative exposure (ppm-days)
0
>0-1,199
1,200-3,679
3,680-13,499
13,500+
Odds ratio (95% CI)
1
0
2
2
o
5
.9(0.16-5.24)
.89 (0.65-12.86)
.74(0.65-11.55)
.76 (1.03-13.64)
Source: Steenland et al. (2004).
The hematopoietic cancer trends were somewhat weaker in this analysis than were those
reported in the earlier studies of the same cohort. This is not unexpected because most of the
cohort was not exposed after the mid-1980s, and the workers who were exposed in more recent
years were exposed to much lower levels because EtO levels decreased substantially in the early
1980s. No association was found in females, although average exposures were only twice as
high in males (37.8 ppm-years) as in females (18.2 ppm-years), and there was enough variability
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-24 DRAFT—DO NOT CITE OR QUOTE
-------
1 in female exposure estimates to expect to be able to see a similar trend if it existed. In later
2 analyses conducted by Dr. Steenland and presented in Appendix D, the difference between the
3 male and female results was found not to be statistically significant, and the same pattern of
4 lymphohematopoietic cancer results observed for males by Steenland et al. (2004) was observed
5 for the males and females combined (i.e., statistically significant positive trends for both
6 hematopoietic and lymphoid cancers using log cumulative exposure and a 15-year lag).
7 This study also reports a significant excess risk of breast cancer in the highest
8 cumulative-exposure quartile, with a 20-year lag (SMR = 2.07, 95% CI 1.1-3.54, n = 13) in
9 female employees. The results using internal Cox regression analyses with a 20-year lag time
10 produced an OR = 3.13 (95% CI 1.42-6.92) in the highest cumulative-exposure quartile. The
11 log of cumulative exposure with a 20-year lag was found to be the best model (p = 0.01) for the
12 analyses of breast cancer. As for hematopoietic cancer in males, cumulative exposure
13 untransformed showed a weaker trend (p = 0.16). A breast cancer incidence study of this cohort
14 is discussed in Steenland et al. (2003).
15
16 A.2.16. Steenland et al. (2003)
17 In a companion study on breast cancer incidence in women employees of the same cohort
18 discussed in Steenland et al. (2004), the authors elaborated on the breast cancer findings in a
19 subgroup of 7,576 women from the cohort (76% of the original cohort). They had to be
20 employed at least 1 year and exposed while employed in commercial sterilization facilities. The
21 average length of exposure was 10.7 years. Breast cancer incidence analyses were based on
22 319 cases identified via interview, death certificates, and cancer registries in the full cohort,
23 including 20 in situ carcinomas. Interviews on 5,139 women (68% of the study cohort) were
24 obtained (next-of-kin interviews were sought for the 18% of the cohort who were deceased);
25 22% could not be located. Using external referent rates (SEER), the SIR was 0.87 for the entire
26 cohort based on a 15-year lag time. When in situ cases were excluded, the overall SIR increased
27 to 0.94. In the top quintile of cumulative exposure, with a 15-year lag time, the SIR was 1.27
28 (95% CI 0.94-1.69, n = 48). A significant positive linear trend of increasing risk with increasing
29 cumulative exposure was noted (p = 0.002) with a 15-year lag time. Breast cancer incidence was
30 believed to be underascertained owing to incomplete response and a lack of coverage by regional
31 cancer registries (68% were contacted directly and 50% worked in areas with cancer registries).
32 An internal nested case-control analysis, which is less subject to concerns about
33 underascertainment, produced a significant positive exposure-response with the log of
34 cumulative exposure and a 15-year lag time (p = 0.05). The top quintile was significant with an
35 OR of 1.74 (CI 1.16-2.65) based on all 319 cases (the entire cohort).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-25 DRAFT—DO NOT CITE OR QUOTE
-------
1 The authors also conducted separate analyses using the subcohort with interviews, for
2 which there was complete case ascertainment and additional information on potential
3 confounders. In the subcohort with interview data, the odds ratio for the top quintile equaled
4 1.87 (CI 1.12-3.1), based on 233 cases in the 5,139 women and controlled for with respect to
5 parity and breast cancer in a first-degree relative. Information on other risk factors was also
6 collected—e.g., body mass index, SES, diet, age at menopause, age at menarche, breast cancer in
7 a first-degree relative, and parity—but only parity and breast cancer in a first-degree relative
8 were significant in the model. Continuous cumulative exposure, as well as the log cumulative
9 exposure, lagged 15 years, produced ^-values for the regression coefficient of 0.02 and 0.03,
10 respectively, for the Cox regression model, taking into account age, race, year of birth, parity,
11 and breast cancer in a first-degree relative.
12 The authors concluded that their data suggest that exposure to EtO is associated with
13 breast cancer, but because of inconsistencies in exposure-response trends and possible biases due
14 to nonresponse and incomplete cancer ascertainment, the case for breast cancer is not conclusive.
15 However, monotonically increasing trends in categorical exposure-response relationships are not
16 always the norm owing to lack of precision in the estimates of exposure. Furthermore, positive
17 trends were observed in both the full cohort and the subcohort with interviews, lessening
18 concerns about nonresponse bias and case underascertainment.
19
20 A.2.17. Kardos et al. (2003)
21 These authors reported on a study completed earlier by Muller and Bertok (1995) of
22 cancer among 299 female workers who were employed from 1976 to 1993 in a pediatric ward at
23 the county hospital in Eger, Hungary, where gas sterilizers were used. Their observation period
24 for cancer was begun in 1987 on the assumption that cancer deaths before 1987 were not due to
25 EtO, based on a paper by Lucas and Teta (1996). Information about the Muller and Bertok
26 (1995) study is unavailable because the paper is in Hungarian and no translated copy is available.
27 Kardos and his colleagues evaluated mortality among these women and found a statistically
28 significant excess of total cancer deaths (n= 11) in the period from 1987 to 1999 when compared
29 with expected deaths generated from three different comparison populations (Hungary, n = 4.38;
30 Heves County, n = 4.03; and city of Eger, n = 4.28). The SMRs are all significant at the
31 p< 0.01 level. Site-specific rates were not calculated. Among the 11 deaths were 3 breast
32 cancer deaths and 1 lymphoid leukemia death. The authors claim that their results confirm
33 "predictions of an increased cancer risk for the Eger hospital staff." They suggest an etiological
34 role for EtO in the excess risk. The observation of 3 breast cancer deaths, with at most 4.4 (with
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-26 DRAFT—DO NOT CITE OR QUOTE
-------
1 Hungarian national rates as the referent) total cancer deaths expected, is indicative of an
2 increased risk of breast cancer2.
O
4 A.2.18. Tompa et al. (1999)
5 The authors reported a cluster of eight breast cancer cases and eight other malignant
6 tumor cases that developed over a period of 12 years in 98 nurses who worked in a hospital in
7 the city of Eger, Hungary, and were exposed to EtO. These nurses were exposed for 5 to
8 15 years in a unit using gas sterilizer equipment. The authors report that EtO concentrations
9 were in the neighborhood of 5 to 150 mg/m3. The authors state that the high breast cancer
10 incidence in the hospital in Eger indicates a combined effect of exposure to EtO and naturally
11 occurring radioactive tap water, possibly due to the presence of radon. This case report study is
12 discussed further in the genotoxicity section.
13
14 A.2.19. Coggon et al. (2004)
15 Descriptive information about this cohort is available from the earlier study (Gardner et
16 al., 1989). In this update, the 1,864 men and 1,012 women described in the (Gardner et al.,
17 1989) study were followed to December 31, 2000. This added 13 more years of follow-up
18 resulting in 565 observed deaths versus 607.6 expected. For total cancer, the observed number
19 of deaths equaled 188 versus 184.2 expected. For NHL, 7 deaths were observed versus
20 4.8 expected. For leukemia, 5 deaths were observed versus 4.6 expected. All 5 leukemia deaths
21 fell into the subset with definite or continual exposure to EtO, where only 2.6 were expected. In
22 fact, the total number of deaths classified to the lymphohematopoietic cancer category was 17
23 with 12.9 expected. This increased risk was not significant. When definite exposure was
24 established, the authors found that the risk of lymphatic and hematopoietic cancer was increased
25 with 9 observed deaths versus 4.9 expected. Deaths from leukemia were also increased in
26 chemical workers with 4 leukemia deaths versus 1.7 expected. No increase was seen in the risk
27 of hematopoietic cancer in the hospital sterilizing unit workers, who are mostly female. Another
28 finding of little significance was that of cancer of the breast. Only 11 deaths were recorded in
29 this cohort up to the cutoff date versus 13.1 expected. Since there were no female workers in the
Hungarian age-standardized female cancer mortality rates reported by the International Agency for Research on
Cancer (http://eu-cancer.iarc.fr/country-348-hungary.html,en) suggest that the ratio of breast cancer deaths to total
cancer deaths in Hungarian females is about 0.16 (28.0/100,000 breast cancer mortality rate versus
180.0/100,000 total cancer mortality rate). Although a comparison of this general population ratio with the ratio of
0.68 for breast cancer to total cancer mortality in the Kardos et al. (2003) study is necessarily crude because the
general population ratio is not based on the age-standardized rates that would correspond to the age distribution of
the person-time of the women in the study, which are unknown, the large difference between the ratios (0.68 for the
study versus 0.16 for the general population) indicates an increased risk of breast cancer in the study.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-27 DRAFT—DO NOT CITE OR QUOTE
-------
1 chemical industry, the results on breast cancer reflect only work in hospital sterilizing units. The
2 researchers concluded that the risk of cancer must be low at the levels sustained by workers in
3 Great Britain over the last 10 or 20 years.
4
5 A.2.20. Swaen et al. (2009)
6 Swaen et al. (2009) redefined and updated the cohort of 1,896 male UCC workers studied
7 by Teta et al. (1993), which was itself a follow-up of the 2,174 UCC workers originally studied
8 by Greenberg et al. (1990), excluding the 278 chlorohydrin unit workers because of potential
9 confounding. (However, confounding by chlorohydrin production has not been established, and
10 49 of those excluded workers were also employed in EtO production and thus had high potential
11 EtO exposures.) Specifically, Swaen et al. (2009) extended the cohort enumeration period from
12 the end of 1978 to the end of 1988 (workers hired after 1988 were not added to the cohort
13 because they were considered to have no appreciable EtO exposure), identifying 167 additional
14 workers, and conducted mortality follow-up of the resulting cohort of 2,063 male workers
15 through 2003. Work histories were also extended through 1988; exposures after 1988 were
16 considered negligible compared to earlier exposure levels. Swaen et al. (2009) used an exposure
17 assessment reportedly based on the qualitative categorizations of potential for EtO exposure in
18 the different departments developed by Greenberg et al. (1990) and time-period exposure
19 estimates from Teta et al. (1993). The exposure assessment matrix for the exposure estimates of
20 Swaen et al. (2009) is presented in Table A-4 below. Cumulative exposures for the individual
21 workers were estimated by multiplying the time (in months) a worker was assigned to a
22 department by the estimated exposure level for the department and summing across the
23 assignments.
24 The exposure assessment used in this study was relatively crude, based on just a small
25 number of department-specific and time-period-specific categories, and with exposure estimates
26 for only a few of the categories derived from actual measurements. For the 1974-1988 time
27 period, based on measurements from environmental monitoring conducted in the (West Virginia)
28 plants since 1976, exposure estimates of 1 ppm and 0.3 ppm were chosen for the high- and
29 low-exposure-potential departments, respectively, and the average of 0.65 ppm was taken for the
30 medium-exposure-potential departments. For the 1957-1973 time period, exposure estimates
31 were based on measurements from an air-sampling survey of three EtO direct-oxidation
32 production units in a UCC plant in Texas in the early 1960s (during this 1957-1973 time period,
33 direct oxidation was the only method used for EtO production at the West Virginia plants as
34 well). The majority of the 8-hour TWA results in these units were between 3 and 20 ppm, with
35 levels between 5 and 10 ppm for operators. Because the West Virginia plants and equipment
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
were much older than for the Texas facility, the high end of the range of values for operators
(10 ppm) was selected as the exposure estimate for the high-exposure-potential departments, and
the low end of the range (5 ppm) was selected for the low-exposure-potential departments (even
though these were not EtO production departments). The average of 7.5 ppm was taken for the
medium-exposure-potential departments.
Table A-4. Exposure assessment matrix from Swaen et al. (2009)—8-hour
TWA exposures in ppm
Time period
1925-1939
1940-1956
1957-1973
1974-1988
Exposure potential category
Low
(most EtO user
departments)
17
7
5
0.3
Medium
(some EtO user
departments)
28
14
7.5
0.65
High
(EtO production
departments)
70
21
10
1
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Source: Swaen et al. (2009).
For the 1940-1956 time period, exposure estimates were derived from "rough" estimates
of exposure reported by Hogstedt et al. (1986) for a chlorohydrin-based EtO production unit in
an enclosed building, as was the West Virginia chlorohydrin-based EtO production. Hogstedt et
al. (1986) reportedly suggested EtO exposures were probably below 14 ppm from 1941 to 1947,
although much higher levels occasionally occurred, and levels from the 1950s to 1963 averaged
5 to 25 ppm. Thus, based on these values, 14 ppm was selected as the exposure estimate for the
medium-exposure-potential departments and values 50% higher (21 ppm) and 50% lower
(7 ppm) were assigned to the high- and low-exposure-potential departments, respectively. For
the 1925-1939 time period, it was assumed that exposures in this earlier, start-up period would
have been higher than those in the subsequent 1940-1956 time period, so the 14 ppm estimate
from the medium-exposure-potential departments in the 1940-1956 time period was used as the
exposure estimate for the low-exposure-potential departments for the 1925-1939 time period.
Then, the same ratio of 1:2 between the low- and medium-exposure-potential departments from
the 1940-1956 time period was used to obtain an estimate of 28 ppm for the medium-exposure-
potential departments for the 1925-1939 time period. A factor of 5 (half an order of magnitude)
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
A-29 DRAFT—DO NOT CITE OR QUOTE
-------
1 was used between the low- and high-exposure-potential departments to obtain a highly uncertain
2 exposure estimate of 70 ppm for the high-exposure-potential departments. Swaen et al. (2009)
3 suggest that despite the high exposure estimates for the 1925-1939 time period, the contribution
4 of this time period to cumulative exposure estimates is limited because only 98 workers (4.8% of
5 the cohort) had employment histories before 1940. It appears, then, that pre-1940 employment
6 histories may have been missing for 13 of the workers, because excluding the 112 pre-1940
7 chlorohydrin production workers (Benson and Teta, 1993) from the original 223 pre-1940
8 workers (Greenberg et al., 1990) leaves 111 pre-1940 workers in the cohort.
9 At the end of the 2003 follow-up, 1,048 of the 2,063 workers had died and 23 were lost to
10 follow-up. In comparison with general population U.S. mortality rates, the all-cause mortality
11 SMR was 0.85 (95% CI = 0.80, 0.90) and the cancer SMR was 0.95 (95% CI = 0.84, 1.06).
12 None of the SMRs for specific cancer types showed any statistically significant increases. In
13 analyses stratified by hire date [pre- (inclusive) or post-1956], the SMR for leukemia was
14 elevated but not statistically significant (1.51; 95% CI 0.69, 2.87) in the early-hire group, based
15 on nine deaths. In analyses stratified by duration of employment, no trends were apparent for
16 any of the lymphohematopoietic cancers, although in the 9+ years of employment subgroup, the
17 SMR for NHL was nonsignificantly increased (1.49; 95% CI 0.48, 3.48), based on 5 deaths. In
18 SMR analyses stratified by cumulative exposure, no trends were apparent for any of the
19 lymphohematopoietic cancers and there were no notable elevations for the highest cumulative
20 exposure category. Note that only 27 lymphohematopoietic cancer deaths (including
21 12 leukemias and 11 NHLs) were observed in the cohort.
22 Internal Cox proportional hazards modeling was also done for some disease categories
23 (all-cause mortality, leukemia mortality, and lymphoid cancer [NHL, lymphocytic leukemia, and
24 myeloma] mortality [17 deaths]), using cumulative exposure as the exposure metric. Year of
25 birth and year of hire were included as covariates in the Cox regression model. Year of hire was
26 reportedly included to adjust for potential cohort effects; however, it is unclear whether or not
27 this covariate was a statistically significant factor in the regression. Furthermore, because age at
28 hire is often correlated with exposure, including it in the regression model could overadjust and
29 attenuate the observed exposure-related effects. These internal analyses showed no evidence of
30 an exposure-response relationship, although, again, these analyses rely on small numbers of
31 cases and a crude exposure assessment, where there is a high potential for exposure
32 misclassification.
33 Swaen et al. (2009) note that one of the strengths of their study is the long average
34 follow-up time of the workers. These authors further note that, because the UCC cohort is a
35 much older population (50% deceased) than the NIOSH cohort (Steenland et al., 2004), the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-3 0 DRAFT—DO NOT CITE OR QUOTE
-------
1 number of expected deaths is less than 3 times larger for the NIOSH cohort even though the
2 sample size is almost 9 times larger. However, the long follow-up and aged cohort might be a
3 limitation, as well. Because the follow-up is extended well beyond the time period of
4 nonnegligible exposures (pre-1989) for workers still employed and, especially, beyond the
5 highest exposures (e.g., pre-1940 or pre-1956), the follow-up is likely observing workers at the
6 high tail end of the distribution of latency times for EtO-associated lymphohematopoietic
7 cancers. In other words, workers that were at risk of developing lymphohematopoietic cancer as
8 a result of their EtO exposures would likely have developed the disease earlier. Meanwhile,
9 having an older cohort means that the background rates of lymphohematopoietic cancers are
10 higher, and thus, relative risks may be attenuated. Such attenuation was observed even in the
11 younger NIOSH cohort between the 1987 follow-up (Steenland et al., 1991) and the 1998
12 follow-up (Steenland et al., 2004), when the follow-up was extended well beyond the period of
13 significant EtO exposures (exposure levels were considered very low by the mid-1980s).
14 Swaen et al. (2009) also note that their estimate of the average cumulative exposure for
15 the UCC cohort was more than twice the average cumulative exposure estimate for the NIOSH
16 cohort. However, there are substantial uncertainties in the exposure assessment, especially for
17 the early years when the highest exposures occurred. And despite the reported strengths of the
18 Swaen et al. (2009) study in terms of follow-up, cohort age, and high exposures, a limitation of
19 the study is the small cohort size. Based on data presented by Greenberg et al. (1990) and
20 Benson and Teta (1993), it appears that fewer than 900 workers were hired before 1956 (1,104 of
21 the original cohort were hired before 1960 and 233 of those were then excluded because they
22 worked in the chlorohydrin unit) and would have been potentially exposed to the higher pre-1956
23 exposures levels. In the full cohort of 2,063 men, only 27 lymphohematopoietic (17 lymphoid)
24 cancers were observed.
25 In alternate analyses of the UCC data, Valdez-Flores et al. (2010) fitted Cox proportional
26 hazards models and conducted categorical exposure-response analyses using a larger set of
27 cancer endpoints. These investigators also performed the same analyses using the data from the
28 last follow-up of the NIOSH cohort (Steenland et al., 2004) and from the two cohorts combined,
29 analyzing the sexes both separately and together. Valdez-Flores et al. (2010) reported that they
30 found no evidence of exposure-response relationships for cumulative exposure with either the
31 Cox model or categorical analyses for all of the cohort/endpoint data sets examined (endpoints
32 included all lymphohematopoietic cancers, lymphoid cancers, and female breast cancer, the latter
33 in the NIOSH cohort only). These investigators suggest that a review of the data from the
34 NIOSH and UCC studies supports combining them, but it should be recognized that the exposure
35 assessment conducted for the UCC cohort is much cruder, especially for the highest exposures,
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-31 DRAFT—DO NOT CITE OR QUOTE
-------
1 (see above) than the NIOSH exposure assessment (which was based on a validated regression
2 model; see A.2.8 above); thus, the results of exposure-response analyses of the combined cohort
3 data are considered to have greater uncertainty than those from analyses of the NIOSH cohort
4 alone, despite the additional cases contributed by the UCC cohort (e.g., the UCC cohort
5 contributes 17 cases of lymphoid cancer to the 53 from the NIOSH cohort; however, as discussed
6 above, it should also be noted that some of these UCC cases are occurring in older workers, with
7 longer postexposure follow-up, and thus, may reflect background disease more than
8 exposure-related disease).
9 Notable differences between the Steenland et al. (2004) and the Valdez-Flores et al.
10 (2010) analyses exist. A major difference is that Valdez-Flores et al. (2010) used only
11 cumulative exposure in the Cox regression model, so they considered only a sublinear
12 exposure-response relationship, whereas Steenland et al. (2004) also used log cumulative
13 exposure, which provides a supralinear exposure-response relationship model structure [e.g., see
14 Figure 4-1, illustrating the difference between the cumulative exposure and log cumulative
15 exposure Cox regression models [RR = ePxexP°sure] for the lymphoid cancers from Steenland et al.
16 (2004)]. Valdez-Flores et al. (2010) objected to the log cumulative exposure model for a number
17 of reasons, the primary one being that the use of log cumulative exposure forces the
18 exposure-response relationship to be supralinear regardless of the observed data. This is correct
19 but no different from the use of cumulative exposure imposing a sublinear exposure-response
20 relationship. Moreover, Steenland et al. (2004) used log cumulative exposure specifically when
21 the cumulative exposure Cox regression model did not yield a statistically significant fit to the
22 exposure-response data and the categorical analyses suggested increases in risk that were more
23 consistent with an underlying supralinear exposure-response relationship. With log cumulative
24 exposure, Steenland et al. (2004) observed statistically significant fits to the exposure-response
25 data for all lymphohematopoietic cancers in males, lymphoid cancers in males, and breast cancer
26 in females, none of which yielded statistically significant fits with the cumulative exposure
27 (sublinear exposure-response) model, supporting the apparent supralinearity of the data.3
28 Another key difference between the Steenland et al. (2004) and the Valdez-Flores et al.
29 (2010) analyses is that Valdez-Flores et al. (2010) present results only for unlagged analyses.
30 Valdez-Flores et al. (2010) state that their Cox regression results with different lag times were
31 similar to the unlagged results. Because the Valdez-Flores et al. (2010) categorical results are
32 for unlagged analyses, however, their referent groups are different from those used by Steenland
3This pattern of findings from the NIOSH cohort data for males (i.e., statistically significant fits with log cumulative
exposure but not with cumulative exposure) was replicated for both the all lymphohematopoietic cancers and the
lymphoid cancers when the NIOSH data on males and females were combined (see Appendix D).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-32 DRAFT—DO NOT CITE OR QUOTE
-------
1 et al. (2004). Valdez-Flores et al. (2010) used the lowest exposure quintile (providing there were
2 sufficient data) as the referent group, whereas Steenland et al. (2004) used the no-exposure
3 (lagged-out) group as the referent. Because the NIOSH cohort data have an underlying
4 supralinear exposure-response relationship, the increased risk in the lowest exposure group is
5 already notably elevated and using the lowest exposure quintile as a referent group would
6 attenuate the relative risk. Nonetheless, Valdez-Flores et al. (2010) observed statistically
7 significant increases in response rates in the highest exposure quintile relative to the lowest
8 exposure quintile for lymphohematopoietic and lymphoid cancers in males in the NIOSH cohort,
9 consistent with the categorical results of Steenland et al. (2004), as well as a statistically
10 significant increase in the highest exposure quintile for lymphoid cancers in males and females
11 combined in the NIOSH cohort, consistent with the results in Appendix D.4
12 Although Valdez-Flores et al. (2010) found no statistically significant exposure-response
13 relationships for any of the cohort/endpoint data sets that they analyzed using the cumulative
14 exposure Cox regression model, these investigators derived risk estimates from the positive
15 relationships for the purposes of comparing those estimates with EPA's 2006 draft risk estimates
16 (U.S. EPA, 2006a). Valdez-Flores et al. (2010) report that their estimate of the exposure level
17 associated with 10 6 risk of lymphohematopoietic cancer based on the male NIOSH cohort data
18 is 1,500 times larger than EPA's 2006 draft estimate (their exposure level estimate based on the
19 NIOSH and UCC male and female data combined was a further 3 times higher). Most of the
20 difference in magnitude between the Valdez-Flores et al. (2010) and the EPA 2006 draft
21 estimates is attributable to the difference in the models used. The Valdez-Flores et al. (2010)
22 estimate is based on the sublinear Cox regression model, which EPA rejected as not providing a
23 good representation of the low-exposure data (EPA's 2006 draft risk estimate is based on a linear
24 model). In addition, Valdez-Flores et al. (2010) used maximum likelihood estimates, while EPA
25 uses upper bounds on risk (or lower bounds on exposure). Valdez-Flores et al. (2010) also
26 modeled down to 10~6 risk, whereas EPA modeled to 10~2 risk and used the LECoi as a point of
27 departure (POD) for linear low-dose extrapolation. Valdez-Flores et al. (2010) suggest that
28 PODs should be within the range of observed exposures, and they chose a 10 6 risk level because
29 the corresponding exposure level was in the range of the observed occupational exposures
30 (converted to equivalent environmental exposures). The intention of EPA's 2005 Guidelines for
31 Carcinogen Risk Assessment (U.S. EPA, 2005a), however, is for the POD to be (or more
32 specifically, to correspond to a response level) at the low end of the observable range of
33 responses (i.e., a response level that might reasonably be observed to have statistical significance
4In Dr. Steenland's analyses of the NIOSH cohort data for both sexes combined, presented in Appendix D, the
categorical results for all lymphohematopoietic cancers were also statistically significantly increased.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-3 3 DRAFT—DO NOT CITE OR QUOTE
-------
1 with respect to background responses). The underlying assumption in this approach is that one
2 can have relative confidence in an exposure-response model in the observable range, but there is
3 less confidence in any empirical exposure-response model for much lower exposures. The
4 estimates also differ because Valdez-Flores et al. (2010) truncated their life-table analysis at
5 70 years, while EPA uses a cutoff of 85 years.
6 A further reason for differences between the risk estimates of Valdez-Flores et al. (2010)
7 and EPA's 2006 draft result is that Valdez-Flores et al. (2010) estimated mortality risks, while
8 EPA estimates incidence risks. In a separate publication, Sielken and Valdez-Flores (2009)
9 disagree with the assumption of similar exposure-response relationships for
10 lymphohematopoietic cancer incidence and mortality used by EPA in deriving incidence
11 estimates and assert that the methods used by EPA in calculating these estimates were
12 inappropriate. Sielken and Valdez-Flores (2009) suggest that, except at high exposure levels, the
13 exposure-response data on all lymphohematopoietic cancers in males in the NIOSH cohort are
14 consistent with decreases in survival time as an explanation for the apparent increases in
15 mortality. For two of the four exposure groups, however, the best-fitting survival times were
16 0 years, which seems improbable. Moreover, Sielken and Valdez-Flores (2009) have not
17 established that the excess mortality is due to decreased survival time; the data are also
18 consistent with increased mortality resulting from increased incidence. Furthermore, the rodent
19 bioassays show that EtO is a complete carcinogen (see Section 3.2), and the mechanistic data
20 demonstrate that EtO is mutagenic (see Section 3.3.3), with sufficient evidence for a mutagenic
21 mode of action (see Section 3.4). Thus, EtO can be expected to act as an initiator in
22 carcinogenesis, and, consequently, be capable of inducing exposure-related increases in
23 incidence. As for the methods used by EPA in calculating the incidence estimates, EPA used
24 adjustments to the life-table analysis where warranted (U.S. EPA, 2006a). EPA did not adjust
25 the all-cause mortality rates in the lymphohematopoietic cancer analyses, because "the
26 lymphohematopoietic cancer incidence rates are small when compared with the all-cause
27 mortality rates" (U.S. EPA, 2006a; Section 4.1.1.3) and, thus, the impact of taking into account
28 lymphohematopoietic cancer incidence when calculating interval "survival" is negligible, as
29 confirmed by Sielken and Valdez-Flores' own calculations, presented in their Table 2 where the
30 "multiplier" = 1 (Sielken and Valdez-Flores, 2009). On the other hand, for the breast cancer
31 incidence analyses, where incidence rates are higher, EPA adjusted the all-cause mortality rates
32 to take into account breast cancer incidence, effectively redefining interval "survival" (and thus
33 the resulting population at risk) as surviving the interval without developing an incident case of
34 breast cancer (U.S. EPA, 2006a; Section 4.1.2.3). Therefore, the concerns raised by Sielken and
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-34 DRAFT—DO NOT CITE OR QUOTE
-------
1 Valdez-Flores (2009) about using life-table analyses to derive incidence estimates do not apply
2 to EPA's calculations.
3 Finally, the risk estimates of Valdez-Flores et al. (2010) and EPA's 2006 draft also differ
4 because Valdez-Flores et al. (2010), based on analyses in a separate publication by Sielken and
5 Valdez Flores (2009), misinterpreted the application of the age-dependent adjustment factors
6 (ADAFs) such that, even though they purported to apply the factors, this application had no
7 impact on the risk estimate. The ADAFs are default adjustment factors intended to be applied
8 directly to the unit risk estimates (i.e., risk per unit constant exposure, or "slope factors") in
9 conjunction with age-specific exposure level estimates (U.S. EPA, 2005b). For the purposes of
10 applying the ADAFs, the unit risk estimate is parsed, as a proportion of an assumed 70-year
11 lifespan, across age groups with different adjustment factors and/or exposure levels. The
12 ADAFs were not designed to be applied in life-table analyses, as was done by Sielken and
13 Valdez Flores (2009). In addition, the use of the 15-year lag in exposure in the life-table
14 analyses does not mean that there is no risk from exposures before age 15 years, as intimated by
15 Sielken and Valdez Flores (2009). Indeed, those exposures do not increase risk for cancer
16 occurring before 15 years of age; however, they do contribute to lifetime risk. The assumption
17 of increased early-life susceptibility that underlies the application of the ADAFs is that early-life
18 exposure increases the lifetime risk of cancer, not just the risk of cancer in early life, so it is
19 inappropriate to apply the ADAFs only to the age-specific hazard rates, as was done by Sielken
20 and Valdez Flores (2009). One might conceivably incorporate the ADAFs into the lifetable
21 analysis by weighting the age-specific exposures before they are aggregated into the cumulative
22 exposure, but such an integrated approach does not allow for the risks associated with less-than-
23 lifetime exposure scenarios to be calculated without redoing the lifetable analysis each time.
24
25 A.3. SUMMARY
26 The initial human studies by Hogstedt and colleagues [Hogstedt (1988); Hogstedt et al.
27 (1986); Hogstedt et al. (1979b); Hogstedt et al. (1979a)], in which positive findings of leukemia
28 and blood-related cancers suggested a causal effect, have been followed by studies that either do
29 not indicate any increased risks of cancer or else suggest a dose-related increased risk of cancer
30 at certain sites. These are chiefly cancers of the lymphohematopoietic system and include
31 leukemia, lymphosarcoma, reticulosarcoma, and NHL. More recently, an association with breast
32 cancer has also been suggested. However, the overall epidemiological evidence is not
33 conclusive because of inadequacies and limitations in the epidemiological database. The main
34 effects and limitations in the epidemiological studies of EtO are presented in Table A-5.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-3 5 DRAFT—DO NOT CITE OR QUOTE
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer
Population/
Industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
Limitations
§•
§
I
Sterilizers,
production
workers, Sweden
Hogstedt (1988);
Hogstedt et al.
(1986)
709
(539 men,
170 women)
Plant 1: mean = 20 ppm in
sterilizer room
Plant 2: mean =14 ppm in
early years, less than 6 ppm
later
Plant 3: less than 8 ppm in
early years, less than 2 ppm
later
33 cancer deaths vs. 20
expected
7 leukemia deaths vs. 0.8
expected (ICD-8 204-207)
9 lymphohematopoietic
cancer deaths vs. 2.0
expected (ICD-8 200-208)
10 stomach cancer deaths
vs. 1.8 expected
Benzene, methyl formate,
bis-(2-chloroethyl) ether, ethylene,
ethylene chlorohydrin, ethylene
dichloride, ethylene glycol,
propylene oxide, amines, butylene
oxide, formaldehyde, propylene,
sodium
No personal exposure
information from which to
estimate dose
No latency analysis
Mixed exposure to other
chemicals
o
a
I
a,
8"
O 5
1
3
o
H
W
Sterilizing workers
in 8 hospitals and
users in 4
companies, Great
Britain
Gardner et al.
(1989)
2,876
(1,864 men,
1,012
women)
In early years, odor
threshold of 700 ppm
noted; in later years, 5 ppm
or less was noted
3 leukemia deaths vs. 2.1
expected (ICD NS)
3 leukemia deaths vs. 0.35
expected (after 20+ years
latency)
4 NHL deaths vs. 1.6
expected
5 esophageal cancer deaths
vs. 2.2 expected
4 bladder cancer deaths vs.
2.04 expected
29 lung cancer deaths vs.
24.6 expected
Aliphatic and aromatic alcohols,
amines, anionic surfactants,
asbestos, butadiene, benzene,
cadmium oxide, dimethylmine,
ethylene, ethylene chlorohydrin,
ethylene glycol, formaldehyde,
heavy fuel oils, methanol,
methylene chloride, propylene,
propylene oxide, styrene, tars, white
spirit, carbon tetrachloride
Insufficient follow-up
Exposure classification
scheme vague, making it
difficult to develop dose-
response gradient
No exposure
measurements prior to
1977, so individual
exposure estimates were
not made
Mixed exposure to several
other chemicals
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Coggon et al.
(2004)
Update of Gardner
etal. (1989)
Number of
subjects
Same cohort
followed
additional
13 years
Extent of exposure to
ethylene oxide
Ibid.
Health outcomes
5 leukemia deaths vs. 4.6
expected (ICD-9 204-208)
5 leukemia deaths vs. 2.6
expected (definite or
continual exposure)
7 NHL vs. 4.8 expected
(ICD-9 200+202)
17 lymphohematopoietic
cancers vs. 12. 9 expected
(ICD-9 200-208)
1 1 breast cancers vs. 13.1
expected
Other chemicals to which subjects
were potentially exposed
Ibid.
Limitations
Ibid, and, in addition, no
latency evaluation
§•
§
I
o
a
I
a,
8"
O 5
3
o
H
W
1
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Production workers
(methods
unspecified) from 8
chemical plants in
West Germany
Kiesselbach et al.
(1990)
Number of
subjects
2,658 men
Extent of exposure to
ethylene oxide
No exposure information
available
Health outcomes
2 leukemia deaths vs. 2.35
expected (ICD-9 204-208)
5 lymphohematopoietic
cancers vs. 5 expected
(ICD-9 200-208)
14 stomach cancer deaths
vs. 10.1 expected
3 esophageal cancer deaths
vs. 1.5 expected
23 lung cancer deaths vs.
19.9 expected
Other chemicals to which subjects
were potentially exposed
Beta-naphthylamine, 4-amino-
diphenyl, benzene, ethylene
chlorohydrin, possibly alkylene
oxide (ethylene oxide/propylene
oxide), based on inclusion of plants
that were part of a cohort study by
Thiessetal. (1981).
Limitations
Insufficient follow-up;
few expected deaths in
cancer sites of
significance with which to
analyze mortality
Production methods not
stated; information vague
on what these plants do
Latency analysis given
only for total cancer and
stomach cancer mortality
Although categories of
exposure are given, they
are nonquantitative and
are not based on actual
measurements
No actual measurement
data are given; dose-
response analysis is not
possible
§•
§
I
o
a
I
a,
8"
O 5
1
3
o
H
W
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
Limitations
§•
§
I
o
a
I
a,
8"
Production workers
and users at 2
chemical plants in
West Virginia
Greenberg et al.
(1990)
2,174 men
Exposure prior to 1976 not
known
1976 survey: average 8-hr
TWA exposure levels less
than 1 ppm; 1-5 ppm 8-hr
TWA for maintenance
workers
7 leukemia and aleukemia
deaths vs. 3 expected;
SMR= 2.3 (ICONS)
2 NHL vs. 2.4 expected
9 lymphohematopoietic
cancers vs. 7.5 expected
3 liver cancer deaths vs. 1.8
expected; SMR= 1.7
7 pancreatic cancer deaths
vs. 4.1 expected; SMR= 1.7
Suggestion of increasing
risk of stomach cancer and
leukemia/aleukemia with
cumulative duration of
potential exposure
Acetaldehyde, acetonitrile, acrolein,
aldehydes, aliphatic and aromatic
alcohols, alkanolamines, allyl
chloride, amines, butadiene,
benzene, bis-(chloroethyl) ether,
ethylene dichloride, diethyl
sulphate, dioxane, epichlorhydrin,
ethylene, ethylene chlorohydrin,
formaldehyde, glycol ethers,
methylene chloride, propylene
chlorohydrin, styrene, toluidine
Low exposure levels:
average 8-hr TWA
exposure levels to EtO
less than 1 ppm (from a
1976 survey)
No actual measurements
of exposure to EtO for
these plants exist prior to
1976
Exposure occurred to
many other chemicals,
some of which may be
carcinogenic
Lack of quantitative
estimates of individual
exposure levels
O 5
3
o
H
W
1
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Same cohort as
Greenberg et al.
(1990) minus all
chlorohydrin-
exposed
employees,
followed an
additional 10 years
Tetaetal. (1993)
Only the
chlorohydrin-
exposed employees
from Greenberg et
al. (1990) cohort,
followed an
additional 10 years
Benson and Teta
(1993)
Number of
subjects
1,896 men
278 men
Extent of exposure to
ethylene oxide
Estimated exposure prior to
1956: 14+ ppm; after
1956: less than 10 ppm
Prior to 1976, estimates
were based on
measurements taken at
similar facilities
Reported to be low
exposure to EtO in the
chlorohydrin process
Health outcomes
5 leukemia and aleukemia
deaths vs. 4.7 expected
(ICD NS)
2 lympho sarcoma and
reticulosarcoma vs. 2.03
expected
7 lympho hematopoietic
cancers vs. 11. 8 expected
Trend of increasing risk of
leukemia and aleukemia
death with increasing
duration of exposure
8 lymphohematopoietic
cancer deaths vs. 2.72
expected (p < 0.05) (ICD
NS); SMR = 2.9
4 leukemia and aleukemia
deaths vs. 1.14 expected
1 lymphosarcoma and
reticulosarcoma vs. 0.50
expected
8 pancreatic cancer deaths
vs. 1.63 expected (p < 0.05)
Other chemicals to which subjects
were potentially exposed
Same (except for chemicals specific
to the chlorohydrin process)
Same
Limitations
Same
Same, and, in addition,
very small cohort
§•
§
I
J^.
o
o
I
o
a
I
a,
8"
O 5
31
o *
H 1
W K-
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Same cohort as for
Tetaetal. (1993)
followed an
additional 15 years
plus cohort
enumeration
extended to end of
1988 (an additional
10 years), adding
167 workers
Swaen et al. (2009)
Number of
subjects
2,063 men
Extent of exposure to
ethylene oxide
Individual exposure
estimates derived from an
exposure matrix based on
potential EtO exposure
categorizations developed
by Greenberg et al. (1990)
and time-period exposure
estimates developed by
Tetaetal. (1993), which
relied on measurements
taken at other facilities and
rough estimates for the
time periods before 1974.
Health outcomes
1 1 leukemia deaths vs. 11.8
expected (ICD NS)
9 leukemia deaths in
workers hired before 1956;
SMR= 1.51
12 NHL vs. 11.5 expected
27 lymphohematopoietic
cancers vs. 30.4 expected
No statistically significant
increases were observed for
any cancer types
No statistically significant
trends were observed for
lymphoid or leukemia
cancer categories examined
using Cox proportional
hazards modeling
Other chemicals to which subjects
were potentially exposed
Same
Limitations
Same
Crude exposure
assessment, especially for
the early time periods
Small cohort; thus, small
numbers of specific
cancers even though long
follow-up time
§•
§
I
o
o
a
I
a,
§•
o
O 5
31
o *
H 1
W K-
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Sterilizers of
medical equipment
and spices; and
manufacturers and
testers of medical
sterilization
equipment, in 14
plants in the United
States
Steenland et al.
(1991);Stayneret
al. (1993)
Number of
subjects
18,254
(45% male,
55% female)
Extent of exposure to
ethylene oxide
1938-1976 (estimated): 16
ppm for sterilizer
operators, 5 ppm for
remainder
1977-1985 (mean): 4.3 for
sterilizers, 2 ppm for
remainder
Individual cumulative
exposure estimates
calculated for workers in
13 of the 14 facilities
Health outcomes
36 lymphohematopoietic
cancer deaths vs. 33.8
expected (ICD NS)
13 leukemia and aleukemia
deaths vs. 13.5 expected
8 lympho sarcoma and
reticulosarcoma deaths vs.
5.3 expected
After 20+ years latency,
SMR= 1.76 for
lymphohematopoietic
cancer; significant trend
with increasing latency
(/?<0.03)
Significantly increasing
lymphohematopoietic
cancer and "lymphoid"
cancer (ICD-9 200, 202,
204) risks with cumulative
exposure (Cox regression
model)
Other chemicals to which subjects
were potentially exposed
No identified exposures to other
chemicals
Limitations
Potential bias due to lack
of follow-up on
"untraceable" members
(4.5%) of the cohort
Short duration of
exposure and low median
exposure levels
Individual exposures were
estimated prior to 1976
before first industrial
hygiene survey was
completed
Short follow-up for most
members of the cohort;
only 8% had attained
20 years latency
Little mortality (6.4%)
had occurred in this large
group of employees
No exposure-response
relationship among female
workers
§•
§
I
o
a
I
a,
8"
O 5
1
3
o
H
W
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Same cohort as
Steenland et al.
(1991) and Stayner
etal. (1993) plus
474 additional
members, followed
1 more yr
Wong and Trent
(1993)
Number of
subjects
18,728
(45% male,
55% female)
Extent of exposure to
ethylene oxide
Same as Steenland et al.
(1991) and Stayner etal.
(1993)
Health outcomes
43 lymphohematopoietic
cancer deaths observed vs.
42 expected (ICD-8 200-
209)
18 NHL deaths vs. 12.7
expected (ICD-8 200+202)
14 leukemia and aleukemia
deaths vs. 16.2 expected
(ICD-8 204-207)
Other chemicals to which subjects
were potentially exposed
No identifiable exposures to other
chemicals
Limitations
All of the limitations of
Steenland etal. (1991)
apply here
Although this group is the
same as Steenland et al.
(1991), an additional
unexplained 474
employees were added
It is questionable that one
additional yr of follow-up
added 392.2 expected
deaths but only 176
observed deaths
No effort was made to
develop exposure-
response data such as in
Stayner etal. (1993) on
the basis of individual
cumulative exposure data
but only on duration of
employment
§•
§
I
S
•3
o
o
a
I
a,
8"
O 5
1
3
o
H
W
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Steenland et al.
(2004)
Update of
Steenland et al.
(1991) and Stayner
etal. (1993)
Number of
subjects
18,254
(45% male,
55% female)
Extent of exposure to
ethylene oxide
Same as Steenland et al.
(1991), with extension of
worker histories based on
job held at end of initial
exposure assessment for
those still employed at end
of 1991 study (25% of
cohort)
Health outcomes
79 lymphohematopoietic
cancer deaths (ICD-9 200-
208): SMR=1.00
31 NHL deaths (ICD-9
200+202): SMR=1.00
29 leukemia deaths (ICD-9
204-208); SMR = 0.99
In males, in internal Cox
regression analyses,
OR=3.42(/?<0.05)in
highest cumulative exposure
group, with 15 -yr lag, for
lymphohematopoietic
cancer; significant
regression coefficient for
continuous log cumulative
exposure (p = 0.02)
Similar results for
"lymphoid" cancers (ICD-9
200, 202, 203, 204) in males
For females, in internal Cox
regression analyses,
OR =3. 13 (p< 0.05) for
breast cancer mortality in
highest cumulative exposure
group, with 20-yr lag;
significant regression
coefficient for continuous
log cumulative exposure
(p = 0.01)
Other chemicals to which subjects
were potentially exposed
No identified exposures to other
chemicals
Limitations
Potential bias due to lack
of follow-up on
"untraceable" members
(4.5% of the cohort)
Individual exposures were
estimated prior to 1976
before first industrial
hygiene survey was
completed
No increase in
lymphohematopoietic
cancer risk with increase
in exposure in women
§•
§
I
o
a
I
a,
8"
O 5
O I
2 x
31
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Women employees
from Steenland et
al. (2004)
employed in
commercial
sterilization
facilities for at least
lyr
Steenland et al.
(2003)
Number of
subjects
7,576
women
Extent of exposure to
ethylene oxide
Same as in Steenland et al.
(2004)
Minimum of 1 yr
Health outcomes
SIR = 0.87
3 19 cases of breast cancer
SIR = 0.94
20 in situ cases excluded
A positive trend in SIRs
with 15-yr lag time for
cumulative exposure
(p = 0.002)
In internal nested case-
control analysis, a positive
exposure-response with log
of cumulative exposure with
15-yr lag; top quintile had
OR = 1.74, p< 0.05
Similar results in subcohort
of 5,139 women with
interviews (233 cases)
Other chemicals to which subjects
were potentially exposed
Same as in Steenland et al. (2004),
Stayneretal. (1993)
Limitations
Interviews were available
for only 68% of the
women; thus, there is
underascertainment of
cancer cases in full
cohort. Also, there are
potential nonresponse
biases in the subcohort
with interviews.
Exposure-response trends
not strictly monotonically
increasing
§•
§
I
o
a
I
a,
8"
O 5
3
o
H
W
1
o
c
o
H
W
-------
This document is a draft for r
/2013
> *
1 able A-5. Lpideimological studies ol ethylene oxide and human cancer (continued)
Population/
Industry
Chemical workers
licensed to handle
EtO and other toxic
chemicals, Italy
Bisantietal. (1993)
Number of
subjects
1,971 men
Extent of exposure to
ethylene oxide
Levels were said to be high
at beginning of
employment; no actual
measurements were
available
637 workers were licensed
only to handle EtO and no
other toxic chemicals
Health outcomes
43 total cancer deaths vs. 33
expected
6 lymphohematopoietic
cancer deaths vs. 2.4
expected (ICD-9 200-208)
4 lympho sarcoma and
reticulosarcoma deaths vs.
0.6 expected (ICD-9 200)
2 leukemia deaths vs. 1.0
expected (ICD-9 204-208)
5 lymphohematopoietic
cancer deaths vs. 0.7
expected in group licensed
to handle only EtO
Other chemicals to which subjects
were potentially exposed
Toxic gases, dimethyl sulphate,
methylene chloride, carbon
disulphide, phosgene, chlorine,
alkalic cyanides, sulfur dioxide,
anhydrous ammonia, hydrocyanic
acid
Limitations
Lack of exposure data
Insufficient follow-up for
this young cohort
Potential selection bias
Possible earlier exposure
than date of licensing
would indicate
O 5
3
o
H
W
1
o
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
Limitations
§•
rj
S
o
a
I
a.
1
Two plants that
produced
disposable medical
equipment, Sweden
Hagmar et al.
(1995);Hagmaret
al. (1991)
2,170
(861 men,
1,309
women)
1964-1966, 75 ppm in
sterilizers, 50 ppm in
packers
1970-1972, 40 ppm in
sterilizers, 20-35 ppm in
packers and engineers
By 1985, levels had
dropped to 0.2 ppm in all
categories except sterilizers
and to 0.75 ppm in
sterilizers
6 lymphohematopoietic
cancer cases vs. 3.37
expected (ICD-7 200-209)
2 NHL cases vs. 1.25
expected (ICD-7 200+202)
2 leukemia cases vs. 0.82
expected (ICD-7 204-205)
Among subjects with at
least 0.14 ppm-years of
cumulative exposure and
10 years latency, the SIR for
leukemia was 7.14, based
on two cases
5 breast cancer cases vs.
10.8 expected
Fluorochlorocarbons, methyl
formate (1:1 mixture withEtO)
Short follow-up period;
authors recommend
another 10 years of
follow-up
Youthful cohort—few
cases and fewer deaths;
unable to determine
significance or
relationships in categories
Only a minority of
subjects had high
exposure to EtO
Sterilizers of
medical equipment
and supplies that
were assembled at
this plant, New
York
Norman et al.
(1995)
1,132
(204 men,
928 women)
In 1980, levels were
50-200 ppm (8-hr TWA);
corrective action reduced
levels to less than 20 ppm
Only 28 cancers were
diagnosed
1 leukemia case vs. 0.54
expected
12 breast cancer cases vs.
4.6 to 7.0 expected
2 pancreatic cancer cases
vs. 0.51 expected
No other chemical exposures cited
Little power to detect any
significant risk chiefly
because a short follow-up
period produced few
cancer cases
Lack of exposure data
Insufficient latency
analysis
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
Limitations
§•
rj
S
Nested case-control
study; cases and
controls from a
large chemical
production plant,
Belgium
Swaenetal. (1996)
10 cases of
Hodgkin
lymphoma
(7 cases
confirmed)
and 200
controls; all
male
Cumulative exposure to
EtO in cases was 500.2
ppm-months vs. 60.2 ppm-
months in controls
3 cases indicated exposure
to EtO, producing an
OR = 8.5 (p < 0.05)
Fertilizers, materials for synthetic
fiber production, PVC, polystyrene,
benzene, methane, acetone,
ammonia, ammonium, sulfate,
aniline, caprolactam, ethylene,
Nah., oleum
This was a hypothesis-
generating study; the
authors were not looking
for EtO exposure alone
but for other chemical
exposures as well to
explain the excess risk
Only one disease—
Hodgkin lymphoma—was
analyzed
>
^
oo
o
a
I
a.
1
Four EtO
production plants
in 3 states utilizing
the chlorohydrin
process (both
ethylene and
propylene)
Olsenetal. (1997)
1,361 men
No actual measurements
were taken
10 lymphohematopoietic
cancer deaths vs. 7.7
expected (ICD-8 200-209)
After 25-yr latency,
SMR= 1.44, based on 6
deaths
2 leukemia and aleukemia
deaths vs. 3.0 expected
(ICD-8 204-207)
No increase in pancreatic
cancer (1 observed vs. 4.0
expected)
Bis-chloroethyl ether, propylene
oxide, ethylene chlorohydrin,
propylene chlorohydrin, ethylene
dichloride, chlorohydrin chemicals
No actual airborne
measurements of EtO or
other chemicals such as
ethylene dichloride were
reported; only length of
employment was used as a
surrogate
An additional 5 to 10
years of follow-up is
needed to confirm the
presence or lack of risk of
pancreatic cancer and
lymphopoietic and
hematopoietic cancers
O
V
O
c
o
H
W
-------
to
OJ
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
Population/
Industry
Female workers
from pediatric
clinic of hospital in
Eger, Hungary
Kardos et al.
(2003)
Number of
subjects
299 female
employees
Extent of exposure to
ethylene oxide
EtO sterilizing units with
unknown elevated
concentrations
Health outcomes
1 1 cancer deaths observed
compared with 4.38, 4.03,
or 4.28 expected (p< 0.01),
based on comparison
populations of Hungary,
Heves County, and city of
Eger, respectively
1 lymphoid leukemia death
3 breast cancer deaths
Other chemicals to which subjects
were potentially exposed
No identifiable exposures to other
chemicals
Limitations
Underlying cause of death
provided on all 1 1 cases
but no expected deaths
available by cause
Possible exposure to
natural radium, which is
common in the region
§•
§
I
'TS
t^
^ 1
O
a
ICONS: ICD codes not specified.
I
a,
§•
fe I
f 1
O I
o §.
OCro
^
H »
O ^
HH ^3
H £.
W ,S'
O
c
o
H
W
-------
1 Exposure information, where available, indicates that levels of EtO probably were not
2 high in these study cohorts. If a causal relationship exists between exposure to EtO and cancer,
3 the reported EtO levels may have been too low to produce a significant finding. Exposures in the
4 earlier years (prior to 1970) in most of the companies, hospitals, and other facilities where EtO
5 was made or used are believed to have been in the range of 20 ppm, with excursions many times
6 higher, although few actual measurements are available during this period. (One exception is the
7 environmental study by Joyner (1964), who sampled airborne levels of EtO from 1960 to 1962 in
8 a Texas City facility owned by Union Carbide.)
9 Almost all actual measurements of EtO were taken in the 1970s and 1980s at most plants
10 and facilities in the United States and Europe, and levels have generally fallen to 5 ppm and
11 below. Some plants may have never sustained high levels of airborne EtO. Assuming that there
12 is a true risk of cancer associated with exposure to EtO, then the risk is not evident at the levels
13 that existed in these plants except under certain conditions, possibly due to a lack of sensitivity in
14 the available studies to detect associated cancers at low exposures.
15 The best evidence of an exposure-response relationship for lymphohematopoietic cancers
16 comes from the large, diverse NIOSH study of sterilizer workers [Steenland et al. (2004);
17 Steenland et al. (1991); Stayner et al. (1993)]. This study estimated cumulative exposure (i.e.,
18 total lifetime occupational exposure to EtO) in every member of the cohort. The investigators
19 estimated exposures from the best available data on airborne levels of EtO throughout the history
20 of the plants and used a regression model to estimate exposures for jobs/time periods where no
21 measurements were available. This regression model predicted 85% of the variation in average
22 EtO exposure levels. An added advantage to this study, besides its diversity, size, and
23 comprehensive exposure assessment, is the absence of other known confounding exposures in
24 the plants, especially benzene.
25 In the recent follow-up of the NIOSH cohort, as in the earlier study, Steenland et al.
26 (2004) observed no overall excess of hematopoietic cancers (ICD-9 codes 200-208). In internal
27 analyses, however, they found a significant positive trend (p = 0.02) for hematopoietic cancers
28 for males only, using log cumulative exposure and a 15-year lag, based on 37 male cases. In the
29 Cox regression analysis using categorical cumulative exposure and a 15-year lag, a positive trend
30 was observed and the OR in the highest exposure quartile was statistically significant
31 (OR = 3.42; 95% CI 1.09-10.73). Similar results were obtained for the "lymphoid" category
32 (lymphocytic leukemia, NHL, and myeloma). No evidence of a relationship between EtO
33 exposure and hematopoietic cancers in females in this cohort was observed. In later analyses
34 conducted by Dr. Steenland and presented in Appendix D, the difference between the male and
35 female results was found not to be statistically significant, and the same pattern of
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-50 DRAFT—DO NOT CITE OR QUOTE
-------
1 lymphohematopoietic cancer results observed for males by Steenland et al. (2004) was observed
2 for the males and females combined (i.e., statistically significant positive trends for both
3 hematopoietic [n = 74] and lymphoid [n = 53] cancers using log cumulative exposure and a
4 15-year lag, as well as statistically significant ORs in the highest exposure quartile for both
5 hematopoietic and lymphoid cancers).
6 In the analysis by Swaen et al. (2009) of male UCC workers, the authors discussed the
7 development of the exposure assessment matrix used in combination with worker histories to
8 estimate cumulative exposures for each worker in West Virginia UCC cohort. The exposure
9 matrix was based on the qualitative categorization of potential EtO exposure in the different
10 departments developed by Greenberg et al. (1990) and the time-period exposure estimates from
11 Teta et al. (1993). Eight-hour TWA concentrations (ppm) were estimated over four time periods
12 (1925-1939, 1940-1956, 1957-1973, and 1974-1978) at the two facilities for three
13 exposure-potential categories (high, medium, and low exposure departments). Average
14 exposures in the latter time period (1974-1978) were based on industrial hygiene monitoring
15 conducted at the locations where the study subjects worked. Estimates for the earlier time
16 periods were inferred from data on airborne exposure levels in "similar" manufacturing
17 operations during the time periods of interest. The estimates for the 1957-1973 time period were
18 inferred from measurements reported for the EtO production facility at Texas City studied by
19 Joyner (1964), and the estimates for the 1940-1956 time period were inferred from "rough"
20 estimates of exposure reported for the Swedish company described by Hogstedt et al. (1979a).
21 Exposures for the 1925-1939 time period were assumed to be greater than for the later time
22 periods, but the exposure estimates for this period are largely guesses.
23 This relatively crude exposure assessment formed the basis of the UCC
24 exposure-response analyses of the UCC study described in Swaen et al. (2009). Swaen et al.
25 (2009) conducted SMR analyses for the UCC workers stratified into those hired before and after
26 December 31, 1956; for three subgroups of employment duration; and for three subgroups of
27 cumulative exposure. These investigators also conducted Cox proportional hazards modeling for
28 leukemia mortality and lymphoid malignancy mortality. No statistically significant excesses in
29 cancer risk or positive trends were reported. Despite the long follow-up of the UCC cohort, its
30 usefulness is limited by its small size (e.g., a total of 27 lymphohematopoietic cancer deaths were
31 observed).
32 Valdez-Flores et al. (2010) used the same exposure assessment to conduct further
33 exposure-response modeling of the UCC data. These authors used the Cox proportional hazards
34 model to model various cancer endpoints, using the UCC data, the NIOSH data (Steenland et al.,
35 2004), or the combined data from both cohorts. Using cumulative exposure as a continuous
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-51 DRAFT—DO NOT CITE OR QUOTE
-------
1 variable, no statistically significant positive trends were observed from any of the analyses.
2 Unlike Steenland et al. (2004), Valdez-Flores et al. (2010) rejected the log cumulative exposure
3 model. Using cumulative exposure as a categorical variable, statistically significant increased
4 risks in the highest exposure quintile were reported for all lymphohematopoietic cancers and for
5 lymphoid cancers in the NIOSH male workers, consistent with results reported by Steenland et
6 al. (2004). Statistically significant increased risks in the highest exposure quintile were also
7 reported for NHL in the NIOSH male workers and for lymphoid cancers and NHL in both sexes
8 combined in the NIOSH cohort.
9 The many different analyses of the UCC data are weakened by the reliance on the crude
10 exposure assessment. The NIOSH investigators, on the other hand, based their exposure
11 estimates on a comprehensive, validated regression model. Furthermore, the NIOSH cohort was
12 a much larger, more diversified group of workers who were exposed to fewer potential
13 confounders.
14 One other study that provides cumulative exposure estimates is the incidence study by
15 Hagmar and colleagues [Hagmar et al. (1995); Hagmar et al. (1991)]. The short follow-up
16 period and relative youthfulness of the cohort produced little morbidity by the end of the study,
17 although some support for an excess risk of leukemia and lymphohematopoietic cancer had
18 appeared.
19 In a separate analysis of the NIOSH cohort by Wong and Trent (1993), duration of
20 exposure to EtO was used as a surrogate for exposure. These authors did not find any positive
21 exposure-response relationships. They did observe an elevated significant risk of "NHL" in
22 males (SMR = 2.47, p < 0.05), based on 16 deaths, which was not dose related or time related.
23 However, a deficit in females remained.
24 Increases in the risk of hematopoietic cancers are also suggested in several other studies
25 (Coggon et al., 2004; Olsen et al., 1997; Swaen et al., 1996; Norman et al., 1995; Bisanti et al.,
26 1993; Gardner et al., 1989). However, in all these studies the deaths were few and the risk ratios
27 were mostly nonsignificant except at higher estimated exposures or after long observation
28 periods. The findings were not robust and there were potentially confounding influences, such as
29 exposure to benzene and/or chlorohydrin derivatives.
30 In those plants with no detectable risks (Norman et al., 1995; Kiesselbach et al., 1990),
31 the cohorts were generally relatively youthful or had not been followed for a sufficient number
32 of years to observe any effects from exposure to EtO. In the study by Olsen et al. (1997),
33 although a slight increase in the risk of cancer of the lymphopoietic and hematopoietic system
34 was evident, the authors stated that their study provided some assurance that working in the
35 chlorohydrin process had not produced significantly increased risks for pancreatic cancer or
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-52 DRAFT—DO NOT CITE OR QUOTE
-------
1 lymphopoietic or hematopoietic cancer, thus contradicting the findings of Benson and Teta
2 (1993). This study lacks any measurement of airborne exposure to any of the chemicals
3 mentioned and the authors indicated that an additional 5 to 10 years of follow-up would be
4 needed to confirm the lack of a risk for the cancers described in their study.
5 Although the strongest evidence of a cancer risk is with cancer of the hematopoietic
6 system, there are indications that the risk of stomach cancer may have been elevated in some
7 studies (Teta et al., 1993; Kiesselbach et al., 1990; Hogstedt et al., 1986; Hogstedt et al., 1979b);
8 however, it attained significance only in the study by Hogstedt et al. (1979b), with 9 observed
9 versus 1.27 expected. It was reported by Shore et al. (1993) that this excess may have been due
10 to the fact that early workers at this plant "tasted" the chemical reaction product to assess the
11 result of the EtO synthesis. This reaction mix would have also contained ethylene dichloride, a
12 suspected carcinogen, and other chemicals. This increased risk of stomach cancer was not
13 supported by analyses of intensity or duration of exposure in the remaining studies, except that
14 Benson and Teta (1993) suggested that exposure to this chemical increased the risk of pancreatic
15 cancer and perhaps hematopoietic cancer but not stomach cancer.
16 A significant risk of pancreatic cancer first reported by Morgan et al. (1981) was also
17 reported by Greenberg et al. (1990) in his cohort of chemical workers, but only in those workers
18 assigned to the ethylene chlorohydrin production process, where the authors reported that
19 exposure to EtO was low. Benson and Teta (1993) attributed the increase in pancreatic cancer
20 seen in Greenberg et al. (1990) to exposure to ethylene dichloride in the chlorohydrin process.
21 However, Olsen et al. (1997) refuted this finding in their study. The pancreatic cancers from the
22 study by Morgan et al. (1981) also occurred in workers in a chlorohydrin process of EtO
23 production. The possibility that exposure to a byproduct chemical such as ethylene dichloride
24 may have produced the elevated risks of pancreatic cancer seen in these workers cannot be ruled
25 out.
26 In addition to the cancer risks described above, some recent evidence indicates that
27 exposure to EtO may increase the risk of breast cancer. The study by Norman et al. (1995) of
28 women who sterilized medical equipment observed a significant twofold elevated risk of breast
29 cancer, based on 12 cases. A study by Tompa et al. (1999) reported on a cluster of breast cancers
30 occurring in Hungarian hospital workers exposed to EtO. In another Hungarian study of female
31 hospital workers by Kardos et al. (2003), 3 breast cancers were noted out of 11 deaths reported
32 by the authors. Although expected breast cancer deaths were not reported, the total expected
33 deaths calculated was just slightly more than 4, making this a significant finding for cancer in
34 this small cohort.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-53 DRAFT—DO NOT CITE OR QUOTE
-------
1 The most compelling evidence on breast cancer comes from the NIOSH cohort. In the
2 recent update of this cohort, no overall excess of breast cancer mortality was observed in the
3 female workers; however, a statistically significant SMR of 2.07 was observed in the highest
4 cumulative exposure quartile, with a 20-year lag. In internal Cox regression analyses, a positive
5 exposure-response (p = 0.01) was observed for log cumulative exposure with a 20-year lag,
6 based on 103 cases. Similar evidence of an excess risk of breast cancer was reported in a breast
7 cancer incidence study of a subgroup of 7,576 female workers from the NIOSH cohort who were
8 exposed for 1 year or longer (Steenland et al., 2003). A significant (p = 0.002) linear trend in
9 SIR was observed across cumulative exposure quintiles, with a 15-year lag. In internal Cox
10 regression analyses, there was a significant regression coefficient with log cumulative exposure
11 and a 15-year lag, based on 319 cases. Using categorical cumulative exposure, the OR of 1.74
12 was statistically significant in the highest exposure quintile. In a subcohort of 5,139 women with
13 interviews, similar results were obtained based on 233 cases, and the models for this subcohort
14 were also able to take information on other potential risk factors for breast cancer into account.
15 Additionally, the coefficient for continuous cumulative exposure was also significant (p = 0.02),
16 with a 15-year lag.
17 Several other studies with female employees in the defined cohorts reported no increased
18 risks of breast cancer due to exposure to EtO (Coggon et al., 2004; Hagmar et al., 1995; Hagmar
19 et al., 1991; Hogstedt et al., 1986). However, these studies have much lower statistical power
20 than the NIOSH studies, as evidenced by the much lower numbers of breast cancer cases that
21 they report. The largest number of cases in any of these other studies is 11 cases in the Coggon
22 et al. (2004) study. Furthermore, none of these other studies conducted internal (or external)
23 exposure-response analyses, which are the analyses that provided the strongest evidence in the
24 NIOSH studies.
25
26 A.4. CONCLUSIONS
27 Experimental evidence demonstrates that exposure to EtO in rodents produces
28 lymphohematopoietic cancers; therefore, an increase in the risk of lymphohematopoietic cancer
29 in humans should not be unexpected. An increase in mammary gland carcinomas was also
30 observed in mice. Although several human studies have indicated the possibility of a
31 carcinogenic effect from exposure to EtO, especially for lymphohematopoietic cancers, the total
32 weight of the epidemiologic evidence is not sufficient to support a causative determination. The
33 causality factors of temporality, coherence, and biological plausibility are satisfied. There is also
34 evidence of consistency and specificity in the elevated risk of lymphohematopoietic cancer as a
35 single entity in the human studies. The earlier significant risk of leukemia seen in the Hogstedt
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-54 DRAFT—DO NOT CITE OR QUOTE
-------
1 studies was supported in some studies and not in others. In fact, not all human studies of EtO
2 have suggested an elevated risk of cancer and in those that do, the marginally elevated risks vary
3 from one site to another within the lymphohematopoietic system. When combined under the
4 rubric "lymphohematopoietic cancers," this loosely defined combination of blood malignancies
5 produces a slightly elevated risk of cancer in some studies but not in all. There is evidence of a
6 biological gradient in the significant dose-response relationship seen in the large, high-quality
7 Steenland et al. (2004) study.
8 The best evidence of a carcinogenic effect produced by exposure to EtO is found in the
9 NIOSH cohort of workers exposed to EtO in 14 sterilizer plants around the country (Steenland et
10 al., 2004; Stayner et al., 1993; Steenland et al., 1991). A positive trend in the risk of
11 lymphohematopoietic and "lymphoid" neoplasms with increasing log cumulative exposure to
12 EtO with a 15-year lag is evident. But there are some limitations to concluding that this is a
13 causal relationship at this time. For example, there was a lack of dose-response relationship in
14 females, although, as presented in Appendix D, later calculations show that the difference in
15 response between females and males is not statistically significant and that significant increases
16 are also observed with both sexes combined.
17 An elevated risk of lymphohematopoietic cancers from exposure to EtO is also apparent
18 in several other studies. In some of these studies, confounding exposure to other chemicals
19 produced in the chlorohydrin process concurrent with EtO may have been partially responsible
20 for the excess risks. In other studies, where the chlorohydrin process was not present, there are
21 no known confounding influences that would produce a positive risk of lymphohematopoietic
22 cancer. Overall, the evidence on lymphohematopoietic cancers in humans is considered to be
23 strong but not sufficient to support a causal association.
24 There is also evidence that exposure to EtO increases the risk of breast cancer, based
25 chiefly on the NIOSH studies [Steenland et al. (2004); Steenland et al. (2003)] discussed earlier,
26 with some corroborating support from the Norman et al. (1995) and Kardos et al. (2003) studies
27 of breast cancer in women exposed to EtO. The risk of breast cancer was analyzed in a few other
28 studies (Coggon et al., 2004; Hagmar et al., 1991; Hogstedt, 1988; Hogstedt et al., 1986), and no
29 increase in the risk of breast cancer was found. However, these studies had far fewer cases to
30 analyze, did not have individual exposure estimates, and relied on external comparisons. The
31 NIOSH studies [Steenland et al. (2004); Steenland et al. (2003)], on the other hand, used the
32 largest cohort of women potentially exposed to EtO and clearly show significantly increased
33 risks of breast cancer incidence and mortality, based on internal exposure-response analyses.
34 The authors suggest that the case is not conclusive of a causal association "due to inconsistencies
35 in exposure-response trends and possible biases due to nonresponse and an incomplete cancer
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-5 5 DRAFT—DO NOT CITE OR QUOTE
-------
1 ascertainment." While these are not decisive limitations—exposure-response relationships are
2 often not strictly monotonically increasing across finely dissected exposure categories, and the
3 consistency of results between the full cohort (less nonresponse bias) and the subcohort with
4 interviews (full case ascertainment) alleviates some of the concerns about those potential
5 biases—the evidence for a causal association between breast cancer and EtO exposure is less
6 than conclusive at this time.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 A-56 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX B.
2 REFERENCES FOR FIGURE 3-3
3 The references in this list correspond to the additional data that were added to Figure 3-3
4 since the IARC (1994b) genetic toxicity profile was published. See the Figure 3-3 legend for
5 details.
6
7 de Serres, FJ; Brockman, HE. (1995) Ethylene oxide: induction of specific-locus mutations in the ad-3 region of
8 heterokaryon 12 of Neurospora crassa and implications for genetic risk assessment of human exposure in the
9 workplace. Mutat Res 328:31-47.
10 Hengstler, JG; Fuchs, J; Gebhard, S; et al. (1994) Glycolaldehyde causes DNA-protein crosslinks: a new aspect of
11 ethylene oxide genotoxicity. Mutat Res 304(2):229-234.
12 Major, J; Jakab, MG; Tompa, A. (1996) Genotoxicological investigation of hospital nurses occupationally exposed
13 to ethylene-oxide: I. chromosome aberrations, sister-chromatid exchanges, cell cycle kinetics, and UV-induced
14 DNA synthesis in peripheral blood lymphocytes. Environ Mol Mutagen 27:84-92.
15 Major, J; Jakab, MG; Tompa, A. (1999) The frequency of induced premature centromere division in human
16 populations occupationally exposed to genotoxic chemicals. Mutat Res 445(2):241-249.
17 Nygren, J; Cedervall, B; Eriksson, S; etal. (1994) Induction of DNA strand breaks by ethylene oxide inhuman
18 diploid fibroblasts. Environ Mol Mutagen 24(3): 161-167.
19 Oesch, F; Hengstler, JG; Arand, M; et al. (1995) Detection of primary DNA damage: applicability to biomonitoring
20 of genotoxic occupational exposure and in clinical therapy. Pharmacogenetics 5 SpecNo:S118-S122.
21 Ribeiro, LR; Salvadori, DM; Rios, AC; et al. (1994) Biological monitoring of workers occupationally exposed to
22 ethylene oxide. MutatRes 313:81-87.
23 Sisk, SC; Pluta, LJ; Meyer, KG; et al. (1997) Assessment of the in vivo mutagenicity of ethylene oxide in the tissues
24 of B6C3F1 lacl transgenic mice following inhalation exposure. MutatRes 391(3): 153-164.
25 Swenberg, JA; Ham, A; Koc, H; et al. (2000) DNA adducts: effects of low exposure to ethylene oxide, vinyl
26 chloride and butadiene. DNA Repair 464:77-86.
27 Tates, AD; vanDam, FJ; Natarajan, AT; et al. (1999) Measurement of HPRT mutations in splenic lymphocytes and
28 haemoglobin adducts in erythrocytes of Lewis rats exposed to ethylene oxide. DNA Repair 431(2):397^U5.
29 van Sittert, NJ; Boogaard, PJ; Natarajan, AT; et al. (2000) Formation of DNA adducts and induction of mutagenic
30 effects in rats following 4 weeks inhalation exposure to ethylene oxide as a basis for cancer risk assessment. Mutat
31 Res - Fundam Mol Mech Mutagen 447:27-48.
32 Vogel, EW; Nivard, MJ. (1997) The response of germ cells to ethylene oxide, propylene oxide, propylene imine and
33 methyl methanesulfonate is a matter of cell stage-related DNA repair. Environ Mol Mutagen 29(2): 124-135.
34 Vogel, EW; Nivard, MJ. (1998) Genotoxic effects of inhaled ethylene oxide, propylene oxide andbutylene oxide on
35 germ cells: sensitivity of genetic endpoints in relation to dose and repair status. MutatRes 405(2):259-271.
36 Walker, VE; Sisk, SC; Upton, PB; et al. (1997) In vivo mutagenicity of ethylene oxide at the hprt locus in T-
37 lymphocytes of B6C3F1 lacl transgenic mice following inhalation exposure. Mutat Res 392(3):211-222.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 B-l DRAFT—DO NOT CITE OR QUOTE
-------
1 Walker, VE; Wu, KY; Upton, PB; et al. (2000) Biomarkers of exposure and effect as indicators of potential
2 carcinogenic risk arising from in vivo metabolism of ethylene to ethylene oxide. Carcinogenesis 21(9): 1661-1669.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 B-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX C.
2 GENOTOXICITY AND MUTAGENICITY OF ETHYLENE OXIDE
3 A summary of the available genotoxicity and mutagenicity data for ethylene oxide (EtO)
4 is presented in Chapter 3 (see Section 3.3.3). This appendix provides further details on the
5 available genotoxicity and mutagenicity data and on some of the studies that are briefly
6 mentioned in Chapter 3. The genotoxic potential of EtO is a key component of the assessment of
7 its carcinogenicity. The relationship between genotoxicity/mutagenicity and carcinogenicity is
8 based on the observations that genetic alterations are observed in almost all cancers and that
9 many of these alterations have been shown to play an important role in carcinogenesis. Exposure
10 to EtO has been found to result in a number of genotoxic effects in laboratory animal studies and
11 in studies of humans exposed in occupational settings. In particular, EtO has been shown to alter
12 or damage genetic material in such a manner that the genetic alterations are transmissible during
13 cell division. Evidence of genotoxicity/mutagenicity provides strong mechanistic support for
14 potential carcinogenicity in humans (Waters et al., 1999).
15 Since the first report of EtO's role in inducing sex-linked recessive lethals in Drosophila
16 (Rapoport, 1948), numerous papers have been published on the mutagenicity of EtO in
17 biological systems, spanning a whole range of assay systems, from bacteriophage to higher
18 plants and animals (see Figure 3-3 in Chapter 3). EtO, being a mono-functional alkylating agent,
19 is DNA-reactive, capable of forming DNA adducts and inducing mutations at both the
20 chromosome and gene levels under appropriate conditions, as evidenced in numerous in vitro
21 and in vivo studies (reviewed in IARC, 2008; Kolman et al., 2002; Thier and Bolt, 2000;
22 Natarajan et al., 1995; Vogel and Natarajan, 1995; Dellarco et al., 1990; Kolman et al., 1986). In
23 prokaryotes (bacteria) and lower eukaryotes (yeasts and fungi), EtO induces DNA damage and
24 gene mutations and conversions. In mammalian cells, EtO induces DNA adducts, unscheduled
25 DNA synthesis, gene mutations, sister chromatid exchanges (SCEs), micronuclei, and
26 chromosomal aberrations (IARC, 2008; Thier and Bolt, 2000; Natarajan et al., 1995; Preston et
27 al., 1995; Dellarco et al., 1990; Walker et al., 1990; Ehrenberg and Hussain, 1981). The results
28 of in vivo studies on the genotoxicity of EtO following ingestion, inhalation or injection have
29 also been consistently positive (IARC, 2008, 1994b). Furthermore, in vivo exposure to
30 EtO-induced gene mutations in the Hprt locus in mouse and rat splenic T-lymphocytes and SCEs
31 in lymphocytes from rabbits, rats, and monkeys, in bone marrow cells from mice and rats, and in
32 rat spleen. Increases in the frequency of gene mutation in the lung (Lad locus) (Recio et al.,
33 2004; Sisk et al., 1997) and in the Hprt locus in T-lymphocytes (Walker et al., 1997) in
34 transgenic mice exposed to EtO via inhalation have been observed at concentrations similar to
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-1 DRAFT—DO NOT CITE OR QUOTE
-------
1 those in carcinogenesis bioassays (NTP, 1987). EtO has also induced heritable mutations or
2 effects in germ cells in rodents (Generoso et al., 1990; Lewis et al., 1986). In addition,
3 significant increases in the frequency of SCEs and chromosomal aberrations in peripheral blood
4 lymphocytes have been consistently reported in workers exposed to concentrations of EtO of
5 greater than 5 ppm (TWA) (IARC, 2008, and references therein). Thus, there is consistent
6 evidence that EtO interacts with the genome from both in vitro studies and in vivo studies of
7 laboratory animals and occupationally exposed humans. Based on these observations, exposure
8 to EtO is considered to cause cancer through a mutagenic mode of action (see Chapter 3,
9 Section 3.4).
10 The following sections provide further details on different genotoxicity test results
11 regarding the mutagenic potential of EtO.
12
13 C.I. DNAADDUCTS
14 Covalent binding of a chemical (direct-acting) or its electrophilic intermediates or
15 metabolites (indirect-acting chemicals following metabolic activation) with the nucleophilic sites
16 in DNA results in the formation of "DNA adducts," which represent the biologically effective
17 dose of the chemical agent in question. Alkylating agents, such as EtO, are direct-acting
18 chemical agents which can transfer alkyl groups (e.g., ethyl groups) to nucleophilic sites in
19 DNA, alkylating the nucleotide bases. Alkylating agents are classified as Snl-type or SN2-type
20 depending on the substitution nucleophilicity (SN). The Snl-type chemicals follow first-order
21 kinetics (e.g., ethylnitrosourea [ENU] and methylnitrosourea or [MNU]), while the SN2-type
22 agents exhibit an intermediate transition state (e.g., EtO and methyl methanesulfonate [MMS]).
23 EtO is a direct-acting SN2 (substitution-nucleophilic-bimolecular)-type alkylating agent that
24 forms adducts with cellular macromolecules such as proteins (e.g., hemoglobin) and DNA. The
25 reactivity of an alkylating agent can be estimated by its Swain Scott substrate constant (s-value),
26 which ranges from 0 to 1 (Warwick, 1963). Alkylating agents such as EtO and MMS, which
27 have high "s" values (0.96 and >0.83, respectively), target the nucleophilic centers of ring
28 nitrogens (e.g., N7 of guanine and N3 of adenine) in DNA, while agents such as ENU with a low
29 "s" values (0.26) target the less nucleophilic centers such as O6 of guanine. EtO has a high
30 substrate constant favoring efficient alkylation at N7 of guanine (Beranek, 1990; Golberg, 1986;
31 Warwick, 1963). Due to the high nucleophilicity and steric availability of the N7 of guanine,
32 EtO predominantly forms the N7-hydroxyethylguanine (N7-HEG) adduct, although minor
33 adducts such as those forming at O6 of guanine, N1, N3, and N6 of adenine, and N3 of cytosine,
34 uracil and thymine are found in some instances (Segerback, 1994).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 Several methods have been developed since 1988 to detect EtO-induced DNA adducts in
2 vitro and in vivo. However, sensitivity and specificity of these methods have been the main
3 concern. These methods include immunochemical assays, fluorescence techniques, high
4 pressure liquid chromatography (HPLC), gas chromatography/mass spectrometry (GC/MS),
5 32P-postlabeling and electrochemical detection, with varying sensitivities for detection of
6 EtO-DNA adducts (Marsden et al., 2009; Huang et al., 2008; Tompkins et al., 2008; Marsden et
7 al., 2007; Bolt et al., 1997; Leclercq et al., 1997; Kumar et al., 1995; Saha et al., 1995; van Delft
8 et al., 1994; van Delft et al., 1993; Uziel et al., 1992; Bolt et al., 1988). In the following
9 paragraphs, a brief summary of available methods is provided to aid in the discussion of the
10 DNA adduct data.
11 van Delft et al. (1993) developed monoclonal antibodies against the imidazole ring of
12 N7-alkyldeoxyguanosine, with the limits of detection being 5-10, 1-2, and 20 adducts per
13 106 nucleotides, respectively, when used in the direct and competitive enzyme-linked
14 immunosorbant assay and in immunofluorescence microscopy. Later the same authors
15 developed an immunoslot-blot assay with increased sensitivity that detected 0.34 N7-HEG
16 adducts per 106 nucleotides (van Delft et al., 1994). Kumar et al. (1995) developed a
^9
17 P-postlabeling method using thin-layer chromatography (TLC) and HPLC, which detected
18 0.1-1.0 fmol 7-alkylguanine adducts in rats exposed to different alkenes. Despite occasional
19 inefficient labeling and poor recovery of adduct due to depurination, this method has potential
20 for use in measuring human exposure to alkenes or their corresponding epoxides as well as the
21 endogenously formed 7-alkylguanine adducts.
22 Bolt et al. (1997) developed a HPLC method involving derivatization with phenylglyoxal
23 and fluorescence detection, using 7-methylguanine as an internal standard, for measuring the
24 physiological background of the N7-HEG adduct in DNA isolated from human blood. Using
25 this method, the authors were able to detect N7-HEG levels in five individuals ranging between
26 2.1 and 5.8 pmol/mg DNA (mean 3.2). Furthermore, Leclercq et al. (1997) developed a method
27 based on DNA neutral thermal hydrolysis, adduct micro-concentration, and HPLC coupled to
28 single-ion monitoring electrospray mass spectrometry which has a detection limit of 1 fmol
29 (10 10 M), allowing the detection of 3 adducts/108 normal nucleotides. Using this method,
30 Leclercq et al. (1997) detected a dose-response relationship for N7-HEG after exposing calf
31 thymus DNA and blood samples to various doses of EtO. Marsden et al. (2007) used a highly
32 sensitive LC-MS/MS assay with selected reaction monitoring that offers a limit of detection of
33 0.1 fmol of N7-HEGto establish background levels of N7-HEG (1.1-3.5 adducts/108
34 nucleotides) in tissues of rats. Huang et al. (2008) developed an isotope-dilution online solid-
35 phase extraction and liquid chromatography coupled with tandem mass spectrometry method
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 with reportedly excellent accuracy, sensitivity, and specificity to analyze N7-HEG in urine
2 samples of nonsmokers. This method also demonstrated high-throughput capacity for detecting
3 EtO-DNA adducts and may be particularly useful for future molecular epidemiology studies of
4 individuals with low-dose EtO exposure. Tompkins et al. (2008) used a high-performance liquid
5 chromatography/electrospray ionization tandem mass spectrometry and reported ~8 N7-HEG
o
6 adducts/10 nucleotides in the livers of control rats. This method was also capable of detecting
7 the less prevalent but potentially more biologically significant Nl -hydroxyethyl-2'-
8 deoxyadenosine (Nl-HEA), O6-hydroxyethyl-2'-deoxyguanosine (O6-HEG), N6-hydroxy ethyl-
9 2'-deoxyadenosine (N6-HEA) and N3-hydroxyethyl-2'-deoxyuridine (N3-HEU) adducts.
10 However, these minor adducts were below the level of detection in control rat tissue DNA.
11 Overall, the sensitivity of EtO adduct detection depends on the method used for analysis.
12 Hence, use of appropriate methods is important when analyzing for these adducts and will be
13 highlighted in the following discussion.
14
15 C.I.I. Detection of EtO Adducts in In Vitro and In Vivo Systems
16 Numerous studies have been conducted to investigate the formation of DNA adducts
17 following EtO exposure, in a wide range of experimental models, including cell-free systems,
18 bacteria, fungi, Drosophila and experimental animals, as well as in exposed human subjects.
19 The following discussion is a review of the available studies of exposure to EtO and DNA adduct
20 formation in in vitro systems, laboratory animals, and humans (Boysen et al., 2009; Pauwels and
21 Veulemans, 1998; Bolt et al., 1988; Van Sittert and de Jong, 1985).
22
23 C.1.2. In Vitro DNA Binding Studies
24 The capacity of EtO to bind to DNA and form DNA adducts has been documented in a
25 few in vitro studies. Segerback (1990) showed that 14C-labeled EtO reacted in vitro with calf
26 thymus DNA to produce N7-HEG adduct as the predominant adduct, with relatively low
27 amounts of O6-HEG and N3-(2-hydroxyethyl)adenine (N3-HEA) adducts. The levels of
28 N3-HEA and O6-HEG are 4.4% and 0.5%, respectively, of the N7-HEG levels. Thus, the ratio
29 of N7-HEG, N3-HEA and O6-HEG produced in vitro was 200:8.8:1, respectively. In the same
30 study, the in vitro reaction products of radiolabeled N-(2-hydroxyethyl)-N-nitrosourea
31 (HOEtNU) with calf thymus DNA exhibited a higher relative amount of O6-HEG, which was
32 63% of the N7-HEG formed. The difference in reactivity towards the N7 and O6 positions in
33 guanine by these two alkylating agents was explained by the difference in their "s" values. EtO,
34 with an s-value of 0.9, has a greater relative preference for reacting with N rather than O atoms
35 than does HOEtNU, with an ^-values of 0.2.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 In another study, Li et al. (1992) observed that EtO in aqueous solution incubated with
2 calf thymus DNA in vitro for 10 hours produced several 2-hydroxyethyl (HE) DNA adducts
3 whose relative yields (nmol/mg DNA) were in the descending order: N7-HEG (330) > N3-HEA
4 (39) > Nl-HEA (28), N6-HEA (6.2) > N3-HE-Cyt (3.1) > N3-HE-dThd (2.0) > N3-HEU (0.8).
5 This in vitro study did not detect the O6-HEG adduct.
6 Recently, Tompkins et al. (2009) treated pSP189 shuttle vector plasmid to a range of EtO
7 concentrations in water and reported that, of the five 2-hydroxy ethyl DNA adducts measurable
8 using their LC-MS/MS analytical method, only the N7-HEG adduct was detectable at EtO
9 concentrations up to 2,000 jiM.5 At the 10 mM concentration, the level of N7-HEG adducts was
10 about 19 times higher than that of Nl-HEA adducts and about 1,000 times higher than that of
11 O6-HEG adducts. At 30 mM, N3-HEU adducts were detectable, but this adduct was not
12 quantifiable due to the lack of a suitable internal standard. Detection of the N3-HEU adduct
13 implies that the N3-HEC adduct is also formed, as the former is the hydrolytic deamination
14 product of the latter Tompkins et al. (2009). No results for the N6-HEA adduct were reported.
15 (N3-HEA, N3-HEC, and N3-HET adducts are not measurable by their method.)
16
17 C.1.3. In Vivo Studies—Animal Experiments
18 Several studies evaluated N7-HEG levels following one or a range of doses with repeated
19 exposures of EtO given by inhalation or intraperitoneal injection in laboratory animals.
20 Segerback (1983) showed that in male CBA mice exposed by inhalation to 14C-labeled EtO
21 N7-HEG adducts are formed in spleen, testes and liver with half-lives of 24, 20, and 12 hours,
22 respectively.
23 Walker et al. (1990) conducted a time-course study to investigate the formation and
24 persistence of N7-HEG adducts in various tissues such as brain, kidney, liver, spleen, lung and
25 kidney of male Fischer 344 rats exposed to one high dose of 300 ppm EtO by inhalation for
26 4 consecutive weeks (6 hours/day, 5 days/week) and sacrificed 1-10 days after the end of
27 exposure. The N7-HEG adduct was detectable in both target (brain, spleen and WBCs) and
28 nontarget (kidney, liver, lung, and testis) tissues with maximum levels (1.5 times control levels)
29 seen in brain compared to other tissues 1 day after exposure. The similarities in N7-HEG levels
30 in various tissues are possibly due to efficient pulmonary uptake of EtO and rapid distribution by
31 the circulatory system. The N7-HEG adduct levels increased linearly for 3-5 days followed by a
32 slow removal from DNA with an apparent half-life of 7 days, suggesting that the adduct was
5The minor adducts may have been present at levels below the limits of detection, which were as follows:
0.001/106 nucleotides for N7-HEG and Nl-HEA; 0.016/106 nucleotides for O6-HEG; and 0.082/106 nucleotides for
N3-HEU (Tompkins et al., 2009).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-5 DRAFT—DO NOT CITE OR QUOTE
-------
1 probably removed by spontaneous depurination. The calculated in vivo half-life for N7-HEG
2 formed by EtO confirms the persistence of this adduct and is consistent with another study in rats
3 exposed to another alkylating agent, N-nitrosomethyl-(2-hydroxyethyl)amine (Koepke et al.,
4 1988). Walker et al. (1990) suggested that the similarity in N7-HEG formation in the target as
5 well as nontarget tissues could also be due to factors such as cell replication, location of the
6 adducts in the genome, and tissue susceptibility genes, which might be critical determinants
7 quantitatively affecting tissue-specific and/or dose-response relationships.
8 Using fluorescence-coupled HPLC, Walker et al. (1992a) measured N7-HEG levels in
9 DNA of target and nontarget tissues from male B6C3Fi mice and F344 rats exposed to 0, 3, 10,
10 33, 100, or 300 (rats only) ppm EtO by inhalation for 4 weeks (6 hours/day, 5 days/week).
11 Another group of mice was exposed to 100 ppm EtO for 1, 3, 7, 14, or 28 days (5 days/week).
12 The authors reported linear dose-response relationships for N7-HEG in rat tissues following EtO
13 exposures between 10 and 100 ppm, with the slope increasing for exposures above 100 ppm. In
14 mice, only exposures to 100 ppm EtO resulted in significant increase in N7-HEG levels. Walker
15 et al. (1992a) observed N7-HEG adduct levels of 2-6 pmols/mg DNA in control mice and rats,
16 while in mice exposed to 100 ppm EtO, N7-HEG levels ranged from 17.5 ± 3.0 (testis) to
17 32.9 ±1.9 (lung) pmol/mg DNA after 4 weeks of exposure. Rats and mice concurrently exposed
18 to 100 ppm EtO for 4 weeks showed two- to threefold lower N7-HEG levels in all tissues of
19 mice compared to rats, suggesting species differences in the susceptibility to EtO-induced
20 genotoxicity. The half-life of N7-HEG in mouse kidney DNA was 6.9 days, and in rat brain and
21 lung it was 5.4-5.8 days. The half-lives of N7-HEG adducts in DNA from other tissues of
22 mouse and rat were 1.0-2.3 days and 2.9-4.8 days, respectively. The authors suggested that the
23 slow linear removal of N7-HEG adducts from the DNA was mainly due to chemical
24 depurination, while the rapid removal was due to loss by depurination and DNA repair. Rats
25 exposed to 300 ppm EtO showed O6-HEG adducts at a steady-state concentration of ~1 pmol/mg
26 DNA. Based on the results from rats and mice, the authors suggested that DNA repair was
27 saturated at the concentration of EtO used in the time-course studies and that repeated exposures
28 to lower concentrations of EtO should lead to species- and tissue-specific differences in the
29 levels of N7-HEG (Walker et al., 1992a).
30 Wu et al. (1999a) analyzed DNA from liver, brain, lung and spleen of B6C3Fi mice and
31 F344 rats for N7-HEG adducts after exposure to EtO (0, 3, 10, 33, or 100 ppm) for 4 weeks
32 (6 h/day, 5 days/week). The authors observed tissue- and species-specific dose-response
33 relationships of N7-HEG adducts in the EtO-exposed animals. Mice showed linear
34 dose-response relationships for N7-HEG adducts in liver, brain and spleen at exposures between
35 3 and 100 ppm, and sublinear responses in lung between 33 and 100 ppm EtO exposure. Rats
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 showed linear increases in adduct levels in liver and spleen DNA between 3 and 100 ppm EtO,
2 and sublinear responses in the brain and lung between 33 and 100 ppm EtO exposure. Overall,
3 rats and mice exposed to 3 ppm EtO showed 5.3- to 12.5- and 1.3- to 2.5-fold higher N7-HEG
4 adducts, respectively, compared to the corresponding unexposed control animals. Thus, results
5 from this study suggest species differences, with rats being more susceptible to adduct formation
6 than mice, at lower levels of EtO exposure. This study also showed a clear difference in
7 N7-HEG levels between unexposed and exposed mice at these lower exposure levels, unlike the
8 study of Walker et al. (1992a) discussed above, which is possibly due to the use of a highly
9 sensitive gas chromatography high-resolution mass spectrometry (GCHRS) assay in the Wu et
10 al. (1999a) study.
11 van Sittert et al. (2000) exposed Lewis rats to 50, 100 and 200 ppm EtO by inhalation
12 (4 weeks, 5 days/week, 6 h/day) and measured N7-HEG adducts 5, 21, 35 and 49 days after
13 cessation of exposure. The authors used mass spectrometry following neutral thermal hydrolysis
14 of DNA to release the N7-HEG adducts, which clearly show a difference between control and
15 EtO-exposed rats. The mean levels of liver N7-HEG immediately after cessation of exposure to
16 50, 100, and 200 ppm were estimated by extrapolation to be 310, 558, and
o
17 1,202 adducts/10 nucleotides, respectively, while the mean level in control rats was
18 2.6 adducts/108 nucleotides. By 49 days postexposure, N7-HEG adducts had returned to near
19 background levels. The N7-HEG levels in liver DNA showed a linear response between 0 and
20 200 ppm EtO, suggesting that detoxification and DNA repair processes were not saturated up to
21 the highest exposure level tested. The authors observed statistically significant linear
22 relationships between mean N7-HEG levels at "day 0" postexposure and (1) Hprt mutant
23 frequencies at expression times of 21/22 and 49/50 days postexposure, (2) SCEs at 5 days
24 postexposure, or (3) high-frequency cells measured 5 days postexposure. The authors also
25 observed that SCEs and high-frequency cells continued to be present at 21-days postexposure
26 and significantly correlated with N7-HEG adducts at that time. However, induction of
27 micronuclei, chromosome breaks or translocations did not show a dose-response relationship.
28 Nivard et al. (2003) showed that in male Drosophila flies EtO exposure (2-1,000 ppm)
29 by inhalation for 24 hours induced a linear dose-response relationship for N7-HEG adduct
30 formation (0.15 to 105.4 adducts/106 nucleotides) over the entire dose range, as detected by
31 32P-postlabeling assay. The N7-HEG adducts were undetectable in controls (i.e., below the
32 detection limit of 1 adduct/108 nucleotides).
33 A study by Rusyn et al. (2005) tested the hypothesis that EtO exposure results in an
34 accumulation of apurinic/apyrimidinic (AP) sites in DNA and induces changes in expression of
35 genes involved in DNA base excision repair (BER). The authors exposed male F344 rats by
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 inhalation to 100 ppm EtO or ethylene (40 or 3,000 ppm) for 1, 3, or 20 days (6 h/day,
2 5 days/week) and sacrificed them 2, 6, 24, or 72 hours after a single-day exposure. Brain and
3 spleen were considered as target sites for EtO-induced carcinogenesis, and liver as a nontarget
4 organ. Rusyn et al. (2005) observed a time-dependent increase in N7-HEG in brain, spleen
5 (target organs) and liver (nontarget organ) and in N-(2-hydroxyethyl)valine (HEVal) adducts in
6 hemoglobin. However, they could not detect any increase in AP sites in control or EtO-exposed
7 rats for any given duration or dose of exposure. Rats exposed to EtO for 1 day showed a
8 threefold to sevenfold decrease in expression of the DNA repair enzyme 3-methyladenine-DNA
9 glycosylase in the brain and spleen, while rats exposed to EtO for 20 days showed increased
10 expression of hepatic 8-oxoguanine DNA glycosylase, 3-methyladenine-DNA glycosylase, AP
11 endonuclease, polymerase beta, and alkylguanine methyltransferase by 20-100%. Levels of
12 brain AP endonuclease and polymerase beta were increased by <20% only in rats exposed to
13 3,000 ppm ethylene for 20 days. Results from this study suggest that EtO-induced DNA damage
14 is repaired without accumulation of AP sites or involvement of the BER pathway in target
15 organs. The authors conclude that accumulation of AP sites is not likely a primary mechanism
16 for mutagenicity and carcinogen!city of EtO, and further suggest that minor DNA adducts such
17 as O6-HEG or Nl-HEA are likely to be involved in mutagenicity. In fact, in a previous study
18 from the same group (Walker et al., 1992a), steady-state concentrations of O6-HEG were
19 reported after 4 weeks of exposure with 300 ppm EtO, a finding which warrants further
20 investigation.
21 Marsden et al. (2007) have shown that intraperitoneal administration of a single or three
22 daily doses of EtO (0.01-1.0 mg/kg) induced dose-related increases in N7-HEG adduct levels in
23 male F344 rats, except at the lowest dose (0.01 mg/kg), where N7-HEG levels were similar to
24 endogenous levels detected in control animals. Further, they observed that N7-HEG adducts did
25 not accumulate in rats given three daily doses of EtO.
26 Recently, using a dual-isotope approach combining HPLC-accelerated mass spectrometry
27 with LC-MS/MS analysis, Marsden et al. (2009) observed linear dose-response relationships for
28 (14C)N7-HEG adducts (0.002 to 4 adducts/108 nucleotides) in spleen, liver and stomach DNA of
29 F344 rats after exposure to low, occupationally relevant concentrations of (14C)EtO (0, 0.0001,
30 0.0005, 0.001, 0.005, 0.01, 0.05, and 0.1 mg/kg daily for 3 consecutive days, with the rats killed
31 4 h after the last exposure). These results suggest that by using a highly sensitive assay, it is
32 possible to measure the N7-HEG adducts resulting from low EtO exposures above the
33 b ackground adduct 1 evel s.
34 Otteneder and Lutz (1999) reviewed the quantitative relationship between DNA adduct
35 levels and tumor incidence in rodents that received repeated administration of EtO. The authors
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-8 DRAFT—DO NOT CITE OR QUOTE
-------
1 observed a correlation with tumor incidence when the DNA adduct levels measured at a given
2 dose were normalized to the TDso dose (the dose which results in 50% tumor incidence in a
3 two-year study). The calculated adduct level in mice associated with the hepatocellular TDso
4 was 812 N7-HEG adducts/108 normal nucleotides.
5
6 C.1.4. In Vivo Studies—Human Subjects
7 A few studies have examined the effect of EtO exposure on humans, particularly in
8 occupational settings, and these have been comprehensively reviewed by Kolman et al. (2002).
9 In that review, the authors examined the use of hemoglobin and DNA adducts as biomarkers of
10 EtO exposure and the roles of genetic polymorphisms and confounding factors. Kolman et al.
11 (2002) also described the genotoxic effects of EtO in mammalian cells and summarized the
12 genotoxic and carcinogenic effects of EtO in humans. Some of the relevant studies in humans
13 are briefly discussed below.
14 An immunoslot blot assay was used to analyze N7-HEG levels in white blood cell DNA
15 from individuals exposed to EtO (2-5 ppm) and from controls van Delft et al. (1994). The
16 authors reported 0.1 and 0.065 N7-HEG adducts/106 nucleotides, respectively, in EtO-exposed
17 individuals (n = 42) and controls (n = 29) by this method. However, these differences were not
18 statistically significant.
19 In a study involving 58 sterilizer operators exposed to low and high levels of EtO (<32
20 and >32 ppm-hour, respectively) and 6 nonexposed controls from different hospitals, Yong et al.
21 (2007) examined N7-HEG adducts in granulocyte DNA. During the four-month study, the
22 cumulative exposure to EtO (ppm-hour) was estimated before the blood sample collection. After
23 adjusting for cigarette smoking and other potential confounders, the mean N7-HEG adduct levels
24 in the nonexposed, low-, and high-exposure groups were 3.8, 16.3, and
25 20.3 adducts/107 nucleotides, respectively, with considerable interindividual variation (range:
26 1.6-241.3 adducts/107 nucleotides). However, these differences in mean adduct level were not
27 statistically significant. The large variability across workers may reflect differences in their
28 recent exposure patterns because granulocytes have a lifespan of less than a day. Also, the study
29 did not find a significant correlation between the levels of N7-HEG adducts and HEVal adducts.
30 Mayer et al. (1991) observed an apparent suppression of DNA repair capacity in
31 EtO-exposed individuals as measured by the DNA repair index, i.e., the ratio of unscheduled
32 DNA synthesis and N-acetoxy-2-acetylaminofluorene (NA-AAF)-DNA binding, (p < 0.01). In
33 this study, 34 sterilization unit workers of a large university hospital and 23 controls working in
34 the university library were used. Overall, this study demonstrates significant correlations
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-9 DRAFT—DO NOT CITE OR QUOTE
-------
1 between EtO-induced hemoglobin adduct levels and SCEs and the number of high frequency
2 cells, at low levels of EtO exposure (<1 ppm), independent of smoking history.
O
4 C.1.5. DNAAdducts—Summary
5 In summary, EtO predominantly forms N7-HEG adducts. Minor adducts are O6-HEG
6 adducts and reaction products with Nl, N3 and TV6 of adenine and with N3 of cytosine, uracil and
7 thymine in vitro. However, the minor adducts are not observed to the same extent in vivo, which
8 may reflect a limitation in the sensitivity of the adduct assays available to date. Repeated
9 inhalation exposure of EtO induces N7-HEG adducts in both target organs (brain, spleen, and
10 white blood cells) and nontarget organs (kidney, liver, and lung) in rodents, with an apparent
11 half-life of 3-6 days in rats and 1-3 days in mice (Walker et al., 1992a). The dose-response
12 relationship of N7-HEG and EtO exposure is influenced by the analytical method used, which
13 also affects the background (endogenous) levels of adducts observed in unexposed rodents.
14 Steady-state levels of O6-HEG adducts (1 pmol/mg DNA) are detected in rats exposed by
15 inhalation to high doses of EtO (300 ppm) which are -250-300 times lower than the N7-HEG
16 levels (Walker et al., 1992a). Although N7-HEG adducts are likely to be removed by
17 depurination forming apurinic/apyrimidinic (AP) sites, Rusyn et al. (2005) showed that DNA
18 damage induced by exposure to EtO is repaired without accumulation of AP sites and without
19 affecting base excision repair (BER) in target organs of Fischer rats. There are only two studies
20 available on EtO-induced DNA adducts in human populations. Although higher levels of
21 N7-HEG DNA adducts were observed in human white blood cells (van Delft et al., 1994) and
22 granulocytes (Yong et al., 2007) of exposed cases compared to controls, these differences were
23 not statistically significant, possibly due to high interindividual variability.
24
25 C.1.6. EtO-Hemoglobin Adducts
26 Several studies have shown that EtO-induced hemoglobin adducts (e.g., HEVal) are good
27 biomarkers of exposure for this compound in human studies and that predicted hemoglobin
28 adduct levels resulting from exposure to ethylene or EtO are in agreement with measured values
29 (Boogaard, 2002; Yong et al., 2001; Fennell et al., 2000; Tates et al., 1999; Walker et al., 1992a;
30 Britton et al., 1991). Csanady et al. (2000) found a good agreement between the predicted and
31 measured hemoglobin adduct levels in humans. However, in rodents, hemoglobin adducts were
32 under-predicted by a factor of 2 to 3, while DNA adduct levels were comparable, suggesting
33 inconsistencies between the two biomarkers. Walker et al. (1993) also observed that the
34 relationships between HEVal and N7-HEG concentrations varied with length of exposure,
35 interval since exposure, species, and tissue, which may be due to differences in formation,
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-10 DRAFT—DO NOT CITE OR QUOTE
-------
1 persistence, repair, and chemical depurination of the DNA adduct. Thus, Walker et al. (1993)
2 suggested that HEVal adducts do not provide accurate prediction of DNA adducts in specific
3 tissues of humans under actual exposure conditions. In summary, HEVal adducts do not appear
4 to be predictable markers for DNA adducts.
5
6 C.2. GENE MUTATIONS
7 EtO has consistently yielded positive results, at both the gene and chromosome levels, in
8 a broad range of in vitro and in vivo mutational assays, including those performed in bacteria,
9 fungi, yeast, insects, plants, Drosophila and rodents, in both repair-deficient and proficient
10 organisms, and in mammalian cell cultures, including cells from humans (reviewed inlARC,
11 2008; Kolman et al., 2002; Thier and Bolt, 2000; Natarajan et al., 1995; Vogel and Natarajan,
12 1995; IARC, 1994b; Dellarco et al., 1990). The results of in vivo studies on the mutagenicity of
13 EtO have also been consistently positive following ingestion, inhalation, or injection (e.g., Tates
14 et al., 1999). Increases in the frequency of gene mutations in the lung (Zac/locus) (Sisk et al.,
15 1997), in T-lymphocytes (Hprt locus) (Walker et al., 1997), and bone marrow and testes in
16 B6C3Fi Zac/transgenic mice (Recio et al., 2004) have been observed in mice exposed to EtO
17 via inhalation at concentrations similar to those used in the carcinogenesis bioassays (NTP,
18 1987), clearly documenting that EtO is a DNA-reactive mutagenic agent. Furthermore,
19 occupational studies provide evidence for the genotoxic potential of EtO.
20
21 C.2.1. Bacterial Systems
22 Studies have been conducted to investigate the ability of EtO to induce gene mutations in
23 bacterial systems. Victorin and Stahlberg (1988) treated Salmonella typhimurium strain TA100
24 with EtO at concentrations of 1-200 ppm for 6 hours and demonstrated that EtO was mutagenic
25 in this system. In another study, Agurell et al. (1991) compared EtO and propylene oxide (two
26 alkylating agents) for genotoxic effectiveness in various test systems. The abilities of the two
27 compounds to induce point mutations in S. typhimurium strains TA 100 and TA1535 were
28 approximately equal. EtO induced a dose-dependent increase in the number of revertants in both
29 tester strains. No toxic effects were observed under the conditions tested.
30 In contrast, Agurell et al. (1991) found EtO to be 5-10 times more effective than
31 propylene oxide with respect to gene conversion and reverse mutation in the Saccharomyces
32 cerevisiae D7 and S. cerevisiae RSI 12 strains. The greater effectiveness of EtO over propylene
33 oxide in inducing these types of mutations was probably due to the difference in these
34 compounds' abilities to cause strand breaks via alkylation of DNA-phosphate groups.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-l 1 DRAFT—DO NOT CITE OR QUOTE
-------
1 Mutagenicity studies of EtO have also been conducted using different Escherichia coli
2 strains. Kolman (1985) investigated the influence of the uvrB and umuC genes on the induction
3 of Zac/-mutants and nonsense mutants by EtO in the Lad gene of E. coli and found that uvrB
4 gene mutation was associated with higher mutation frequencies whereas umuC mutation did not
5 significantly affect the induction of Lad mutations. Thus, mutations induced by EtO were
6 enhanced by a lack of excision repair but not influenced by changes in error-prone repair. In
7 another study by the same group of authors (Kolman and Naslund, 1987), the mutagenicity of
8 EtO in E. coli B strains with different repair capacities was investigated. Deficiencies in
9 excision repair (uvrA, polA) led to considerable increases in mutation frequency compared to the
10 wild-type strain and strains deficient in error-prone repair (recA, lexA).
11 The induction of specific-locus mutations in the adenine-3 (ad-3) region of a
12 two-component heterokaryon (H-12) of Neurospora crassa by EtO was studied by de Serres and
13 Brockman (1995). The objective of this study was to compare EtO's mutational spectrum for
14 induced specific-locus mutations with those of other chemical mutagens. Conidial suspensions
15 were treated with five different concentrations of EtO (0.1-0.35%) for 3 hours. The results from
16 these experiments showed (1) the dose-response curve for EtO-induced specific-locus mutations
17 in the ad-3 region was linear, with an estimated slope of 1.49 ± 0.07, and (2) the maximum
18 forward-mutation frequency was between 10 and 100 ad-3 mutations per 106 survivors. The
19 overall data demonstrate that EtO-induced ad-3 mutations were the result of a high percentage
20 (96.9%) of gene/point mutations at the ad-3A and ad-3B loci.
21
22 C.2.2. Mammalian Systems
23 EtO has yielded positive results in virtually all in vitro mammalian cell culture systems
24 tested, including human cells (IARC, 2008; Kolman et al., 2002; Thier and Bolt, 2000; Preston,
25 1999; Natarajan et al., 1995; Vogel and Natarajan, 1995; IARC, 1994b; Dellarco et al., 1990).
26 Only select in vitro studies of human cells will be reviewed here. For reviews of other in vitro
27 studies using mammalian cell cultures, see the aforementioned references.
28
29 C.2.2.1. In Vitro Studies
30 Single base pair deletion and base substitution (both transitions and transversions)
31 mutations were observed in the HPRTgene in human diploid fibroblasts exposed to EtO
32 (Bastlova et al., 1993). Sequence analysis revealed that EtO induces many different kinds of
33 //Permutations—several mutants had large HPRTgene deletions, a few mutants showed
34 deletion of the entire HPRT gene, and other mutants had a truncated HPRTgene; overall, as
35 many as 50% were large deletions. In another study by the same group of authors (Lambert et
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-12 DRAFT—DO NOT CITE OR QUOTE
-------
1 al., 1994), comparisons of the //Permutations in human diploid fibroblasts were made for three
2 urban air pollutants (acetaldehyde, benzo[a]pyrene, and EtO). Large genomic deletions in the
3 HPRT gene were observed for acetaldehyde and EtO, whereas benzo[a]pyrene induced point
4 mutations. The authors concluded that the HPRT locus could be a useful target for the study of
5 chemical-specific mutational events (Lambert et al., 1994).
6 The effect of EtO as a pretreatment or posttreatment to ionizing radiation was studied by
7 Kolman and Chovanec (2000). Human diploid VH-10 fibroblasts were either preexposed to
8 gamma rays (0.66 Gy/minute or 10 Gy/minute) and then treated with EtO (2.5 mMh) or
9 pretreated with EtO and then exposed to gamma rays. Cell killing/cytotoxicity, DNA
10 double-strand breakage, and mutagenicity were studied in both types of exposures. The results
11 of the study indicate that preexposure of the cells to gamma radiation (1 Gy) followed by
12 treatment with EtO (2.5 mMh) led to an additive interaction, irrespective of the dose rate. On the
13 other hand, pretreatment with EtO followed by gamma ray exposure resulted in an antagonistic
14 effect, which was most pronounced in the high-dose group (10 Gy/minute). In this group, the
15 mutant frequency was half that of the sum of the mutant frequencies after the individual
16 treatments. The authors suggest that one possible explanation for the difference in the results is
17 that DNA damage induced by preexposure to gamma radiation persisted into the EtO treatment
18 phase, and EtO might also prohibit DNA repair enzymes from operating, thus both treatments
19 contributed to the mutant frequency. However, when cells were exposed to gamma radiation
20 following EtO treatment, the cells may have been able to repair, at least in part, the promutagenic
21 lesions induced by the gamma rays.
22 Tompkins et al. (2009) investigated the mutagenicity of EtO-derived DNA adducts in a
23 supF forward mutation assay. Aliquots of pSP 189 plasmid containing the supF gene were
24 exposed to various concentrations of EtO in water to induce the formation of DNA adducts. The
25 plasmids were then transfected into human embryonic adenovirus-transformed kidney (Ad293)
26 cells and allowed to replicate to propagate any mutations. Replicated plasmids were isolated and
27 used to treat E. Coli indicator bacteria under conditions in which only bacteria containing the
28 plasmid can grow; nonmutant colonies appear dark blue and mutant colonies appear white or
29 pale blue. Two studies were conducted: Study 1, in which the plasmid was incubated with EtO
30 concentrations ranging from 10-2,000 jiM at 22°C for 4 hours, and Study 2, in which the
31 plasmid was treated under "refined" conditions optimised to produce more of the minor
32 2-hydroxyethyl adducts, which involved incubation of the plasmid with EtO concentrations
33 ranging from 10-100 mM at 37°C for 24 hours. For Study 1, Tompkins et al. (2009) reported
34 that N7-HEG was the only detectable adduct of the five they measured (before transfection; see
35 Section C. 1.2 above) and there was no clear exposure-response relationship for the relative
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-13 DRAFT—DO NOT CITE OR QUOTE
-------
1 mutation frequency. In Study 2, Nl-HEA and O6-HEG adducts were also quantifiable, but at
2 lower levels than the N7-HEG adduct, and there was an apparent exposure-response relationship
3 for the relative mutation frequency for plasmids exposed to the 10 and 30 mM EtO
4 concentrations. Plasmids exposed to higher concentrations of EtO failed to produce any E. Coli
5 colonies; this was attributed to excessive strand breaks in the plasmid DNA at those
6 concentrations. For the DNA damage induced by EtO-derived adducts, this limitation in the
7 assay imposes a short response range for the relative mutation frequency for the mutations
8 measured by the assay—the relative mutation frequency was 5.34 for plasmids exposed to
9 30 mM and no E. Coli colonies were produced with plasmids exposed to the next highest EtO
10 concentration of 50 mM, due to excessive DNA strand breaks.
11 Tompkins et al. (2009) concluded that EtO is a relatively weak mutagen and that their
12 results suggest that a certain level of total DNA adducts or of specific promutagenic adducts
13 must be achieved before mutations become detectable above background levels. However,
14 several issues pertaining to the study raise concerns about the interpretation of the results. For
15 example, two solvent controls were used in the study—Solvent Control 1 was prepared in "a
16 separate fume hood to totally exclude any possibility of [EtO] contamination" and Solvent
17 Control 2 was prepared "alongside the [EtO] reactions." Solvent Control 1 was used as the
18 referent group for the relative mutation frequency determinations. In two replicates, Solvent
19 Control 2 had a relative mutation frequency of 3.0 and 2.6 compared to Solvent Control 1. If this
20 difference reflects a real difference between the two different solvent control preparations, it
21 raises the possibility that cross-contamination may have been a problem and, if any
22 cross-contamination also occurred across the different EtO concentrations, this could have
23 dampened any exposure-response relationship. In addition, if the "refined conditions" for
24 plasmid treatment used to produce more of the minor (more directly promutagenic) adducts in
25 Study 2, which included incubation at a temperature more comparable to mammalian body
26 temperatures, had also been used for Study 1, a different adduct profile, and different relative
27 mutation frequencies, might have resulted. The authors themselves acknowledged that "[in]
28 order to categorically determine whether a threshold exists for [EtO] in this system, a more
29 detailed examination of the dose-response relationship using the optimised reaction protocol and
30 including more concentrations around the mutagenic range is needed" (Tompkins et al., 2009).
31 Moreover, there is uncertainty about the generalizability of mutagenicity results from this in vitro
32 experimental system to the mutagenicity and genotoxicity induced by EtO exposure in vivo; for
33 example, human embryonic adenovirus-transformed kidney cells were used for plasmid
34 replication and mutation production, but embryonic kidneys are not a known target for EtO
35 carcinogenesis.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-14 DRAFT—DO NOT CITE OR QUOTE
-------
1 C.2.2.2. In Vivo Studies—Experimental Animals
2 The results of in vivo studies on the mutagenicity of EtO following ingestion, inhalation,
3 or injection have also been consistently positive (e.g., Tates et al., 1999). For example, increases
4 in the frequency of gene mutations in T-lymphocytes (Hprt locus) (Walker et al., 1997) and in
5 bone marrow and testes (Lad locus) (Recio et al., 2004) have been observed in transgenic mice
6 exposed to EtO via inhalation at concentrations similar to those in carcinogenesis bioassays with
7 this species (NTP, 1987). At somewhat higher concentrations than those used in the
8 carcinogenesis bioassays (200 ppm, but for only 4 weeks), increases in the frequency of gene
9 mutations have also been observed in the lung of transgenic mice (Lad locus) (Sisk et al., 1997)
10 and in T-lymphocytes of rats (Hprt locus) (van Sittert et al., 2000; Tates et al., 1999). These and
11 other key in vivo studies are discussed in more detail below.
12 An approach for determining mutational spectra in ex on 3 of the Hprt gene in splenic
13 T-lymphocytes of B6C3Fi mice was developed by Walker and Skopek (1993). Mice (12 days
14 old) were given 2, 6, or 9 single intraperitoneal (i.p.) injections of 100 mg/kg EtO every other
15 day or 30, 60, 90, or 120 mg/kg of EtO for 5 consecutive days to achieve different cumulative
16 doses. In mice exposed every other day, cumulative doses of 200, 600, and 900 mg/kg produced
17 average mutant frequencies of 15 x 10 6, 45 x 10 6, and 73 x 10 6, respectively, 8 weeks after
18 dosing began. However, in mice exposed daily, cumulative doses of 150, 300, 450, and
19 600 mg/kg yielded average mutant frequencies of 4 x 10~6, 8 x 10~6, 11 x icf6, and 16 x 10~6,
20 20 weeks after initiation of dosing. Hprt mutants obtained from mice exposed to 600 or
21 900 mg/kg EtO were isolated and analyzed for mutations, specifically in exon 3. DNA
22 sequencing showed base-pair substitutions, transitions, and transversions. The results suggested
23 both modified guanine and adenine bases being involved in EtO-induced mutagenesis.
24 The same group of authors (Walker et al., 1997) studied the in vivo mutagenicity of EtO
25 at the Hprt locus of T-lymphocytes following inhalation exposure of male B6C3Fi Lad
26 transgenic mice. Big Blue mice at 6-8 and 8-10 weeks of age were exposed to 0, 50, 100, or
27 200 ppm EtO for 4 weeks (6 h/day, 5 days/week). T-cells were isolated from the thymus and
28 spleen and cultured in the presence of concanavalin A, IL-2, and 6-thioguanine. Mice were
29 sacrificed at 2 hours, 2 weeks, and 8 weeks after exposure to 200 ppm EtO to determine a time
30 course for the expression of Hpr^-negative lymphocytes in the thymus. The results of this study
31 showed that following 2 hours of exposure, the Hprt mutant frequency in the thymic
32 lymphocytes of the exposed mice was increased and reached an average maximum mutant
33 frequency of7.5±0.9xlQ6at2 weeks postexposure when compared to2.3±0.8 x 10 6in the
34 thymic lymphocytes of control mice. Dose-related increases in Hprt mutant frequency were
35 found in thymic lymphocytes from mice exposed to 100 and 200 ppm EtO. Furthermore, a
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-15 DRAFT—DO NOT CITE OR QUOTE
-------
1 greater mutagenic efficiency (mutations per unit dose) was found at higher concentrations than at
2 lower concentrations of EtO in splenic T-cells. The average induced mutant frequencies in
3 splenic T-cells were 1.6, 4.6, and 11.9 x 10 6 following exposures to 50, 100, or 200 ppm EtO,
4 respectively. For the analysis of the Lad mutations, lymphocytes (both B- and T-cells) were
5 isolated from the spleen in the same animals. Two of three EtO-exposed mice at the 200 ppm
6 exposure level demonstrated an elevated Lad mutant frequency. The authors suggest that these
7 elevations were probably due to the in vivo replication of preexisting mutants and not to the
8 induction of new mutations associated with EtO exposure. The results of this study indicate that
9 repeated inhalation exposures to high concentrations of EtO produce dose-related increases in
10 mutations at the Hprt locus of T-lymphocytes in male Lad transgenic mice.
11 Lad mutant frequencies as a result of exposure to EtO were further investigated by Sisk
12 et al. (1997). Male transgenic Zac/B6C3Fi mice (n = 15) were exposed to 0, 50, 100, or
13 200 ppm EtO for 4 weeks (6 hours/day, 5 days/week) and were sacrificed at 0, 2, or 8 weeks
14 after the last EtO exposure. To determine the Lad mutant frequency, the Lad transgene was
15 recovered from several tissues, including lung, spleen, germ cells and bone marrow, selected
16 because they were the target sites for tumor formation (particularly lung tumors and lymphomas)
17 in chronic bioassays or germ cells. The results of this study indicate that the Lad mutant
18 frequency in lung was significantly increased at 8 weeks postexposure to 200 ppm EtO. In
19 contrast, no significant increase in the Lad mutant frequencies was observed in the spleen, bone
20 marrow or germ cells at either 2 or 8 weeks following exposure. These results suggest that a
21 4-week inhalation exposure to EtO is mutagenic in lung but not in other tissues examined under
22 similar conditions. The authors predict that the lack of mutagenic response in other tissues
23 examined is probably because of large deletions that were either not detected or recovered in the
24 current lambda-based shuttle vector systems. Based on the above study, the authors also suggest
25 that the primary mechanism of EtO-induced mutagenicity in vivo is likely through the induction
26 of deletions.
27 Tates et al. (1999) exposed rats to EtO via three routes: a single i.p. injection
28 (10-80 mg/kg), ingestion of drinking water (4 weeks at concentrations of 2, 5, and 10 mM), or
29 inhalation (50, 100, or 200 ppm for 4 weeks, 5 days/week, 6 hours/day). The goal of this study
30 was to measure the induction of Hprt mutations in splenic lymphocytes using a cloning assay.
31 Mutagenic effects of EtO following EtO administration via the three routes were compared in the
32 Hprt assay based on blood doses, which were determined from HEVal adduct levels in
33 hemoglobin. Exposure to EtO via both injection and ingestion of drinking water led to a
34 statistically significant dose-dependent induction of mutations (up to 2.3- and 2.5-fold increases
35 in mutant frequency compared to background, respectively). Exposure via inhalation also caused
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-16 DRAFT—DO NOT CITE OR QUOTE
-------
1 a statistically significant increase in mutant frequency, although to a lesser extent (up to 1.4-fold
2 over background). Plotting of the mutagenicity data for the three exposure routes against blood
3 doses as a common denominator indicated that, at equal blood doses, the order of increased
4 mutant frequency was i.p. injection > ingestion (drinking water) > inhalation. In the injection
5 experiments, there was evidence for a saturation of detoxification processes at the highest doses,
6 although such effects were not seen following subchronic administration. Taken together, the
7 mutagenicity data from this study provide consistent results, showing that exposure to EtO gives
8 rise to a linear dose-dependent increase in mutant frequency.
9 In a study by Recio et al. (2004), male Big Blue (Zac/transgenic) B6C3Fi mice were
10 exposed to 0, 25, 50, 100, or 200 ppm EtO (6 hours per day, 5 days per week) for 12, 24, and
11 48 weeks. An unambiguous mutagenic response in the bone marrow was observed only after
12 48 weeks, with dose-related Lad mutant frequencies of 7.3 x 10~5, 11.3 x 10~5, 9.3 x 10~5,
13 14.1 x 10 5, and 30.3 x 10~5. The mutagenic response in bone marrow is consistent with a linear
14 exposure-response relationship, contrary to the assertion by Recio et al. (2004) which appears to
15 be based on a misleading plotting scale. Mutant frequencies from testes (seminiferous tubules)
16 were significantly greater than in controls at 25, 50, and 100 ppm (48-week exposure). No
17 difference between the control and treated groups was observed in the Lad mutant frequency
18 after 48 weeks of 200 ppm EtO exposure. The authors suggest that this was probably due to
19 testicular toxicity. Furthermore, a mutation spectrum analysis of induced mutations in bone
20 marrow indicated a decrease in mutations at G:C base pairs and an increase at A:T base pairs,
21 exclusively in A:T to T: A transversions; however, the mutation spectrum from testes was similar
22 to that of the untreated animals. The difference in mutation spectrum between the two tissues
23 was probably due to differences in the repair of the DNA adducts formed.
24 Mutations in oncogenes {Kras, Hras) and in thep53 tumor suppressor gene have been
25 studied in tumor tissues of several types from B6C3Fi mice exposed to EtO. Hong et al. (2007)
26 obtained tumor tissues from lung, harderian gland and uterus from a 2-year study (NTP, 1987) in
27 which male and female mice were exposed to 0, 50, or 100 ppm EtO by inhalation 6 hours/day,
28 5 days/week and from control mice from other NTP 2-year bioassays. The authors analyzed the
29 tissues for Kras mutations in codons 12, 13, and 61. A high frequency of Kras mutations
30 (23/23 examined, 100%) was observed in EtO-induced lung neoplasms compared to spontaneous
31 lung neoplasms (27/108, 25%). EtO-induced lung neoplasms predominantly exhibited
32 GGT-GTT mutations in codon 12 (21/23), a transversion that was rare in spontaneous lung
33 tumors (1/108). A similar spectrum of Kras mutations was detected in EtO-induced lung
34 neoplasms regardless of histological subtype (adenomas or carcinomas) or dose group. In the
35 case of Harderian gland neoplasms, a high frequency (18/21, 86%) of Kras mutations was
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-17 DRAFT—DO NOT CITE OR QUOTE
-------
1 detected in EtO-induced neoplasms compared to spontaneous tumors (2/27, 7%). The
2 predominant mutations in EtO-induced harderian gland neoplasms consisted of GGC to CGC
3 transversions at codon 13 and GGT to TGT transversions at codon 12, neither of which was
4 observed in the spontaneous tumors. When the six EtO-induced uterine neoplasms were
5 examined (there were no uterine tumors in the controls), the predominant mutation was a GGC to
6 GGT transition in codon 13 (5/6, 83%). Based on the above results, the authors propose that the
7 prominent targeting of guanine bases in the lung and harderian gland neoplasms suggests that the
8 formation of N7-HEG adducts by EtO plays a role in the induction of these tumors. The authors
9 further propose that EtO can specifically target the Kras gene in multiple types of tissues and that
10 this is a critical component of EtO-induced tumorigenesis and is of potential relevance to
11 humans.
12 In an earlier study by the same group of authors (Houle et al., 2006), mammary
13 carcinoma tissues from the same NTP study of mice exposed to EtO (0, 50, or 100 ppm)
14 mentioned above were examined for p53 protein expression and forp53 (exons 5-8) and Hras
15 (codon 61) mutations. The authors supplemented the number of spontaneous mammary
16 carcinomas with tissues from female control mice in other NTP studies. P53 protein expression
17 was detected in 67% (8/12) of the EtO-induced mammary carcinomas and 42% (8/19) of the
18 spontaneous tumors; however, expression levels were about 6-times higher in the EtO-induced
19 than in the spontaneous tumors. P53 mutations were observed in 67% (8/12) of the EtO-induced
20 mammary carcinomas and 42% (8/19) of the spontaneous tumors. Hras mutations were detected
21 in 33% (4/12) of the EtO-induced mammary carcinomas and 26% (5/19) of the spontaneous
22 tumors of the samples. While the mutation levels for these two genes were not substantially
23 elevated in the EtO-induced mammary carcinomas compared to the spontaneous tumors, a shift
24 in the mutational spectrum was observed, with EtO-induced Hras mutations exhibiting a
25 preference for A-to-G and A-to-T transversions while spontaneous Hras mutations exhibited a
26 preference for C-to-A transversions and EtO-inducedp53 mutations exhibiting a base preference
27 for guanine while spontaneous p53 mutations exhibited a preference for cytosine. In addition,
28 concurrent Hras andp53 mutations were more common in the EtO-induced tumors than in the
29 spontaneous tumors. Based on the results of the above two studies, it is suggested that the purine
30 bases serve as primary targets for mutations induced by EtO, while mutations of these genes
31 involving cytosine appears to be a more common spontaneous event.
32 In vivo exposure to EtO also induced heritable mutations or effects in germ cells in
33 rodents (IARC, 1994b). EtO induces dominant lethal effects in mice and rats and heritable
34 translocations in mice (Generoso et al., 1990; Lewis et al., 1986). Generoso et al. (1988);
35 Generoso et al. (1986) have reported that short bursts of EtO at high concentrations, such as
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-18 DRAFT—DO NOT CITE OR QUOTE
-------
1 those that may occur in the workplace, may present a greater risk to germ cell damage than
2 cumulative, long-term exposure to lower levels.
3 Dominant-lethal mutations were investigated by Generoso et al. (1986) by conducting
4 two studies (dose response and dose rate) in mice exposed to different doses of EtO.
5 Dominant-lethal responses were assessed based on matings involving sperm exposed as late
6 spermatids and early spermatozoa, since these are the stages most sensitive to EtO exposure. In
7 the dose-response study, male mice were exposed by inhalation to 300 ppm, 400 ppm, or
8 500 ppm EtO, 6 hours per day, for 4 consecutive days. A dose-related increase in
9 dominant-lethal mutations was observed. In the dose-rate study, mice were given a total
10 exposure of 1,800 ppm x hours per day, also for 4 consecutive days, delivered either as 300 ppm
11 in 6 hours, 600 ppm in 3 hours, or 1,200 ppm in 1.5 hours. Dominant-lethal responses increased
12 with increasing concentration level, indicating a dose-rate effect for the production of
13 dominant-lethal mutations.
14
15 C.2.2.3. in Vivo Studies—Humans
16 In humans, workers occupationally exposed to EtO have been studied using different
17 physical and biological measures (Tates et al., 1991). Blood samples from 9 hospital workers
18 and 15 factory workers engaged in sterilization of medical equipment with EtO and from
19 matched controls were collected. Average exposure levels during 4 months (the lifespan of
20 erythrocytes) prior to blood sampling were estimated from levels of HEVal adducts in
21 hemoglobin. The adduct levels were significantly increased in hospital workers and factory
22 workers and corresponded to a 40-hour time-weighted average of 0.025 ppm in hospital workers
23 and 5 ppm in factory workers. Exposures were usually received in bursts, with EtO
24 concentrations in air ranging from 22 to 72 ppm in hospital workers and 14 to 400 ppm in factory
25 workers. All blood samples were analyzed for HPRT mutant frequencies, chromosomal
26 aberrations, micronuclei and SCEs. Mutant frequencies were significantly increased in factory
27 workers but not in hospital workers. The chromosomal aberration and SCE results are discussed
28 in the respective sections below.
29 The same authors (Tates et al., 1995) conducted another study of workers in an EtO
30 production facility. //Permutations were measured in three exposed groups and one unexposed
31 group (seven workers per group). Contrary to the earlier study, no significant differences in
32 mutant frequencies were observed between the groups; however, the authors stated that about
33 50 subjects per group would have been needed to detect a 50% increase.
34 Major et al. (2001) measured //Permutations in female nurses employed in hospitals in
35 Eger and Budapest, Hungary. This study was conducted to examine a possible causal
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-19 DRAFT—DO NOT CITE OR QUOTE
-------
1 relationship between EtO exposure and a cluster of cancers (mostly breast) in nurses exposed to
2 EtO in the Eger hospital. Controls were female hospital workers in the respective cities. The
3 mean peak levels of EtO were 5 mg/m3 (2.7 ppm) in Budapest and 10 mg/m3 (5.4 ppm) in Eger.
4 HPRT variant frequencies in both controls and EtO-exposed workers in the Eger hospital were
5 higher than either group in the Budapest hospital, but there was no significant increase among
6 the EtO-exposed workers in either hospital when compared with the respective controls.
7
8 C.2.3. Gene Mutations—Summary
9 In summary, there is sufficient evidence for mutagenicity of EtO in various organisms
10 (prokaryotes, eukaryotes, in vitro and in vivo in rodents and in vitro in human cells) tested in a
11 variety of mutational assays. In addition, increases in mutations in specific oncogenes and tumor
12 suppressor genes in EtO-induced mouse tumors have been reported. Dominant-lethal mutations
13 have also been observed in several in vivo studies. Although data in humans are limited, there is
14 some evidence of increased frequencies of mutations from occupational studies.
15
16 C.3. CHROMOSOMAL ABERRATIONS
17 The induction and persistence of EtO-induced chromosomal alterations have been studied
18 both in in vitro and in vivo systems in rodent and monkey models (Lorenti Garcia et al., 2001;
19 Farooqi et al., 1993; Lynch et al., 1984b; Kligerman et al., 1983). In addition, several studies
20 examined the association of chromosomal aberrations and EtO exposure in humans (WHO,
21 2003; Lerda and Rizzi, 1992; Galloway et al., 1986; Clare et al., 1985; Sarto et al., 1984a;
22 Stolley et al., 1984; Pero et al., 1981; Thiess et al., 1981). Chromosomal aberrations have been
23 linked to an increased risk of cancer in several large prospective studies (e.g., Boffetta et al.,
24 2007; Rossner et al., 2005; Hagmar et al., 2004; Liou et al., 1999). This section discusses key
25 studies on EtO and chromosomal aberrations.
26 Lorenti Garcia et al. (2001) studied the effect of EtO on the formation of chromosomal
27 aberrations in rat bone-marrow cells and splenocytes following in vivo exposure. Rats were
28 exposed to EtO either chronically by inhalation (50-200 ppm, 4 weeks, 5 days/week,
29 6 hours/day) or acutely by i.p. injection at dose levels of 50-100 ppm. Frequencies of both
30 spontaneous and EtO-induced chromosomal aberrations (and other endpoints, such as
31 micronucleus formation and SCEs, which are discussed in Sections 3.3.2.4 and 3.3.2.5) were
32 determined in the splenocytes and bone-marrow cells following in vivo mitogen stimulation. No
33 significant increase in chromosomal aberrations was observed from the chronic or acute
34 exposures. In another study, by Kligerman et al. (1983) , no increase in chromosomal
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-20 DRAFT—DO NOT CITE OR QUOTE
-------
1 aberrations was observed in peripheral blood lymphocytes from rats exposed to EtO by
2 inhalation at concentrations of either 50, 150, or 450 ppm, for 6 hours per day, for 1 and 3 days.
3 A recent study by Donner et al. (2010) in mice, however, showed clear, statistically
4 significant increases in chromosomal aberrations with longer durations of exposure (>12 weeks).
5 Male B6C3Fi mice were exposed by inhalation to 0, 25, 50, 100, or 200 ppm EtO, 5 days/week,
6 6 hours/day, for 6, 12, 24, or 48 weeks. The frequency of total chromosomal aberrations in
7 peripheral blood lymphocytes was statistically significantly increased after 12 weeks exposure to
8 100 or 200 ppm EtO. By 48 weeks, statistically significant increases were observed for all the
9 exposure groups. In addition, reciprocal translocation frequencies were statistically significantly
10 increased in spermatocytes for all the exposure groups at 48 weeks. Ribeiro et al. (1987)
11 similarly observed chromosomal aberrations in mouse bone marrow cells and spermatocytes
12 following 1-day and 2-week inhalation exposures to higher levels of EtO. Male Swiss Webster
13 mice were exposed to 0, 200, 400, or 600 ppm EtO for 6 hours in 1 day or to 0, 200, or 400 ppm
14 EtO for 6 hours/day, 5 days/week, for 2 weeks. Statistically significant increases in
15 chromosomal aberrations were observed in bone marrow cells and in spermatocytes following a
16 1-day exposure of 400 or 600 ppm EtO or a 2-week exposure of 200 or 400 ppm EtO.
17 Chromosomal aberrations in bone marrow cells were also reported in a study of acute EtO
18 exposure in mice (Farooqi et al., 1993). Female Swiss albino mice were administered single
19 doses of EtO in the range of 30-150 mg/kg by i.p. injection. A dose-related increase in
20 chromosomal aberrations in the bone marrow cells was observed.
21 Chromosomal aberrations induced by long-term exposures to inhaled EtO were also
22 investigated in the peripheral lymphocytes of cynomolgus monkeys (Lynch et al., 1984b).
23 Groups of 12 adult male monkeys were exposed at 0, 50, or 100 ppm EtO (7 hours/day,
24 5 days/week) for 2 years. Exposure to EtO at 100 ppm resulted in statistically significant
25 increases in chromosome-type aberrations in monkey lymphocytes, and exposure at both 50 and
26 100 ppm resulted in statistically significant increases in chromatid-type aberrations and in
27 chromosome- and chromatid-type aberrations in combination. No differences in the number of
28 gaps were found.
29 Increases in chromosomal aberrations in peripheral blood lymphocytes have been
30 consistently reported in studies of workers exposed to high occupational concentrations of EtO
31 (>5 ppm, TWA). Effects observed at lower concentrations have been mixed (WHO, 2003).
32 Chromosomal aberrations that have been detected in the peripheral blood lymphocytes of
33 workers include breaks, gaps, and exchanges and supernumerary chromosomes (Lerda and Rizzi,
34 1992; Galloway et al., 1986; Clare et al., 1985; Sarto et al., 1984a; Pero et al., 1981; Thiess et al.,
35 1981).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-21 DRAFT—DO NOT CITE OR QUOTE
-------
1 Clare et al. (1985) conducted chromosomal analyses of lymphocytes from 33 workers
2 employed in the manufacture of EtO. A slightly higher frequency of chromatid aberrations was
3 observed in workers exposed to EtO than in controls. Further, a positive correlation between
4 length of employment in the EtO-exposed group and the number of aberrations was observed. In
5 another study, Galloway et al. (1986) analyzed chromosomal aberration frequencies in
6 61 employees potentially exposed to EtO. Three work sites (I, II and III) with different historical
7 ambient levels of EtO were chosen for the study. Blood samples were drawn over a 24-month
8 period and aberrations were analyzed in 100 cells per sample after culture for 48-51 hours. At
9 work sites I and II, no consistent differences in aberration frequencies were found. However, at
10 work site III, aberration frequencies in potentially exposed individuals were significantly
11 increased when compared with controls. A previous study by the same group (Stolley et al.,
12 1984) showed an association between SCE frequency and EtO exposure. When the aberrations
13 were compared with the levels of SCEs, the authors found a weak overall association. In
14 addition, Lerda and Rizzi (1992) showed a significant increase in chromosomal aberration
15 frequencies in EtO-exposed individuals when compared with controls. Major et al. (1996)
16 studied hospital nurses exposed to low doses and high doses of EtO to identify changes in
17 structural and numerical chromosomal aberrations. Chromosomal aberrations were found to be
18 significantly elevated in both the low-dose and the high-dose exposure groups. Deletions and, to
19 a lesser extent, chromatid exchanges and dicentrics were detected in the low-dose exposure
20 group; however, in the high-dose group, in addition to the increased number of deletions, the
21 frequencies of dicentrics and rings showed a significant excess when compared with controls.
22 The authors suggest that a natural radioactivity from local tap water may have been a
23 confounding factor.
24 A study by Sarto et al. (1984a) showed significant increases in chromosomal aberrations
25 after exposure to EtO. Chromosomal aberrations were detected in the peripheral lymphocytes of
26 41 workers exposed to EtO in the sterilizing units of eight hospitals in the Venice region
27 compared to 41 age- and smoking-matched controls. In another study of 28 EtO-exposed
28 sterilizer workers and 20 unexposed controls, Hogstedt et al. (1983) reported a statistically
29 significant increase in total chromosomal aberrations and gaps, but not breaks, in the peripheral
30 blood lymphocytes of the exposed workers, adjusted for age, smoking, drug intake, and exposure
31 to ionizing radiation; no significant increases in chromosomal aberrations were observed in bone
32 marrow cells. Tates et al. (1991) reported a significant increase in chromosomal aberrations in
33 hospital workers and in factory workers (details of this study are provided in the section on gene
34 mutations above). Tompa et al. (2006) reported statistically significant increases in
35 chromosomal aberrations and SCEs in 66 Hungarian hospital nurses exposed to sterilizing gases
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-22 DRAFT—DO NOT CITE OR QUOTE
-------
1 in uncontrolled environments compared to 94 nonexposed controls; however, it is difficult to sort
2 out any effects of EtO exposure from possible effects from smoking or exposure to ionizing
3 radiation or to formaldehyde or other possible sterilizing gases in this study.
4 In summary, the above data clearly indicate that EtO is genotoxic and can cause a variety
5 of chromosomal aberrations, including breaks, gaps and exchanges (reviewed in detail in
6 Preston, 1999). Chromosomal aberrations have been observed in both in vitro and in vivo
7 studies in rodent models and mammalian cells. Increases in chromosomal aberrations in
8 peripheral blood lymphocytes have been consistently reported in studies of workers exposed to
9 EtO.
10
11 C.4. MICRONUCLEUS FORMATION
12 Micronucleus formation also demonstrates the genotoxic effects of a chemical. When
13 appropriate methods are used to identify the origin of the micronucleus (kinetochore-positive or
14 kinetochore-negative), this assay can provide information about a chemical's mechanism of
15 action (e.g., if a chemical causes direct DNA damage resulting from strand breaks [clastogen] or
16 indirect numerical changes [aneugen] resulting from spindle disruption). An association between
17 increased micronucleus frequency and cancer risk has been reported in at least one large
18 prospective study (Bonassi et al., 2007). Several in vitro and in vivo studies in both laboratory
19 animals (Lorenti Garcia et al., 2001; Jenssen and Ramel, 1980; Appelgren et al., 1978) and
20 humans (Ribeiro et al., 1994; Schulte et al., 1992; Mayer et al., 1991; Tates et al., 1991; Sarto et
21 al., 1990; Hogstedt et al., 1983) have been conducted to explore the induction of micronuclei as a
22 result of exposure to EtO.
23 Lorenti Garcia et al. (2001) studied the effect of EtO on the formation of micronuclei in
24 rat bone marrow cells and splenocytes following in vivo exposure. Rats were exposed to EtO
25 either subchronically by inhalation (50-200 ppm, 5 days/week, 6 hours/day, for 4 weeks) or
26 acutely by i.p. injection at dose levels of 50 or 100 mg/kg. Spontaneous and induced frequencies
27 of micronuclei were determined in the bone marrow cells (only for acute EtO exposure) and
28 splenocytes following in vitro mitogen stimulation. Following chronic exposure, no significant
29 increase in micronuclei was observed in rat splenocytes. Following acute exposure, micronuclei
30 increased significantly in rat bone marrow cells as well as splenocytes.
31 In the Hogstedt et al. (1983) study of 28 EtO-exposed sterilizer workers and
32 20 unexposed controls discussed in Section C.3, a statistically significant increase in micronuclei
33 was observed in bone marrow cells (erythroblasts and polychromatic erythrocytes), but not in
34 lymphocytes, in the exposed workers, adjusted for age, smoking, drug intake, and exposure to
35 ionizing radiation.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-23 DRAFT—DO NOT CITE OR QUOTE
-------
1 The frequency of micronuclei in peripheral blood cells was increased in workers exposed
2 to relatively high (3.7-60.4 mg/m3) levels of EtO (Ribeiro et al., 1994; Tates et al., 1991).
3 Schulte et al. (1992) did not observe increased micronuclei in the lymphocytes of hospital
4 workers with low levels of EtO exposure (up to 2.5 mg/m3 8-hour TWAs). Sarto et al. (1990)
5 studied micronucleus formation in human exfoliated cells of buccal and nasal cavities to monitor
6 the genotoxic risk in a group of workers (n = 9) chronically exposed to EtO (concentrations
7 lower than 0.38 ppm as time-weighted average). The mean frequencies of micronucleated
8 buccal cells were similar to control values. The frequency of nasal micronucleated cells was
9 higher than in controls (0.77 vs. 0.44); however, the difference was not statistically significant.
10 In another group of three subjects that were acutely exposed (concentration not provided) to EtO,
11 buccal cavity and nasal mucosa samples were taken 3, 9, or 16 days after acute exposure. The
12 frequencies of micronucleated buccal cells did not change, while the frequencies of
13 micronucleated nasal cells significantly increased.
14 Peripheral blood cells of 34 EtO-exposed workers at a sterilization plant and
15 23 unexposed controls were assessed for different biological markers, such as EtO-hemoglobin
16 adducts, SCEs, micronuclei, chromosomal aberrations, DNA single-strand breaks and an index
17 of DNA repair (Mayer et al., 1991). Neither chromosomal aberrations nor micronuclei differed
18 significantly by exposure status, whether or not adjusted for smoking status.
19 In summary, increases in the frequency of micronuclei have been observed in in vivo
20 animal studies. The frequency of micronuclei in peripheral blood cells was also increased in
21 workers exposed to relatively high (3.7-60.4 mg/m3) levels of EtO (Ribeiro et al., 1994; Tates et
22 al., 1991). However, in the majority of human studies involving exposures at lower levels, no
23 effects on the frequency of micronuclei were observed. Apparent inconsistencies in the data
24 could reflect the influence of peak exposures, differences in exposure measurement errors,
25 duration of exposure and/or smoking status.
26
27 C.5. SISTER CHROMATID EXCHANGES (SCES)
28 There is a significant body of evidence for the induction of SCEs as a result of exposure
29 to EtO. Studies have been conducted both in laboratory animals (Lorenti Garcia et al., 2001;
30 Ong et al., 1993; Kelsey et al., 1988; Lynch et al., 1984b; Kligerman et al., 1983; Yager and
31 Benz, 1982) and in humans (Agurell et al., 1991; Galloway et al., 1986; Laurent et al., 1984;
32 Sarto et al., 1984a, b; Stolley et al., 1984; Yager et al., 1983; Garry et al., 1979). In particular,
33 several occupational exposure studies have yielded positive results when EtO-exposed workers
34 were studied. The following is a summary of both the animal and human studies.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-24 DRAFT—DO NOT CITE OR QUOTE
-------
1 Inhalation studies with rats have shown that exposures to EtO at 50 ppm or more for
2 3 days result in an increase in SCEs in peripheral blood lymphocytes (Kligerman et al., 1983).
3 Increased incidences of SCEs in the peripheral blood lymphocytes of monkeys exposed to EtO at
4 500 or 100 ppm were also reported by Lynch et al. (1984b). A follow-up study in these same
5 monkeys by Kelsey et al. (1988) indicated that the high SCE counts persisted for 6 years after
6 exposure.
7 Lorenti Garcia et al. (2001) studied the effect of EtO on the persistence of SCEs in rat
8 bone marrow cells and splenocytes following in vivo exposure. Rats were exposed to EtO either
9 chronically by inhalation (50-200 ppm, 5 days/week, 6 h/day, for 4 weeks) or acutely by i.p.
10 injection at dose levels of 50 or 100 mg/kg. Frequencies of SCEs were determined in the bone
11 marrow cells and splenocytes after in vitro mitogen stimulation. Following chronic exposure,
12 cytogenetic analyses were carried out at days 5 and 21 in the splenocytes. In these experiments,
13 EtO was effective in inducing SCEs, and marked increases in cells with high frequency SCEs
14 were observed which persisted until day 21 postexposure. Following acute exposure, SCEs were
15 increased significantly in rat bone marrow cells as well as splenocytes.
16 New Zealand white male rabbits (n = 4) were exposed in inhalation chambers to 0, 10,
17 50, and 250 ppm EtO for 6 hours a day, 5 days a week, for 12 weeks (Yager and Benz, 1982).
18 Peripheral blood samples were drawn in three regimes (before the start of exposure, at intervals
19 during exposure, and up to 15 weeks after the end of exposure) to measure SCE rates. No
20 change in SCE rates was observed from exposure to 10 ppm; however, an increase was seen after
21 exposure to 50 and 250 ppm. Above-baseline levels were observed even after 15 weeks
22 postexposure, although the levels were not as high as during exposure. These results indicate
23 that inhalation exposure to the EtO results in a dose-related increase in SCEs.
24 The ability of long-term exposures to inhaled EtO to induce SCEs in peripheral
25 lymphocytes of monkeys was investigated by Lynch et al. (1984b). Groups of 12 adult male
26 cynomolgus monkeys were exposed at 0, 50, or 100 ppm EtO (7 hours/day, 5 days/week) for
27 2 years. Statistically significant increases in SCE rates were observed in monkey lymphocytes in
28 both exposure groups. Both exposure groups had increased numbers of SCEs/metaphase as
29 compared to controls, and these numbers increased in a dose-dependent manner.
30 In an in vitro study of human cells, peripheral lymphocyte cultures were exposed to
31 methyl bromide, EtO, and propylene oxide, as well as diesel exhaust (Tucker et al., 1986). SCE
32 frequency was measured, and the frequency more than doubled in the cultures treated with EtO.
33 Agurell et al. (1991) also studied the effect of EtO on SCEs in human peripheral blood
34 lymphocytes in vitro. An increase in SCE frequency was observed as a result of exposure
35 (0-20 mMh) to EtO. Similarly, Hallier et al. (1993) observed that the frequency of SCEs in
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-25 DRAFT—DO NOT CITE OR QUOTE
-------
1 human peripheral blood lymphocytes exposed in vitro to EtO was higher in cells isolated from
2 individuals expressing low levels of glutathione S-transferase Tl than in cells from subjects
3 expressing higher levels of this enzyme.
4 Several studies of EtO-exposed workers have also reported an increased incidence of
5 SCEs in peripheral lymphocytes (e.g., Schulte et al., 1992; Galloway et al., 1986; Sarto et al.,
6 1984a, b; Yager et al., 1983; Garry et al., 1979), although the Hogstedt et al. (1983) study
7 discussed in Sections C.3 and C.4 did not report significant increases in SCEs in the lymphocytes
8 of the exposed workers.
9 Garry et al. (1979) analyzed SCEs in lymphocytes cultured from EtO-exposed individuals
10 as well as comparable controls. Significant increases in SCEs were observed at 3 weeks and at
11 8 weeks following exposure. Although this study does not describe the exact exposure estimates,
12 EtO was recognized as a mutagenic or genotoxic agent. Laurent et al. (1984) studied SCE
13 frequency in workers exposed to high levels of EtO in a hospital sterilization service. Blood
14 samples were obtained retrospectively from a group of 25 subjects exposed to high levels of EtO
15 for a period of 2 years. A significant increase in SCEs was observed in the exposed group when
16 compared with the control group. The authors concluded that the effect of exposure to EtO was
17 sufficient to produce a cumulative and, in some cases, a persistent genetic change.
18 Peripheral blood lymphocytes of nurses exposed to low and high concentrations of EtO
19 were studied by Major et al. (1996). SCEs were slightly elevated in the low-exposure group but
20 were significantly increased in the high-exposure group. Similarly, several studies by Sarto et al.
21 (1991); Sarto et al. (1990); Sarto et al. (1987); Sarto et al. (1984a, b) showed significant
22 increases in SCEs.
23 Tates et al. (1991) studied workers occupationally exposed to EtO using different
24 physical and biological measures. Blood samples from 9 hospital workers and 15 factory
25 workers engaged in sterilization of medical equipment with EtO and from matched controls were
26 collected. Exposures were usually received in bursts, with EtO concentrations in air ranging
27 from 22 to 72 ppm in hospital workers and 14 to 400 ppm in factory workers. The mean
28 frequency of SCEs was significantly elevated by 20% in hospital workers and by almost 100% in
29 factory workers. In contrast, no significant increase in SCEs was observed in lymphocytes of
30 workers who were accidentally exposed to high concentrations of EtO or of workers with low
31 exposure concentrations (Tates et al., 1995).
32 Schulte et al. (1992) observed a statistically significant increase in SCEs in 43 workers
33 exposed to EtO in U.S. hospitals compared to 8 unexposed hospital workers. The frequency of
34 SCEs was also significantly associated with cumulative EtO exposure in a regression analysis
35 that controlled for various potential confounding factors, including smoking. A similar
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-26 DRAFT—DO NOT CITE OR QUOTE
-------
1 relationship was not observed in 22 Mexican hospital workers. Schulte et al. (1992)
2 hypothesized that the difference may have been due to longer shipping times of the Mexican
3 specimens for the cytogenetic assays.
4 In summary, significant increases in the frequency of SCEs were observed in rats and in
5 monkeys both by inhalation and i.p. injection. In humans, multiple occupational studies have
6 reported positive responses, with significant increases in frequency of SCEs in peripheral blood
7 lymphocytes having been observed among individuals exposed to higher levels of EtO. In some
8 studies, increases in the frequency of SCEs have been observed to persist after exposure has
9 ceased. The results of studies of individual workers exposed to very low levels (<0.9 mg/m3) of
10 EtO have been mixed.
11
12 C.6. OTHER ENDPOINTS (GENETIC POLYMORPHISM, SUSCEPTIBILITY)
13 Dose-dependent effects of polymorphisms in the genes for epoxide hydrolase (EPHXJ\
14 different subfamilies of glutathione-^-transferase (GSTM1, GSTP1, GSTT1) and various DNA
15 repair enzymes (hOGGl, XRCC1, XRCC3) on EtO-induced genotoxicity were evaluated by
16 Godderis et al. (2006). Peripheral blood mononuclear cells from 20 individuals were exposed to
17 3 doses of EtO (0.45, 0.67, 0.9 mM), and genotoxicity was evaluated by measuring comet tail
18 length and micronucleus frequencies in binucleated cells (MNBC). A dose-dependent increase
19 in tail length (indicating DNA strand breaks) was observed in exposed individuals compared to
20 controls. No change in MNBC was observed. None of the epoxide hydrolase or glutathione-^-
21 transferase polymorphisms had a significant influence on the tail length or MNBC results for any
22 EtO dose. Further analysis revealed a significant contribution of the hOGGl (involved in base
23 excision repair) and XRCC3 (involved in repair of cross-links and chromosomal double-strand
24 breaks) genotypes to the interindividual variability of EtO-induced increases in tail length.
25 Homozygous hOGGl326 wild-type cells showed significantly lower effects of EtO on tail length
26 compared to the heterozygous cells. Also, significantly higher tail lengths were found in
27 EtO-exposed cells carrying at least one variant XRCC3241 Met allele. For the latter effect, there
28 was a significant interaction between theXRCCS241 polymorphism and dose, signifying a greater
29 impact of the polymorphism on DNA damage at higher doses.
30 In contrast to the findings of no significant effect of glutathione-^-transferase
31 polymorphisms on DNA breaks and micronuclei production by Godderis et al. (2006), Hallier et
32 al. (1993) observed that the frequency of SCEs in human peripheral blood lymphocytes exposed
33 in vitro to EtO was higher in cells isolated from individuals expressing low levels of GSTT1 than
34 in cells from subjects expressing higher levels of this enzyme. Similarly, Yong et al. (2001)
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-27 DRAFT—DO NOT CITE OR QUOTE
-------
1 measured approximately twofold greater EtO-hemoglobin adduct levels in occupationally
2 exposed persons with a G-STr7-null genotype than in those with positive genotypes.
3 In a study involving small numbers (n = 4-12 per group) of nonsmoking males and
4 females exposed to EtO through the sterilization of medical equipment, Fuchs et al. (1994)
5 reported 1.5-, 2.2-, and 1.5-fold increases in DNA single-strand breaks in peripheral blood
6 mononuclear cells obtained from individuals exposed to EtO concentrations of 0.1-0.49 mg/m3,
7 0.5-2.0 mg/m3, and >2 mg/m3, respectively. Fuchs et al. (1994) further noted that these
8 nonsmokers could be divided into two distinct susceptibility groups, with 67% of the subjects
9 exhibiting approximately fivefold higher levels of DNA single-strand breaks in response to EtO
10 exposure than the remaining subjects, and that the bimodal nature of the differential
11 susceptibility suggested that the susceptibility was attributable to an unidentified polymorphism.
12 Primary and secondary cultures of lymphoblasts, breast epithelial cells, peripheral blood
13 lymphocytes, keratinocytes and cervical epithelial cells were exposed to 0-100 mM EtO, and
14 DNA damage was measured using the comet assay (Adam et al., 2005). A dose-dependent
15 increase in DNA damage was observed in all cell types without notable cytotoxicity. Breast
16 epithelial cells (26% increase in tail length) were more sensitive than keratinocytes (5% increase)
17 and cervical epithelial cells (5% increase) but less sensitive than lymphoblasts (51% increase)
18 and peripheral lymphocytes (71% increase) at the same dose of 20 mM.
19
20 C.7. ENDOGENOUS PRODUCTION OF ETHYLENE AND ETO
21 Ethylene, a biological precursor of EtO, is ubiquitous in the environment as an air
22 pollutant and is produced in plants, animals and humans (Abeles and Heggestad, 1973).
23 Ethylene is generated in vivo endogenously during normal physiological processes such as (1)
24 oxidation of methionine, (2) oxidation of hemoglobin, (3) lipid peroxidation, and (4) metabolism
25 of intestinal bacteria (reviewed by Thier and Bolt, 2000; IARC, 1994a). Recently, Marsden et al.
26 (2009) proposed that oxidative stress can induce the endogenous formation of ethyl ene, which
27 can in turn be metabolized to EtO. Endogenous production of ethyl ene has been documented in
28 laboratory animals and in humans (Filser et al., 1992; Shen et al., 1989; Ehrenberg et al., 1977;
29 Chandra and Spencer, 1963).
30 Shen et al. (1989) reported an endogenous production rate of 2.8 and 41 nmol/h ethylene
31 in Sprague-Dawley rats and humans, respectively, with similar thermodynamic partition
32 coefficients between the two species. Filser et al. (1992) reported a low degree of endogenous
33 production of ethylene (32 ±12 nmol/h) in healthy volunteers based on exhalation data. The
34 authors indicated that the endogenous levels of ethylene would account for -66% of the
35 background level of EtO-hemoglobin adducts (HEVal), while the remaining one-third (15 ppb) is
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-28 DRAFT—DO NOT CITE OR QUOTE
-------
1 contributed by exogenous environmental ethylene exposure. Although the percentage of
2 endogenous ethylene converted to EtO is not known, Tornqvist et al. (1989) have shown that in
3 fruit-store workers exposed to 0.3 ppm ethylene, only 3% is metabolized to EtO. Thus, the
4 amount of endogenous ethylene converted to EtO would be minimal. Furthermore, with
5 inadequate laboratory animal and human evidence available for ethylene as a carcinogen (IARC,
6 1994a), exogenous ethylene exposure may not produce enough EtO to contribute significantly to
7 carcinogenicity under standard bioassay conditions (Walker et al., 2000).
8 Ethylene formed from endogenous sources is converted to EtO by cytochrome
9 P450-mediated metabolism (Tornqvist, 1996; IARC, 1994a). EtO formed from the endogenous
10 conversion of ethylene leads to 2-hydroxyethylation of DNA and forms N7-HEG adducts
11 contributing to the background levels of this adduct in unexposed humans and rodents. As
12 shown in Table C-l, improvements in analytical methodology have led to the detection and
13 quantification of background N7-HEG adducts in DNA of unexposed experimental animals and
14 humans (Marsden et al., 2009; Swenberg et al., 2008; Tompkins et al., 2008; Marsden et al.,
15 2007; Swenberg et al., 2000; van Sittert et al., 2000; Walker et al., 2000; Eide et al., 1999;
16 Farmer and Shuker, 1999; Wu et al., 1999b; Wu et al., 1999a; Zhao et al., 1999; Bolt et al., 1997;
17 Zhao et al., 1997; Kumar et al., 1995; van Delft et al., 1994; Farmer et al., 1993; van Delft et al.,
18 1993; Leutbecher et al., 1992; Walker et al., 1992b; Cushnir et al., 1991; Fost et al., 1989).
19 However, there is a wide variation in the levels of adducts detected in rodents and humans which
20 appears to depend on the type of the analytical method used. Even with the most advanced
21 techniques (Tompkins et al., 2008), minor DNA adducts such as O6-F£EG and N3-HEA were
22 below the level of detection. Also, some researchers consistently demonstrated higher
23 background levels of DNA adducts (Wu et al., 1999a; Walker et al., 1992b). However, the
24 higher background levels in some of these studies are possibly due to the methodology used,
25 which may have caused an artifactual increase in the adduct levels.
26
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-29 DRAFT—DO NOT CITE OR QUOTE
-------
to
OJ
Table C-l. Levels of endogenous (background) N7-HEG adducts in unexposed human and experimental rodent
tissues
Species
Human
Human
Human
Human
Human
Human
Human
Rat
Mice/Rats
Rat
Mice/Rats
Rat
Rat
Rat
Rat
Rat
Tissue
Lymphocytes
WBC
Blood
Lymphocytes
WBC
WBC
Lung
Lymphocytes
Control tissues
Liver, kidney, spleen
Spleen
Control tissues
Liver
Control tissues
Liver
Spleen
Detection method
GC/MS
Immuno-slotblot
HPLC-fluorescence
GC/MS
32P/TLC/HPLC
32P/TLC/HPLC
32P/TLC/HPLC
GC/MS
HPLC-fluorescence
32P/GC/MS
GC/EC/NCI-HRMS
32P/TLC/HPLC
GC/MS
LC-MS/MS
HPLC/ESI TMS
HPLC/LC-MS/MS
Adduct levels reported
8.5 pmol/mg DNA
0.34 adducts/106 nucleotides
3. 2 pmol/mg DNA
2-19 adducts per 107 nucleotides
0.6 adducts/107 nucleotides
2.9 adducts/107 nucleotides
4.0 adducts/107 nucleotides
5.6 pmol/mg DNA
2-6 pmol/mg DNA
0.4 to 1.1 adducts/107 nucleotides
0.2 to 0.3 pmol/mmol guanine
0.6 to 0.9 adducts/107 nucleotides
2.6 adducts/108 nucleotides
1.1-3.5 adducts/108 nucleotides
8 adducts/108 normal nucleotides
0.08 adducts/1010 nucleotides
Adducts/107
nucleotides*
28.05
3.4
10.56
2.0-19
0.6
2.9
4
18.48
8.58
0.4-1.1
0.6-0.9
0.26
0.11-0.35
0.8
0.00008
Reference
(Fostetal., 1989)
(van Delft etal., 1994)
(Boltetal., 1997)
(Wuetal., 1999b)
(Zhao etal., 1999)
(Zhao etal., 1999)
(Zhao etal., 1999)
(Fostetal., 1989)
(Walker etal., 1992b)
(Eideetal., 1999)
(Wuetal., 1999a)
(Zhao etal., 1999)
(van Sittert etal., 2000)
(Marsden etal., 2007)
(Tompkins etal., 2008)
(Marsden etal., 2009)
§•
§
I
o ^
i> 1
0
o
a
I
a,
8"
O 5
31
o ^
H t
W ^S'
o •
o
c
o
H
W
-------
^5 Table C-l. Levels of endogenous (background) N7-HEG adducts in unexposed human and experimental rodent
2 tissues (continued)
OJ
*Adduct levels are normalized using the formula: 1 pmol adducts/mg DNA =3.3 adducts/107 normal nucleotides.
^ GC/MS, gas chromatography mass spectrometry; HPLC, high performance liquid chromatography; 32P, 32P-postlabeling assay; TLC, thin-layer chromatography;
5' LC-MS, liquid chromatography mass spectrometry; ESI TMS, electrospray ionization tandem mass spectrometry; GC/EC/NCI-HRMS, gas
o' chromatography/electron capture/negative chemical ionization high-resolution mass spectrometry.
s
3
§
<-t.
>!'
s
I
o s
o
a
I
a,
8"
O 5
31
o ^
H t
W ^S'
o
c
o
H
W
-------
1 Using sensitive detection techniques and an approach designed to separately quantify
2 both endogenous N7-HEG adducts and "exogenous" N7-HEG adducts induced by EtO treatment
3 in F344 rats, Marsden et al. (2009) recently reported increases in exogenous adducts in DNA of
4 spleen and liver consistent with a linear dose-response relationship (p < 0.05), down to the
5 lowest dose administered (0.0001 mg/kg injected i.p. daily for 3 days). Note that the whole
6 range of doses studied by Marsden et al. (2009) lies well below the dose corresponding to the
7 lowest LOAEL from an EtO cancer bioassay. For example, an approximate calculation indicates
8 that the low exposure level of 10 ppm for 6 hours/day used in the Snellings et al. (1984) bioassay
9 of F344 rats is equivalent to a daily dose of about 1.7 mg/kg, which is over 10 times higher than
10 the largest daily dose of 0.1 mg/kg used by Marsden et al. (2009).6
11 In summary, endogenous ethylene and EtO production, which contribute to background
12 N7-HEG DNA adducts indicative of DNA damage, have been observed in unexposed rodents
13 and humans. Although a constant reduction in DNA damage in vivo is carried out by DNA
14 repair and DNA replicative synthesis, a certain steady-state background level of adducts is
15 measurable at all times. The quantitative relationships between the background DNA damage
16 and the spontaneous rates of mutation and cancer are not well established. Experimental
17 evidence is needed that can unequivocally measure artifact-free levels of background DNA
18 damage, including effects other than adducts, clearly establish mutagenic potency of such
19 background lesions, and demonstrate the organ- and cell type-specific requirements for the
20 primary DNA damage to be expressed as heritable genetic changes (Gupta and Lutz, 1999).
21 Some investigators have posited that the high and variable background levels of
22 endogenous EtO-induced DNA damage in the body may overwhelm any contribution from
23 exogenous EtO exposure (Marsden et al., 2009; SAB, 2007). It is true that the existence of these
24 high and variable background levels may make it hard to observe statistically significant
25 increases in risk from low levels of exogenous exposure. However, there is clear evidence of
26 carcinogenic hazard from the rodent bioassays and strong evidence from human studies (see
27 Chapter 3, Section 3.5), and the genotoxicity/mutagenicity of EtO (Section 3.4) supports low-
28 dose linear extrapolation of risk estimates from those studies (U.S. EPA, 2005a). In fact, as
29 discussed above, Marsden et al. (2009) reported increases in exogenous adducts in DNA of
30 spleen and liver consistent with a linear dose-response relationship (p < 0.05), down to the
6This calculation uses the mean alveolar ventilation rate for rats of 52.9 niL/minute/100 g reported by Brown et al.
(1998). Changing the units, this rate is equivalent to approximately 0.032 m3/hour/kg. For a 6-hour exposure, this
results in an alveolar inhalation of 0.19 m3/kg. 10 ppm EtO is equivalent to 18.3 mg/m3, so a 6-hour exposure
equates to about 3.48 mg/kg. IARC (2008) reports that measurements from Johanson and Filser (1992) indicate that
only 50% of alveolar ventilation is available to be absorbed into the bloodstream, so the 6-hour exposure to 10 ppm
EtO would approximate an absorbed daily dose of 1.7 mg/kg.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-32 DRAFT—DO NOT CITE OR QUOTE
-------
1 lowest dose administered (0.0001 mg/kg injected i.p. daily for 3 days, which is a very low dose
2 compared to the LOAELs in the carcinogenicity bioassays). Furthermore, while the
3 contributions to cancer risk from low exogenous EtO exposures may be relatively small
4 compared to those from endogenous EtO exposure, low levels of exogenous EtO may
5 nonetheless be responsible for levels of risk (above background risk) that exceed de minimis risk
6 (e.g., >10 6). This is not inconsistent with the much higher levels of background cancer risk, to
7 which endogenous EtO may contribute, for the two cancer types observed in the human
8 studies—lymphoid cancers have a background lifetime incidence risk on the order of 3%,
9 whereas the background lifetime incidence risk for breast cancer is on the order of 15%.
10
11 C.8. CONCLUSIONS
12 The overall available data from in vitro studies, laboratory animal studies, and human
13 studies indicate that EtO is both a mutagen and a genotoxicant. In addition, increases in
14 mutations in specific oncogenes and tumor suppressor genes in EtO-induced mouse tumors have
15 been reported. Stable translocations seen in human leukemias may arise from similar DNA
16 adducts that produce chromosome breaks, micronuclei, SCEs, and even gene mutations observed
17 in peripheral lymphocytes. Dominant lethal mutations, heritable translocations, chromosomal
18 aberrations, DNA damage, and adduct formation in rodent sperm cells have been observed in a
19 number of studies involving the exposure of rats and mice to EtO. Based upon the likely role for
20 DNA alkylation in the production of the genotoxic effects in germ cells in laboratory animals
21 exposed to EtO, as well as the lack of qualitative differences in the metabolism of EtO between
22 humans and laboratory animals, EtO can also be considered a likely human germ cell mutagen
23 (WHO, 2003). There is consistent evidence that EtO interacts with the genome of cells within
24 the circulatory system in occupationally exposed humans and overwhelming evidence of
25 carcinogenicity and genotoxicity in laboratory animals. Based on these considerations, there is a
26 strong weight of evidence suggesting that EtO would be carcinogenic to humans (see Chapter 3,
27 Section 3.4).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 C-33 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX D.
2 REANALYSES AND INTERPRETATION OF ETHYLENE OXIDE
3 EXPOSURE-RESPONSE DATA
4 Kyle Steenland
5 May 27, 2010
6
7 (EDITORIAL NOTE: This Appendix contains the report submitted by Dr. Steenland
8 summarizing the results of analyses that he conducted under contract to U.S. EPA. The
9 terminology originally used by Dr. Steenland to designate the different exposure-response
10 model forms has been changed to be consistent with the terminology used in EPA's Ethylene
11 Oxide Carcinogenicity Assessment. Models that are linear in log RR and which were
12 previously referred to as "linear" models have been renamed "log-linear" models (except
13 where it is stated that they are log RR models), and models of the form RR = 1 + p x
14 exposure, which were previously referred to as "excess relative risk" (ERR) models have
15 been renamed "linear" models. In addition, section headings, figures, and tables have been
16 renumbered for the table of contents. Finally, some supplemental results received from Dr.
17 Steenland after the original completion of this Appendix have been inserted in the relevant
18 sections.)
19
20 This report contains the results of reanalyses of the National Institute for Occupational Safety
21 and Health cohort of workers exposed to ethyl ene oxide conducted for the U.S.
22 Environmental Protection Agency. The report begins with an overview of the modeling
23 strategy used, followed by the results of reanalyses of the breast cancer incidence, breast
24 cancer mortality, lymphoid cancer mortality, and, finally, hematopoietic cancer mortality
25 databases. Various models were used for these reanalyses, as discussed in this report. The
26 report concludes with the results of some sensitivity analyses and discussions of the possible
27 influences of the healthy worker survivor effect and exposure mismeasurement.
28
29 Introduction. Modeling strategy for ethylene oxide (ETO) risk assessment
30
31 The modeling strategy adopted here for ETO risk assessment relies principally on the usual
32 epidemiologic models in which the log of the rate ratio (RR) is some function of exposure, in
33 this case cumulative exposure with a lag to reflect a length of time which is likely necessary
34 before an exposure can result in (observable or fatal) cancer. We have relied primarily on
35 Cox regression as a flexible method of modeling the log RR; however we have also included
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-1 DRAFT—DO NOT CITE OR QUOTE
-------
1 some linear relative risk models. Cumulative exposure is typically the exposure metric of
2 interest in predicting chronic disease.
O
4 For breast cancer incidence, we have relied principally on 2-piece linear models, in which log
5 RR (in the log-linear model) or RR (in the linear model) is a function of two lines which join
6 smoothly at a single point of inflection. Two-piece linear models may also be thought of as
7 linear splines with one knot, or point of inflection. They have been described as part of a
8 general description of exposure-response modeling by Steenland and Deddens (2004) and
9 have been used previously in risk assessment (e.g., see the risk assessment for dioxin by
10 Steenland et al., 2001). The 2-piece log-linear model has the form log RR = PO + PI x
11 cumexp + P2 x (max(0,cumexp-knot)), where cumexp is cumulative exposure, the last term
12 equals either 0 or cumexp-knot, whichever is greater, and the knot is the point of inflection or
13 point of change of slope for the 2 linear pieces. The slope of the last term is Pi + fa for
14 cumulative exposure values above the knot.
15
16 Log RR models are not linear when the log RR function is transformed via exponentiation
17 back to a nonlogarithmic function, but they are nearly so in the low dose region of interest.
18 The splines are linear using the linear RR model.
19
20 "Plateau-like" exposure-response curves, in which the exposure-response curve begins
21 steeply but is attenuated at higher exposure, have been seen for many occupational
22 carcinogens. This may occur for a variety of reasons, including depletion of susceptible
23 subpopulations, mismeasurement at high exposure resulting in attenuation, and the healthy
24 worker survivor effect (Stayner et al., 1993). Attenuation of the exposure-response
25 relationship occurs for the breast cancer and (lympho) hematopoietic endpoints of interest for
26 ETO. For these endpoints, a simple linear model (often considered the default model), where
27 the log RR (for the log-linear model) or the RR increases linearly with cumulative exposure,
28 does not fit the data well, based on simple visual inspection of the categorical data.
29
30 Frequently, such plateau-like curves may be modeled by using the log of cumulative
31 exposure rather than cumulative exposure itself, but this has the disadvantage that the curve
32 is usually highly supra-linear at low doses. Two-piece linear spline models are particularly
33 useful in modeling exposure-response relationships in which the log RR or RR increases
34 initially with increasing exposure but then tends to increase less or plateau at high exposures.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 The 2-piece linear models avoid this supra-linearity in the low-dose region (Steenland and
2 Deddens, 2004).
O
4 The shape of the 2-piece linear spline model, in particular the slope of the curve in the low-
5 dose region, depends on the choice of the point of inflection where the two linear pieces are
6 joined. Here we have chosen the point of inflection based on the best model likelihood,
7 trying a range of points of inflection (knots) across the range of exposure starting from 0 and
8 incrementing by 100 ppm-days (or 1000 ppm-days) intervals. The model likelihood often
9 does not change much across these different points of inflection, but it does change some and
10 we have chosen the point of inflection resulting in the best model likelihood. The model
11 likelihood used to find the best fit in all models used in this analysis is the usual partial
12 likelihood (Langholz and Richardson, 2010), as used with the Cox models, which maximizes
13 the probability, across all the cases, that a case fails (the numerator) relative to its case-
14 control risk set (which includes the case) (the denominator) and has the form
15
16 L( P) = (pcase (Z;P)/ Ej cases and controls
-------
1 D.I. BREAST CANCER INCIDENCE BASED ON THE DATA WITH INTERVIEWS
2 a. Distribution of exposure among ETO-exposed women in breast cancer incidence
3 cohort with interviews (« = 5139)
4
5 The estimated daily exposure to ETO across different jobs and time periods ranged from
6 0.05 ppm to 77 ppm. Exposure intensities from this broad range were multiplied by the
7 length of time in different jobs to get estimates of cumulative exposure. The duration of
8 exposure had a mean of 10.8 years (std dev 9.1), and a median of 7.4 years. The range was
9 from 1.00 to 50.3 years. The 25* percentile was 2.8 years and the 75* percentile was
10 17.6 years. Multiplying exposure intensity and exposure duration results in a wide range of
11 cumulative exposures.
12
13 Cumulative exposure at the end of follow-up, with no lag, had a mean of 13,524 ppm-days
14 (37.0 ppm-years), with a standard deviation of 13,254 ppm-days. These data are highly
15 skewed, with a range from 5 to 253,848 ppm-days. The 25* percentile is 926 ppm-days,
16 while the 75th is 10,206 ppm-days. Log transformation of these data results in an
17 approximately normal distribution of the data.
18
19 As a caveat, it should be remembered that cumulative exposure at the end of follow-up may
20 be misleading, as it is not relevant to standard analyses, all of which treat cumulative
21 exposure as a time-dependent variable which must be assessed at specific points in time. For
22 example, standard life-table analyses calculate cumulative exposure at different times during
23 follow-up for each person. Subsequently, both person-time and disease events are put into
24 categories of cumulative exposure. A given person may pass through many such categories,
25 contributing person-time to each. Poisson regression, analogous to life-table analyses (and
26 often based directly on output from life table programs), similarly relies on person-time (and
27 disease occurrence) categorized by cumulative exposure. Both these types of analyses are
28 inherently categorical.
29
30 In the analyses presented here, we have used Cox regression in which age is the time
31 variable. The basic approach is to compare each case to a set of 100 randomly chosen
32 controls, whose exposure is evaluated at the same age at which the case fails (gets disease or
33 dies of disease). Using 100 controls generally would be expected to give the same result as
34 the full risk set and shortens analysis time (Steenland and Deddens, 1997). Hence, again
35 cumulative exposure is time dependent. For the case who fails at an early age, the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 cumulative exposure of the case and many of his or her controls at that same age may be low.
2 For the case who fails late in life, the cumulative exposure of the case and his or her controls
3 will be higher. When cumulative exposure is lagged so that no exposure is counted until
4 after a lag period (e.g., 15 years) is fulfilled, many cases and their respective controls will be
5 "lagged out" (i.e., will have no cumulative exposure, if the case fails at an early age). Note
6 Note that Cox regression uses individual data, and there is no inherent categorization typical
7 of life table analyses and Poisson regression, although categorical analyses can still be done
8 in Cox regression and are often useful.
9
10 For these reasons, it is difficult to describe the cumulative exposure distribution of all
11 subjects in the Cox regression. Controls may appear more than once matched to different
12 cases, and their cumulative exposure will differ each time depending on the age of the case.
13 However, cases only appear once in the data and their exposure distribution can be easily
14 presented. In our situation, we have used Cox regression with a 15-year lag to analyze breast
15 cancer incidence. The exposure distribution of the cases, by deciles above the lagged out
16 category, is shown below. Creating deciles such that cases are equally distributed is a good a
17 priori way of creating categories in which rate ratios will have approximate equal variance, a
18 desirable feature. The lagged out cases are women who got incident breast cancer within
19 15 years of first exposure.
20
21
22
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-la. Distribution of cases in Cox regression for breast cancer
morbidity analysis after using a 15-year lag
Cumulative exposure,
15-year lag
0 (Lagged out)
>0-355 ppm-days
356-842 ppm-days
843-1361 ppm-days
1362-2187 ppm-days
2188-3772 ppm-days
3773-5522 ppm-days
5523-7891 ppm-days
7892-14483 ppm-days
14484-25 112 ppm-days
>25 1 12 ppm-days
Mean cumulative exposure
(ppm-days)
157
580
1097
1725
2899
4546
6554
14384
18859
48807
Number of incident breast
cancer cases
62
17
16
17
17
17
18
16
17
17
18
4
5
6
7
b.l. Results of Cox regression analysis of breast cancer incidence using a variety of (log
RR) models
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Analyses used a case-control approach, with 100 controls per case, as in Steenland et al.
(2003). Age was the time variable in proportional hazards (Cox) regression. For breast
cancer incidence, family history of breast cancer, date of birth (quartiles), and parity were
included in models along with exposure variables. For our exposure variable, we used
cumulative exposure lagged 15 years, which was found in prior analyses to provide the best
fit to the data (Steenland et al., 2003).
Using log RR models, we used a categorical model, a linear model, a 2-piece linear model, a
log transform model, a cubic spline model, and a square-root transform model. We also ran a
number of analogous models using linear RR models.
The categorical analysis (log RR model) used deciles, as indicated in Table D-lb. Deciles
were used instead of the original quintiles from the publication (Steenland et al., 2003)
because the relatively large sample size enabled more extensive categorization. Results of
the categorical decile analysis are in Table D-lb below.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-6 DRAFT—DO NOT CITE OR QUOTE
-------
Table D-lb. Categorical analysis of breast cancer incidence by deciles (log
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
RR model)
Analysis of Maximum Likelihood Estimates
Parameter
Variable
CAT1
CAT 2
CATS
CAT 4
CATS
CAT 6
CAT 7
CATS
CAT 9
CAT 10
-2 LOG
We then fit a
Standard
Estimate
-0
-0
0
0
0
0
0
0
0
0
L
.09015
.08363
.18536
.12606
.07900
.37651
.38177
.25179
.57845
.80396
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
1936
cubic spline (restricted
Error
29318
30341
29757
29995
29968
29675
31168
30640
31120
30766
.910,
at the
Chi-Square Pr
0.
0.
0.
0.
0.
1.
1.
0.
3.
6.
df=15 (10
ends to be li
0945
0760
3880
1766
0695
6097
5003
6753
4551
8284
exposure
inear) which
>
0
0
0
0
0
0
0
0
0
0
ChiSq
.7585
.7828
.5333
.6743
.7921
.2045
.2206
.4112
.0631
.0090
terms, 5
prei
sents a
Hazard
Ratio
0
0
1
1
1
1
1
1
1
2
.914
.920
.204
.134
.082
.457
.465
.286
.783
.234
covariates )
description of
24
25
26
27
28
29
30
31
32
33
34
35
36
37
the data similar to the categorical analyses but using a smooth curve. The exposure metric
was cumulative exposure with a 15-year lag, which was found in earlier analyses to be the
optimal lag (Steenland et al., 2003). Five knots for the cubic spline were chosen using every
other midpoint from the categorical analysis (598, 1774, 4647, 11187, and 37668 ppm-days).
We then ran a 2-piece linear (log RR) model. The knot, or inflection point, was chosen to be
the one where the model likelihood was highest, which was at 5,800 ppm-days. To choose
this knot we looked at possible inflection points over the range 100 to 15,000 ppm-days by
100 ppm-day increments. Figure D-la shows the -2 log likelihood graphed against the
knots. In this figure the lower peak corresponds to the highest likelihood.7
Figures D-lb and D-lc show the results of the 2-piece linear, the categorical, the linear, and
the cubic spline (log RR) models. In these figures the categorical points are the mid-points
of the categories in Table D-la, with final category assigned the final cut point plus 50%.
Editorial note: -2 x (natural) log likelihood is reported because the difference in this value for any two models is the
value of the test statistic commonly used to compare model fit (likelihood ratio test). Under certain assumptions, the
probability distribution for this statistic is approximately chi-squared with degrees of freedom equal to the difference
in degrees of freedom between the two (nested) models.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 It appears that the two-piece log-linear curve in Figure D-lb approximates the shape of the
2 exposure-response seen in the decile and cubic spline (log RR) analyses, better than the log-
3 linear curve in Figure D-1 c.
4
5 The log-linear curve appears to have a low slope versus the other models, suggesting possible
6 influential observations in the upper tail of exposure. To further explore this, we excluded
7 from the analysis increasing amounts of the upper tail of the data using the log-linear model,
8 i.e., via excluding the upper 1%, 2.5%, 5%, 10%, 15%, 20%, and 27% of exposure, based
9 on the exposure distribution of the cases (the last amount, 27%, corresponds to excluding
10 subjects with cumulative exposure above 6000 ppm-days, which was close to the knot in the
11 2-piece log-linear model [5800 ppm-days]). The ratios of the slope (coefficient) for the
12 linear term (log RR model) with these exclusions vs. the slope for the linear term (log RR
13 model) with no exclusions were 1.5, 2.3, 3.2, 3.2, 2.5, 3.1, 6.1, 9.2, respectively. As
14 expected, the slope increases markedly as the data are restricted to the lower range of
15 exposure. For example, a modified log-linear curve after excluding the upper 5% of the data
16 is seen in Figure D-ld, along with the full log-linear curve from Figure D-lc. Nonetheless,
17 even the log-linear curve from these truncated data has a markedly lower slope in the low-
18 exposure region than the 2-piece log-linear (or spline) curves. For example, inspection
19 shows that the RR for 6000 ppm-days is about 1.2 for the log-linear curve from the truncated
20 data and 1.6 from the 2-piece log-linear model. Use of the log-linear curve based on
21 truncated data has the disadvantage of having to choose rather arbitrarily where to truncate
22 the data. This disadvantage is avoided by using the 2-piece log-linear model.
23
24 A 2-piece log-linear model, then, is preferred for estimating risk parsimoniously in the low-
25 exposure region. For comparison purposes, we also show the model using the logarithm of
26 exposure (Figure D-le), which we have not used for risk assessment because it is supralinear
27 in the low-dose region.
28
29 We also fit a square-root transformation (square root of cumulative exposure, 15-year lag)
30 log RR model, which is shown in Figure D-lf. This model also fit the breast cancer
31 morbidity well (it did not fit the other outcomes well and is not shown for them), and can be
32 used for risk assessment, but with the disadvantage that it is not linear or approximately
33 linear in the low-dose region. For this reason, we prefer the 2-piece log-linear curve, with is
34 approximately linear in the low-dose region (and strictly linear in the linear RR models
35 discussed below). Excess lifetime risk does not vary greatly between all these models (see
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
below), with the exception of the log RR model with a linear term for cumulative exposure,
which is below other excess risk estimates.
5
6
7
0
5
1
1
Log RR Model
-\
V
\
V
^
\
^•^ _^_-
DOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
DOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
^H^H^H^Hr^r^rNjrnrnm^^^^Lni^LniXJiXJiXJr^r^r^-r^-DOooDOCTic^c^o
Knot Location, CUMEXP15
Figure D-la. Likelihoods vs knots, 2-piece linear log RR model for breast
cancer morbidity.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-9 DRAFT—DO NOT CITE OR QUOTE
-------
HR,
1
2
3
4
5
6
7
4QDOO
Figure D-lb. Breast cancer incidence. Plot of the dose-response relationship
for continuous exposure generated using a 2-piece log-linear spline overlaid with
a plot using restricted cubic (log RR) splines. Dots that represent the effect of
exposure grouped in deciles (log RR categorical model) are also presented in the
plot. Deciles formed by allocating cases approximately equally in ten groups,
above lagged-out cases, see Table D-la above. Y-axis is rate ratio, X-axis is
cumulative exposure lagged 15 years, in ppm-days.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-10 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
430DO
Figure D-lc. Breast cancer incidence. Plot of a log-linear dose-response
relationship overlaid with a dose-response relationship generated using restricted
cubic log RR model with continuous exposure. Dots that represent the effect of
exposure grouped in deciles (log RR categorical model) are also presented in the
plot. Deciles formed by allocating cases approximately equally in ten groups,
above lagged-out cases.
Comparing log linear models, model with higher slope omits highest 5% of exposure
10
11
12
13
5000
10000 15000
CUMEXP15
20000
25000
Figure D-ld. Breast cancer incidence. Comparison of log-linear curve (log
RR = P x cumexp) with all the data and the log-linear curve (higher slope) after
excluding those in the top 5% of exposure (>27,500 ppm-days).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-11 DRAFT—DO NOT CITE OR QUOTE
-------
Breast cancer morbidity log transformed
-21 qg I i te i hood i s 1914153
O
3000
nmnqprr
10QQD
1300D
2GGQO
2SOOD 3QQQD 3GOOD 4000D
1
2
3
4
5
6
7
Figure D-le. Breast cancer incidence. Plot of a logarithmic transformation log
RR dose-response model (log RR = P x log(cumexp)) overlaid with a dose-
response relationship generated using categorical log RR analyses (deciles).
Deciles formed by allocating cases approximately equally in ten groups, above
lagged-out cases.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-12 DRAFT—DO NOT CITE OR QUOTE
-------
Breast cancer morbidity sqrt root transformed
i
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
-21 qg I i kEH hood i s 1911. COB
and\sas cxarl ^ed
3DDO
3BDQO
4000D
Figure D-lf. Breast cancer incidence. Plot of a square-root transformation log
RR dose-response model overlaid with a dose-response relationship generated
using categorical log RR analyses (deciles). Deciles formed by allocating cases
approximately equally in ten groups, above lagged-out cases.
Tables D-lc, D-ld, D-le, and D-lf below present the model fit statistics for the 2-piece log-
linear, the log-linear, the square root log RR model, and the log transform log RR model seen
above. Table D-lg summarizes the goodness-of-fit data with regard to the exposure term.
Table D-lg shows that the addition of exposure terms to the various models results in similar
model fits. The exposure terms in the 2-piece log-linear improve model fit marginally better
than those in the other models except the square root log RR model, with which the 2-piece
log-linear model is tied. If one adds a degree of freedom to the chi-square test for the 2-piece
log-linear model, on the assumption that the choice of the knot is equivalent to estimating
another parameter, the/?-value increases to 0.04, in the same range as the log-linear and log-
transform log RR models. Our argument here, however, is not that the 2-piece log-linear
model fits the data dramatically better than other models in purely statistical terms. Rather
we believe that the fit conforms to the categorical and cubic spline models well in the low-
exposure region of interest, and that the nearly linear exposure-response relationship in that
region (strictly linear with the linear RR model) is a reason to prefer the 2-piece log-linear
model to the other models. In particular, among the parametric models, the log transform
and square root log RR models are supralinear in the low-exposure region.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-13 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
10
11
12
13
The effects of these departures from linearity in the low-exposure region can be seen in the
risk assessment results for the ECoi (estimate of effective concentration resulting in 1% extra
risk) in the next sections (c, d, and e). In these sections we use some of the results from the
exposure-response models to calculate ECois. We restrict these calculations to models which
appear most reasonable based on our results above, namely the 2-piece log-linear model, the
square root transform log RR model, and the cubic spline log RR model. While we do not
recommend the use of the cubic spline model for risk assessment due to its complexity, the
ECoi based on the cubic spline model provides a good comparison to other parametric
models.
Table D-lc. Fit of 2-piece log-linear model to breast cancer incidence data,
Cox regression8
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
Criterion
-2 LOG L
AIC
SBC
Testing
Test
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1940.485
1954.485
1978.612
Global Null Hypothesis: BETA=0
Chi-Square
Likelihood Ratio 27.3281
Score
Wald
Analysis
Parameter
Variable Estimate
LIN 0 (PI) 0.0000770
LIN 1 -0.0000724
DOB1 0.08770
DOB2 0.41958
DOBS 0.55168
PARITY1 -0.23398
FREL BR CAN1 0.47341
Covariance linO and linl -
29.0949
28.4426
DF Pr > ChiSq
7 0.0003
7 0.0001
7 0.0002
of Maximum Likelihood Estimates
Standard
Error Chi-Square Pr > ChiSq
0.0000317
0.0000334
0.21805
0.24430
0.29096
0.18793
0.17934
1 x 1CT9
5.4642 0.0194
4.1818 0.0409
0.1618 0.6875
2.9496 0.0859
3.5950 0.0580
1.5502 0.2131
6.9686 0.0083
Hazard
Ratio
1.000
1.000
1.092
1.521
1.736
0.791
1.605
8For environmental exposures, only exposures below the knot are of interest. Below the knot, RR = e(|31 * exP°sure).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-14 DRAFT—DO NOT CITE OR QUOTE
-------
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1944.675
1956.675
1977.356
1 Table D-ld. Fit of log-linear model to breast cancer incidence data, Cox
2 regression (RR = e(p x exposure))
3
4
5
6
7
8
9
10
11
12 Testing Global Null Hypothesis: BETA=0
13
14 Test Chi-Square DF Pr > ChiSq
15
16 Likelihood Ratio 23.1374 6 0.0008
17 Score 25.8389 6 0.0002
18 Wald 25.3594 6 0.0003
19
20
21 Analysis of Maximum Likelihood Estimates
22
23
24
25
26
27
28
29
30
31 FREL BR CAN1 0.46449 0.17928 6.7126 0.0096 1.591
32
33
Parameter
Variable
CUMEXP15 (P)
DOB1
DOB2
DOB3
PARITY1
Standard
Estimate
9.54826E-6
0,
0,
0,
-0,
.13558
.53147
.74477
.23394
Error
4.09902E-6
0
0
0
0
.21676
.23741
.27425
.18882
Chi-Square
5
0
5
7
1
.4261
.3912
.0116
.3748
.5351
Pr
0
0
0
0
0
> ChiSq
.0198
.5316
.0252
.0066
.2154
Hazard
1.
1.
1.
2.
0.
Ratio
000
145
701
106
791
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-15 DRAFT—DO NOT CITE OR QUOTE
-------
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Table D-le. Fit of the square root transformation log RR model to breast
cancer incidence data, Cox regression (RR = e(p x sirt(exP°sure)))
Model Fit Statistics
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1941.028
1953.028
1973.708
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
26.7851
28.9446
28.5277
0.0002
<.0001
<.0001
Analysis of Maximum Likelihood Estimates
Variable
dobl
dob2
dob3
sqrtcumexplS
PARITY1
FREL BR CAN1
DF
1
1
1
(P) 1
1
1
Parameter
Estimate
0.09778
0.43872
0.58623
0.00349
-0.22539
0.46937
Standard
0.
0.
0.
0.
0.
0.
Error
21756
24177
28404
00118
18787
17922
Chi-Square
0.2020
3.2929
4.2596
8.7489
1.4393
6.8589
Pr > ChiSq
0. 6531
0.0696
0.0390
0.0031
0.2302
0.0088
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-16 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table D-lf. Fit of the log transform model to breast cancer incidence data,
2 Cox regression (RR = e(p x ln(exPosure»)
3
4 Model Fit Statistics
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26 Parameter
27
28 dobi
29 dob2
30 dob3
31 LCUMEXP15 (P)
32 PARITY1
33 FREL BR CAN1
Without
Criterion Covariates
-2 LOG L
AIC
SBC
Testing
Test
Likelihood Ratio
Score
Wald
Analysis
Parameter
DF Estimate
1 0.08605
1 0.38780
1 0.47303
1 0.04949
1 -0.25908
1 0.47620
1967.813
1967.813
1967.813
With
Covariates
1944.176
1956.176
1976.856
Global Null Hypothesis: BETA=0
Chi-Square
23.6371
24.0044
23.5651
DF Pr >
6
6
6
ChiSq
0.0006
0.0005
0.0006
of Maximum Likelihood Estimates
Standard
Error Chi-Square Pr >
0.21943
0.25363
0.31528
0.02288
0.18638
0.17923
0.1538 0
2.3378 0
2.2509 0
4.6787 0
1.9322 0
7.0595 0
ChiSq
.6949
.1263
.1335
.0305
.1645
.0079
Hazard
Ratio
1.090
1.474
1. 605
1.051
0.772
1.610
34
35
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-17 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-lg. Change in -2 log likelihood for log RR models for breast cancer
incidence, with addition of exposure term(s)a
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Log RR model
Log transform
Linear
Categorical
Cubic spline
2-piece linear
Square root
Change (chi square)
4.8
4.2
12.0
8.8
8.4
7.7
d.f.
1
1
10
4
2
1
/7-value
0.03
0.04
0.29
0.07
0.01
0.006
aAll models had 3 variables for date of birth, 1 for family history, and 1 for parity.
b.2. Linear relative risk models for breast cancer incidence
We also ran linear relative risk models for breast cancer incidence, using the techniques
described recently by Langholz and Richardson (2010) to use SAS to fit these models, using
the same data as used for the log RR models. The form of these linear RR models is
RR = 1 + PX, where x can be cumulative dose, the log of cumulative dose, a 2-piece linear
function of cumulative dose, etc.
Figure D-lg below shows the different curves with the linear RR model, using cumulative
exposure lagged 15 years as the exposure metric. The categorical points in Figure D-lg
come from the published categorical results for the log RR model (Steenland et al., 2003).
The midpoints for the 5 categories (above the lagged out referent, at 0 exposure) are the
medians of cumulative exposure (lagged 15 years), which were 253, 1193, 3241, 7741, and
26,597 ppm-days.
Figure D-lh shows the likelihood profile for different possible knots for the 2-piece linear
spline, with the search conducted by using increments of 100 ppm-days. For the 2-piece
linear spline model the best knot was 5800 ppm-days, as was the case for the 2-piece log-
linear model.
Table D-lh shows the model fit statistics for the linear RR models. These models tend to fit
slightly better than their log RR counterparts, although generally the improvement in the chi
square does not attain significance at the 0.05 level. For the 2-piece linear model, the model
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-18 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
likelihood is 1936.9 vs a likelihood of 1940.5 for the 2-piece log-linear model. Among the linear
RR models, the 2-piece spline model fits better than the other models, although not significantly
so. Table D-li gives the exposure parameter values for the linear RR models. Table D-lj
presents the parameter estimates for the exposure variables for the categorical (decile) linear RR
model.
2.25
1.Z5
o./s
• Categorical
Spline ERR, Knot=5SOO.CUMEXP15
IRK,
ERR, LoglCUMtXPlS)
'..IMHJ
1U.UUU
20.000
2j,OUU
30.000
CUMEXP15
9
10
11
Figure D-lg. Breast cancer incidence exposure-response curves, linear RR
models (units are ppm-days, 15-year lag).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-19 DRAFT—DO NOT CITE OR QUOTE
-------
1
1102
1101.5
1101 -
T3
O
0
•= 1100.5
01
g> 1100
1099.5
1099 -
1098.5
' V
^ —
^v —
5,800
i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i
3OOOOOOOOOOOOOOOOOOOOOOOOOO
3OOOOOOOOOOOOOOOOOOOOOOOOOO
Knot Location
2 Figure D-lh. Knot location for Figure D-lg above, 2-piece linear spline
3 model, breast cancer incidence (units are ppm-days, 15-year lag).
4
5
6 Table D-lh. Model fit statistics for linear RR models, breast cancer
7 incidence"
Linear RR Model
CUMEXP15
Log(CUMEXP15)
Spline, knot = 5,800,
CUMEXP15
d.f. (full
model)b
6
6
7
-2 Log
likelihood
(full model)
1940.260
1942.267
1936.935
-2 LL (model
without
exposure)
1949.06
1949.06
1949.06
-2LL
(model
without any
covariates)
1967.813
1967.813
1967.813
/7-value
(full model)
< 0.0001
0.0003
< 0.0001
/7-value
(for addition
of exposure
terms)0
0.0030
0.0096
0.0023
9
10
11
12
13
14
15
16
17
18
19
Tor the linear RR models, all covariates were included linearly (i.e., additively). Including the nonexposure
covariates in the model multiplicatively instead did not improve model fit (e.g., for the 2-piece spline model,
inclusion of the non-exposure covariates multiplicatively instead of additively gave a -2 LL of 1940.4 (vs. 1936.9
for additive inclusion).
bDegrees of freedom for full model.
°Based on change in likelihood for breast cancer incidence linear RR models with addition of exposure term(s) to
model with date of birth, parity, and breast cancer in first degree relative. Degrees of freedom for addition of
exposure terms is (degrees of freedom for the full model - 5).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-20 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
Table D-li. Model coefficients for linear RR models, breast cancer incidence
Linear RR Model
CUMEXP15
Log(CUMEXP15)
Spline, knot = 5,800,
CUMEXPIS3-13
Parameter(s)
B = 0.000030402
B = 0.071322
Bl= 0.000119,
B2 = -0.000105
SE
SE = 0.000017549
SE = 0.039227
SE1 = 0.000067727,
SE2 = 0.000070478
Profile likelihood 95% (one-
sided) upper and lower bounds0
UB = 0.0000745
LB = 0.00000975
UB 1=0.000309
LB 1 = 0.000032
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
aCovariance of 2 pieces of linear spline, -4.64 x 10 9.
bFor estimating risks from occupational exposures (Section 4.7 of the Carcinogenicity Assessment Document), both
pieces of the 2-piece linear spline model were used. For the maximum likelihood estimate, for exposures below the
knot, RR = 1 + (Bl x exp); for exposures above the knot, RR = 1 + (Bl x exp + B2 x (exp - knot)). For the 95%
upper confidence limit, for exposures below the knot, RR = 1 + (((31+ 1.645 x SE1) x exp); for exposures above the
knot, RR = 1 + ((31 x exp + (32 x (exp-knot) + 1.645 x sqrt(exp2 x Varl + (exp-knot)2 x Var2 + 2 x exp x (exp-knot)
x covar)), where exp = cumulative exposure, var = variance, covar = covariance.
0Editorial note: As discussed in footnotes iandj of Table 4-7 in Section 4.1.2.3 of this assessment, confidence
intervals were determined using the Wald approach. Confidence intervals for linear RR models, however, in
contrast to those for the log-linear RR models, may not be symmetrical. EPA also evaluated application of a
profile likelihood approach (Langholz and Richardson, 2010), which allows for asymmetric CIs, for selected linear
RR models, for comparison with the Wald approach. 95% (one-sided) upper and lower bounds on the parameter
estimate (regression coefficient) derived using the profile likelihood method are presented here. For the continuous
linear model (CUMEXP15), the profile likelihood upper bound is about 29% higher than the upper bound obtained
using the Wald approach. For the low-exposure segment of the linear spline model, the profile likelihood upper
bound is about 34% higher than the upper bound obtained using the Wald approach. Calculating the profile
likelihood bounds for the second spline segment parameter estimate is computationally difficult and was not
pursued here.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-21 DRAFT—DO NOT CITE OR QUOTE
-------
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
Table D-lj. Supplemental Results: Parameter estimates for exposure
variables for categorical (decile) linear RR model (RR = 1 + P), breast cancer
incidence
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
N
1
2
3
4
5
6
7
8
9
10
Parameter Estimate
betala
betalb
betalc
betald
betale
betalf
betalg
betalh
betali
betalj
0
0
0
0
-0
0
0
0
0
1
.039745
.133788
.304056
.166744
.071617
.749956
.919113
.487590
.985298
.575028
Value of Objective
0
0
0
0
0
0
0
0
0
0
Approx
Std Err
. 340310
. 371450
.438525
.402813
. 347348
.583672
.643333
.505453
.753167
. 960886
Function (log
0
0
0
0
-0
1
1
0
1
1
t Value
.116792
. 360177
.693361
.413950
.206182
.284893
.428674
. 964660
. 308206
.639141
likelihood) = -966.
Approx
Pr > |t|
0.907133
0.719065
0.488824
0.679319
0.836842
0.200200
0.154536
0.335789
0.192187
0.102633
9720784
6
1
3
7
9
7
5
1
1
Gradient
Objective
Function
2.4264E-10
.847537E-11
.119129E-10
.011577E-10
.617004E-11
.568004E-11
.337724E-11
.850301E-11
.207377E-10
.341514E-10
22
23
24 c. Risk assessment for breast cancer incidence using the 2-piece log-linear spline
25
We used the 95% upper bound of the coefficient for the 1st piece of the 2-piece log-linear
model from Table D-lc, which is 0.0000770 + 1.64 x 0.0000317 or 0.0001290, to calculate
the LECoi via the life-table analysis of excess risk used by EPA in Appendix C of their risk
assessment. Here we used the same data on background breast cancer incidence and
background all-cause mortality as used by EPA in their 2006 calculations. The rate ratio
then, as a function of exposure, is RR = e(0-0001290 x cumexP15). Note that the 2-piece log-linear
model is linear for the log RR. Once this is exponentiated, it is no longer strictly linear, but
is still approximately so, as can be seen in Figure D-la.
Use of the function RR = e(°-0001290 x cumexpi5) m ^e iife_table analysis results in an excess risk
of 0.01 when the daily exposure is 0.0090 ppm, which is the LECoi. This is slightly lower
than the previous LECoi of 0.01 10 ppm in EPA's 2006 draft risk assessment (U.S. EPA,
2006a, Table 14).
Similar calculations were done for the ECoi, which resulted in a value of 0.0152 ppm.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-22 DRAFT—DO NOT CITE OR QUOTE
-------
1 d. Risk assessment for breast cancer incidence using the square root transformation log
2 RR model
3
4 Use of the 95% upper bound of the relative risk function, ie, RR = e((0-000349 + -00118 x L64) x square
5 root(cumexpi5))^ -^ ^g life-table analysis results in an excess risk of 0.01 when the daily exposure
6 is 0.00225 ppm, which is the LECoi. This is about 5 times lower than the previous LECoi of
7 0.0110 ppm in EPA's 2006 draft risk assessment (U.S. EPA, 2006a, Table 14). The ECoi is
8 0.0060 ppm, which is about four times lower than the EPA's 2006 ECoi. The reason these
9 estimates are much lower than the EPA's is that the square root curve, as can be seen in
10 Figure D-ld, rises very sharply (is supralinear) in the low-dose region. In this sense, it shares
11 the disadvantage of the log transform model, and we recommend against using it as a basis
12 for risk assessment for that reason.
13
14 e. Risk assessment for breast cancer incidence using the cubic spline curve log RR
15 model
16
17 Risk assessment using the spline curve is more difficult due to the semi-parametric
18 complicated nature of the restricted cubic spline function. The cubic spline function for the
19 breast cancer incidence rate ratio is
20
21 RR=exp((ns_0*cumexp!5) + ns_l*(((cumexpl5-598)**3)*(cumexp!5>= 598) -
22 ((37668-598) /(37668-11187)) *(((cumexpl5-11187)**3) *(cumexp!5>= 11187)) +
23 ((11187 -598)7(37668 - 11187)) *(((cumexplS-37668 )**3) *(cumexp!5>= 37668))
24 ) + ns_2*(((cumexpl5-1774)**3)*(cumexp!5>= 1774) - ((37668-1774) 7(37668-
25 11187)) *(((cumexpl5-11187)**3) *(cumexpl5>= 11187)) + ((11187 -1774) 7(37668
26 - 11187))*(((cumexplS-37668 )**3) *(cumexp!5>= 37668)) ) + ns_3*(((cumexplS-
27 4647)**3)*(cumexp!5>= 4647) - ((37668-4647) 7(37668-11187)) *(((cumexplS-
28 11187)**3) *(cumexpl5>= 11187)) + ((11187 -4647)7(37668 - 11187))
29 *(((cumexplS-37668 )**3) *(cumexp!5>= 37668)) )).
30
31 The coefficients ns_0, ns_l, ns_2, and ns_3 used in this function are 0.00008294999811, -
32 0.00000000000310 0.00000000000425, and -0.00000000000114, respectively. The
33 expression "cumexpl5>=" is a logical statement whereby the term is 0 when "cumexp" is less
34 than the specified value.
35
36 Here we calculate only the ECoi (without the LECoi) for comparison with the corresponding
37 ECoi from the 2-piece log-linear model. The point is to show that the cubic spline log RR model
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-23 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
and the 2-piece log-linear spline give similar answers, not to use the cubic spline for risk
assessment, given its relatively complicated formula above. Calculation of the LECoi is also
particularly complicated because to do it correctly one must use not only the standard error for
four coefficients but also their covariances.
For breast cancer incidence, the ECoi using the cubic spline log RR model is 0.0138 ppm, similar
to the value of 0.0152 ppm using the 2-piece log-linear model.
9 f. Risk assessment for breast cancer incidence using the 2-piece linear spline model
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
Use of the function RR = 1 + (0.0001 19 + 1.64 x 0.000067) x cumexplS in the life-table
analysis results in
an excess risk of 0.01 when the daily exposure is 0.0052 ppm, which is
the
LECoi, which is about half of the value of 0.01 10 ppm from the 2-piece log-linear spline
model. The corresponding ECoi is 0.0100 ppm.
g. Supplemental
results: results for cumulative exposure and log cumulative exposure Cox
regression models with different lag times (no lag, 5 years, 10 years, 15 years, and 20
years)
(i) cumulative exposure model, no lag
Variable
dobl
dob2
dob3
CUMEXP
PARITY1
FREL_BR_CAN1
Without With
Criterion Covariates Covariates
- 2 LOG L 1967.813 1946.492
AIC 1967.813 1958.492
SBC 1967.813 1979.172
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 21.3211 6 0.0016
Score 22.2448 6 0.0011
Wald 22.0301 6 0.0012
Analysis of Maximum Likelihood Estimates
Parameter Standard Hazard
DF Estimate Error Chi-Square Pr > ChiSq Ratio
1 0.17056 0.21590 0.6241 0.4295 1.186
1 0.59054 0.23671 6.2242 0.0126 1.805
1 0.83494 0.27295 9.3573 0.0022 2.305
1 5.93879E-6 3.52892E-6 2.8321 0.0924 1.000
1 -0.25022 0.18784 1.7746 0.1828 0.779
1 0.47120 0.17920 6.9144 0.0086 1.602
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-24 DRAFT—DO NOT CITE OR QUOTE
-------
1
)cumulative exposure model, 5-year lag
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1945.875
1957.875
1978.555
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
21.9381
23.1833
22.9563
0.0012
0.0007
0.0008
Analysis of Maximum Likelihood Estimates
Variable
dobl
dob2
dob3
CUMEXP5
PARITY1
FREL BR CAN1
DF
1
1
1
1
1
1
Parameter
Estimate
0.16362
0.57250
0.80642
6.8565E-6
- 0.24489
0.47063
Standard
Error
0.21604
0.23698
0.27311
3.59626E-6
0.18810
0.17919
Chi-Square
0.5736
5.8363
8.7184
3.6350
1.6951
6.8981
Pr > ChiSq
0 . 4488
0.0157
0.0032
0.0566
0.1929
0.0086
Hazard
Ratio
1.178
1.773
2.240
1.000
0.783
1.601
(in) cumulative exposure model, 10-year lag
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1945.521
1957.521
1978.201
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Variable
dobl
dob2
dob3
CUMEXP10
PARITY1
FREL BR CAN1
Score
Wald
DF
1
1
1
1
1
1
Analysis of
Parameter
Estimate
0.15185
0.55144
0.77339
7.75726E-6
- 0.24110
0.46864
22.2922
23.9807
23.6876
6
6
6
0.0011
0.0005
0.0006
Maximum Likelihood Estimates
Standard
Error
0.21633
0.23733
0.27377
3.80799E-6
0.18839
0.17921
Chi-Square
0.4927
5.3986
7.9805
4.1498
1.6379
6.8385
Pr > ChiSq
0.4827
0.0202
0 . 0047
0.0416
0.2006
0.0089
Hazard
Ratio
1.164
1.736
2.167
1.000
0.786
1.598
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-25 DRAFT—DO NOT CITE OR QUOTE
-------
1
)cumulative exposure model, 15-year lag
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1944.675
1956.675
1977.356
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Variable
dobl
dob2
dob3
CUMEXP15
PARITY1
FREL BR CAN1
Score
Wald
DF
1
1
1
1
1
1
Analysis of
Parameter
Estimate
0.13558
0.53147
0.74477
9.54826E-6
- 0.23394
0 . 46449
23.1374
25.8389
25.3594
6
6
6
0.0008
0.0002
0.0003
Maximum Likelihood Estimates
Standard
Error
0.21676
0.23741
0.27425
4.09902E-6
0.18882
0.17928
Chi-Square
0.3912
5.0116
7.3748
5.4261
1.5351
6.7126
Pr > ChiSq
0.5316
0.0252
0.0066
0.0198
0.2154
0.0096
Hazard
Ratio
1.145
1.701
2.106
1.000
0.791
1.591
(v) cumulative exposure model, 20-year lag
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1946.040
1958.040
1978.720
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
Analysis of
Parameter
Variable
dobl
dob2
dob3
CUMEXP20
PARITY1
FREL_BR_CAN1
DF
1
1
1
1
1
1
21.7730 6 0.0013
24.0576 6 0.0005
23.5506 6 0.0006
Maximum Likelihood Estimates
Standard
Estimate
0.
0.
0.
,13721
,53985
,76037
0.0000101
- 0.
0.
,23887
,46310
0.
0.
0.
Error
,21682
,23711
,27371
5.27041E-6
0.
0,
,18905
.17935
Chi-Square
0.
5.
7.
3.
1.
6.
,4005
,1839
,7177
,6890
5966
,6673
Pr >
0
0
0
0
0
0
ChiSq
.5268
.0228
.0055
.0548
.2064
.0098
Hazard
Ratio
.147
.716
.139
.000
0.788
1.589
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-26 DRAFT—DO NOT CITE OR QUOTE
-------
1
(vi)log cumulative exposure model, no lag
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1943.662
1955.662
1976.343
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Variable
dobl
dob2
dob3
Icumexp
PARITY1
FREL BR CAN1
Score
Wald
DF
1
1
1
1
1
1
Analysis of
Parameter
Estimate
0.17618
0.59516
0.83783
0.09294
- 0.25682
0.47417
24.1508
24.4372
24.1563
6
6
6
0.0005
0.0004
0.0005
Maximum Likelihood Estimates
Standard
Error
0.21596
0.23703
0.27359
0.04097
0.18640
0.17923
Chi-Square
0.6655
6.3045
9.3780
5.1458
1.8984
6.9991
Pr > ChiSq
0 . 4146
0.0120
0.0022
0.0233
0.1683
0.0082
Hazard
Ratio
1.193
1.813
2.311
1.097
0.774
1.607
log cumulative exposure model, 5-year lag
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1946.843
1958.843
1979.523
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Variable
dobl
dob2
dob3
LCUMEXP5
PARITY1
FREL_BR_CAN1
Score
Wald
DF
1
1
1
1
1
1
Analysis of
Parameter
Estimate
0.15082
0.53156
0.72413
0.04458
- 0.26745
0.47497
20.9703
21.0320
20.7379
6
6
6
0.0019
0.0018
0.0020
Maximum Likelihood Estimates
Standard
Error
0.21658
0.24038
0.28191
0.03135
0.18630
0.17922
Chi-Square
0.4850
4.8900
6.5981
2.0222
2.0608
7.0241
Pr > ChiSq
0.4862
0.0270
0.0102
0.1550
0.1511
0.0080
Hazard
Ratio
1.163
1.702
2.063
1.046
0.765
1.608
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-27 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
O
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
(viii)
log cumulative exposure model, 10-year lag
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1944.040
1956.040
1976.721
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Variable
dobl
dob2
dob3
LCUMEXP10
PARITY1
FREL BR CAN1
Score
Wald
DF
1
1
1
1
1
1
Analysis of
Parameter
Estimate
0.11282
0.43207
0.53777
0.05654
- 0.26063
0.47636
23.7728
23.5846
23.1565
6
6
6
0.0006
0.0006
0.0007
Maximum Likelihood Estimates
Standard
Error
0.21798
0.24800
0.30203
0.02594
0.18629
0.17921
Chi-Square
0.2679
3.0352
3.1702
4.7498
1.9573
7.0653
Pr > ChiSq
0 . 6048
0.0815
0.0750
0.0293
0.1618
0.0079
Hazard
Ratio
1.119
1.540
1.712
1.058
0.771
1.610
(ix)log cumulative exposure model, 15-year lag
Criterion
2 LOG L
AIC
SBC
Without
Covariates
1967.813
1967.813
1967.813
With
Covariates
1944.176
1956.176
1976.856
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Variable
dobl
dob2
dob3
LCUMEXP15
PARITY1
FREL_BR_CAN1
Score
Wald
DF
1
1
1
1
1
1
Analysis of
Parameter
Estimate
0.08605
0.38780
0.47303
0.04949
- 0.25908
0.47620
23.6371
24.0044
23.5651
6
6
6
0.0006
0.0005
0.0006
Maximum Likelihood Estimates
Standard
Error
0.21943
0.25363
0.31528
0.02288
0.18638
0.17923
Chi-Square
0.1538
2.3378
2.2509
4.6787
1.9322
7.0595
Pr > ChiSq
0 . 6949
0.1263
0.1335
0.0305
0.1645
0.0079
Hazard
Ratio
1.090
1.474
1.605
1.051
0.772
1.610
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-28 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
O
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
(x) log cumulative exposure model, 20-year lag
Without With
Criterion Covariates Covariates
- 2 LOG L 1967.813 1947.020
AIC 1967.813 1959.020
SBC 1967.813 1979.700
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 20.7930 6 0.0020
Score 21.5306 6 0.0015
Wald 21.1847 6 0.0017
Analysis of Maximum Likelihood Estimates
Parameter
Variable
dobl
dob2
dob3
LCUMEXP20
PARITY1
FREL BR CAN1
DF
1
1
1
1
1
1
Standard
Estimate
0.
0.
0.
0.
- 0.
0.
,10961
,46136
,61353
,02970
,26623
,47060
0,
0,
0,
0,
0,
0,
Error
.22008
.25203
.30969
.02151
.18642
.17925
Chi-Square
0.
3.
3.
1.
2.
6.
,2481
,3509
,9248
,9068
,0397
,8927
Pr >
0
0
0
0
0
0
ChiSq
.6184
.0672
.0476
.1673
.1532
.0087
Hazard
Ratio
1.
1.
1.
1.
0.
1.
,116
,586
,847
,030
,766
,601
D.2. BREAST CANCER MORTALITY
a. Exposure distribution among women and breast cancer deaths in the cohort
mortality study (« = 9544)
In the Cox regression analyses of Steenland et al. (2004), the data on breast cancer mortality
was found to be fit best using cumulative exposure with a 20-year lag. Below is the
distribution of the 102 breast cancer deaths used in the analysis. The cut points are those
used in the published data (Steenland et al., 2004).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-29 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-2a. Distribution of cases in Cox regression analysis of breast cancer
mortality after using a 20-year lag
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Cumulative exposure,
20-year lag"
0 (Lagged out)
>0-646 ppm-days
647-2779 ppm-days
2780-12321 ppm-days
>12321 ppm-days
Number of breast
cancer deaths
42
17
16
15
12
aMean exposures for females with a 20-year lag for the categorical exposure quartiles were
276; 1,453; 5,869; and 26,391 ppm x days. Median values were 250; 1,340; 5,300; and
26,676 ppm x days. These values are for the risk sets but should provide a good
approximation to the full cohort values.
Regarding the women in the cohort as a whole, cumulative exposure at the end of follow-up,
with no lag, had a mean of 8.2 ppm-years, with a standard deviation of 38.2. This
distribution was highly skewed; the median was 4.6 ppm-years.
b. Results of Cox regression analysis of breast cancer mortality using a variety of log
RR models
Analyses used a case-control approach, with 100 controls per case, as in Steenland et al.
(2004). Age was the time variable in proportional hazards (Cox) regression. For breast
cancer mortality, only exposure variables were included in models. Cases and controls were
matched on sex (all female), date of birth, and race.
Using log RR models, we used a categorical model, a linear model, a 2-piece linear model, a
log transform model, and a cubic spline model. We also ran a number of analogous models
using linear RR models (Section D-2.c below).
The categorical log RR model for breast cancer mortality was run using the originally
published cut points to form four categories above the lagged-out group, as shown in
Table D-2a. To graph the categorical points, each category was assigned the mid-point of the
category as its exposure level, except for the last one which was assigned 50% more than the
last cut point 12,322 ppm-days.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 0 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
For the 2-piece log-linear model, the single knot was chosen at 700 ppm-days based on a
comparison of likelihoods assessed every 100 ppm-days from 0 to 7,000 (Figure D-2a). We
also explored knots beyond 7,000 ppm-days by looking at increments of 1,000 ppm-days
from 0 to 25,000 (Figure D-2a shows the results for knots up to 15,000 ppm-days). None of
these outperformed the knot at 700 ppm-days, although Figure D-2a' suggests a local
maximum likelihood near 13,000 ppm-days.
-2 log likelihood for different knots for breast cancer mortality
9
10
11
12
13
14
15
16
17
1000
2000
3000
\
4000
\
5000
\
6000
7000
KNOT
Figure D-2a. Likelihoods vs knots for the 2-piece log-linear model, breast
cancer morality.
In Figure D-2b below, we show the categorical and 2-piece log-linear spline models, as well
as the log-linear model and the log-linear model after cutting out the top 5% of exposed
subjects.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-31 DRAFT—DO NOT CITE OR QUOTE
-------
-2 log likelihood for different knots for breast cancer mortality
1
2
3
4
5
6
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
919.400:
919.300-
919.200:
919.100:
919.000:
918.900-
918.800-
918.700-
918.600-
918.500-
918.400-
918.300:
918.200:
1000 2000 3000 4000 5000 6000 7000 8000
KNOT
9000 10000 11000 12000 13000 14000 15000
Figure D-2b. Likelihoods vs knots for the 2-piece log-linear model, breast
cancer morality.
The log-linear model was clearly highly sensitive to exclusion of the most highly exposed.
As a sensitivity analysis, we excluded 1%, 2.5%, 5%, and 10% of the upper tail of exposure.
The 5% cutoff was at 15,000 ppm-days, while the 10% cutoff was at 13,000 ppm-days. The
slope of the linear exposure-response relationship increased by 1.2, 1.6, 5.9, and 4.5 times,
respectively, with the exclusion of progressively more data. It would appear that the upper
5% of the exposure range most affects the linear slope, and it is responsible for the
attenuation seen in the exposure-response at high exposures.
The 2-piece log-linear spline model in Figure D-2b fits reasonably well but appears to
underestimate the categorical RRs at higher exposures. This may be due to the influence of
the top 5% of the exposed, which appear to have a strong attenuating influence on the slope
(see below).
For comparison purposes, we also show the logarithmic transformation log RR model in
Figure D-2c (which we have not used for risk assessment because it is supralinear in the low
dose region).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-32 DRAFT—DO NOT CITE OR QUOTE
-------
2.5
LogRR
• Categorical
Log RR, 95% cutoff
^—LogRR, Spline w/ Knot @ 700
5,000
10,000 15,000 20,000 25,000 30,000
CUMEXP20
1
2
3
4
5
6
7
Figure D-2c. Plot of the dose-response relationship of continuous exposure
(lagged 20 years) for breast cancer mortality, with the 2-piece linear spline,
the categorical, and the linear log RR models (labeled "log RR"). Also shown
is the log-linear curve (log RR = P x cumexp20) after cutting out the top 5% of
exposure subjects ('log RR 95% cutoff).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 3 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Breast cancer mortality log transformed
-2 1 qg I i ksl i hand i s 917 ~71&
1-'
o
5QDD
1QDDD
18DDD
'jnmn
Figure D-2d. Plot of the dose-response relationship of continuous exposure
(lagged 20 years) for breast cancer mortality, generated using a logarithmic
transformation log RR model. Dots that represent the effect of exposure
grouped in categories are also presented in the plot.
Outputs from the categorical, 2-piece linear spline, and linear log RR models are given
below. The 2-piece log-linear model performed similarly to the log-linear model, but
appeared to fit the categorical log RR model points and the cubic spline log RR model much
better. The log-linear spline model is at the border of statistical significance (p = 0.07). In
any case, models with relatively sparse data may not achieve conventional statistical
significance (at the 0.05 level) but still provide a good fit to the data, judged by conformity
with categorical and cubic spline analysis, and may still be useful for risk assessment.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-34 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Table D-2b. Categorical output breast cancer mortality, 20-year lag (log RR
model)
Model
Criterion
-2 LOG L
AIC
SBC
Fit Statistics
Without
Covariates
923.433
923.433
923.433
Testing Global Null
Test
Likelihood Ratio
Score
Wald
Analysis
Parameter Standard
Variable DF Estimate Error
CUM201 1 0.56653 0.33920
CUM202 1 0.57236 0.35505
CUM203 1 0.67537 0.37632
CUM204 1 1.14110 0.40446
Chi-Square
7.9244
8.5160
8.3993
of Maximum
Chi-Square
2.7894
2.5987
3.2207
7.9598
With
Covariates
915.509
923.509
934.009
Hypothesis: BETA=0
DF Pr >
4 0
4 0
4 0
ChiSq
.0944
.0744
.0780
Likelihood Estimates
Pr > ChiSq
0.0949
0.1070
0.0727
0.0048
Hazard
Ratio
1.762
1.772
1.965
3.130
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 5 DRAFT—DO NOT CITE OR QUOTE
-------
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Table D-2c. 2-piece log-linear spline, breast cancer mortality, 20-year lag,
knot at 700 ppm-days
Model Fit Statistics
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
923.433
923.433
923.433
With
Covariates
918.037
922.037
927.287
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
5.3967
6.0153
5.8857
0.0673
0.0494
0.0527
Analysis of Maximum Likelihood Estimates
Parameter
LIN 0
LIN 1
Parameter
Estimate
0.0006877
-0.0006782
Standard
Error
0.0004171
0.0004188
Chi-Square
2.7178
2.6229
Pr > ChiSq
0.0992
0.1053
Hazard
Ratio
1.001
0.999
*covariance linO and linl -1.75 x 10
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 6 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Table D-2d. Log-linear model, breast cancer mortality, 20-year lag
Model Fit Statistics
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
923.433
923.433
923.433
With
Covariates
920.647
922.647
925.272
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
2.7865
3.7383
3.6046
0.0951
0.0532
0.0576
Analysis of Maximum Likelihood Estimates
Variable
CUMEXP20
Parameter
Estimate
0.0000122
Standard
Error Chi-Square Pr > ChiSq
6.40812E-6
3.6046
0.0576
Hazard
Ratio
1.000
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 7 DRAFT—DO NOT CITE OR QUOTE
-------
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Table D-2e. Log transform log RR model, breast cancer mortality, 20-year
lag
Model Fit Statistics
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
923.433
923.433
923.433
With
Covariates
917.743
919.743
922.368
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
5.6908
5.7676
5.7688
0.0171
0.0163
0.0163
Parameter DF
Icum20 1
Analysis of Maximum Likelihood Estimates
Parameter Standard
Estimate Error Chi-Square Pr > ChiSq
0.08376
0.03487
5.7688
0.0163
Hazard
Ratio
1.087
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-3 8 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table D-2f. 2-piece log-linear spline model, breast cancer mortality, 20-year
2 lag, knot at 13,000 ppm-days
3
4 Model Fit Statistics
5
6 Without With
7 Criterion Covariates Covariates
8
9 -2 LOG L 923.433 918.237
10 AIC 923.433 922.237
11 SBC 923.433 927.487
12
13
14 Testing Global Null Hypothesis: BETA=0
15
16 Test Chi-Square DF Pr > ChiSq
18 Likelihood Ratio 5.1963 2 0.0744
19 Score 5.9044 2 0.0522
20 Wald 5.7813 2 0.0555
21
22
23 Analysis of Maximum Likelihood Estimates
24
25 Parameter Standard Hazard
26 Variable Estimate Error Chi-Square Pr > ChiSq Ratio
27
28 LIN_0 0.0000607 0.0000309 3.8539 0.0496 1.000
29 LIN_1 -0.0000583 0.0000371 2.4761 0.1156 1.000
30
31
32 c. Linear relative risk models for breast cancer mortality
33
34 Finally, we also ran linear RR models for these data, as shown in Figure D-2d below
35 (denoted "ERR" models), which also includes the RRs from the log RR categorical model as
36 shown in other graphs. Again, the linear curve, highly influenced by the upper 5% tail of
37 exposure, underestimates the categorical points, while the log transform and 2-piece spline
38 capture better the initial increase in risk followed by an attenuation. Parameter estimates for
39 these models can be found in Table D-2g.
40
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-3 9 DRAFT—DO NOT CITE OR QUOTE
-------
• Categorical
SplineERR. Knot=700.CUMEXP20
- ERR, OJMlWiQ
Log(CUMtXP20)
.'V(IIK)
iD.lHII)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Figure D-2e. Linear RR models for breast cancer mortality.
Table D-2g. Model results for breast cancer mortality, linear RR models
Linear RR Model
CUMEXP20
Log(CUMEXP20)
Spline, knot = 700,
CUMEXP203
Parameter(s)
B = 0.000026779
B = 0.122090
Bl = 0.000830, B2 = -
0.000807
SE
0.000021537
SE = 0.061659
SE1 = 0.000614, SE2 =
0.000619
-2 Log Likelihood
920.122
917.841
918.058
aCovariance 2 pieces of spline, -3.80 x 10 .
bEditorial note: As discussed in footnotes i andj of Table 4-7 in Section 4.1.2.3, Confidence intervals were
determined using the Wald approach. Confidence intervals for linear RR models, however, in contrast to
those for the log-linear RR models, may not be symmetrical. For breast cancer incidence, EPA also
evaluated application of a profile likelihood approach for the linear RR models (Langholz and Richardson,
2010), which allows for asymmetric CIs, for comparison with the Wald approach. The unit risk estimate for
breast cancer mortality presented in this assessment does not rely on any of the linear RR models, thus
revised CIs calculated using the profile likelihood method are not shown here.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-40 DRAFT—DO NOT CITE OR QUOTE
-------
1 d. Risk assessment for breast cancer mortality using the 2-piece log-linear spline model
2
3 We next used the 95% upper bound of the coefficient for the 1st piece of the 2-piece log-
4 linear model from Table D-2c, which is 0.0006877 + 1.64 x 0.0004171, to calculate the
5 LECoi via the life-table analysis of excess risk used by EPA in Appendix C of their 2006
6 draft risk assessment. Here we used the same data on background breast cancer mortality
7 and background all cause mortality as used by EPA in their 2006 calculations. The rate ratio,
8 then, as a function of exposure, is RR = e(0-00137 x cumexP2°). Note that the 2-piece log-linear
9 model is linear for the log of the rate ratio. Once this is exponentiated, it is no longer strictly
10 linear, but is still approximately so, as can be seen in Figure D-2b.
11
12 Use of this function in the life-table analysis results in an excess risk of 0.01 when the daily
13 exposure is 0.0048 ppm, which is the LECoi. This is substantially lower than the previous
14 LECoi of 0.0195 ppm in EPAs 2006 draft risk assessment (U.S. EPA, 2006a, Table 12).
15
16 Similar calculations were done to derive the ECoi which was 0.0095 ppm.
17
18 e. Risk assessment for breast cancer mortality using the 2-piece linear spline model.
19
20 The slope of the first segment of the 2-piece linear model was 21% higher than the slope of
21 the corresponding 2-piece log-linear spline (knot at 700 ppm-days). The slope coefficient
22 was 0.0008300, with a std. err. of 0.000614. The resulting ECoi and LECoi were 0.0080 and
23 0.0037 ppm, respectively.
24
25 D.3. LYMPHOID CANCER MORTALITY (SUBSET OF ALL HEMATOPOIETIC
26 CANCERS COMBINED) (N= 18,235).
27
28 a. Exposure distribution in cohort and among lymphoid cases in the cohort mortality
29 study
30
31 The estimated daily exposure to ETO across different jobs and time periods ranged from
32 0.05 to 77 ppm. Exposure intensities from this broad range were multiplied by the length of
33 time in different jobs to get estimates of cumulative exposure. The duration of exposure for
34 the full cohort at the end of follow-up had a mean of 8.7 years and a standard deviation of
35 9.3 years. Cumulative exposure at the end of follow-up, with no lag, had a mean of 27 ppm-
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-41 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
years and a median of 6 ppm-years, indicating that these data are highly skewed. Log
transformation of these data results in an approximately normal distribution of the data.
As noted in Section D.I.a, cumulative exposure at the end of follow-up may be misleading,
as it is not relevant to standard analyses, all of which treat cumulative exposure as a time-
dependent variable which must be assessed at specific points in time. See Section D. 1 .a for
more discussion.
9
10
11
12
13
14
15
16
17
18
In modeling lymphoid cancer, a subset of all (lympho)hematopoietic cancer, we used a 15-
year lag for cumulative exposure as in the prior publication (Steenland et al., 2004), and we
also used the same cut points as in the publication. Lymphoid cancer consists of nonHodgkin
lymphoma, lymphocytic leukemia, and myeloma (ICD-9 200, 202, 203, 204). The
distribution of cases for lymphoid cancer mortality is seen below.
Table D-3a. Exposure categories and case distribution for lymphoid cancer
mortality
Cumulative exposure,
15-year lag3
0 (Lagged out)
>0-1200 ppm-days
120 1-3680 ppm-days
3681-13,500 ppm-days
>13,500 ppm-days
Male lymphoid
cancer deaths
6
2
4
5
10
Female lymphoid
cancer deaths
3
8
7
5
o
J
Total lymphoid
cancer deaths
9
10
11
10
13
19
20
21
22
23
24
25
26
27
28
29
30
31
aThe means of the categories were 0, 446, 2,143, 7,335, and 39,927 ppm-days, respectively. The medians were
374, 1,985, 6,755, and 26,373 ppm-days, respectively. These values are for the full cohort.
b. Results of Cox regression analysis of lymphoid cancer mortality using categorical, 2-
piece linear, log transform, and linear log RR models
While the published results in Steenland et al. (2004) focused on males (Table 7 in Steenland
et al., 2004), in fact males and females do not differ greatly in categorical results using a 15-
year lag. A formal chunk test for four interaction terms between exposure and gender is not
close to significance (p = 0.58), although such tests are not very powerful in the face of
sparse data such as these. Table D-3b below shows the categorical odds ratio results for men
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-42 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
and women separately and combined. In the analyses presented here, males and female are
combined.
Table D-3b. Lymphoid cancer mortality results by sex
Cumulative exposure,
15-year lag
0 (Lagged out)
>0-1200 ppm-days
1,201-3,680 ppm-days
3,681-13,500 ppm-days
>13,500 ppm-days
Odds ratios
(95% CI)
males
1.00
0.91 (0.16-5.23)
2.89 (0.65-12.86)
2.71(0.65-11.55)
3.76(1.03-13.64)
Odds ratios
(95% CI)
females
1.00
2.25 (0.41-12.45)
3.26 (0.56-18.98)
2.16(0.34-13.59)
1.83 (0.25-13.40)
Odds ratios
(95% CI)
combined
1.00
1.75 (0.59-5.25)
3.15(1.04-9.49)
2.44 (0.80-7.50)
3.00 (1.02-8.45)
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Analyses used a case-control approach, with 100 controls per case, as in Steenland et al.
(2004). Age was the time variable in proportional hazards (Cox) regression. For lymphoid
cancer mortality, only exposure variables were included in the model. Cases and controls
were within risk sets matched on age, gender, and race.
Using log RR models, we used a categorical model, a linear model, a 2-piece linear model,
and a log transform model. We also ran a number of analogous models using linear RR
models (Section D-3.c below).
The categorical log RR model for lymphoid cancer mortality was run using the originally
published cut points to form four categories above the lagged-out group, as shown in Table
D-3b. To graph the categorical points, each category was assigned the mid-point of the
category as its exposure level, except for the last one which was assigned 50% more than the
last cut point.
For the 2-piece log-linear model, the single knot was chosen at 100 ppm-days based on a
comparison of likelihoods assessed every 100 ppm-day from 100 to 15,000. The best
likelihood was at 100 ppm-days. Figure D-3a below shows the likelihood vs the knots.
Figure D-3a also suggests a local maximum likelihood near 1600 ppm-days. Figure D-3b
shows the log RR models.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-43 DRAFT—DO NOT CITE OR QUOTE
-------
1 Model results for the categorical and 2-piece linear log RR models are shown in Tables D-3c
2 and D-3d. Tables D-3e and D-3f give the results for the log transform model and linear log
3 RR models; the latter does not fit the data well. Table D-3g shows the model results for the
4 2-piece log-linear spine model with the knot at the local maximum likelihood of 1600 ppm-
5 days.
6
7 Figure D-3b shows the graphical results for the categorical, 2-piece linear, and log transform
8 log RR models. There is a very steep increase in risk at very low exposures. The knot for
9 the 2-piece log-linear curve is a low 100 ppm-days. The steep slope at low exposures may be
10 unrealistic as a basis for risk assessment, dependent as it is on relatively sparse data in the
11 low-exposure region (e.g., only 19 cases in the non-exposed lagged-out referent group and
12 the lowest cumulative exposure group, up to 1200 ppm-days, combined).
13
14 We further explored the sensitivity of the log-linear model to high exposures, by excluding
15 progressively more of the upper tail of exposure. We excluded 5%, 10%, 20%, 30%, 40%,
16 and 55% of the upper tail of exposure. The 55% cutoff was at 2,000 ppm-days. The slope of
17 the log-linear exposure-response model increased by 0.4, 1.7, 7.9, 5.6, 26.7, and 113.7 times,
18 respectively, with the exclusion of progressively more data. It is clear that the curve changes
19 substantially once the top 20% of the exposure range is truncated.
20
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-44 DRAFT—DO NOT CITE OR QUOTE
-------
-2 log likelihood for different knots for lymphoid cancer mortality
1
2
3
4
5
6
7
8
9
10
461.000-
460.000 -
459.000 -
458.000 -
457.000 -
1000
nr
2000
3000 4000
KNOT
5000
6000
Figure D-3a. Likelihoods vs knots for 2-piece log-linear model, lymphoid
cancer mortality.
i -•>
7000
• Categorical
LogRR, CUMEXP15
- LOJRR, LoglCUMtXPls)
logRR Spline, Knot = 100.CUMEXP15
5,000
la. mm
CUMEXP15
11,000
20,000
Figure D-3b. Plot of the exposure and lymphoid cancer mortality rate ratios
generated using a 2-piece log-linear spline model overlaid with other log RR
curves and categorical log RR model points.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-45 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Table D-3c. Categorical results for lymphoid cancer mortality (log RR
model), men and women combined
Model Fit Statistics
Without With
Criterion Covariates Covariates
-2 LOG L 463.912 458.069
AIC 463.912 466.069
SBC 463.912 473.950
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
5.8435
5.7397
5.6220
0.2111
0.2195
0.2292
Analysis of Maximum Likelihood Estimates
Parameter
Variable
CUM151
CUM152
CUM153
CUM154
DF
1
1
1
1
Standard
Estimate
0.
1.
0.
1.
56036
14581
89001
09998
0
0
0
0
Error
.55981
.56351
.57391
.55112
Chi-Square
1.
4.
2.
3.
0020
1344
4049
9837
Pr >
0
0
0
0
ChiSq
.3168
.0420
.1210
.0459
Hazard
Ratio
1
3
2
3
.75
.15
.44
.00
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-46 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Table D-3d. Results of 2-piece log-linear spline model for lymphoid cancer
mortality, men and women combined, knot at 100 ppm-days
Model Fit Statistics
Without With
Criterion Covariates Covariates
-2 LOG L 463.912 457.847
AIC 463.912 461.847
SBC 463.912 465.787
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 6.0658 2 0.0482
Score 5.9648 2 0.0507
Wald 5.8246 2 0.0544
Analysis of Maximum Likelihood Estimates
Parameter Standard
Parameter Estimate Error Chi-Square Pr > ChiSq
LIN 0 0.01010 0.00493 4.1997 0.0404
LIN 1 -0.01010 0.00493 4.1959 0.0405
Hazard
Ratio
1.010
0.990
35
36
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-47 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
Table D-3e. Results of the log transform log RR model for lymphoid cancer
mortality, both sexes combined
Model Fit Statistics
Without With
Criterion Covariates Covariates
-2 LOG L 463.912 458.426
AIC 463.912 460.426
SBC 463.912 462.396
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 5.4868 1 0.0192
Score 5.3479 1 0.0207
Wald 5.2936 1 0.0214
Analysis of Maximum Likelihood Estimates
Parameter Standard
Parameter DF Estimate Error Chi-Square Pr > ChiSq
IcumlS 1 0.11184 0.04861 5.2936 0.0214
Table D-3f. Results of the log-linear model for lymphoid cancer mortality,
both sexes combined
Model Fit Statistics
Without With
Criterion Covariates Covariates
-2 LOG L 463.912 462.413
AIC 463.912 464.413
SBC 463.912 466.383
Testinq Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 1.4998 1 0.2207
Score 2.0403 1 0.1532
Wald 1.9959 1 0.1577
Analysis of Maximum Likelihood Estimates
Parameter Standard
Parameter DF Estimate Error Chi-Square Pr > ChiSq
CUMEXP15 1 4.73679E-6 3.35285E-6 1.9959 0.1577
Hazard
Ratio
1.118
Hazard
Ratio
1.000
60 This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-48 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
Table D-3g. Results of 2-piece log-linear spline model for lymphoid cancer
mortality
Parameter
LIN 0
LIN_1
, men and women combined, knot at 1600 ppm-days
Model Fit Statistics
Without With
Criterion Covariates Covariates
-2 LOG L 463.912 458.640
AIC 463.912 462.640
SBC 463.912 466.581
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 5.2722 2 0.0716
Score 5.2666 2 0.0718
Wald 5.1436 2 0.0764
Analysis of Maximum Likelihood Estimates
Parameter Standard Hazard
DF Estimate Error Chi-Square Pr > ChiSq Ratio
1 0.0004893 0.0002554 3.6713 0.0554 1.000
1 -0.0004864 0.0002563 3.6014 0.0577 1.000
c. Results for linear relative risk models
Table D-3h shows the model fit statistics and coefficients for the linear RR models
(Supplemental Results). Results for linear RR models are seen in Figure D-3c (denoted as
"ERR" models).
They are quite similar to the log RR results in Figure D-2b. Again there is
a very steep rise in the exposure-response curve at very low exposures. The knot for the 2-
piece linear curve is again at 100 ppm-days.
43
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-49 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
Table D-3h. Supplemental Results: Model fit statistics and coefficients for
linear RR models, lymphoid cancer mortality
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Linear RR Model
CUMEXP15
Log(CUMEXP15)
Spline, knot = 100,
CUMEXP15b'c
-2 Log
likelihood
(full model)
461.62
458.54
457.48
AIC
463.62
460.54
461.48
/rvalue"
0.13
0.02
0.04
Parameter(s)
B = 0.0000 1226
B = 0.2083
Bl= 0.010090
B2 = -0.010086
SEd
SE = 0.00001214
SE = 0.1434
SE1 = 0.004458
SE2 = 0.004458
aFrom likelihood ratio test.
bCovariance of 2 pieces of linear spline: -2.52 x 10~5.
Tor the maximum likelihood estimate, for exposures below the knot, RR = 1 + (B 1 x exp); for exposures
above the knot, RR = 1 + (Bl x exp + B2 x (exp - knot)). For the 95% upper confidence limit, for
exposures below the knot, RR = 1 + (((31+ 1.645 x SE1) x exp); for exposures above the knot, RR = 1 +
((31 x exp + (32 x (exp-knot) + 1.645 x sqrt(exp2 x Varl + (exp-knot)2 x Var2 + 2 x exp x (exp-knot) x
covar)), where exp = cumulative exposure, var = variance, covar = covariance.
dEditorial note: Confidence intervals for linear RR models, in contrast to those for the log-linear RR models,
may not be symmetrical. EPA did not apply the profile likelihood approach (Langholz and Richardson,
2010), which allows for asymmetric CIs, to develop CIs for this model because the model was not used
further in the assessment.
3.:.
2.!)
1 5
1
• Categorical
----tKli. CUMtXPlb
- ERR, Log(CUMfcXPlb)
ERR Spline. Knot=100. CU M EXP15
10.000 lb.000
CUMEXP15
20,000
18
19
20
Figure D-3c. Linear RR models for lymphoid cancer.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-50 DRAFT—DO NOT CITE OR QUOTE
-------
1 d. Risk assessment for all lymphoid cancer mortality using the 2-piece log-linear spline
2 model
3
4 We consider that none of the parametric models (either log RR or linear RR) generated for
5 the lymphoid cancer data (and the same is true for all hematopoietic cancer) are suitable for
6 EPA risk assessment because of the overly steep exposure-response relationship in the low-
7 dose range for the 2-piece models and log transform models (highly influenced by the sparse
8 number of deaths in the low-exposure region), and the overly shallow exposure-response
9 relationship for the linear and log-linear models, which are influenced highly by the upper
10 tail of exposures. A reasonable alternative approach is a weighted regression through the
11 categorical points (excluding the highest exposure group), an approach adopted originally by
12 EPA.
13
14 Nonetheless, we have used the 2-piece log-linear model to calculate the LECoi and the ECoi,
15 by way of illustrating the effect of the very steep exposure-response curve in the low-dose
16 region.
17
18 We used the 95% upper bound of the coefficient for the 1st piece of the 2-piece log-linear
19 model from Table D-3d, which is 0.01010 + 1.64 x 0.00493, to calculate the LECoi via the
20 life-table analysis of excess risk used by EPA in Appendix C of their 2006 draft risk
21 assessment. Here we used the same data on lymphoid cancer mortality and background all-
22 cause mortality as used by EPA in their 2006 calculations. The predicted rate ratio, then, as a
23 function of exposure, is RR = e((0-01010 + L64 x 0'00493) x ™P15). Use of this RR model in the
24 life-table analysis results in an excess risk of 0.01 when the daily exposure (15-year lag) is
25 0.0006 ppm, which is the LECoi. This is much lower than the previous LECoi of 0.0165 ppm
26 for lymphoid cancer mortality in EPA's 2006 draft risk assessment (U.S. EPA, 2006a, Table
27 9).
28
29 A similar calculation was done for the ECoi, which resulted in a value of 0.0012 ppm.
30
31
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-51 DRAFT—DO NOT CITE OR QUOTE
-------
1 e. Supplemental results: results for log cumulative exposure Cox regression models with no
2 lag
3
4
5 Without With
6 Criterion Covariates Covariates
7
8 -2 LOG L 463.912 462.014
9 AIC 463.912 464.014
10 SBC 463.912 465.984
11
12
13 Testing Global Null Hypothesis: BETA=0
14
15 Test Chi-Square DF Pr > ChiSq
16
17 Likelihood Ratio 1.8987 1 0.1682
18 Score 1.8589 1 0.1728
19 Wald 1.8530 1 0.1734
20
21
22 Analysis of Maximum Likelihood Estimates
23
24 Parameter Standard Hazard
25 Parameter DF Estimate Error Chi-Square Pr > ChiSq Ratio
26
27 Icumexp 1 0.10230 0.07515 1.8530 0.1734 1.108
28
29
30
31 D.4. HEMATOPOIETIC CANCER MORTALITY (ALL HEMATOPOIETIC CANCERS
32 COMBINED)
33
34 a. Exposure distribution in cohort and among all (lympho)hematopoietic cases in the
35 cohort mortality study
36
37 In modeling hematopoietic cancer, we used a 15-year lag for cumulative exposure, as in the
38 prior publication (Steenland et al., 2004), and we also used the same cut points as in that
39 publication. The distribution of cases for hematopoietic cancer mortality is seen below.
40
41
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-52 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-4a. Exposure categories and case distribution for hematopoietic
cancer mortality
Cumulative exposure,
15 year lag
0 (Lagged out)
>0-1200 ppm-days
120 1-3680 ppm-days
3681-13,500 ppm-days
>13,500 ppm-days
Male hematopoietic
cancer deaths
9
4
5
8
11
Female
hematopoietic cancer
deaths
4
13
10
7
o
J
Total
hematopoietic
cancer deaths
13
17
15
15
14
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
aMean exposures for both sexes combined with a 15-year lag for the categorical exposure quartiles were 446;
2,143; 7,335; and 39,927 ppm x days. Median values were 374; 1,985; 6,755; and 26,373 ppm x days. These
values are for the full cohort.
b. Results of Cox regression analysis of hematopoietic cancer mortality using
categorical, 2-piece linear, linear and log transform log RR models
While the published results of these data in Steenland et al. (2004) focused on males (Table 8
in Steenland et al., 2004), in fact males and females do not differ greatly in categorical results
using a 15-year lag. A formal chunk test for four interaction terms between exposure and
gender is not close to significance (chi square 4.5, 4 df;p = 0.34), although such tests are not
very powerful in the face of sparse data such as these. Table D-4b below shows the
categorical odds ratio results for men and women separately and combined. Males and
females were combined in all analyses for hematopoietic cancer here.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-53 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-4b. All hematopoietic cancer mortality categorical results by sex
(log RR model)
Cumulative exposure,
15 year lag
0 (Lagged out)
>0-1200 ppm-days
120 1-3680 ppm-days
3681-13,500 ppm-days
>13,500 ppm-days
Odds ratio
(95% CI)
males
1.00
1.23 (0.32-4.74)
2.53 (0.69-9.27)
3.14(0.95-10.37)
3.42 (1.09-10.73)
Odds ratio
(95% CI)
females
1.00
3.76(1.01-17.23)
4.93 (1.01-23.99)
3.31,(0.64-17.16)
2.11(0.33-13.74)
Odds ratio
(95% CI)
combined
1.00
2.33 (0.93-5.86)
3.46 (1.33-8.95)
3.02(1.16-7.89)
2.96(1.12-7.81)
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Analyses used a case-control approach, with 100 controls per case, as in Steenland et al.
(2004). Age was the time variable in proportional hazards (Cox) regression. For lymphoid
cancer mortality, only exposure variables were included in the model. Cases and controls
were matched within risk sets on age, gender, and race.
Using log RR models, we used a categorical model, a linear model, a 2-piece linear model,
and a log transform model. We also ran a number of analogous models using linear RR
models (Section D-4.c below).
The categorical log RR model for hematopoietic cancer mortality was run using the originally
published cut points to form four categories above the lagged-out group, as shown in Table D-
4b. To graph the categorical points, each category was assigned the mid-point of the category as
its exposure level, except for the last one which was assigned 50% more than the last cut point.
For the 2-piece log-linear model, the single knot was chosen based on a comparison of
likelihoods assessed every 100 ppm-days from 0 to 7,000 ppm-days. The best likelihood was
at 500 ppm-days (Figure D-4a). In Figure D-4b below we show the categorical, 2-piece
linear spline and log transform log RR model results.
Model results for the categorical and 2-piece linear log RR models are shown in Tables D-4c
and D-4d, and the results of the log transform and linear log RR models in Table D-4e and
Table D-4f. Again the linear model appears to substantially underestimate the exposure-
response relationship and does not provide a good model fit.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-54 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
We further explored the sensitivity of the log-linear model to high exposures by excluding
progressively more of the upper tail of exposure. We excluded 5%, 10%, 20%, 30%, 40%,
and 53% of the upper tail of exposure. The 53% cutoff was at 2,000 ppm-days. The slope of
the log-linear exposure-response model increased by 0.8, 1.0, 9.3, 28.6, 58.2, and 191.4
times, respectively, with the exclusion of progressively more data. It appears the curve is flat
in the top 20% of exposure.
-2 log likelihood for different knots for all hematopoetic cancer mortality
654.000 -
653.000 :_
652.000 :
651.000 -_
650.000 7
649.000 :
648.000 :_
647.000 :
1000
2000
3000 4000
KNOT
5000
6000
7000
10
11
12
13
14
Figure D-4a. Likelihood vs knots for 2-piece log-linear model, all
hematopoietic cancer.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-5 5 DRAFT—DO NOT CITE OR QUOTE
-------
3.5
2.5
1
2
3
4
5
6
7
• Categorical
LogRR, CUMfcXPli
--LogRR, Loe(CUMtXPlb)
Spline Log RR. Knot=500. CU MEXP15
s, IKK)
10,000
CUMEXP15
Figure D-4b. Plot of exposure and rate ratios for all hematopoietic cancer
generated using a 2-piece log-linear spline model and log transform, linear,
and categorical log RR models.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-56 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Table D-4c. Categorical results for all hematopoietic cancer mortality (log
RR model), men and women combined, cumulative exposure with a 15-year
lag
Model Fit Statistics
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
655.643
655.643
655.643
With
Covariates
647.806
655.806
665.022
Testing Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio
Score
Wald
7.8371
7.3994
7.2354
0.0977
0.1162
0.1240
Analysis of Maximum Likelihood Estimates
Parameter
Variable
CUM151
CUM152
CUM153
CUM154
DF
1
1
1
1
Standard
Estimate
0
1
1
1
.84746
.23989
.10664
.08360
0
0
0
0
Error
.46956
.48571
.48943
.49603
Chi-Square
3
6
5
4
.2573
.5166
.1126
.7723
Pr >
0
0
0
0
ChiSq
.0711
.0107
.0238
.0289
Hazard
Ratio
2.
3.
3.
2.
33
46
02
96
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-57 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table D-4d. Results of 2-piece log-linear spline model for all hematopoietic
2 cancer mortality, men and women combined, cumulative exposure with a 15-
3 year lag
4
5
6 Model Fit Statistics
7
8 Without With
9 Criterion Covariates Covariates
10
11 -2 LOG L 655.643 647.581
12 AIC 655.643 651.581
13 SBC 655.643 656.189
14
15
16 Testinq Global Null Hypothesis: BETA=0
17
18 Test Chi-Square DF Pr > ChiSq
19
20 Likelihood Ratio 8.0615 2 0.0178
21 Score 7.5092 2 0.0234
22 Wald 7.3467 2 0.0254
23
24
25 Analysis of Maximum Likelihood Estimates
26
27 Parameter Standard Hazard
28 Parameter DF Estimate Error Chi-Square Pr > ChiSq Ratio
29
30 spll 1 0.00201 0.0007731 6.7457 0.0094 1.002
31 sp!2 1 -0.00201 0.0007738 6.7249 0.0095 0.998
32
33
34
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-5 8 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Table D-4e. Results of log-transform log RR model for all hematopoietic
cancer mortality, men and women combined, cumulative exposure with a 15-
year lag
Model Fit Statistics
Criterion
-2 LOG L
AIC
SBC
Without
Covariates
655. 643
655.643
655.643
With
Covariates
648.825
650.825
653.129
Testing Global Null Hypothesis: BETA=0
Test
Likelihood Ratio
Score
Wald
Chi-Square
6.8177
6.6260
6.5593
DF
1
1
1
Pr > ChiSq
0.0090
0.0100
0.0104
Analysis of Maximum Likelihood Estimates
Parameter
1cuml5
DF
Parameter
Estimate
0.10706
Standard
Error Chi-Square Pr > ChiSq
0.04180
6.5593
0.0104
Hazard
Ratio
1.113
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-59 DRAFT—DO NOT CITE OR QUOTE
-------
Table D-4f. Results of log-linear model for all hematopoietic cancer
morality, men and women combined, cumulative exposure with a 15-year lag
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Model Fit Statistics
Without With
Criterion Covariates Covariates
-2 LOG L 655.643 654.922
AIC 655.643 656.922
SBC 655.643 659.226
Testinq Global Null Hypothesis: BETA=0
Test Chi-Square DF Pr > ChiSq
Likelihood Ratio 0.7213 1 0.3957
Score 0.8783 1 0.3487
Wald 0.8739 1 0.3499
Analysis of Maximum Likelihood Estimates
Parameter Standard Hazard
Parameter DF
Estimate Error Chi-Square Pr > ChiSq
CUMEXP15 1 3.26052E-6 3.48788E-6 0.8739 0.3499
c. Results for linear
For completeness, we
relative risk models for hematopoietic cancer mortality
Ratio
1.000
also present the results of the linear RR models below (Table D-4g and
Figure D-4c; linear RR models are denoted "ERR" models in the figure). They look much
like their counterparts
for the log RR models. Again, the high slope of the exposure-response
relationship in the low-dose region for the 2-piece linear and log transform curves, and
the
low overall slope of the linear curve, call into question the use of these models for risk
assessment.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-60 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
Table D-4g. Supplemental Results: Model fit statistics and coefficients for
linear RR models, hematopoietic cancer mortality
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Linear RR Model
CUMEXP15
Log(CUMEXP15)
Spline, knot = 100,
CUMEXP15b'c
-2 Log
likelihood
(full model)
654.64
648.13
646.95
AIC
656.64
650.13
650.95
/rvalue"
0.32
0.006
0.01
Parameter(s)
B = 0.000006257
B = 0.2322
Bl= 0.003673
B2 = -0.003668
SEd
SE = 0.000008187
SE = 0.1437
SE1 = 0.002345
SE2 = 0.002345
aFrom likelihood ratio test.
bCovariance of 2 pieces of linear spline: - 5.70 x 10~6.
Tor the maximum likelihood estimate, for exposures below the knot, RR = 1 + (B 1 x exp); for exposures
above the knot, RR = 1 + (Bl x exp + B2 x (exp - knot)). For the 95% upper confidence limit, for
exposures below the knot, RR = 1 + (((31+ 1.645 x SE1) x exp); for exposures above the knot, RR = 1 +
((31 x exp + (32 x (exp-knot) + 1.645 x sqrt(exp2 x Varl + (exp-knot)2 x Var2 + 2 x exp x (exp-knot) x
covar)), where exp = cumulative exposure, var = variance, covar = covariance.
dEditorial note: Confidence intervals for linear RR models, in contrast to those for the log-linear RR models,
may not be symmetrical. EPA did not apply the profile likelihood approach (Langholz and Richardson,
2010), which allows for asymmetric CIs, to develop CIs for this model because the model was not used
further in the assessment.
18
19
20
21
10.000 15.000
CUMEXP1S
Figure D-4c. Linear RR models for hematopoietic cancer mortality.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-61 DRAFT—DO NOT CITE OR QUOTE
-------
1 d. Risk assessment for all hematopoietic cancer mortality using the 2-piece log-linear
2 spline model
3
4 As was the case for lymphoid cancer (which is a subset of the hematopoietic cancers), we
5 consider that none of the parametric models (either log RR or ERR) generated for the
6 hematopoietic cancer data are suitable for EPA risk assessment because of the overly steep
7 exposure-response relationship in the low-dose range for the 2-piece models and the log
8 transform models (highly influenced by the sparse number of deaths in the low-exposure
9 region), and the overly shallow exposure-response relationship for the linear models, which
10 are influenced highly by the upper tail of exposures. A reasonable alternative approach is a
11 weighted regression through the categorical points (excluding the highest exposure group),
12 an approach adopted originally by EPA.
13
14 Nonetheless, we have used the 2-piece log-linear model to calculate the LECoi and the ECoi,
15 by way of illustrating the effect of the very steep exposure-response curve in the low-dose
16 region.
17
18 We used the 95% upper bound of the coefficient for the 1st piece of the 2-piece log-linear
19 model from Table D-4d, which is 0.00201 + 1.64 x 0.000773, or 0.003277, to calculate the
20 predicted LECoi via the life-table analysis of excess risk used by EPA in Appendix C of their
21 2006 draft risk assessment. Again, here we used the data on hematopoietic cancer mortality
22 and background all-cause mortality as used in EPA's 2006 calculations. The predicted RR,
23 then, as a function of exposure, is RR = e(0-003277 x cumexP15) (up to the knot of 500 ppm-days).
24
25 This results in an excess risk of 0.01 when the daily exposure (15-year lag) is 0.0032 ppm,
26 which is the LECoi. This is notably lower than the previous LECoi of 0.0109 ppm for
27 hematopoietic cancer mortality in EPA's 2006 draft risk assessment (U.S. EPA, 2006a,
28 Table 7).
29
30 Similar calculations were done for the ECoi, which resulted in a value of 0.0043 ppm.
31
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-62 DRAFT—DO NOT CITE OR QUOTE
-------
1 D.5. SUMMARY TABLE OF ECoiS FOR DIFFERENT OUTCOMES, USING 2-PIECE
2 SPLINE MODELS
3
4 Table D-5 below provides a summary of the current findings for ECoi and the prior EPA
5 findings for ECoi.
6
7 In general, findings are similar. As described above, the ECoi values based on the 2-piece
8 spline models were obtained by multiplying the background cancer rate by e^eta x cumexp) for
9 log RR models or by (1 + beta x cumexp) for linear RR models, where the beta coefficient
10 was for the first piece of the 2-piece linear models, and cumexp was determined such that a
11 daily exposure would result in an excess risk of 1% above background, with risk calculated
12 through age 85 years (BIER methodology, spreadsheet obtained from EPA). In the case of
13 breast cancer incidence, following EPA's methods in the risk assessment, the life-table
14 values for all-cause mortality (within each 5-year age interval) were adjusted to account for
15 incident cases being withdrawn from the pool at risk entering the next age interval, by adding
16 the breast cancer incidence rate to the all-cause mortality rate and then subtracting breast
17 cancer mortality rate so that fatal breast cancer cases are not "counted" twice in this
18 adjustment.
19
20 As noted above, we believe the 2-piece spline models (either log RR or linear RR versions
21 are reasonable bases for risk assessment for the breast cancer incidence and mortality data.
22 They also result in ECoi values that are lower than but in the ballpark of the previous EPA
23 estimates using weighted regression for categorical points, excluding the highest exposure
24 quintile. However, this is not the case for the hematopoietic/lymphoid cancer data.
25
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-63 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-5a. Summary of ECoi results (in ppm) in current analysis and
previous EPA risk assessment
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Breast cancer incidence13
(log RR model, 15 -year lag)
Breast cancer incidence (linear RR
model, 15-yearlag)b
Breast cancer mortality (log RR
model, 20 -year lag)
Breast cancer mortality (linear RR
model, 20 -year lag)
Hematopoietic cancer mortality
(log RR model, 15-yr lag)c
lymphoid cancer mortality (log RR
model, 15-yr lag)0
U.S. EPA (2006a)
EC01a
0.0238
-
0.0387
-
0.0238
0.0427
Steenland3
LECoi
2-piece spline
0.009
0.0052
0.0048
0.0037
0.0032
0.0006
Steenland
EC01
2-piece spline
0.0152
0.0100
0.0096
0.0080
0.0043d
0.0012e
aEPA uses regression through categorical points (U.S. EPA, 2006a), Steenland uses 2-piece spline models.
bBreast cancer incidence for the subgroup with interviews, see Steenland et al. (2004).
Tor hematopoietic and lymphoid cancer, EPA ECM calculated for males only, Steenland includes both men and
women.
dUsing at knot at 500 ppm-days. 2-piece linear RR model results similar but not presented.
eUsing knot at 100 ppm-days. 2-piece linear RR model results similar but not presented.
D.6. SENSITIVITY OF 2-PIECE SPLINE CURVES TO PLACEMENT OF KNOT
By way of sensitivity analysis, we ran 2-piece log-linear models for all breast cancer incidence
with knots chosen at 5000, 5800 (optimal) and 7000 ppm-days, and for hematopoietic cancer
mortality for knots of 500 (optimal) and 1000. Results show the relatively large sensitivity to the
knot placement in the ECoi.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-64 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table D-6a. Exposure-response coefficients and ECois based on selection of
different knots, using 2-piece log-linear models
Breast cancer incidence knot at 5000 ppm-days
Breast cancer incidence knot at 5800 ppm-daysa
Breast cancer incidence knot at 7000 ppm-days
Hematopoietic cancer mortality knot at 500 ppm-days
Hematopoietic cancer mortality knot at 1000 ppm-days
Coefficient first
piece
0.0000860
0.0000770
0.0000653
0.00201
0.00089
-2 log-likelihoodb
1940.6
1940.5
1940.7
647.6
648.4
EC01
0.0133
0.0151
0.0176
0.0043
0.0098
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
aKnot used in analysis.
bLower numbers equal better fit, linear RR model likelihoods not comparable to log RR likelihoods and are not
shown here.
D.7. POSSIBLE INFLUENCE OF THE HEALTHY WORKER SURVIVOR EFFECT
The healthy worker survivor effect is the effect of healthy workers remaining in the
workforce as sick workers leave, independently of any damaging effects of exposure. It is a
selection bias via which healthier workers remain in the workforce. It tends to create a
downward bias in exposure-response coefficients when the exposure metric is cumulative
exposure, which is by definition correlated with duration of exposure and almost always with
duration of employment Steenland et al. (1996). Given a true effect of exposure on disease
incidence or mortality in the case of ethylene oxide, it is possible that the health worker
survivor effect has caused some negative bias in observed exposure-response coefficients.
However, there are no standard methods to correct for this bias, because leaving work is both
a confounder and an intermediate variable on a pathway between exposure and disease.
Therefore, standard analyses would need to adjust for employment status as a confounder,
but should not adjust for it because it is an intermediate variable. Robins et al. (1992) has
proposed some solutions using G-estimation to address this problem, but to date these
solutions are not commonly used and can be difficult to implement. The degree to which the
health worker survivor effect confounds measured exposure-response trends is not known,
but it is likely that lagging exposure, as has been done here, diminishes such confounding
(Arrighi and Hertz-Picciotto, 1994)
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
D-65 DRAFT—DO NOT CITE OR QUOTE
-------
1 D.8. POSSIBLE INFLUENCE OF EXPOSURE MIS-MEASUREMENT
2
3 Exposure estimation in the ETO studies considered here is subject to errors in measurement.
4 The method for exposure estimation used here involved assigned estimated average
5 exposures in a given job, at a given time period in a given plant, to each worker in that job.
6 Estimated average exposures were taken from observed measurements in a given job, or
7 estimated likely average exposures in that job derived from a regression model based on
8 observed measurements (Hornung et al., 1994). Errors in measurement in this type of
9 situation are typically errors of the Berkson type, rather than classical errors (Armstrong,
10 1998,1990). In Berkson errors, the model for errors is
11
12 Exposuretrae = exposure0bserved + error,
13
14 and the error is independent of the observed exposure. The classical error model is
15
16 Exposure0bserved = exposuretme + error,
17
18 and the error is independent of the true exposure. Assuming the errors are unbiased, i.e.,
19 their expected value is 0, in the classical error model it is well known that measurement error
20 will bias exposure-response coefficients towards the null in regression analyses. However, in
21 the Berkson error model, exposure-response coefficients will be unbiased in linear regression
22 models, although their variance may be increased. In log-linear regression models, such as
23 used here, Berkson error in some instances may result in biased exposure-response estimates
24 (Deddens and Hornung, 1994; Prentice, 1982). This may occur when the variance of the
25 errors increases with the true exposure level, which is often the case in occupational studies,
26 when the disease is relatively rare (also typical), and when the true exposure is distributed
27 log-normally (again typical of occupational exposures). In this situation, (Steenland et al.,
28 2000) have shown that exposure-response coefficients using cumulative exposure can be
29 biased either upward or downward. The direction and degree of bias depends on the degree
30 of increase in the variance of exposure error as exposure level increases and on the variance
31 of duration of exposure. When the standard deviation of duration of exposure is less than or
32 equal to its mean, as is the case in the ETO cohort studied here, simulations have shown that
33 the exposure-response coefficients are approximately unbiased (Steenland et al., 2000). An
34 added complication not considered in the simulations conducted by (Steenland et al., 2000) is
35 the possible correlation between measurement error and outcome. If this correlation is
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-66 DRAFT—DO NOT CITE OR QUOTE
-------
1 strong, which may occur when there is a strong exposure-response relationship, it is
2 important to take it into account. Estimating the effect of exposure measurement in the
3 presence of this correlation can be done using Bayesian models and special software
4 (WINBUGS), but the calculations are complex and require a good deal of time.
5
6 Hornung et al. (1994) provide an estimate of the log-normal distribution of measured
7 exposure based on personal samples, as well as the likely distribution of error in assigning
8 the j ob-specific means to estimate individual exposures. Assignment of such j ob-specific
9 means was shown to involve some bias as well as random error. This provides a rich source
10 of information with which one could simulate the effect of measurement error on exposure-
11 response coefficients. Based on the exposure estimates used in the study, and some
12 assumptions about the error of such measurement in terms of bias and random error, as well
13 as the assumption of a Berkson error model, one could simulate what the true job-specific
14 exposure means were likely to have been, and then in turn simulate likely true personal
15 exposure distributions. Using the latter in exposure-response analysis, one could estimate the
16 true exposure-response coefficient. However, such analyses are rather involved and beyond
17 the scope of the current task.
18
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-67 DRAFT—DO NOT CITE OR QUOTE
-------
1 D.9. REFERENCES
2
3 Armstrong, B. (1990) The effects of measurement errors on relative risk regressions. Am J Epidemiol
4 132:1176-1184.
5 Armstrong, B. (1998) Effect of measurement error on epidemiological studies of environmental and occupational
6 exposures [Review]. Occup EnvironMed 55(10):651-656.
7 Arrighi, HM; Hertz-Piccioto, I. (1994) The evolving concept of the healthy worker effect. Epidemiology
8 5(2):189-196.
9 Deddens, J; Hornung, R. (1994) Quantitative examples of continuous exposure measurement errors that bias risk
10 estimates away from the null. In: CM Smith; DC Christiani; KT Kelsey; (Eds.) Chemical risk assessment of
11 occupational health: current applications, limitations, and future prospects (pp.77-85). Westport, CT: Auburn
12 House.
13 Hornung, RW; Greife, AL; Stayner, LT; et al. (1994) Statistical model for prediction of retrospective exposure to
14 ethylene oxide in an occupational mortality study. Am J Ind Med 25(6):825-836.
15 Langholz, B; Richardson, DB. (2010) Fitting general relative risk models for survival time and matched case-control
16 analysis. Am J Epidemiol 171:377-383.
17 Prentice, RL. (1982) Covariate measurement errors and parameter estimation in a failure time regression model.
18 Biometrika69(2):331-341.
19 Robins, J; Blevins, D; Ritter, G; et al. (1992) G-estimation of the effect of prophylaxis therapy for Pneumoocystis
20 carinii pneumonia on the survival of AIDS patients. Epidemiology 3:319-335.
21 Stayner, L; Steenland, K; Dosemeci, M; et al. (2003) Attenuation of exposure-response curves in occupational
22 cohort studies at high exposure levels. Scand J Work Environ Health 29:317-324.
23 Steenland, K; Deddens, J. (1997) Increased precision using counter-matching in nested case-control studies.
24 Epidemiology 8(3):238-242.
25 Steenland, K; Deddens, J. (2004) A practical guide to dose-response analyses and risk assessment in occupational
26 epidemiology [Review]. Epidemiol 15:63-70.
27 Steenland, K; Deddens, J; Piacitelli, L. (2001) Risk assessment for 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCCD)
28 based on an epidemiologic study. Am J Epidemiol 154:451-458.
29 Steenland, K; Deddens, J; Salvan, A; et al. (1996) Negative bias in exposure-response trends in occupational studies:
30 modeling the healthy worker effect. Am J Epi 143(2):202-210.
31 Steenland, K; Deddens, J; Zhao, S. (2000) Biases in estimating the effect of cumulative exposure in log-linear
32 models when estimated exposure levels are assigned. Scan J Work Environ Health 26:37-43.
33 Steenland, K; Stayner, L; Deddens, J. (2004) Mortality analyses in a cohort of 18,235 ethylene oxide-exposed
34 workers: follow up extended from 1987 to 1998. Occup EnvironMed 61:2-7.
3 5 Steenland, K; Whelan, E; Deddens, J; et al. (2003) Ethylene oxide and breast cancer incidence in a cohort study of
36 7576 women (United States). Cancer Causes Control 14: 531-539.
37 U.S. EPA (Environmental Protection Agency). (2006a). Evaluation of the carcinogenicity of ethylene oxide:
38 external review draft [EPA Report]. (EPA/635/R-06/003). National Center for Environmental Assessment,
39 Washington, DC.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 D-68 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX E.
2 LIFE-TABLE ANALYSIS
3 A spreadsheet illustrating the extra risk calculation for the derivation of the LECoi for
4 lymphoid cancer incidence is presented in Table E-l.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 E-l DRAFT—DO NOT CITE OR QUOTE
-------
to
OJ
Table E-l. Extra risk calculation" for environmental exposure to 0.0114 ppm (the LECoi for lymphoid cancer
incidence)"5 using the weighted linear regression model based on the categorical cumulative exposure results of
Steenland et al. (2004), reanalyzed by Steenland for both sexes combined (see Appendix D of this assessment),
with a 15-year lag, as described in Section 4.1.1
A
Interval
number
(i)
1
2
3
4
5
6
1
8
9
10
11
12
13
14
15
16
B
Age
interval
<1
1^
5-9
10-14
15-19
20-24
25-29
30-34
35-39
40^4
45^9
50-54
55-59
60-64
65-69
70-74
C
All cause
mortality
(xl05/yr)
685.2
29.9
14.7
18.7
66.1
94
96
107.9
151.7
231.7
352.3
511.7
734.8
1140.1
1727.4
2676.4
D
lymphoid
cancer
incidence
(xl05/yr)
1.9
8.1
4.2
3.2
3.5
3.2
4.1
6.0
9.0
13.2
20.9
32.5
49.2
70.1
101.1
128.7
E
All
cause
hazard
rate
(h*)
0.0069
0.0012
0.0007
0.0009
0.0033
0.0047
0.0048
0.0054
0.0076
0.0116
0.0176
0.0256
0.0367
0.0570
0.0864
0.1338
F
Prob of
surviving
interval
(q)
0.9932
0.9988
0.9993
0.9991
0.9967
0.9953
0.9952
0.9946
0.9924
0.9885
0.9825
0.9747
0.9639
0.9446
0.9173
0.8747
G
Prob of
surviving
up to
interval
(S)
1.0000
0.9932
0.9920
0.9913
0.9903
0.9871
0.9824
0.9777
0.9725
0.9651
0.9540
0.9373
0.9137
0.8807
0.8319
0.7631
H
lymphoid
cancer
hazard
rate (h)
0.0000
0.0003
0.0002
0.0002
0.0002
0.0002
0.0002
0.0003
0.0005
0.0007
0.0010
0.0016
0.0025
0.0035
0.0051
0.0064
I
Cond
prob of
lymphoid
cancer
incidence
in
interval
(RO)
0.00002
0.00032
0.00021
0.00016
0.00017
0.00016
0.00020
0.00029
0.00044
0.00063
0.00099
0.00150
0.00221
0.00300
0.00403
0.00460
J
Exp
duration
mid
interval
(xtime)
0
0
0
0
2.5
7.5
12.5
17.5
22.5
27.5
32.5
37.5
42.5
47.5
52.5
57.5
K
Cum
exp mid
interval
(xdose)
0.00
0.00
0.00
0.00
31.64
94.92
158.20
221.49
284.77
348.05
411.33
474.61
537.90
601.18
664.46
727.74
L
Exposed
lymphoid
cancer
hazard
rate (hx)
0.00002
0.00032
0.00021
0.00016
0.00018
0.00017
0.00022
0.00034
0.00052
0.00079
0.00128
0.00205
0.00319
0.00467
0.00691
0.00902
M
Exposed
all cause
hazard
rate
(h*x)
0.0069
0.0012
0.0007
0.0009
0.0033
0.0047
0.0048
0.0054
0.0077
0.0117
0.0179
0.0260
0.0375
0.0582
0.0882
0.1364
N
Exposed
prob of
surviving
interval
(qx)
0.9932
0.9988
0.9993
0.9991
0.9967
0.9953
0.9952
0.9946
0.9924
0.9884
0.9823
0.9743
0.9632
0.9435
0.9156
0.8725
O
Exposed
prob of
surviving
up to
interval
(Sx)
1.0000
0.9932
0.9920
0.9913
0.9903
0.9871
0.9824
0.9777
0.9724
0.9650
0.9538
0.9369
0.9128
0.8793
0.8296
0.7595
P
Exposed
cond prob
of
lymphoid
cancer in
interval
(Rx)
0.00002
0.00032
0.00021
0.00016
0.00018
0.00017
0.00022
0.00033
0.00050
0.00075
0.00121
0.00190
0.00286
0.00399
0.00549
0.00640
§•
§
I
^
'TS
o
a
I
a,
O 5
31
o *
H 1
W K-
o
C
o
H
W
-------
to
OJ
Table E-l. Extra risk calculation" for environmental exposure to 0.0114 ppm (the LECoi for lymphoid cancer
incidence)1* using the weighted linear regression model based on the categorical cumulative exposure results of
Steenland et al. (2004), reanalyzed by Steenland for both sexes combined (see Appendix D of this assessment), with
a 15-year lag, as described in Section 4.1.1 (continued)
A
Interval
number
(i)
17
18
B
Age
interval
75-59
80-84
C
All cause
mortality
(xl05/yr)
4193.2
6717.2
D
lymphoid
cancer
incidence
(xl05/yr)
163.0
179.8
E
All
cause
hazard
rate
(h*)
0.2097
0.3359
F
Prob of
surviving
interval
(q)
0.8109
0.7147
G
Prob of
surviving
up to
interval
(S)
0.6675
0.5412
H
lymphoid
cancer
hazard
rate (h)
0.0082
0.0090
Ro =
I
Cond
prob of
lymphoid
cancer
incidence
in
interval
(RO)
0.00491
0.00413
0.02797
J
Exp
duration
mid
interval
(xtime)
62.5
67.5
K
Cum
exp mid
interval
(xdose)
791.02
854.31
L
Exposed
lymphoid
cancer
hazard
rate (hx)
0.01171
0.01323
M
Exposed
all cause
hazard
rate
(h*x)
0.2132
0.3401
N
Exposed
prob of
surviving
interval
(qx)
0.8080
0.7117
0
Exposed
prob of
surviving
up to
interval
(Sx)
0.6627
0.5354
Rx =
P
Exposed
cond prob
of
lymphoid
cancer in
interval
(Rx)
0.00699
0.00601
0.03769
extra risk = (Rx-Ro)/(l-Ro) = 0.01001
§•
§
I
^
'TS
I
o
a
I
a,
O
O
2
O
H
W
O
C
o
H
W
Column A: interval index number (i).
Column B: 5-yr age interval (except <1 and 1-4) up to age 85.
Column C: all-cause mortality rate for interval i (x 105/yr) (2004 data from NCHS).
Column D: lymphoid cancer incidence rate for interval i (x 105/yr) (2000-2004 SEER data).c
Column E: all-cause hazard rate for interval i (h*0 (= all-cause mortality rate x number of years in age interval)."1
Column F: probability of surviving interval i (without being diagnosed with lymphoid cancer) (qO [= exp(-h*!)]- This column is intended to represent the
probability of surviving the interval without a diagnosis of lymphoid cancer; however, because lymphoid cancer incidence rates are negligible compared to
all-cause mortality rates, no adjustment was made to the population at risk to account for the probability of a lymphoid cancer diagnosis. For breast cancer
incidence, on the other hand, the age-specific "mortality" rates (representing the rates at which the population at risk is decreased in each interval) were adjusted
to include the age-specific breast cancer incidence rates and to exclude the age-specific breast cancer mortality rates, this latter adjustment so that the probability
of death from lymphoid cancer is not counted twice, i.e., both as an incident case and as a component of the all-cause mortality.
Column G: probability of surviving up to interval i (without having been diagnosed with lymphoid cancer) (SO (Si = 1; Si = SM x q^i, for i > 1).
Column H: lymphoid cancer incidence hazard rate for interval i (h^ (= lymphoid cancer incidence rate x number of years in interval).
Column I: conditional probability of being diagnosed with lymphoid cancer in interval i [= (h/h*!) x Si x (1-qO], i.e., conditional upon surviving up to interval i
(without having been diagnosed with lymphoid cancer) (Ro, the background lifetime probability of being diagnosed with lymphoid cancer = the sum of the
conditional probabilities across the intervals).
-------
jjj Table E-l. Extra risk calculation" for environmental exposure to 0.0114 ppm (the LECoi for lymphoid cancer
2 incidence)"5 using the weighted linear regression model based on the categorical cumulative exposure results of
Steenland et al. (2004), reanalyzed by Steenland for both sexes combined (see Appendix D of this assessment),
with a 15-year lag, as described in Section 4.1.1 (continued)
^ Column J: exposure duration at midinterval (taking into account 15-yr lag) (xtime).
o Column K: cumulative exposure midinterval (xdose) (= exposure level (i.e., 0.0114 ppm) x 365/240 x 20/10 x xtime x 365) [365/240 x 20/10 converts
S continuous environmental exposures to corresponding occupational exposures; xtime x 365 converts exposure duration in years to exposure duration in days).
§ Column L: lymphoid cancer incidence hazard rate in exposed people for interval i (hxO (= h; x (i + p x xdose), where P = 0.0002472 + (1.645 x
5' 0.0001854) = 0.0005522) (0.0002472 per ppm x day is the regression coefficient obtained from the weighted linear regression model of the categorical results
^ [see Section 4.1.1.2]). To estimate the LEC0i, i.e., the 95% lower bound on the continuous exposure giving an extra risk of 1%, the 95% upper bound on the
<2 regression coefficient is used, i.e., MLE + 1.645 x SE].
X Column M: all-cause hazard rate in exposed people for interval i (h*x^ [= h*; + (hxj - hi)].
^ Column N: probability of surviving interval i (without being diagnosed with lymphoid cancer) for exposed people (qx^ [= exp(-h*Xi)].
5 Column O: probability of surviving up to interval i (without having been diagnosed with lymphoid cancer) for exposed people (Sxi) (Sxi = 1; Sx; = Sx;-i x qx.i_i,
| for i > 1).
^3 Column P: conditional probability of being diagnosed with lymphoid cancer in interval i for exposed people [= (hx/h*^) x Sx; x (1-qxO] (Rx, the lifetime
probability of being diagnosed with lymphoid cancer for exposed people = the sum of the conditional probabilities across the intervals).
1
"Using the methodology of BEIR (1988).
<5. bThe estimated 95% lower bound on the continuous exposure level that gives a 1% extra lifetime risk of lymphoid cancer incidence.
a Background cancer incidence rates are used to estimate extra risks for cancer incidence under the assumption that the exposure-response relationship for cancer
a, incidence is the same as that for cancer mortality (see Section 4.1.1.3).
o' dFor the cancer incidence calculation, the all-cause hazard rate for interval i should technically be the rate of either dying of any cause or being diagnosed with
^ the specific cancer during the interval, i.e., (the all-cause mortality rate for the interval + the cancer-specific incidence rate for the interval — the cancer-specific
S. mortality rate for the interval [so that a cancer case isn't counted twice, i.e., upon diagnosis and upon death]) x number of years in interval. For the lymphoid
§ cancer incidence calculations, this adjustment was ignored because the lymphoid cancer incidence rates are small when compared with the all-cause mortality
§ rates. For the breast cancer incidence calculations, on the other hand, this adjustment was made in the all-cause hazard rate (see Section 4.1.2.3).
S" MLE = maximum likelihood estimate, SE = standard error.
31
o *
H 1
W K-
o
c
o
H
W
-------
1 APPENDIX F.
2 EQUATIONS USED FOR WEIGHTED LINEAR REGRESSION
3 (Source: Rothman (1986, p. 343-344))
4
5
6 linear model: RR=\ + bX
1
8 where RR = rate ratio, X = exposure, and b = slope
9
10 b can be estimated from the following equation:
11
7=2
12 * - n ~ (F-l)
7=2
13
14 where y specifies the exposure category level and the reference category (/ = 1) is ignored.
15
16 the standard error of the slope can be estimated as follows:
17
18 SE(b)« (F-2)
. i=2
19
20 the weights, Wj, are estimated from the confidence intervals (as the inverse of the variance):
21
2
22 Var(RR}) « RR]Var[\n(RR])] « RRj
(F-3)
2x1.96
23
24 where RR j is the 95% upper bound on the RRj estimate (for the jth exposure category) and RR± is
25 the 95% lower bound on the RRj estimate.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 F-l DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX G.
2 MODEL PARAMETERS IN THE ANALYSIS OF ANIMAL TUMOR INCIDENCE
4
5
Table G-l. Analysis of grouped data, NTP mice study (NTP, 1987);a
multistage model parameters
Tumor
Multistage11
polynomial
degree
qo
qic
(mg/m3)-1
Q2
(mg/m3)-2
qs
(mg/m3)-2
p value
(chi-square
goodness of
fit)
Males
Lung adenomas
plus carcinomas
1
2.52 x KT1
1.52 x 10~2
0.92
Females
Lung adenomas
plus carcinomas
Malignant
lymphoma
Uterine carcinoma
Mammary
carcinoma
2
3
2
ld
3.87 x 1(T2
1.74 x 10'1
0.0
2.27 x 10'2
0.0
0.0
0.0
1.09 x 10'2
4.80 x 10'4
0.0
9.80 x 10'5
1.13 x 10'5
0.39
0.18
0.90
-
6
7
8
9
10
11
aThe exposure concentrations were at 0, 50 ppm, and 100 ppm. These were adjusted to continuous exposure.
bP(d) a 1 - exp[-(q0 + qid + q2d2 + ... + qkdk)], where d is inhaled ethylene oxide exposure concentration.
°Even though qj is zero in some cases, the upper bound of q! is nonzero.
dThe 100-ppm dose was deleted; the fit was perfect with only two points to fit.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
G-1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table G-2. Analysis of grouped data, Lynch et al. (1984a); Lynch et al.
(1984c) study of male F344 rats;a multistage model parameters
Tumor
Splenic mononuclear cell leukemia
Testicular peritoneal
mesothelioma
Brain mixed-cell glioma
Multistage1"
polynomial degree
lc
1
1
qo
3.12 x ID'1
3.54 x ID'2
0
qi
(mg/m3)-1
1.48 x ID'2
6.30 x -3
1.72 x ID'4
/7-value
(chi-square
goodness of fit)
-
0.34
0.96
4
5
6
7
8
9
10
11
12
aThe exposure concentrations were at 0, 50 ppm, and 100 ppm. These were adjusted to continuous exposure.
bP(d) a 1 - exp[-(q0 + qid + q2d2 + ... + qkdk)], where d is inhaled ethylene oxide exposure concentration.
°The 100-ppm dose was deleted; the fit was perfect with only two points to fit.
Table G-3. Analysis of grouped data, Garman et al. (1985) and Snellings et
al. (1984) reports on F344 rats;a multistage model parameters
Tumor
Multistage11
polynomial degree
qo
qi
(mg/m3)-1
/7-value
(chi-square
goodness of fit)
Males
Splenic mononuclear cell leukemia
Testicular peritoneal mesothelioma
Primary brain tumors
1
1
1
1.63 x 10'1
2.38 x 1Q-2
5.88 x 1Q-3
8.56 x 10'3
4.74 x 1Q-3
2.92 x 1Q-3
0.34
0.68
0.46
Females
Splenic mononuclear cell leukemia
Primary brain tumors
1
1
1.08 x 1Q-1
5.94 x 1Q-3
2.37 x 1Q-2
1.65 x 1Q-3
0.75
0.80
13
14
15
16
17
18
19
"The exposure concentrations were at 0, 10 ppm, 33 ppm, and 100 ppm. These were adjusted to continuous
exposure.
bP(d) a 1 - exp[-(q0 + qid + q2d2 + ... + qkdk)], where d is inhaled ethylene oxide exposure concentration.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
G-2 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table G-4. Time-to-tumor analysis of individual animal data, NTP mice
study (NTP, 1987);a multistage-Weibull model"5 parameters
Tumor
Multistage
polynomial degree
qo
qi
(mg/m3)-1
z
Males
Lung adenomas plus
carcinomas
1
3.44 x ID'1
2.03 x 1Q-2
5.39
Females
Lung adenomas plus
carcinomas
Malignant lymphoma
Uterine carcinoma
Mammary carcinoma
1
1
1
1
5.35 x ID'2
1.91 x ID'1
0.0
3.78 x 10'2
1.76 x 1Q-2
8.80 x 1Q-3
3.81 x 10'3
5.10 x 10'3
7.27
1.00
3.93
1.00
4
5
6
7
8
9
10
aThe exposure concentrations were at 0, 50 ppm, and 100 ppm. These were adjusted to continuous exposure.
bP(d, t) = 1 - exp[-(q0 + qi d + q2d2 + .... + qkdk)*(t - t0)z], where d is inhaled ethylene oxide exposure
concentration.
The length of the study was 104 weeks. The times t and t0 as expressed in the above formula are scaled so that the
length of the study is 1.0. Then, q0 is dimensionless, and the coefficients qk are expressed in units of (mg/m3)~k.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
G-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX H.
2 SUMMARY OF 2007 EXTERNAL PEER REVIEW AND
3 PUBLIC COMMENTS AND DISPOSITION
4 A draft of this assessment document entitled Evaluation of the Carcinogenicity of
5 Ethylene Oxide (dated August 2006) (U.S. EPA, 2006a) was available for public comment and
6 underwent a formal external peer review in accordance with EPA guidance on peer review (U.S.
7 EPA, 2006b) At the request of EPA's Office of Research and Development, the EPA Science
8 Advisory Board (SAB) convened a panel of 15 experts external to the Agency to review the
9 ethylene oxide (EtO) assessment document. An external peer review meeting was held in
10 January 2007, and a Final Peer Review Report was released in December 2007 (SAB, 2007).
11 The purpose of this assessment was to review and characterize the available data on the
12 carcinogenicity of EtO and to estimate the lifetime unit cancer risk from inhalation exposure.
13 The SAB panel was asked to comment primarily on three main issues including carcinogenic
14 hazard, cancer risk estimation, and uncertainty associated with the hazard characterization and
15 quantitative risk estimation. The SAB panel was charged with answering a number of specific
16 questions that addressed key scientific issues relevant to the assessment. A summary of
17 significant comments made by the panel in response to the charge questions and EPA's response
18 to these comments arranged by charge question are provided below.
19 In addition, a number of comments from the public were received during the public
20 comment period. A summary of the public comments and EPA's responses are also included in
21 a separate section of this appendix.
22
23 SAB Panel Comments:
24 The statement of the issues as contained in the Agency's charge to the SAB panel are
25 listed below in italics followed by (1) the Panel's summary comments quoted directly from the
26 Executive Summary of the Panel's report (SAB, 2007) and (2) the Agency's response to the
27 comments.
28
29 Issue 1: Carcinogenic Hazard (Section 3 and Appendix A of the EPA Draft Assessment)
30 Do the available data and discussion in the draft document support the hazard conclusion that
31 EtO is carcinogenic to humans based on the weight-of-evidence descriptors in EPA's
32 2005 Guidelines for Carcinogen Risk Assessment? In your response, please include
33 consideration of the following:
34
35
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-1 DRAFT—DO NOT CITE OR QUOTE
-------
1 La, EPA concluded that the epidemiological evidence on EtO carcinogenicity was strong, but
2 less than completely conclusive. Does the draft document provide sufficient description of the
3 studies, balanced treatment of positive and negative results, and a rigorous and transparent
4 analysis of the data used to assess the carcinogenic hazard ofethylene oxide (EtO) to
5 humans? Please comment on the EPA's characterization of the body of epidemiological data
6 reviewed. Considerations include: a) the consistency of the findings, including the
1 significance of differences in results using different exposure metrics, b) the utility of the
8 internal (based on exposure category) versus external (e.g., SMR and SIR) comparisons of
9 cancer rates, c) the magnitude of the risks, and d) the strength of the epidemiological evidence.
10
11 SAB Panel Comment: A majority of the Panel agreed with the conclusion in the draft document
12 that the available evidence supports a descriptor of "Carcinogenic to Humans" although some
13 Panel members concluded that the descriptor "Likely to be Carcinogenic to Humans" was more
14 appropriate. There was consensus that the epidemiological data regarding ethylene oxide
15 carcinogenicity were not in and of themselves sufficient to provide convincing evidence of a
16 causal association between human exposure and cancer. Differing views as to the appropriate
17 descriptor for ethylene oxide were based on differences of opinion as to whether criteria
18 necessary for designation as "Carcinogenic to Humans" in the absence of conclusive evidence
19 from epidemiologic studies were met. The majority of Panel members thought that the
20 combined weight of the epidemiological, experimental animal, and mutagenicity evidence was
21 sufficient to conclude that EtO is carcinogenic to humans.
22 The Panel concluded that the assessment would be improved by: (1) a better introduction
23 to the hazard characterization section, including a brief description of the information that will be
24 presented; (2) a clear articulation of the criteria by which epidemiologic studies were judged as
25 to strengths and weaknesses; (3) addition of a more inclusive summary figure and/or table at the
26 beginning of section 3.0; and (4) inclusion of material now provided in Appendix A of the draft
27 assessment to within the main body of that assessment.
28 The Panel agreed with the EPA in their reliance on "internal" estimates of cancer rates
29 rather than "external" comparisons (SMR, SIR) due to well recognized limitations to the latter
30 method of analysis. The Draft Assessment characterizes the magnitude of the unit risk estimate
31 associated with EtO as "weak". This finding is substantiated by the epidemiologic evidence
32 where a relatively small number of excess cancers are found above background even among
33 highly exposed individuals. However, the magnitude of risk suggested by the unit risk estimate
34 is somewhat at odds with this concept. Subsequent recommendations in our report try to address
35 this apparent inconsistency.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA Response: EPA is retaining the conclusion that the combined weight of the
2 epidemiological, experimental animal, and mutagenicity evidence is sufficient to conclude that
3 EtO is carcinogenic to humans, which was supported by the majority of the Panel. Some Panel
4 members were of the opinion that the descriptor "Likely to be Carcinogenic to Humans" was
5 more appropriate, as they found the epidemiological evidence to be weak and the data
6 insufficient to conclude that key precursor events were observed in humans (SAB, 2007, p. 10).
7 EPA and the majority of the SAB Panel disagree that the epidemiological evidence is weak.
8 EPA has strengthened the summary review of these data in the human evidence section (Section
9 3.1) and in the hazard characterization section (Section 3.5.1). In addition, the assessment
10 specifically addresses the precursor data for rodents and humans, and while the databases for
11 humans and rodents contain different types of studies, EPA did not find any inconsistency and
12 concluded that the data support a finding of a mutagenic mode of action (relevant to humans), a
13 finding with which the SAB concurred. EPA has expanded the discussion of these data,
14 specifically in Sections 3.3.3.2, 3.3.3.3, and 3.4.1.
15 In response to the Panel recommendations, EPA has added an introduction at the
16 beginning of Chapter 3 that provides a brief description of the information presented in the
17 Chapter and has provided a clearer explanation of the criteria used to evaluate the strengths and
18 weaknesses of epidemiological studies (at the beginning of Section 3.1). With respect to the
19 recommendation to put material from Appendix A into the main body of the document, EPA has
20 added two shorter summary tables of the lymphohematopoietic cancer (Table 3-1) and breast
21 cancer (Table 3-2) findings in the various epidemiology studies to Section 3.1.1. EPA has also
22 added a cross-reference to summary Table A-5 in Appendix A at the beginning of Section 3.1.
23 The main body of the document provides a summary of the findings of all the epidemiological
24 studies, referencing Appendix A for further details. EPA considered the recommendation to
25 move more of the material in Appendix A of the draft assessment into the main body of the
26 document, but judged that the in-depth level of detail in Appendix A was not appropriate for the
27 main body of the document.
28 EPA notes that the Panel agreed with EPA's use of "internal" estimates rather than
29 "external" comparisons.
30 The Draft Assessment did not refer to or characterize the magnitude of the unit risk
31 associated with EtO exposure as "weak." Rather, it was with respect to the Hill considerations
32 for causality (Hill, 1965) in the weight-of-evidence analysis for hazard characterization (Section
33 3.5.1) that the Draft Assessment noted that there was little strength in the association, as reflected
34 by the modest magnitude of the (relative) risk estimates from the epidemiology studies. The
35 exposure-response models used to develop the unit risk estimates are derived from the NIOSH
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 data and are thus consistent with the results of the NIOSH epidemiology study, as can be seen in
2 the figures depicting RR versus exposure for the various exposure-response models. The unit
3 risk estimates are derived from these exposure-response models and are thus similarly consistent
4 with the results of the NIOSH study, as long as they are used in the low-exposure range, as
5 intended. Because the exposure-response relationships for the cancers of interest in the NIOSH
6 study are generally supralinear, the unit risk estimates will overpredict the NIOSH results if
7 applied to exposure levels that correspond to the region of the exposure-response relationships
8 where the responses plateau.
9
10 l.b. Are there additional key published studies or publicly available scientific reports that are
11 missing from the draft document and that might be useful for the discussion of the
12 carcinogenic hazard ofEtO ?
13
14 SAB Panel Comment: The Panel agreed that the discussion of endogenous metabolic
15 production of ethylene oxide and the formation of background adducts should be expanded. The
16 Panel believed that the description of studies of DNA adduct formation resulting from EtO
17 exposure appears incomplete and superficial. This discussion should be expanded—both in
18 terms of the number of studies cited and the depth of the discussion. Since ethylene is
19 metabolized to EtO, some members recommended the inclusion of the ethylene body of literature
20 for consideration. Most members were hesitant about adding them to the document, but if added,
21 they cautioned that a discussion of the caveats associated with their interpretation relative to
22 ethylene oxide should be included.
23
24 EPA Response: The discussion of endogenous metabolic production of EtO and its significance
25 and contribution to the formation of background adducts in rodents and humans has been
26 expanded (Sections 3.3.2 and 3.3.3.1 and Section C.7 of Appendix C). The discussion of DNA
27 adduct formation resulting from EtO exposure has also been expanded to add depth and breadth
28 (Section 3.3.3.1 and Section C.I of Appendix C). Section C.I of Appendix C includes
29 discussion of general DNA adduct formation, sensitivity of the methods used to detect DNA
30 adducts, and DNA adduct studies, both in vitro and in vivo, that have been conducted in animals
31 and humans. A discussion of the endogenous production of ethylene during normal
32 physiological processes and its metabolism to EtO under certain conditions has been added
33 (Section C.7 of Appendix C). It should be noted that the endogenous production of EtO due to
34 the metabolism of endogenous ethylene will be present in all test animals or subjects (including
35 controls) and hence this factor is considered inherently in the analysis of effects of EtO exposure.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA agrees with the majority of the Panel that data on (exogenous) ethyl ene should not
2 be included in the assessment. One caveat provided on page 12 of the SAB report is that the
3 ethylene bioassays administered ethylene concentrations with such low EtO equivalents that they
4 would appear "to be below the limit of detection for a tumor response over the spontaneous
5 background in the F344 rat." Thus, the ethylene data would not be very informative for the EtO
6 assessment, for which there are already adequate EtO bioassays.
7
8 I.e. Do the available data and discussion in the draft document support the mode-of-action
9 conclusions?
10
11 SAB Panel Comment: The Panel agreed with the Draft Assessment conclusion of a mutagenic
12 mode of action. However, an expanded discussion of the formation of DNA adducts and
13 mutagenicity is warranted.
14
15 EPA Response: EPA has expanded the discussion of DNA adduct formation (Section 3.3.3.1
16 and Section C.I of Appendix C) and mutagenicity (Section 3.3.3 and Sections C.2-C.5 of
17 Appendix C) in the revised assessment document.
18
19 LtL Does the hazard characterization discussion for EtO provide a scientifically balanced and
20 sound description that synthesizes the human, laboratory animal, and supporting (e.g., in
21 vitro) evidence for human carcinogenic hazard?
22
23 SAB Panel Comment: While some members of the Panel found the hazard characterization
24 section of the Draft Assessment to be satisfactory, a majority expressed concerns that this section
25 did not achieve the necessary level of rigor and balance. An issue in this characterization,
26 particularly in the face of epidemiological data that are not strongly conclusive, is whether the
27 presumed precursor events leading to cancer in animals, such as mutations and/or chromosomal
28 aberrations, are observed in humans. This issue needs to be addressed in greater detail.
29
30 EPA Response: The genotoxicity (Section 3.3.3 and Appendix C), mode of action (Section
31 3.4.1), and hazard characterization (Section 3.5.1) sections have been revised to provide a more
32 complete and balanced discussion of EtO-induced precursor events in animals and humans. As
33 addressed in the EPA response under charge question 1 .a above, while the databases for humans
34 and rodents contain different types of studies, EPA did not find an inconsistency in EtO-induced
35 precursor events (Sections 3.3.3.2, 3.3.3.3, 3.4.1, and 3.5.1).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-5 DRAFT—DO NOT CITE OR QUOTE
-------
1
2 Issue 2: Risk Estimation (Section 4 and Appendices C and D of the EPA Draft Assessment)
3 Do the available data and discussion in the draft document support the approaches taken by
4 EPA in its derivation of cancer risk estimates for EtO? In your response, please include
5 consideration of the following:
6
1 2. a. EPA concluded that the epidemiological evidence alone was strong but less than
8 completely conclusive (although EPA characterized the total evidence—-from human,
9 laboratory animal, and in vitro studies—as supporting a conclusion that EtO is "carcinogenic
10 to humans"). Is the use of epidemiological data, in particular the Steenland et al. (2004);
11 Steenland et al. (2003) data set, the most appropriate for estimating the magnitude of the
12 carcinogenic risk to humans from environmental EtO exposures? Are the scientific
13 justifications for using this data set transparently described? Is the basis for selecting the
14 Steenland et al. data over other available data (e.g., the Union Carbide data) for quantifying
15 risk adequately described?
16
17 SAB Panel Comment: The Panel concurred that the NIOSH cohort is the best single
18 epidemiological data set with which to study the relationship of cancer mortality to the full range
19 of occupational exposures to EtO. That said, the Panel encouraged the EPA to broadly consider
20 all of the epidemiological data in developing its final Assessment. In particular, the Panel
21 encourages the EPA to explore uses for the Greenberg et al. (1990) data including leukemia and
22 pancreatic cancer mortality and EtO exposures for 2,174 Union Carbide workers from its two
23 Kanawha Valley, West Virginia facilities. (Also described in Teta et al., 1999; Teta et al., 1993).
24 The Panel encouraged the EPA to investigate potential instability that may result from
25 interaction between the chosen time metric for the dose response model and the treatment of time
26 in the estimated exposure (i.e., log cumulative exposure with 15 year lag) that is the independent
27 variable in that dose-response model.
28
29 EPA Response: EPA has revised the assessment to include an expanded discussion of study
30 selection, including a summary table of important considerations, in Section 4.1, as well as
31 expanded discussions of the exposure assessments for the Union Carbide (Appendix A, Section
32 A.2.20) and NIOSH (Appendix A, Section A.2.8) studies.
33 In regard to the possible use of other epidemiologic data, the assessment document
34 includes a detailed discussion of the studies of workers at the Union Carbide facilities in West
35 Virginia. The Greenberg et al. (1990) data are quite limited in terms of the number of cancers
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 observed. Teta et al. (1993) extended the follow-up of the Union Carbide cohort for an
2 additional 10 years and excluded the 278 chlorohydrin unit workers, in which a significant three-
3 fold excess of lymphohematopoietic cancer was observed (8 vs. 2.7 expected, SMR 2.94, see
4 Benson and Teta, 1993), on the grounds that the chlorohydrin unit workers were exposed to other
5 potential carcinogens and likely had low exposures to EtO. Teta et al. (1993) studied the
6 remaining 1,896 EtO production workers who did not work in the chlorohydrin unit. This cohort
7 is about a tenth of the size of the NIOSH cohort. This cohort did not show an excess of
8 lymphohematopoietic cancer (7 observed vs. 11.8 expected) but the cohort continues to be
9 limited by small numbers (e.g., fewer than 6 expected deaths for non-Hodgkin lymphoma
10 [NHL], although the exact number is not given). Furthermore, the Union Carbide study has a
11 less extensive exposure assessment than the NIOSH study. In part, the deficiency is inherent in a
12 chemical production setting, where it is difficult to find workers with relatively uniform work
13 histories that involve relatively constant exposure to EtO. As such, the exposure assessment
14 used in the Union Carbide study was relatively crude, based on just a small number of
15 department-specific and time-period-specific categories, and with exposure estimates for only a
16 few of the categories derived from actual measurements (see Section A.2.20 of Appendix A for
17 the details). This is in contrast to sterilization plants, where the NIOSH study was done, where
18 workers can be grouped into relatively common jobs/work zones, facilitating assignment of
19 exposure. Furthermore, extensive sampling data (2,350 measurements from 1975 to 1986,
20 reduced to 205 annual job-specific means, representing 80% of the data; another 20% were not
21 included but used as a validation sample) were used in the NIOSH study to estimate exposure in
22 different jobs and years. Such sampling data were not used in estimating exposures in the Union
23 Carbide cohort. Finally, the NIOSH regression model for estimating EtO exposure included data
24 not only on job/work zone, but also on variables such as size of sterilizer, type of product,
25 freshness of product, and exhaust systems for sterilizers. This regression model explained 85%
26 of the variance in the EtO validation data set. As a result, the exposure estimates in the NIOSH
27 study are likely to be more accurate. Because of the lack of comparability in the exposure
28 estimates across the two studies, it is not possible to group together the NIOSH cohort and the
29 Union Carbide cohort for a rigorous combined quantitative exposure-response analysis.
30 Teta et al. (1993) do not include any exposure-response analyses, but a later paper (Teta
31 et al., 1999) does. Teta et al. (1999) divide exposure into high, medium, and low intensity of
32 exposure and four time periods (1925-1939, 1940-1956, 1957-1973, and 1974-1988). The
33 paper does not give the exposure level assigned to each of the resulting 12 cells, nor any
34 justification for the chosen exposure levels. No published data describing how these estimates
35 were derived could be found.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 Teta et al. (1999) also do not provide the number of observed leukemia deaths, but
2 models leukemia as a function of exposure using three categories of cumulative exposure and a
3 variety of models using continuous exposure. Assuming, as indicated, that the data are the same
4 as the 1988 follow-up reported by Teta et al. (1993), there are only five observed leukemia
5 deaths which suggests that the extensive modeling of the data that was done is highly uncertain.
6 The published (through 2006) Union Carbide data and analyses were not sufficient for
7 dose-response assessment of lymphohematopoietic cancer due to small numbers and the inherent
8 problem posed by the general assignment of exposure levels to subjects, adequate details of
9 which were not provided.
10 Since the peer review, follow-up of the Union Carbide cohort, without the chlorohydrin
11 production workers, has now been extended through 2003, and analyses of the data have been
12 published by Swaen et al. (2009) and Valdez-Flores et al. (2010). Swaen et al. (2009) used an
13 exposure assessment based on the qualitative categorizations of potential EtO exposure in the
14 different departments developed by Greenberg et al. (1990) and time-period exposure estimates
15 from Teta et al. (1993). These are the same generalized exposure estimates described above
16 based on a small number of department-specific and time-period-specific categories, and with
17 exposure estimates for only a few of the categories derived from actual measurements (additional
18 detailed discussion is provided in Section A.2.20 of Appendix A of the final assessment
19 document). At the end of the 2003 follow-up, only 27 lymphohematopoietic cancer deaths
20 (including 12 leukemias and 11 NHLs) were observed in the cohort. Thus, even after extended
21 follow-up, the number of cases is small compared to the NIOSH study, which had 74
22 lymphohematopoietic cancer deaths, 53 from lymphoid cancers. More importantly, as discussed
23 above, the exposure assessment is much more rudimentary than that used for the NIOSH cohort.
24 The lack of comparability in the exposure estimates precludes a rigorous combined exposure-
25 response analysis of data from the two cohorts.
26 EPA requested that Professor Kyle Steenland, the principal investigator of the NIOSH
27 study, respond to the following excerpt from this comment from the SAB Panel:
28
29 "The Panel encouraged the EPA to investigate potential instability that may result from
30 interaction between the chosen time metric for the dose response model and the treatment of
31 time in the estimated exposure (e.g. log cumulative exposure with 15 year lag) that is the
32 independent variable in that dose-response model."
33
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-8 DRAFT—DO NOT CITE OR QUOTE
-------
1 Professor Steenland's response:
2
3 "This comment is difficult to understand, but appears to be a concern that the 15-year lag in the
4 exposure metric, which discounts the most recent exposure, may cause an over-reliance in the
5 exposure-response analysis on exposures which were estimated prior to 1979, which possibly are
6 less accurate. The reason they may be less accurate is because the NIOSH exposure model
7 assumed that the effect of calendar year was constant before 1979. There are a couple of
8 comments to be made here. First, it is certain the much higher exposures took place before the
9 early 1980s when engineering controls were implemented, and that these exposures are likely to
10 compose the majority of the metric "cumulative exposure." Second, such early exposures would
11 often, but not always, also be more biologically relevant than later exposures, given that there is
12 likely to be some latency period before a given exposure causes a cancer (the best fitting lag was
13 15 years in the analysis), and cancers occurred during the period 1980-2004, so that later lower
14 exposures were often discounted by the lag. But were such early exposures estimated
15 appreciably worse than later exposures by the NIOSH regression model? The NIOSH regression
16 model was based on seven variables, one of which had 8 levels (job), one of which had 5 levels
17 (product types), and one of which was time or year. All these variables were statistically
18 significant at the p< 0.05 level except one (aeration), which had ap value of 0.10. Given that
19 engineering controls were included in the model, the effect of calendar year was thought to
20 reflect improved work practices which got better year by year as employees and managers
21 became more conscious of the dangers of exposure. The effect of year only began in 1979, and
22 was not apparent in the period 1975-1978 when there much less concern about the dangers of
23 EtO. It would seem logical that prior to 1975 (when there were no sampling data to include in
24 the model), work practices also would have changed little year to year, given that worker and
25 management concern about the dangers of EtO was minimal or nonexistent. Furthermore, data
26 for the other variables in the model were available for years before 1979, and hence were able to
27 play a role in prediction of EtO prior to 1979, independent of the year effect, which was constant
28 prior to 1979. Hence, the model would be expected to perform reasonably well in the period
29 before sampling data were available, i.e., prior to 1975, regardless of the assumption that
30 calendar year had no effect independent of the other variables in the model."
31
32 "In summary, there is obviously more uncertainty about the estimation of exposures prior to
33 1975 when there were no sampling data. This uncertainty is of some concern in the sense that
34 the majority of cumulative exposure metric for most workers is probably contributed by earlier,
35 higher exposures. The use of a 15-year lag does not, however, necessarily increase this
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-9 DRAFT—DO NOT CITE OR QUOTE
-------
1 uncertainty, given that exposure in the lagged out period for most workers would be appreciably
2 lower than exposure before the lag came into effect. Furthermore, while the validity of the
3 NIOSH estimates before 1975 cannot be tested against sampling data, the NIOSH model would
4 be expected to permit reasonable estimation of exposure prior to 1975 based on other variables in
5 the model (job, type of product, size of sterilizer, exhaust of sterilizer, etc)."
6
7 "What if exposures prior to 1975 were estimated poorly? This raises the general question of
8 measurement error, which is more likely to have occurred in years before sampling data existed.
9 Measurement error is a complicated issue and its effects cannot be easily predicted. It does not
10 seem likely that the use of the 15-year lag, however, would appreciably increase whatever
11 measurement error occurred for early years of exposure before 1975. While it is possible that the
12 EPA should formally evaluate the likely effect of measurement error, this is a large task which
13 would take considerable amount of time and would necessarily depend on a large number of
14 assumptions about the error in the period before sampling data existed (as I have argued, it is
15 also largely independent of the use of a 15-year lag)."
16
17 2.b. Assuming that Steenland et al. (2004); Steenland et al. (2003) is the most appropriate
18 data set, is the use of a linear regression model fit to Steenland et al. 's categorical results for
19 all lymphohematopoietic cancer in males in only the lower exposure groups scientifically and
20 statistically appropriate for estimating potential human risk at the lower end of the observable
21 range? Is the use of the grouping of all lymphohematopoietic cancer for the purpose of
22 estimating risk appropriate? Are there other appropriate analytical approaches that should be
23 considered for estimating potential risk in the lower end of the observable range? Is EPA's
24 choice of a preferred model adequately supported and justified? In particular, has EPA
25 adequately explained its reasons for not using a quadratic model approach such as that of
26 Kirman et al. (2004)? What recommendations would you make regarding low-dose
27 extrapolation below the observed range?
28
29 SAB Panel Comment: The Panel identified several important shortcomings in the linear
30 regression modeling approach used to establish the point of departure for low dose extrapolation
31 of cancer risk due to EtO. The Panel was unanimous in its recommendation that the EPA
32 develop its risk models based on direct analysis of the individual exposure and cancer outcome
33 data for the NIOSH cohort rather than the approach based on published grouped data that is
34 presently used. The suggested analysis will require EPA to acquire or otherwise access
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-10 DRAFT—DO NOT CITE OR QUOTE
-------
1 individual data and develop appropriate methods of analysis. The panel recommends that the
2 Agency allocate the appropriate resources to conduct this analysis.
3 The Panel was divided on whether low dose extrapolation of risk due to environmental
4 EtO exposure levels should be linear (following Cancer Guideline defaults for carcinogenic
5 agents operating via a mutagenic mode of action) or whether plausible biological mechanisms
6 argued for a nonlinear form for the low dose response relationship. With appropriate discussion
7 of the statistical and biological uncertainties, several Panel members strongly advocated that both
8 linear and nonlinear calculations be considered in the final EtO Risk Assessment.
9 In conjunction with its recommendation to use the individual NIOSH cohort data to
10 model the relationship of cancer risk to exposures in the occupational range, the Panel
11 recommended that the Agency explore the use of the full NIOSH data set to estimate the cancer
12 slope coefficients that will in turn be used to extrapolate risk below the established point of
13 departure. The use of different data to estimate different dose response curves should be avoided
14 unless there is both strong biologic and statistical justification for doing so. The Panel believed
15 this justification was not made in the Agency's draft assessment.
16 Although the analysis based on total lymphohematopoietic (LH) cancers might have
17 value as part of a complete risk assessment, the rationale for this aggregate grouping needs to be
18 better justified. The Panel recommends that data be analyzed by subtype of LH cancers (e.g.
19 lymphoid, myeloid) and strong consideration be given to these more biologically justified
20 groupings as primary disease endpoints.
21 The Panel was divided in its views concerning the appropriateness of estimating the
22 population unit risk for LH cancer based only on the NIOSH data for males. Several Panel
23 members pointed out that a standard approach in cancer epidemiology and risk analysis begins
24 by conducting separate dose-response analyses on males and females and combining the data
25 only if the results are similar. Conducting separate analyses for males and females is also the
26 standard practice when analyzing data from animal carcinogenicity bioassays. A second
27 approach to dealing with the possibility of gender differences in response is to include gender as
28 a fixed effect in the statistical modeling of the data and determine whether gender or its
29 interaction with other predictors (e.g., gender x exposure) are significant explanatory variables.
30 If so, the combined model with the estimated gender effects could be used directly or separate,
31 gender-specific dose response analysis would be performed. If not, the gender effects could be
32 dropped and the model re-estimated for the combined male and female data. In addition, the
33 Agency should test whether the male/female differences are mitigated by use of alternate disease
34 endpoints discussed in the previous paragraph.
35
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-11 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA Response: The above comment from the Panel addresses a variety of issues and EPA's
2 responses to some of these issues are comparatively detailed; thus, EPA has subdivided its
3 response into separately titled subsections to make it easier to read.
4
5 EPA Response on the modeling of the individual-level data: In response to the SAB
6 comments, EPA conducted considerable additional analysis using the individual-level
7 (continuous) exposure and cancer outcome data for the NIOSH cohort. These analyses are
8 described in Section 4.1.1.2 for lymphoid cancer modeling and Section 4.1.2.3 for breast cancer
9 incidence modeling. These Sections also include summary tables of the key models examined
10 (Table 4-4) and the factors considered in model selection (Table 4-12). More details on the
11 various models and the model results are provided in Appendix D.
12 The underlying problem that makes the EtO datasets from the NIOSH cohort difficult to
13 model (for the purposes of environmental risk assessment) is that the exposure-response
14 relationships, particularly for lymphoid cancer and breast cancer mortality, are supralinear, i.e.,
15 the responses rise relatively steeply at low exposures and then attenuate or "plateau."
16 Supralinear exposure-response relationships are inherently difficult to model for the purposes of
17 environmental risk assessment, i.e., to estimate risk at low exposures, because the standard
18 single-parameter exposure-response models tend to exaggerate the low-exposure slope in order
19 to simultaneously fit the plateauing at higher exposures. One approach attempted by EPA, in
20 consultation with Dr. Steenland, to address this difficulty was to use two-piece spline models,
21 allowing for the lower exposure and higher exposure data to be fit with different spline segments.
22 For the breast cancer incidence data, EPA was able to develop several continuous models
23 that provided reasonable fits to the individual-level exposure data across the entire range of the
24 data, consistent with the SAB recommendations. The best-fitting of these models, the two-piece
25 linear spline model, now forms the basis for EPA's unit risk estimate for breast cancer incidence
26 (Section 4.1.2.3).
27 For lymphoid cancer, however, despite the extensive modeling efforts, the various
28 alternative continuous models investigated, including the two-piece spline models, proved
29 problematic, as explained in detail in the text (Section 4.1.1.2). In particular, the adequately
30 fitting models predicted extremely steep slopes in the low-dose region. In consideration of these
31 results, EPA has retained the approach used in the Draft Assessment and has based the risk
32 estimates for lymphoid cancer on a linear regression using the categorical data, excluding the
33 highest exposure group.
34 While EPA understood and appreciated the SAB's recommendation and did much work to
35 model the individual-level data for lymphoid cancer, it should be noted that modeling of grouped
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-12 DRAFT—DO NOT CITE OR QUOTE
-------
1 data is also an important and well-recognized statistical methodology and its use is consistent with
2 EPA guidance, policy, and past practice. For example, EPA's 2005 Guidelines for Carcinogen Risk
3 Assessment specifically recognize the use of linear modeling of grouped epidemiological data ("For
4 epidemiologic studies, including those with grouped data, analysis by linear models in the range of
5 observation is generally appropriate unless the fit is poor.", p. 3-11). In addition, EPA's approach
6 of using a weighted linear regression through the categorical relative risk estimates follows
7 established statistical procedures (van Wijngaarden and Hertz-Picciotto, 2004; Rothman, 1986).
8 With regard to modeling without the high-dose category, the lymphoid cancer data show
9 a rise and then plateauing of response such that an overall linear relationship is not an
10 appropriate description of the exposure-response relationship across the entire exposure range, in
11 particular in the low-exposure region of interest for the derivation of low-exposure risk
12 estimates. Restricting the linear regression to the lower categorical exposure groups provides a
13 better representation of the exposure-response relationship in that lower exposure region. EPA's
14 Benchmark Dose Technical Guidance (U.S. EPA, 2012) recognizes analyses omitting high-dose
15 data points, when these data are not compatible with the development of suitable descriptive
16 statistical analyses, as a viable analytical approach.
17 The breast cancer mortality data displayed similar extreme supralinearity, and the two-
18 piece spline model yielded an unrealistically steep low-dose slope estimate; thus, EPA again
19 used a linear regression of the categorical data, excluding the highest exposure group (Section
20 4.1.2.2). The breast cancer mortality data, however, are not critical to the assessment because
21 the breast cancer incidence data set is preferred (Section 4.1.2.3).
22
23 EPA Response on the use of a nonlinear approach to low-exposure extrapolation:
24 EPA has given careful consideration to the range of perspectives provided in the SAB report on
25 the issue of low-dose extrapolation, including the viewpoint expressed by several Panel members
26 who advocated that both linear and nonlinear calculations be considered in the EtO assessment.
27 It is EPA's judgment, as detailed below, that the inclusion of a nonlinear approach is not
28 warranted.
29 As discussed in Chapter 3 of the assessment, EtO is a DNA-reactive, mutagenic, multi-
30 site carcinogen in humans and laboratory animal species; as such, it has the hallmarks of a
31 compound for which low-dose linear extrapolation is strongly supported. EPA's Guidelines for
32 Carcinogen Risk Assessment (U.S. EPA, 2005a) specifically note the use of low-dose linear
33 extrapolation for "agents that are DNA-reactive and have direct mutagenic activity." By
34 comparison, the Guidelines recommend that, "A nonlinear approach should be selected when
35 there are sufficient data to ascertain the mode of action and conclude that it is not linear at low
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-13 DRAFT—DO NOT CITE OR QUOTE
-------
1 doses and the agent does not demonstrate mutagenic or other activity consistent with linearity at
2 low doses." EPA's analysis indicates that EtO does not meet any of those conditions. ForEtO,
3 there is sufficient weight of evidence to support a mutagenic/genotoxic MO A, without evidence
4 of additional or alternative MO As being operative (Section 3.4.1).
5 EPA specifically considered a proposed 2-hit MOA hypothesized to support a (nonlinear)
6 quadratic model for lymphohematopoietic cancer (leukemia specifically) and concluded that the
7 evidence for this MOA was inadequate, as discussed in detail in Section 3.4 of the assessment.
8 Appendix A of the SAB report also provides more general evidence for why a 2-hit process does
9 not imply a quadratic exposure-response relationship for leukemia at low exposures.
10 With regard to the particular comments of the SAB members advocating presentation of a
11 nonlinear approach, the reasons for using such an approach presented in Appendix C of the SAB
12 report were largely that (1) DNA adducts may show a nonlinear response when identical adducts
13 are formed endogenously and (2) mutations do not have linear relationships with exposure but
14 exhibit an "inflection point." However, recent data from Marsden et al. (2009) support a linear
15 exposure-response relationship for EtO exposure and DNA adducts (p < 0.05) and demonstrate
16 increases of DNA adducts from exogenous EtO exposure above those from endogenous EtO for
17 very low exposures to exogenous EtO, as discussed in detail in the assessment (Section 3.3.3.1
18 and 4.5), providing direct evidence against the first reason proposed in support of a nonlinear
19 approach in Appendix C of the SAB report. In support of the second reason, Appendix C of the
20 SAB report presents two EtO-specific mutation datasets; however, EPA's analysis of these
21 datasets, summarized below, finds that they are in fact consistent with low-dose linearity. In
22 summary, EPA's review of studies addressing dose-response patterns for adduct formation and
23 mutagenesis by EtO finds these data to be supportive of the inferences made in the EtO
24 assessment (and more broadly in EPA's Guidelines for Carcinogen Risk Assessment) regarding
25 the plausibility of linear, nonthreshold, low-dose dose-response relationships for the biological
26 effects of EtO, which is mutagenic and directly damages DNA.
27 EPA further notes that the supralinear exposure-response relationships from the NIOSH
28 data at low occupational exposures argue against the existence of a "threshold," practical or
29 otherwise, at exposure levels anywhere near the POD. Also, the rodent bioassays do not suggest
30 an absence of increased cancer risk at their lowest exposure levels.
31
32 Analysis of the EtO mutagenicity datasets presented in Appendix C of the SAB Report:
33
34 Appendix C in the SAB report provides slides (numbers 25 and 26) showing dose-
35 response data for hprt mutations in mice exposed to either EtO or to ethylene. For ethylene, a
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-14 DRAFT—DO NOT CITE OR QUOTE
-------
1 model estimate of an EtO-equivalent concentration was used to represent metabolism of ethylene
2 to EtO. In both cases, mutations at the hprt locus of T-cells isolated from spleens of Big Blue
3 mice were quantified. The EtO study results are from Walker et al. (1997), and it appears that
4 the ethylene results are derived from experiments presented in Walker et al. (2000). In the latter
5 case, there are some differences in the estimated EtO equivalents and the hprt mutation
6 frequencies between the values given in the slide and those reported by Walker et al. (2000). We
7 performed statistical analyses using the data presented in slide 26 of Appendix C.
8 To examine these data, we first analyzed the EtO dataset (Walker et al., 1997) using
9 maximum likelihood estimation (MLE). We then looked at the consistency of the ethylene
10 dataset (Walker et al., 2000) with the EtO dataset. The EtO data were fit with a linear model
11 utilizing a log-normal distribution of the individual animal response measurements due to the
12 low mutant frequency that causes skewness of the data. As shown in Figure H-l, this model
13 provided an adequate fit to the EtO data (open circles represent individual animal data for the
14 EtO exposures; model goodness-of-fit/> = 0.09; variance fit assuming homogeneous variance in
15 log scale,/? = 0.64). The MLE of the model is plotted (geometric mean [solid line] as an
16 estimation of the median response along with the lower and upper 2.5 percentiles of the model
17 [dashed lines]). The second, ethylene-derived, dataset is plotted on the same graph (closed
18 circles). The predicted EtO-equivalents from the ethylene dataset fall well below the lowest dose
19 level used in the EtO experiment, a range in which the EtO-based model would predict only a
20 small response (i.e., no more than a 25% increase in mutation rate above background, a level that
21 cannot be expected to be detectable given the variability in the EtO experimental data; see Figure
22 H-l). The fact that the ethylene results did not show measureable increases in hprt mutations is
23 consistent with the modeled EtO results.
24 Note, however, that all medians of the ethyl ene-derived data are at or below the EtO-
25 based model and one of the points is below the lower 2.5 percentile of the model, indicating that
26 this point is unlikely to be consistent with the same model. To further investigate the
27 compatibility of the data from the two experiments, we analyzed the combined dataset by
28 including a term that represents the source of the data (the EtO vs. ethylene experiments) into the
29 modeling (as above). This experimental variable was significant (p < 0.05), indicating that there
30 is a systematic difference in response between the EtO and ethyl ene-derived data. As a further
31 check, we refit the data using an exponential model that provided a MLE fit with a degree of
32 upward curvature (but still having low-dose linear behavior). Using a categorical experimental
33 variable within this experiment also indicated a systematic dependence of results on data source
34 (EtO vs. ethylene), indicating that this finding was not dependent on the choice of a straight-line
35 dose-response model. As an additional sensitivity analysis, we reran the modeling using the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-15 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
values of EtO equivalents from ethylene exposure and hprt results directly from Walker et al.
(2000) (rather than the values shown in the SAB Appendix C slide); the modeling results were
essentially unchanged. Accordingly, we conclude that it is not appropriate to combine the
ethylene data with EtO data in evaluating dose-response relationships for the hprt mutations.
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
20
o
X
cr
ID
it
10
Q_
I
50
100 Ethylene Oxide (ppm) 150
200
Figure H-l. Induction of hprt mutations by EtO (open circles and modeled
fit) with data from ethylene (using estimated EtO equivalents) shown (solid
circles). Source: SAB (2007) Appendix C (slides 25 and 26); original
experiments of Walker et al. (1997).
Slide 27 of the SAB report presents data from Nivard et al. (2003) on the frequency of
recessive lethal (RL) mutations in Drosophila exposed to EtO (full data set presented in Vogel
and Nivard, 1998). Plotting of mutation rate versus EtO concentration for wild-type Drosophila
on non-log-transformed axes shows a downward curving (supralinear) relationship indicating
greater potency of EtO (per unit exposure) at low exposures as compared with high exposures
(Figure H-2). These data are adequately fit by a Michaelis-Menten-type relationship (downward
curving, linear at low dose); the fit is somewhat improved with a fractional power Hill model,
which would indicate even steeper low-dose response.
In conclusion, our review of the EtO mutagenicity data presented in Appendix C of the
SAB report finds that these data do not show a disproportionate fall-off of mutagenic effects at
low doses of EtO; that is, they do not indicate a low-dose nonlinear or threshold-type dose-
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-16 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
response pattern. Thus, our review finds these data to be supportive of the inferences made in
the assessment (and more broadly in EPA's 2005 Guidelines for Carcinogen Risk Assessment)
regarding the plausibility of linear, nonthreshold, low-dose dose-response relationships for the
carcinogenic effects of EtO, which is mutagenic and directly damages DNA.
CD
^
CD
Q.
>i
O
C
CD
D
CT
C
o
0
200
400
600
800
1000
Ethylene oxide concentration, ppm
Data: Nivard (2003)/Vogel and Nivard (1998). Model: Michaelis Menten
Figure H- 2. Induction of recessive lethal mutations by EtO in Drosophila
(wild-type). Standard deviations are calculated as the square root of the number
of mutations, assuming a Poisson distribution, and plotted as ± (SD x percent
mutation frequency).
EPA response on using different data to estimate different dose-response curves:
With respect to using different data to estimate different dose-response curves, that Panel
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
H-17 DRAFT—DO NOT CITE OR QUOTE
-------
1 comment pertains only to the occupational exposure scenarios. This is addressed in EPA's
2 response to the SAB comment on charge question 2.d below.
O
4 EPA response on lymphohematopoietic cancer groupings: As recommend by the
5 Panel, the primary risk estimates are now based on the lymphoid cancers (Section 4.1.1.2).
6 Analysis based on total lymphohematopoietic cancers is also included for completeness and
7 comparison purposes.
8
9 EPA response on the use of only the male data for lymphohematopoietic cancers:
10 Analyses by Dr. Steenland determined that there was not a statistically significant difference
11 between the lymphohematopoietic cancer results for males and females. Thus, in the revised
12 assessment, unit risk estimates based on lymphohematopoietic cancer in males only are not used.
13 Unit risk estimates are now based on lymphoid cancers for males and females combined and
14 breast cancer in females.
15
16 The following additional comments on page 31 of the SAB Panel report under "2.b.
17 Methods of Analysis," "7. Statistical issues," are quoted verbatim below followed by EPA's
18 responses:
19
20 SAB Panel Comment:
21 7. Statistical issues
22 Pages 29-49 of the draft Evaluation outline the EPA's proposed approach to estimation of the
23 Inhalation Unit Risk for EtO. In addition to the general issues of estimation and model-based
24 extrapolation described above, there are a number of statistical assumptions and methods used in
25 this approach that deserve mention. Conditional on the cancer slope factor results from the
26 weighted least squares regression analysis, the life table (BEIRIV) approach to the
27 determination of the LEC01 is programmed correctly. The life table methodology that is the
28 basis for the BEIR IV algorithm is designed to estimate excess mortality and is not readily
29 adapted to modeling excess risk for events (incidence) that do not censor observation on the
30 individual in population under study. The methodology for substituting the mortality slope to an
31 excess risk computation for HL cancer incidence requires the assumption of a proportional rate
32 of incidence/mortality across the cancer types that are included in the grouped analysis. This is
33 generally not a viable assumption. The Panel therefore discourages the use of the BEIR IV
34 algorithm for extrapolation of the cancer mortality algorithm to estimation of excess cancer
35 incidence.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-18 DRAFT—DO NOT CITE OR QUOTE
-------
1 Several Panel members commented on the use of the upper confidence limit for the
2 estimated slope coefficient as the basis for estimating an LECoi. The Panel encourages the EPA
3 to present unit risk estimates based on the range of ECoi values corresponding to the lower 95%
4 confidence limit, the point estimate, and the upper 95% confidence limit for the estimated cancer
5 slope coefficients from the final dose-response models.
6
7 EPA Response on using the BEIR approach to estimate incidence risks: In this assessment
8 EPA is developing estimates of the risk of cancer incidence, not mortality, as the cancers
9 associated with EtO exposure (lymphohematopoietic, in particular lymphoid, and breast cancers)
10 have substantial survival rates. The breast cancer incidence estimates are not at issue here
11 because they are based on incidence data. Regarding the lymphoid cancers, the SAB provided
12 the relevant comment that mathematically the BEIR formula would apply to the case where there
13 is a proportional rate of incidence/mortality across the cancer types that are included in the
14 grouped analysis. EPA considered this in its application of the BEIR formula. The fact that the
15 ratios of incidence to mortality are not strictly proportional contributes some uncertainty to the
16 incidence estimates for the grouping of lymphoid cancers, but not a large amount. Uncertainties
17 in using the life-table analysis approach to seek to develop reasonable estimates for incidence
18 risk, including those noted by the SAB, are acknowledged in the assessment, and the impact of
19 nonproportionality between cancer types is one of the uncertainties discussed (Section 4.1.1.3).
20 As illustrated in the assessment, these uncertainties do not have a major impact on the final risk
21 estimates. The incidence unit risk estimate is about 120% higher than (i.e., 2.2 times) the
22 mortality-based estimate, which is consistent with the relatively high survival rates for lymphoid
23 cancers. Potential concern that the incidence estimates might be overestimated would come
24 primarily from the inclusion of multiple myeloma, because that subtype has the lowest
25 incidence:mortality ratios (and, thus, if that subtype were driving the increased mortality
26 observed for the lymphoid cancer grouping, then including the incidence rates for the other
27 subtypes, which have higher incidence:mortality ratios, might inflate the incidence estimates).
28 Multiple myelomas, however, constitute only 25% of the lymphoid cancer cases, and there is no
29 evidence that multiple myeloma is driving the EtO-induced excess in lymphoid cancer mortality
30 (25% is below the proportion of multiple myeloma deaths one would expect in the cohort based
31 on age-adjusted background mortality rates of multiple myeloma, NHL, and chronic lymphocytic
32 leukemia, and these 3 subtypes have the same pattern of mortality rates increasing as a function
33 of age mostly above age 50, so the comparison with lifetime background rates is reasonable).
34 Thus, using the total lymphoid cancer incidence rates is not expected to result in an
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-19 DRAFT—DO NOT CITE OR QUOTE
-------
1 overestimation of the incidence risk estimates; if anything, the incidence risks would likely be
2 diluted with the inclusion of the multiple myeloma rates.
3 EPA decided that the Panel's suggestion to not use the BEIR approach for development
4 of cancer incidence estimates for lymphohematopoietic cancer would not allow EPA to develop
5 the desired cancer incidence risk estimates. One possible alternative approach involving a crude
6 survival adjustment to the mortality-based estimates would yield results with greater uncertainty
7 than use of the BEIR approach. No alternative approaches were identified by the SAB. In the
8 absence of an appropriate alternative approach to estimate risks of cancer incidence, EPA has
9 retained the application of the BEIR approach, which it judges to provide a reasonable estimate
10 of incidence risks. EPA recognizes the uncertainties and assumptions outlined by the Panel and
11 has expanded the discussion of these in the carcinogenicity assessment (Section 4.1.1.3).
12 However, EPA notes that deriving mortality estimates as the sole cancer risk estimates for
13 lymphohematopoietic cancer would substantially underestimate cancer risk. In addition, EPA
14 presents the mortality-based estimates as well for comparison, and as discussed above, the
15 lymphoid cancer incidence unit risk estimate is about 120% higher than (i.e., 2.2 times) the
16 mortality-based estimate, which is considered reasonable, given the high survival rates for
17 lymphoid cancers.
18
19 EPA Response on the use of upper and lower confidence limits: In the EtO assessment, EPA
20 presents 95% (one-sided) lower bounds and central estimates of the ECois as well as standard
21 errors for the regression coefficients used in the modeling, which provide information about the
22 variability in the modeled slope estimate. EPA's Guidelines for Carcinogen Risk Assessment
23 also recommend the calculation of a 95% upper bound on the central estimate (in this case the
24 ECoi) related to the POD "to the extent practicable" (U.S. EPA, 2005a, p. 1-14), and such a
25 value has been added for the selected breast cancer incidence model (Section 4.1.2.3, Table 4-7,
26 footnote i, based on the profile likelihood confidence limits for the regression coefficient).
27 However, for the linear regression model used as the basis for the lymphoid cancer unit risk
28 estimate, it was not practicable to calculate such a value, as it was undefined. Although there
29 were models for lymphoid cancer from which upper bounds could have been calculated, the
30 linear regression model was selected as the basis for the POD for the expressed purpose of
31 obtaining a realistic slope estimate for the low-exposure region (Section 4.1.1.2) and not for
32 providing a realistic upper bound estimate for the ECoi.
33 EPA considered the SAB Panel comment encouraging the EPA "to present unit risk
34 estimates based on the range of ECoi values corresponding to the lower 95% confidence limit,
35 the point estimate, and the upper 95% confidence limit." However, as a consequence of the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-20 DRAFT—DO NOT CITE OR QUOTE
-------
1 2-step approach used by EPA to generate cancer potency estimates from a POD rather than
2 directly from the statistical model used to estimate the POD, potency estimates below the
3 response level corresponding to the POD are no longer associated with the statistical model.
4 While linear extrapolation from a POD that is the 95% (one-sided) lower bound on the central
5 estimate of the exposure concentration associated with the selected (benchmark) response level
6 (e.g., the LECoi) might be generally expected to yield a reasonable upper bound on cancer risk
7 for that data set (though not strictly a statistical "95%" upper bound), estimates involving a linear
8 extrapolation from the upper bound on that central estimate are not generally meaningful and
9 could be misleading if they are mistaken for lower bounds on potency, as the actual exposure-
10 response relationship may exhibit some sublinearity below the response level corresponding to
11 the POD. Thus, it has not been EPA practice to develop such potency estimates, and EPA did
12 not undertake to develop any for this assessment.
13
14 2.c. Is the incorporation of age-dependent adjustment factors in the lifetime cancer unit risk
15 estimate, in accordance with EPA's Supplemental Guidance (U.S. EPA, 2005b), appropriate
16 and transparently described?
17
18 SAB Panel Comment: In accordance with EPA guidance, the Draft Assessment applied an Age
19 Dependent Adjustment Factor (ADAF) to adjust the unit risk for early life exposure. While the
20 majority of the Panel felt that the application of a default value by the Agency was appropriate
21 due to lack of data, the description in the Draft Assessment was not adequate, particularly for
22 those not familiar with the EPA's Supplemental Guidance.
23
24 EPA Response: EPA has added a new subsection (Section 4.4) detailing the application of the
25 ADAFs.
26
27 2.d. Is the use of different models for estimation of potential carcinogenic risk to humans
28 from the higher exposure levels more typical of occupational exposures (versus the lower
29 exposure levels typical of environmental exposures) appropriate and transparently described
30 in Section 4.5?
31
32 SAB Panel Comment: While the method was transparently described, most of the Panel did not
33 agree with the estimation based on two different models for two different parts of the dose
34 response curve (see response to 2b). The use of different data to estimate different dose response
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-21 DRAFT—DO NOT CITE OR QUOTE
-------
1 models curves should be avoided unless there is both strong biological and statistical justification
2 for doing so. The Panel believed this justification was not made in the Agency's draft report.
O
4 EPA Response: For the breast cancer incidence risk estimates, a single model, the 2-piece linear
5 model is now recommended for the occupational exposure scenarios. The 2-piece linear model
6 is a unitary model comprised of two linear pieces or segments with different slopes that are
7 joined at a point referred to as a "knot." The 2-piece linear model has the flexibility to represent
8 situations, such as with EtO, where the relationship between exposure level and response
9 changes over the range of exposure. For lymphoid cancer risk estimates, two models are
10 presented for the lower-exposure exposure scenarios, but just one of the models is recommended
11 for the higher-exposure exposure scenarios; users have the option of using a single model across
12 the range of exposure scenarios or of transit!oning across models, depending on the exposure
13 scenarios of interest, and some guidance on choice of approach is provided in Section 4.7 of the
14 revised assessment. As discussed in the assessment, the log-cumulative exposure model, which
15 provides a good fit to the data in the plateau and is suitable for exposure scenarios with
16 cumulative exposures in that region, is not appropriate for the low-exposure region (i.e., below
17 the range of the occupational scenarios presented in this assessment) because such a steep
18 increase in slope is considered to be biologically implausible and the good statistical global fit of
19 the model should not be over-interpreted to infer that the model provides a meaningful fit to the
20 low-exposure region. Likewise, the linear regression used to model the lower-dose exposure
21 groups is not intended to reflect the exposure-response relationship in the higher-exposure
22 region. Hence, for lymphoid cancer, the use of both models may be required to cover a range of
23 occupational exposure scenarios. Table 4-19 of the assessment shows how results from the two
24 models compare over a range of exposure scenarios for which either model might be used.
25
26 2.e. Are the methodologies used to estimate the carcinogenic risk based on rodent data
27 appropriate and transparently described? Is the use of "ppm equivalence" adequate for
28 interspecies scaling of EtO exposures from the rodent data to humans?
29
30 SAB Panel Comment: The ppm equivalence method is a reasonable approach for interspecies
31 scaling of EtO exposures from rodent data to humans. If the use of animal data becomes more
32 important (i.e., the principal basis for the ethylene oxide unit risk value), more sophisticated
33 approaches such as PBPK modeling should be considered.
34
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-22 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA Response: EPA notes the Panel's support for the use of the ppm equivalence method. As
2 the unit risk value is based on human data, the use of more sophisticated models is not necessary.
O
4 Issue 3: Uncertainty (Sections 3 and 4 of the EPA Draft Assessment)
5 EPA's Risk Characterization Handbook requires that assessments address in a transparent
6 manner a number of important factors. Please comment on how well this assessment clearly
1 describes, characterizes and communicates the following:
8 a. The assessment approach employed;
9 b. The use of assumptions and their impact on the assessment;
10 c. The use of extrapolations and their impact on the assessment;
11 d. Plausible alternatives and the choices made among those alternatives;
12 e. The impact of one choice versus another on the assessment;
13 / Significant data gaps and their implications for the assessment;
14 g. The scientific conclusions identified separately from default assumptions and policy calls;
15 h. The major risk conclusions and the assessor's confidence and uncertainties in them; and
16 i. The relative strength of each risk assessment component and its impact on the overall
17 assessment.
18
19 SAB Panel Comment: The Panel's report contained specific responses to charge questions 1
20 and 2. The report did not contain specific responses to question 3 and instead contained the
21 following statements regarding question 3:
22
23 "The Panel has responded to Charge Questions 1 and 2 and has tried to incorporate their
24 comments regarding Charge Question 3 within those responses. A separate response for
25 Charge Question 3 was not deemed necessary since issues of uncertainty were addressed
26 in the responses to charge questions 1 and 2."
27
28 The following are detailed comments on the regression modeling used in the draft ethylene
29 oxide assessment quoted from the SAB Ethylene Oxide Panel report and the EPA response:
30
31 SAB Panel Comment:
32 2. Linear regression model for categorical data
33
34 The Panel identified several important shortcomings in the linear regression modeling
35 approach used to establish the point of departure for low dose extrapolation of cancer risk due to
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-23 DRAFT—DO NOT CITE OR QUOTE
-------
1 EtO. Based on its review of the methods and results presented at the January 17,18, 2007
2 meeting, the Panel was unanimous in its recommendation that the EPA develop its risk models
3 based on direct analysis of the individual exposure and cancer outcome data for the NIOSH
4 cohort. The Panel understands that these data are available to EPA analysts upon request to the
5 CDC/NIOSH. The Panel recognizes the burden that a reanalysis of the individual data places on
6 the EPA ORD staff but given the important implications of the risk assessment, this burden is
7 well justified to achieve the best scientific and statistical treatment of all the available
8 epidemiological data.
9 The following paragraphs present the statistical basis for the Panel's assessment of the
10 linear regression model approach and the use of categorized exposure and outcome data.
11 The approach described in the Draft Assessment uses a model based on categories
12 defined by cumulative exposure ranges for male subjects in the NIOSH cohort. Steenland et al.
13 identified several models that provide a significant (p < 0.05) fit to the exposure data; however,
14 the EPA has elected to use model-based relative rate parameter estimates for categories of 15
15 year lagged, cumulative exposure. In Steenland et al. (2004) this model was not one that
16 provided a significant fit to the NIOSH data (p = 0.15 for the likelihood ratio test of P = [01, 02,
17 03, 04] = 0). The use of the weighted least squares regression fit of a linear regression line
18 through the three data points defined by the estimated rate ratios and mean cumulative exposures
19 for the first three exposure categories of the Steenland et al. 15 year lag, cumulative exposure
20 category model is not a robust application of this technique. The Panel identified four
21 weaknesses in the approach.
22 a) Model-based dependent variable: The dependent variables are model-based estimates
23 of rate ratios for exposure categories. The rate ratio values used in the weighted least squares
24 regression are derived from a cumulative exposure model (15 year lag) in which the estimated
25 regression parameters in the proportional hazards regression model are not significantly different
26 from 0 at a = 0.05 (p = 0.15). In Steenland et al. (2004), the only individually based
27 (proportional hazards) model that fits the data for males in the NIOSH cohort is a model for log
28 of individual exposure through t-15 years.
29 b) Grouped data regression: The weighted least squares fit applies estimates of variance
30 for the individual rate ratios under that assumption that these inverse weighting corrections
31 correctly adjust for heteroscedasticity of residuals in the underlying regression model.
32 Historically, models for grouped proportions applied adjustments of this type but it is by no
33 means a preferred technique when the underlying individual data are available. The "ecological
34 regression" model per Rothman (Rothman and Greenland, 1998) is subject to bias due to within
35 group heterogeneity of predictors and unmeasured confounders. The heterogeneity in the
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-24 DRAFT—DO NOT CITE OR QUOTE
-------
1 grouped model involves the range of exposures within the collapsed categories. The unmeasured
2 confounders include variables (other than gender) that affect the potency of exposure or may
3 have produced gross misclassification based on the original exposure model estimation for the
4 individual (Hornung et al., 1994).
5 c) The model fitting does not conform exactly to the Rothman (1986) procedure: The
6 1998 (Second edition) of Rothman (Rothman and Greenland, 1998) describes the technique for
7 estimating this risk from grouped data in Chapter 23. In that updated version of the original
8 monograph the model that is fitted is:
9
1 0 Expected(Rate I Exposure} = + * Mean(Exposure~)
11
12 The objective is to estimate the rate ratio (for exposure 0=no, l=yes, or equivalently for a one
13 unit increase in the exposure metric). That estimator is then:
14
15 rr = 1 + 4/4
16
17 The model estimated by the EPA method is:
18
1 9 Expected(rr I Exposure) = B* * Mean(Exposure~)
20
21 In the former, the variance in the estimation of the rate ratio is a function of the variance of the
22 estimated slope and the variance in the estimated baseline hazard, represented by the estimated
23 intercept. This variance is present in the estimation of the baseline hazard in the Steenland et al.
24 (2004) estimation of the rate ratios but is not present in the EPA adaptation to the linear rate ratio
25 model. The EPA approach permits no intercept (>0) for the background exposure or any
26 allowance for an effect of true non-zero exposures in the internal control group (exposures less
27 than 15 years).
28
29 In general, the use of categorical exposure ranges is not the optimal strategy for using
30 epidemiologic data. When continuous data are categorized and then used in dose response
3 1 modeling, it amounts to starting with a full range of exposures, collapsing that range into
32 somewhat arbitrary boundaries and then deriving a continuous dose response model for an even
33 larger range of exposures.
34
35 Categorizing continuous variables results in a host of issues:
This document is a draft for review purposes only and does not constitute Agency policy.
7/20 13 H-25 DRAFT— DO NOT CITE OR QUOTE
-------
1 • Assumption that the risk within the category boundaries is constant.
2 • It is not known whether a given categorization is representative of the data since there are many
3 ways of categorizing.
4 • Loss of power and precision by spending degrees of freedom on each category.
5 • Misclassification at category boundaries (this can be minimized by choosing cutpoints
6 where relatively few observations are present).
7 • Categorizations can be manipulated to show the desired results.
8
9 The Panel acknowledged that techniques such as the linear regression method described
10 by Rothman and Greenland (1998) or Poisson regression may be the most appropriate techniques
11 when only grouped or categorized data are available for estimating the dose/response model.
12 However, the original NIOSH cohort data are available at the individual level and this permits
13 the use of models such as the Cox regression models employed by Steenland et al. (2004) that
14 utilize the full information in the individual observations. If categories of exposure (as opposed
15 to individual exposure estimates) must be used, the crude rates should be computed for a large
16 number of equally spaced exposure ranges and the Rothman and Greenland (1998) model fitted
17 to these multiple points.
18
19 EPA Response: EPA agrees that it may be preferable to develop risk models on the basis of
20 direct analysis of individual exposure and cancer outcome data. In fact, the Draft Assessment
21 document included the presentation of models based on fitting Cox regression models to
22 individual exposure-outcome data for EtO. These models provided reasonable fits to the data, as
23 described by Steenland et al. (2004) and in the Draft Assessment document. However, it was the
24 judgment of EPA that these models represented exposure-response relationships that were
25 excessively sensitive to changes in exposure level in the low-dose region and thus were not
26 biologically realistic. That is, in the low-dose region, these models would yield extremely large
27 changes in response for small changes in dose level. Accordingly, the judgment was that these
28 models would not be suitable as the basis for low-dose unit risk values. This is what led EPA to
29 use the regression methodology with the published grouped data. The grouped data regression
30 methodology is considered to be a valid procedure for analysis of such data, and, as mentioned
31 above with respect to charge question 2.b, EPA's 2005 Guidelines for Carcinogen Risk Assessment
32 specifically recognize the use of linear modeling of grouped epidemiological data (U.S. EPA,
33 2005a); therefore, EPA has retained its use for some endpoints in the final assessment and
34 implemented it as described by Rothman (1986) (also described in van Wijngaarden and Hertz-
35 Picciotto, 2004).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-26 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA followed the Panel's recommendation and performed additional analyses of the
2 individual data in collaboration with Professor Steenland. The work performed by Professor
3 Steenland is described in Appendix D of the final assessment. Working with Professor
4 Steenland, alternative models based on direct analysis of all individual data using (1) linear
5 relative risk models (Langholz and Richardson, 2010) and (2) two-piece linear and log-linear
6 spline models (e.g., Rothman et al., 2008) were developed and evaluated. In the final
7 assessment, linear low-dose risk estimates based on the two-piece linear spline model (using the
8 Langholz-Richardson linear relative risk approach) were used for breast cancer incidence risk
9 estimates. Additional responses to specific comments follow:
10 a) Model-based dependent variable: The rate ratios for the exposure categories were not
11 statistically significant, likely due to loss of power and the use of a cumulative Cox regression
12 model within the categories that was not statistically significant across the full exposure range, as
13 noted in the comment. Although the cumulative Cox regression model was not a good fit across
14 the full range of exposures, the RR estimates based on bounded cumulative exposure ranges can
15 still reflect the overall supralinear exposure-response relationship across the categories while
16 representing a different subrelationship within the categories. It is doubtful that categorical
17 results based on log cumulative exposure would have been better because, while a log
18 cumulative exposure Cox regression model provided a statistically significant fit and made it
19 possible to represent the supralinearity (e.g. the plateauing at high exposures) across the range of
20 the data, one would not expect the exposure-response relationship within each discrete category
21 to be supralinear. For example, for lymphoid cancer, the exposure-response relationship in the
22 first exposure category is in the steeply increasing part of the overall exposure-response
23 relationship and is not expected to have its own plateau (Figure 4-1). EPA used the cumulative
24 exposure categorical results because they should provide adequate estimates of the RRs for the
25 limited exposure ranges reflected in each category, particularly the three lowest quartiles (the
26 highest exposure quartiles were excluded from the linear regression models), and this was the
27 approach taken to obtain the categorical results that were reported in the peer-reviewed,
28 published paper of Steenland et al. (2004).
29 b) Grouped data regression: These comments correctly identify assumptions inherent in
30 the method. The assumptions do not, however, preclude the use of the Rothman model in the
31 context of the EtO cancer risk estimation. EPA disagrees with the suggestion that unmeasured
32 confounders may have produced gross misclassification and somehow impaired the exposure
33 model estimation for individuals. The estimation performed by NIOSH to estimate individual
34 worker exposure (Hornung et al., 1994) was extensive and detailed. The resulting model used to
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-27 DRAFT—DO NOT CITE OR QUOTE
-------
1 estimate worker exposure accounted for 85% of the variation in average EtO exposure (see
2 Section 4.1 and Section A.2.8 of Appendix A). EPA agrees with the Panel that the exposure
3 analysis of Hornung et al. (1994) is an example of an "exemplary quantitative analysis of likely
4 errors in exposure estimates." In response to the Panel's suggestion that the Hornung analysis
5 represents an "invaluable opportunity" for further analysis of the impact of possible errors in
6 exposure estimation, EPA investigated the possible use of the "errors in variables" approach
7 (page 27 of the Panel report). Professor Steenland visited the NIOSH offices in Cincinnati in
8 order to review the data and assess whether it would support an "errors in variables" analysis.
9 Unfortunately, the electronic data files used in the exposure analysis were no longer available, so
10 that analysis based on the "errors in variables" approach was not possible.
11 c) EPA reviewed the statistical procedure for modeling categorical data using the
12 methodology in Rothman (1986). This review confirmed that the Rothman procedure was
13 followed closely. The equations used, which are the same as those in Rothman (1986, pp.
14 341-344), are described in Appendix F. The equations are also provided in van Wijngaarden
15 and Hertz-Picciotto (2004). The Rothman (1986) procedure, which is appropriate for case-
16 control data such as the NIOSH data, is based on estimating the effect at each response level
17 relative to the reference or baseline level. Thus, the effect estimates are relative rates (odds
18 ratios), not absolute rates as used in the approach of Rothman and Greenland (1998) cited by the
19 SAB. The rate ratio in the referent group (i.e., those with cumulative exposure = 0) is 1.0, by
20 definition, hence, there is no intercept term in the model. As described by Rothman (1986, page
21 345), variability in the reference category is necessarily entrained in estimates of the slope. As
22 Rothman points out, this can result in loss of estimation efficiency but nevertheless yields a valid
23 estimate of trend. Thus, while it is true, as the comment states, that this procedure may not be
24 optimal in a theoretical sense, it can provide a useful mechanism for estimating linear trend. The
25 Panel acknowledges that a linear regression may be the most appropriate approach when only
26 grouped data are available. EPA agrees but would add that when the objective is low-dose risk
27 estimation, the approach may yield the most useful results from a pragmatic perspective. The
28 availability of individual data does not preclude the use of the Rothman grouped data regression
29 methodology.
30 In the case of the EtO data, it was possible to derive theoretically correct models via
31 direct analysis of the individual data. In the case of the breast cancer incidence data, this
32 approach yielded a model that provided a suitable basis for risk estimation. For the other
33 endpoints (breast cancer mortality, lymphoid cancer mortality), however, the models derived
34 using all individual data were not suitable for risk estimation because of excessive sensitivity in
35 the low-dose range. The large sensitivity of the models to small changes in low-dose values
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-28 DRAFT—DO NOT CITE OR QUOTE
-------
1 results in unstable low-dose risk estimates lacking in biological plausibility and, thus, the
2 Rothman procedure was used.
O
4 Responses to SAB Panel 'bullet' comments:
5
6 • Assumption that the risk within the category boundaries is constant.
7
8 EPA Response: EPA is not assuming that within-category risk is constant. Instead, the
9 assumption is that observed risk within a category may be averaged over a category even though
10 there may be a trend within the category. This is a conventional approach in epidemiological
11 analyses in which categorical analysis is used.
12
13 • It is not known whether a given categorization is representative of the data since there are many
14 ways of categorizing.
15
16 EPA Response: The data groupings used in the EPA analyses were based on sound statistical
17 principles and standard epidemiological practice and were subject to peer review through the
18 publications of Steenland et al. (2004); Steenland et al. (2003). The categories were generally
19 quartiles based on the distribution of cumulative exposures for the cases of the cancer of interest,
20 resulting in essentially the same number of cancer cases per quartile, a typical approach in
21 epidemiological studies.
22
23 • Loss of power and precision by spending degrees of freedom on each category.
24
25 EPA Response: There is some loss of power and precision in categorization. This can result in a
26 failure to find a statistically significant effect when in fact there is a meaningful effect in the
27 data.
28
29 • Misclassification at category boundaries (this can be minimized by choosing cut points where
30 relatively few observations are present)
31
32 EPA Response: Misclassification can occur at category boundaries; however, this is expected to
33 have a small impact on overall results. Moreover, the likely consequences of misclassification
34 across boundaries are that if an RR is overestimated in one category, the RR in an adjacent
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-29 DRAFT—DO NOT CITE OR QUOTE
-------
1 category will be underestimated. Using a linear regression model across the categories may
2 serve to smooth out some of this misclassification, if there is any.
O
4 • Categorizations can be manipulated to show the desired results.
5
6 EPA Response: This may be possible, but no manipulation of the EtO data was performed to
7 show "desired results." The data categories used in the EPA analyses were established a priori in
8 the Steenland et al. publications. The Panel's recommendation to use "a large number of equally
9 spaced exposure ranges" was not deemed feasible for lymphoid cancer because of the relatively
10 small number of deaths.
11
12 PUBLIC COMMENTS:
13
14 A number of public comments were received that addressed a range of technical issues
15 related to the inhalation carcinogen!city of EtO. A number of comments were also received that
16 are generally directed at what are referred to as "Risk Management" issues and, as such, are not
17 addressed here. In the following, summaries of comments on technical risk assessment issues
18 submitted by the public are provided followed by EPA's responses (note that some duplicate
19 comments were omitted).
20
21 Comment 1.0: The Draft Cancer Assessment Fails to Meet the Rigorous Standard of
22 Quality Required Under the Information Quality Act and Cancer Guidelines. The Draft
23 Cancer Assessment is "influential information" as set forth under the Information Quality Act
24 (IQA) and therefore is subject to a rigorous standard of quality. EPA guidance and the
25 Guidelines for Carcinogen Risk Assessment (Cancer Guidelines) require a rigorous standard of
26 quality, which necessitates ensuring that the Draft Cancer Assessment uses scientifically
27 defensible analytical and statistical methods and has a higher degree of transparency than
28 information considered noninfluential, particularly regarding the application of uncertainty
29 factors in EPA's dose-response assessment and risk characterization. The Draft Cancer
30 Assessment demonstrably fails to meet either the standard set forth under the IQA or the Cancer
31 Guidelines. EPA must, therefore, substantially revise the assessment before the final EtO
32 Integrated Risk Information System (IRIS) Risk Assessment (IRIS Assessment) is publicly
33 disseminated or relied upon for any regulatory purposes.
34
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 0 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA RESPONSE: Comments received from the SAB and from the public have been addressed
2 and the EtO carcinogenicity assessment has been revised. It is EPA's position that as a result of
3 the extensive development, review, reanalysis and revision, the final assessment follows EPA's
4 2005 Guidelines for Carcinogen Risk Assessment, uses scientifically defensible analytical and
5 statistical methods, and meets a high standard of transparency. As such, the final assessment is
6 consistent with Information Quality Guidelines.
7
8 Comment 2.0: EPA failed to use all available epidemiologic data, including the Union Carbide
9 Corporation (UCC) data and all the National Institute of Occupational Safety and Health
10 (NIOSH) data that were available at the time EPA conducted its assessment.
11
12 EPA RESPONSE: The assessment describes and considers all relevant epidemiological data
13 available at the time the assessment was conducted, including all the NIOSH data and the UCC
14 data. The Union Carbide data and the publications that this public commentator referred to were
15 evaluated and included in the assessment. EPA also reviewed articles describing additional
16 follow-up and analysis of the Union Carbide data that have been published after the Panel's
17 report was finalized. Ultimately, EPA came to the conclusion that the shortcomings inherent in
18 the Union Carbide data, particularly the crude assignment of exposure levels to subjects in the
19 UCC cohort, are fundamental, and, as a consequence, the data are not suitable for credible
20 quantitative analysis of the carcinogenic risk due to exposure to EtO. In the NIOSH data,
21 exposure estimates were based on a very large number of exposure measurements and a
22 sophisticated modeling approach (Hornung et al., 1994) which took into account job category
23 and other factors such as product type, exhaust controls, age of product, cubic feet of sterilizer,
24 and degree of aeration. Hence, prediction and assignment of exposure levels for different
25 workers in the NIOSH study would be expected to be much better than the crude assignment
26 methods used in the Union Carbide study. Although the recent follow-up of the UCC cohort has
27 now been reported, there still remains a rather small number of cancers (27 lymphohematopoietic
28 cancers, vs. 79 in the NIOSH cohort, 12 vs. 31 NHLs). Consequently, for example, there was a
29 50% excess of NHL in the 9+ years of employment category in the Union Carbide study (Swaen
30 et al., 2009), but it was based on only five cases and was thus not statistically significant. Also,
31 the UCC cohort is restricted to men, making impossible an analysis of breast cancer, which was
32 seen to have a significant increase among those with high EtO exposures in the NIOSH cohort.
33 In sum, the Union Carbide and NIOSH cohorts are not comparable on a number of levels, and
34 the NIOSH cohort remains superior as a basis for risk assessment analyses. In the NIOSH
35 cohort, exposure-response analyses are likely to involve much less misclassification of exposure
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-31 DRAFT—DO NOT CITE OR QUOTE
-------
1 and are based on greater numbers, and thus would be expected to be more reliable. Analyses of
2 the important breast cancer endpoint are only possible with the NIOSH cohort. See also EPA's
3 response to comments on charge question 2.a above.
4
5 Comment 3.0: EPA inappropriately based its evaluation on summaries of statistics available in
6 various publications, rather than the primary source data, review of which and reliance upon are
7 essential to conduct valid dose-response modeling. EPA should have based its calculations on
8 readily available NIOSH data for individual subjects from the cohort mortality study.
9
10 EPA RESPONSE: The statistics used in draft assessment were obtained from published journal
11 articles describing the analysis of the NIOSH data. They are summary and categorical statistics
12 that are commonly used in epidemiological research. The methodology for using such
13 categorical data to perform dose-response analysis is well established in the epidemiological
14 literature and is described in Rothman (1986, pp. 343-344) and van Wijngaarden and Hertz-
15 Picciotto (2004). The categorical and summary statistics used by EPA are constructed from the
16 individual data in the NIOSH study. It is possible to perform analyses and construct models via
17 direct analysis of the individual data and in some cases this is a preferable approach. In fact, the
18 draft EPA assessment presented the results of such analyses in the form of the Cox regression
19 models that were based on direct analysis of the individual data with exposure as a continuous
20 variable. These models provided reasonable fits to the data. However, it was the judgment of
21 EPA that these models generated estimates of risk in the low-dose region that were excessively
22 sensitive to changes in exposure level and therefore would not be suitable as the basis for low-
23 dose unit risk values. This is what led EPA to use the regression methodology with the
24 published grouped data. EPA, in consultation with Professor Steenland, did perform analyses to
25 fit additional models to the continuous NIOSH data. The work performed by Professor
26 Steenland is described in Appendix D of the final assessment. Working with Professor
27 Steenland, EPA developed and evaluated sets of models using the individual data, including (1)
28 linear relative risk models (Langholz and Richardson, 2010) and (2) two-piece linear and log-
29 linear spline models (e.g., Rothman et al., 2008). In the final assessment, linear low-dose
30 estimates based on the two-piece spline model and using the Langholz-Richardson linear
31 approach were used for breast cancer incidence risk estimates. See also EPA's response to
32 comments on charge question 2.b above.
33
34 Comment 4.0: EPA Statistical Analysis of the Data Is Flawed and Other Incorrect
35 Procedures Grossly Overestimate Risk. Key flaws include:
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-32 DRAFT—DO NOT CITE OR QUOTE
-------
1 Comment 4.1: EPA's risk assessments are invalid, based on linear regressions on odds ratios
2 (ORs), rather than on individual subject data;
O
4 EPA RESPONSE: The odds ratios referred to are summary statistics. Regression on
5 categorical or summary statistics such as odds ratios is a valid statistical approach. See the
6 response to comment 1.2 and response to the SAB Panel comment on this issue (charge question
7 2.b above).
8
9 Comment 4.2: EPA fails to include all available epidemiologic data;
10
11 EPA RESPONSE: This comment refers to the Union Carbide data. See response to Comment
12 2.0 and response to the SAB Panel comment on this issue (charge question 2.b above).
13
14 Comment 4.3: EPA's rationale and methodology for exclusion of the highest exposure group
15 is inappropriate;
16
17 EPA RESPONSE: EPA did not use the data from the highest exposure group in estimating the
18 unit risk because it was evident that the relationship between exposure and response changed
19 over the range of exposure. The general pattern in the data indicated a steep increase in response
20 in the low exposure range with a leveling or plateau in the high exposure range. Inclusion of the
21 data from the highest exposure levels in either a Cox regression model or a linear regression
22 yielded overall estimated relationships that were not suitable for risk assessment. Although the
23 Cox regression models with log cumulative exposure provided adequate fits to the data,
24 estimates of risk in the low-dose region were overly sensitive to changes in dose level and thus
25 not biologically realistic. In order to obtain a suitable result for risk estimation at low exposures,
26 in the draft assessment, EPA used a linear regression model and excluded the highest exposure
27 group. For the final assessment, EPA investigated the use of two-piece linear models that
28 modeled the data as a combination of two linear relationships or segments, one that increased
29 steeply in the lower dose region joined with a second that increased at a lower rate in the higher
30 dose region. This approach has the advantage of including all the (individual) data and
31 incorporating into the overall model the change in the relationship over the observed range of
32 exposure. EPA's Benchmark Dose Technical Guidance (U.S. EPA, 2012) recognizes analyses
33 omitting high-dose data points, when these data are not compatible with the development of
34 suitable descriptive statistical analyses, as a viable analytical approach.
35
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 3 DRAFT—DO NOT CITE OR QUOTE
-------
1 Comment 4.4: EP A's use of the heterogeneous broad category of distinct diseases of
2 lymphohematopoietic (LH) cancers as the response increases sample size at the expense of
3 validity and, thereby, reduces the ability to identify a valid positive dose-response relationship.
4
5 EPA RESPONSE: EPA uses the narrower category of lymphoid cancer data for the primary risk
6 estimates in the final assessment.
7
8 Comment 5.0: Certain Policy Decisions EPA Implements in the Draft Cancer Assessment
9 Are Scientifically Unsupported, Overly Conservative, Inappropriate and Have Not Been
10 Reviewed by a Science Advisory Board. EPA made several policy decisions that compounded
11 greatly the inherent conservatism in the risk estimates. These include, among others: (1) EPA's
12 reliance on the lower bound of the point of departure, rather than the best estimate when using
13 human data; (2) use of background incidence rates with mortality-based relative rates, thereby
14 relying on unsupported assumptions that bias results; (3) EPA's assumption of an 85-year
15 lifetime of continuous exposure and cumulative risk, rather than the more traditional 70-year
16 lifetime; and (4) the application of adjustment factors for early-life exposures.
17
18 EPA RESPONSE: The EtO assessment has been reviewed by the SAB and EPA has responded
19 to their comments and revised the assessment. With regard to: (1), use of the lower bound on the
20 point of departure is consistent with EPA's 2005 Guidelines for Carcinogen Risk Assessment
21 (U.S. EPA, 2005a); (2), background incidence rates were used with mortality-based relative rates
22 because EPA's objective is to estimate incidence risk not mortality risk (see also EPA's response
23 to this issue under the further statistical issues subsection at the end of charge question 2.b
24 above); (3), EPA did not assume an 85-year lifetime, rather exposures were considered up to age
25 85 (i.e., actual age-specific mortality and disease rates to age 85 were used in a life-table
26 analysis; because most individuals die before age 85 years, the overall average lifespan from the
27 analysis is about 75 years); (4), EPA's application of adjustment factors for early life exposures
28 in the EtO assessment was in accordance with the recommendations in EPA's Supplemental
29 Guidelines and the scientific data supporting the Supplemental Guidelines (U.S. EPA, 2005b).
30 The application of these adjustment factors was endorsed by the SAB.
31
32 Comment 6.0: EPA Improperly Relies Entirely on Males in Its Assessment of
33 Lymphohematopoietic (LH) Cancer Mortality. To be scientifically defensible, EPA's LH
34 cancer risk characterization must include both males and females, consistent with a "weight-of-
35 evidence" approach that relies on all relevant information. In the NIOSH retrospective study,
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-34 DRAFT—DO NOT CITE OR QUOTE
-------
1 increased risks of LH cancer were observed in males but not females, even though the NIOSH
2 cohort was large and diverse, and consisted of more women than men. EPA's exclusive reliance
3 on male data is scientifically unsound without a mechanistic justification for treating males and
4 females differently with respect to LH, which the analysis lacks.
5
6 EPA RESPONSE: In the final assessment, the lymphohematopoietic cancer unit risk estimates
7 are based on data for both sexes.
8
9 Comment 7.0: EPA's Draft Risk Estimates for Occupational Exposure Levels Rely on
10 Invalid and/or Inappropriate Models. The models used to estimate risks from occupational
11 exposure are flawed because they generate supralinear results, regardless of the observed data.
12 These estimates also suffer from the same invalid methodology used in the environmental risk
13 estimates. EPA must employ a dose-response model that would generate results consistent with
14 the observed data.
15
16 EPA RESPONSE: It is the underlying data that indicate a supralinear exposure-response
17 relationship, particularly for lymphohematopoietic cancer and breast cancer mortality, as
18 suggested by the categorical results as well as by the poorer fits of the Cox regression models
19 with untransformed cumulative exposure data.
20
21 Comment 8.0: EtO is Considered by Many to be a Weak Mutagen and EPA Should
22 Consider This in Proposing a Unit Risk Factor. A chemical's mutagenic potency is
23 necessarily related to its carcinogenic potency. If genotoxicity is considered the means by which
24 a chemical induces cancer, it follows that it will not induce cancer under conditions where it does
25 not induce mutations, at either the chromosome or gene level, thus providing a mechanistic basis
26 for estimating carcinogenicity. EtO has been shown only to be a weak mutagen; therefore, it
27 should not be automatically considered a human carcinogen and certainly not a potent
28 carcinogen. In addition, no treatment-related tumors were observed in rats exposed to EtO, even
29 at the 100 ppm concentration level, at the 18 month sacrifice, and the most sensitive tumor type
30 (i.e., splenic mononuclear cell leukemia) did not significantly increase in the exposed rats until
31 23 months, almost the end of their lifetime of exposures (Snellings et al., 1984). EPA's analysis
32 should have reconciled these findings with its estimation of EtO's carcinogenic potency, but the
33 analysis does not do so.
34
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 5 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA RESPONSE: EPA does not consider the mutagenicity and carcinogenicity findings to be
2 in conflict with the potency estimates. EtO is a relatively weak mutagen when compared to
3 strong mutagens such as cancer chemotherapeutic agents and diepoxides but not necessarily
4 when compared to other environmental mutagens. And EtO is clearly carcinogenic in mice and
5 rats. The inhalation unit risk estimate based on human data is notably larger than that based on
6 rodent data (about 23 times larger), and the reasons for this discrepancy are unknown; however,
7 such species differences are not unusual.
8 It is not surprising that that there was no statistically significant increase in tumors at 18
9 months in the Snellings et al. (1984) study. Because of the latency for cancer development,
10 tumors generally occur later in life. Furthermore, only 20 animals per sex per dose group were
11 killed at 18 months (and tissues from the animals in the low- and mid-dose group only got
12 microscopically examined in the presence of a gross lesion), so there is low power to detect an
13 effect.
14
15 Comment 9.0: EPA's Risk Estimates Do Not Pass Simple Reality Checks.
16
17 Comment 9.1: The results of the Draft Cancer Assessment (resulting in negligible risk only at
18 levels less than a part per trillion), are not reasonable when compared with the results generated
19 for other substances that are considered potent mutagens and/or potent carcinogens, and do not
20 comport with the results of other assessments EPA has undertaken.
21
22 EPA RESPONSE: The procedures used in this assessment comport with those used in other
23 assessments EPA has undertaken. Differences in relative potency across chemicals based on
24 exposure levels may reflect differences in absorption, distribution, metabolism, excretion, or
25 pharmacodynamics of the chemicals.
26
27 Comment 9.2: The Draft Cancer Assessment grossly over predicts the observed number of
28 cancer mortalities in the study upon which it is based by more than 60-fold.
29
30 EPA RESPONSE: The unit risk estimates are derived from, and are consistent with, the results
31 of the NIOSH epidemiology study, as long as they are used in the low-exposure range, as
32 intended. Because the exposure-response relationships for the cancers of interest in the NIOSH
33 study are generally supralinear, the unit risk estimates will overpredict the NIOSH results if
34 applied to the region of the exposure-response relationships where the responses plateau. The
35 potency estimates derived in the assessment are constructed for use with low dose levels
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 6 DRAFT—DO NOT CITE OR QUOTE
-------
1 consistent with environmental exposure and are not appropriate for use with exposures in
2 occupational settings, as stated explicitly in the document. Occupational exposure scenarios are
3 addressed in Section 4.7 of the assessment document. Extra risks associated with occupational
4 exposures are in the "plateau" region of the exposure-response relationships and thus increase
5 proportionately less than risks in the low-dose region.
6
7 Comment 9.3: EPA's de minimis value from the Draft Cancer Assessment is 2 to 3 orders of
8 magnitude below the endogenous level of EtO that is produced naturally in humans.
9
10 EPA RESPONSE: EPA's risk estimates are for risk above background. The issue of
11 endogenous levels is addressed in the final assessment. See Section 4.5 for a discussion of the
12 specific issue raised in this comment.
13
14 Comment 9.4: EPA's draft unit risk values for EtO are unreasonably large, given the evidence
15 of carcinogenicity in a large body of epidemiology studies that is not conclusive, the weak
16 mutagenicity data, and the lack of cancer response in rodents until very late in life. EPA must
17 make the best use of all of the epidemiology, toxicology and genotoxicity data for EtO that
18 provide valid information on the relationship between exposure and cancer response to improve
19 the reasonableness of the unit risk values for EtO.
20
21 EPA RESPONSE: EPA believes that it has made the best use of the available information in
22 revising the assessment. EPA's evaluation of the weight of evidence concludes that the
23 epidemiological evidence is strong (Section 3.5.1). In addition, the unequivocal evidence of
24 rodent carcinogenicity and the supporting mechanistic evidence add sufficient weight for the
25 characterization of "carcinogenic to humans" (Section 3.5.1), which is beyond what is needed to
26 support the derivation of quantitative risk estimates. This is thoroughly presented in the
27 assessment and was supported by the SAB review. The unit risk estimates are derived from, and
28 are consistent with, the results of the large, high-quality NIOSH epidemiology study. See also
29 the response to Comment 8.0 above.
30
31 Comment 10.0: The Draft Cancer Assessment Does Not Use the Best Available Science as
32 Required under the Information Quality Act and Cancer Guidelines.
33
34 Comment 10.1: EPA based its evaluation on summaries of statistics available in various
35 publications. These data, however, are not sufficient to conduct valid dose-response modeling.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 7 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA should have based its calculations on readily available National Institute of Occupational
2 Safety and Health (NIOSH) data for individual subjects from the cohort mortality study.
O
4 EPA RESPONSE: See response to Comment 3.0.
5
6 Comment 10.2: EPA did not use all available epidemiologic data, including the Union Carbide
7 Corporation (UCC) data and all NIOSH data that were available at the time EPA conducted its
8 assessment. In particular, the Greenberg et al. (1990) UCC study reported the consistency of the
9 death certificate diagnosis with a pathology review of medical records for leukemia cases, a
10 validation not conducted for cases in the NIOSH study.
11
12 EPA RESPONSE: EPA considered all the available epidemiological data, including NIOSH
13 data and the Union Carbide data and the publications that the ACC Panel referred to in its
14 comments. See response to Comment 2.0 for more details on why the UCC data were not used
15 for the derivation of quantitative risk estimates.
16
17 Comment 11.0: EPA Should Recognize That EtO Is Both a Weak Mutagen and Weak
18 Animal Carcinogen.
19
20 EPA RESPONSE: The full text of this comment was essentially the same as Comment 8.0 and
21 is addressed in EPA's response to that comment above.
22
23 Comment 11.1: Among 26 alkylating agents studies by (Vogel and Nivard, 1998), EtO
24 showed the second lowest carcinogenic potency.
25
26 EPA RESPONSE: The (Vogel and Nivard, 1998) study is not relevant to EPA's assessment of
27 the carcinogenicity of EtO. Most of the substances considered by (Vogel and Nivard, 1998) are
28 chemotherapeutic chemicals that are, by design, intended to be strong alkylating agents.
29
30 Comment 11.2: Previous assessments of EtO inhalation time to tumor in rats showed that the
31 increased risks observed at higher experimental doses did not extend to the lowest experimental
32 dose. To comply with the Cancer Guidelines, EPA should include these and other relevant
33 animal data in a weight-of-evidence characterization of EtO.
34
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 8 DRAFT—DO NOT CITE OR QUOTE
-------
1 EPA RESPONSE: The carcinogen!city data reviewed in Section 3.2 reveal that, of 13 exposure-
2 response relationships for the tumor types associated with EtO exposure from the 3 rodent
3 bioassays, all but one show an increased incidence at the lowest exposure level, though not all
4 the increases are statistically significant at that level.
5
6 Comment 12.0: EPA's Risk Estimates Do Not Pass Simple Reality Checks.
7
8 Comment 12.1: [This was the same as Comment 9.1 above.]
9
10 Comment 12.2: The results of the Draft Cancer Assessment are at odds with EPA's conclusion
11 that EtO is a potent (de minimis level < 1 ppt) human carcinogen and EtO's potency seen in
12 animal studies.
13
14 EPA RESPONSE: The risk estimates based on the rodent data are over an order of magnitude
15 lower than (-1/23) the estimate based on the human data, for unknown reasons, but species
16 differences are not unusual and human data are generally preferred over rodent data for
17 quantitative risk estimates because the uncertainties due to interspecies extrapolation are
18 avoided.
19
20 Comment 12.3: EPA's draft unit risk values for EtO are not applicable to the general public.
21 The Draft Cancer Assessment grossly over predicts the observed number of LH cancer
22 mortalities in the study upon which it is based by more than 60-fold. Further, EPA's de minimis
23 value is about 50 times lower than the lowest ambient concentration found at remote coastal
24 locations. Based upon PBPK simulations, endogenous concentrations of EtO in humans are
25 approximately 400-1700 times greater than EPA's proposed de minimis value of 0.00036 parts
26 per billion.
27
28 EPA RESPONSE: The unit risk estimates are derived from, and are consistent with, the results
29 of the NIOSH epidemiology study, as long as they are used in the low-exposure range, as
30 intended; see response to Comment 9.2 above. Endogenous and ambient concentrations of EtO
31 could be contributing to background rates of lymphohematopoietic cancer and breast cancer
32 incidences, which are appreciable. The EPA values are not implausible upper bound estimates.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 H-3 9 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX I.
2 LIST OF REFERENCES ADDED AFTER THE 2006 EXTERNAL REVIEW DRAFT
3 Note: These references were added to the Carcinogenicity Assessment in response to the 2007
4 peer reviewers' and public comments, and for completeness. The added references have not
5 changed the overall qualitative or quantitative conclusions. These references are also included in
6 the reference list at the end of the main body of the assessment or the reference list at the end of
7 this appendix volume; see those reference lists for the HERO links.
8
9
10 Abeles, FB; Heggestad, HE. (1973) Ethylene: an urban air pollutant. J Air Waste Manag Assoc 23:517-521.
11 Adam, B; Bardos, H; Adany R. (2005) Increased genotoxic susceptibility of breast epithelial cells to ethylene oxide.
12 Mutat Res 585(1-2): 120-126.
13 Agurell, E; Cederberg, H; Ehrenberg, L; et al. (1991) Genotoxic effects of ethylene oxide and propylene oxide: a
14 comparative study. Mutat Res 250(l-2):229-237.
15 Applebaum, KM; Malloy, EJ; Eisen, EA. (2007) Reducing healthy worker survivor bias by restricting date of hire in
16 a cohort of Vermont granite workers. Occup Environ Med 64:681-687.
17 Applegren, LE; Eneroth, G; Grant, C; et al. (1978) Testing of ethylene oxide for mutagenicity using the
18 micronucleus test in mice and rats. Act Pharmacol Toxicol 43:69-71.
19 Arias, E. (2007). United States life tables, 2004. Atlanta, GA: Centers of Disease Control and Prevention; National
20 Center for Health Statistics. http://www.cdc.gov/nchs/data/nvsr/nvsr56/nvsr56_09.pdf
21 Bastlova, T; Andersson, B; Lambert, B; et al. (1993) Molecular analysis of ethylene oxide-induced mutations at the
22 HPRT locus in human diploid fibroblasts. Mutat Res 287:283-292.
23 Boffetta, P; van der Hel, O; Norppa, H; et al. (2007) Chromosomal aberrations and cancer risk: results of a cohort
24 study from Central Europe. Am JEpidemiol 165:36-43.
25 Bolt, HM; Leutbecher, M; Golka, K. (1997) A note on the physiological background of the ethylene oxide adduct
26 7-(2-hydroxyethyl) guanine in DNA from human blood [Letter]. Arch Toxicol 71(11):719-721.
27 Bolt, HM; Peter, H; Post, U. (1988) Analysis of macromolecular ethylene oxide adducts [Review]. Int Arch Occup
28 Environ Health 60:141-144.
29 Bonassi, S; Znaor, A; Ceppi, M; et al. (2007) An increased micronucleus frequency in peripheral blood lymphocytes
3 0 predicts the risk of cancer in humans [Review]. Carcinogenesis 28:625-631.
31 Boogaard, PJ. (2002) Use of haemoglobin adducts in exposure monitoring and risk assessment. J Chromatogr B
32 Analyt Technol Biomed Life Sci 778(l-2):309-322.
33 Boysen, G; Pachkowski, BF; Nakamura, J; et al. (2009) The formation and biological significance of N7-guanine
34 adducts [Review]. Mutat Res 678:76-94.
35 Britton, DW; Tornqvist, M; van Sittert, NJ; et al. (1991) Immunochemical and GC/MS analysis of protein adducts:
36 dosimetry studies with ethylene oxide [Review]. Prog ClinBiol Res 372:99-106.
37 Chandra, GR; Spencer, M. (1963) A micro apparatus for absorption of ethylene and its use in determination of
3 8 ethylene in exhaled gases from human subjects. Biochim Biophys Acta 69:423-425.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 I-1 DRAFT—DO NOT CITE OR QUOTE
-------
1 Christiansen, DH; Andersen, MK; Desta, F; et al. (2005) Mutations of genes in the receptor tyrosine kinase
2 (RTK)/RAS-BRAF signal transduction pathway in therapy-related myelodysplasia and acute myeloid leukemia.
3 Leukemia 19:2232-2240.
4 Christiansen, DH; Andersen, MK; Pedersen-Bjergaard, J. (2001) Mutations with loss of heterozygosity ofp53 are
5 common in therapy-related myelodysplasia and acute myeloid leukemia after exposure to alkylating agents and
6 significantly associated with deletion or loss of 5q, a complex karyotype, and a poor prognosis. J Clin Oncol
7 19:1405-1413.
8 Cushnir, JR; Lamb, JH; Parry, A; et al. (1991) Tandem mass spectrometric approaches for determining exposure to
9 alkylating agents. IARC Sci Publ 105:107-12.
10 de Serres, FJ; Brockman, HE. (1995) Ethylene oxide: induction of specific-locus mutations in the ad-3 region of
11 heterokaryon 12 of Neurospora crassa and implications for genetic risk assessment of human exposure in the
12 workplace. Mutat Res 328:31-47.
13 Dormer, EM; Wong, B A; James, RA; et al. (2010) Reciprocal translocations in somatic and germ cells of mice
14 chronically exposed by inhalation to ethylene oxide: implications for risk assessment. Mutagenesis 25:49-55.
15 Ehrenberg, L; Osterman-Golkar, S; Segerback, D; et al. (1977) Evaluation of genetic risks of alkylating agents. III.
16 alkylation of haemoglobin after metabolic conversion of ethene to ethene oxide in vivo. Mutat Res 45(2): 175-184.
17 Eide, I; Zhao, C; Kumar, R; et al. (1999) Comparison of (32)P-postlabeling and high-resolution GC/MS in
18 quantifying N7-(2-Hydroxyethyl)guanine adducts. Chem Res Toxicol 12(10):979-984.
19 Farmer, PB; Bailey, E; Naylor, S; et al. (1993) Identification of endogenous electrophiles by means of mass
20 spectrometric determination of protein and DNA adducts [Review]. Environ Health Perspect 99:19-24.
21 Farmer, PB; Shuker, DE. (1999) What is the significance of increases in background levels of carcinogen-derived
22 protein and DNA adducts? Some considerations for incremental risk assessment [Review]. Mutat Res
23 424(l-2):275-286.
24 Farooqi, Z; Tornqvist, M; Ehrenberg, L; et al. (1993) Genotoxic effects of ethylene oxide and propylene oxide in
25 mouse bone marrow cells. Mutat Res 288(2):223-228.
26 Fennell, TR; MacNeela, JP; Morris, RW; et al. (2000) Hemoglobin adducts from acrylonitrile and ethylene oxide in
27 cigarette smokers: effects of glutathione S-transferase Tl-null and Mi-null genotypes. Cancer Epidemiol
28 Biomarkers Prev 9(7):705-712.
29 Post, U; Marczynski, B; Kasemann, R; et al. (1989) Determination of 7-(2-hydroxyethyl)guanine with gas
3 0 chromatography/mass spectrometry as a parameter for genotoxicity of ethylene oxide. Arch Toxicol Suppl
31 13:250-253.
32 Post, U; Hallier, E; Ottenwalder, H; etal. (1991) Distribution of ethylene oxide in human blood and its implications
33 for biomonitoring. Hum Exp Toxicol 10:25-31.
34 Fuchs, J; Wullenweber, U; Hengstler, JG; et al. (1994) Genotoxic risk for humans due to workplace exposure to
35 ethylene oxide: remarkable individual differences in susceptibility. Arch Toxicol 68(6):343-348.
36 Generoso, WM; Cain, KT; Hughes, LA; et al. (1986) Ethylene oxide dose and dose-rate effects in the mouse
37 dominant-lethal test. Environ MolMutagen 8(1): 1-7.
3 8 Generoso, WM; Rutledge, JC; Cain, KT; et al. (1988) Mutagen-induced fetal anomalies and death following
39 treatment of females within hours after mating. DNA Repair 199:175-181.
40 Generoso, WM; Cain, KT; Cornett, CV; et al. (1990) Concentration-response curves for ethylene-oxide-induced
41 heritable translocations and dominant lethal mutations. EnvironMol Mutagen 16(2):126-131.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 1-2 DRAFT—DO NOT CITE OR QUOTE
-------
1 Godderis, L; Aka, P; Matecuca, R; et al. (2006) Dose-dependent influence of genetic polymorphisms on DNA
2 damage induced by styrene oxide, ethylene oxide and gamma-radiation. Toxicology 219(l-3):220-229.
3 Golberg, L. (1986) Chemical and physical properties. In Hazard assessment of ethylene oxide. Boca Raton, FL:
4 CRC Press.
5 Gupta, RC; Lutz, WK. (1999) Background DNA damage for endogenous and unavoidable exogenous carcinogens: a
6 basis for spontaneous cancer incidence? DNA Repair 424(1-2): 1-8.
7 Hallier, E; Langhof, T; Dannappel, D; et al. (1993) Polymorphism of glutathione conjugation of methyl bromide,
8 ethylene oxide and dichloromethane in human blood: influence on the induction of sister chromatid exchanges
9 (SCE) in lymphocytes. Arch Toxicol 67(3): 173-178.
10 Harris, NL: Jaffe, ES; Diebold, J; etal. (1999) World Health Organization classification of neoplastic diseases of the
11 hematopoietic and lymphoid tissues report of the Clinical Advisory Committee—Airlee House, Virginia, November
12 1997. J Clin Oncol 17:3835-3849.
13 Haufroid, V; Merz, B; Hofmann, A; et al. (2007) Exposure to ethylene oxide in hospitals: biological monitoring and
14 influence of glutathione S-transferase and epoxide hydrolase polymorphisms. Cancer Epidemiol Biomarkers Prev
15 16(4):796-802.
16 Hong, HH; Houle, CD; Ton, TV; et al. (2007) K-ras mutations in lung tumors and tumors from other organs are
17 consistent with a common mechanism of ethylene oxide tumorigenesis in the B6C3F1 mouse. Toxicol Pathol
18 35:81-85.
19 Horner, MJ; Ries, LAG; Krapcho, M; Neyman, N; Aminou, R; Howlader, N; Altekruse, SF; Feuer, EJ; Huang, L;
20 Marietta, A; Miller, BA; Lewis, DR; Eisner, MP; Stinchcomb, DG; Edwards, BK. (2009). SEER cancer statistics
21 review, 1975-2006. Bethesda, MD: National Cancer Institute. http://seer.cancer.gov/csr/1975_2006/
22 Houle, CD; Ton, TV; Clayton, N; et al. (2006) Frequent p53 and H-ras mutations in benzene- and ethylene oxide-
23 induced mammary gland carcinomas from B6C3F1 mice. Toxicol Pathol 34(6):752-762.
24 Huang, CC; Shih, WC; Wu, CF; et al. (2008) Rapid and sensitive on-line liquid chromatographic/tandem mass
25 spectrometric determination of an ethylene oxide-DNA adduct, N7-(2-hydroxyethyl)guanine, in urine of
26 nonsmokers. Rapid Commun Mass Spectrom22(5):706-710.
27 IARC (International Agency for Research on Cancer). (1994a) Some industrial chemicals. Ethylene. In: IARC
28 monographs on the evaluation of carcinogenic risks to humans and their supplements. Vol. 60. Some industrial
29 chemicals. Lyon, France: World Health Organization; pp 45-71.
30 IARC (International Agency for Research on Cancer). (2008) Ethylene oxide. In: IARC monographs on the
31 evaluation of carcinogenic risks to humans. Vol. 97. 1,3-Butadiene, ethylene oxide, and vinyl halides (vinyl fluoride,
32 vinyl chloride and vinyl bromide). Lyon, France: World Health Organization; pp. 185-311.
33 Ingvarsson, S. (1999) Molecular genetics of breast cancer progression [Review]. Semin Cancer Biol 9(4):277-288.
34 Jenssen, D; Ramel, C. (1980) The micronucleus test as part of a short-term mutagenicity test program for the
35 prediction of carcinogenicity evaluated by 143 agents tested. MutatRes 75(2):191-202.
36 Kelsey, KT; Wiencke, JK; Eisen, EA; et al. (1988) Persistently elevated sister chromatid exchanges in ethylene
37 oxide-exposed primates: the role of a subpopulation of high frequency cells. Cancer Res 48(17):5045-5050.
3 8 Kligerman, AD; Erexon, GL; Phelps, ME; et al. (1983) Sister-chromatid exchange induction in peripheral blood
39 lymphocytes of rats exposed to ethylene oxide by inhalation. Mutat Res Lett 120:37-44.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 1-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 Koepke, SR; Kroeger-Koepke, MB; Bosan, W; et al. (1988) Alkylation of DNA in rats by
2 N-nitrosomethyl-(2-hydroxyethyl)amine: dose response and persistence of the alkylated lesions in vivo. Cancer Res
3 48(6):1537-1542.
4 Kolman, A. (1985) Effect of deficiency in excision repair and umuC function on the mutagenicity with ethylene
5 oxide in the lacl gene of E. coli. MutatRes 146(l):43-46.
6 Kolman, A; Chovanec, M. (2000) Combined effects of gamma-radiation and ethylene oxide in human diploid
7 fibroblasts. Mutagenesis 15(2):99-104.
8 Kolman, A; Naslund, M. (1987) Mutagenicity testing of ethylene oxide in Escherichia coli strains with different
9 repair capacities. Environ Mol Mutagen 10(3):311-315.
10 Kolman, A; Naslund, M; Calleman, CJ. (1986) Genotoxic effects of ethylene oxide and their relevance to human
11 cancer [Review]. Carcinogenesis 7(8): 1245-1250.
12 Kolman, A; Chovanec, M; Osterman-Golkar, S. (2002) Genotoxic effects of ethylene oxide, propylene oxide and
13 epichlorohydrin in humans: update review (1990-2001) [Review]. DNA Repair 512(2-3): 173-194.
14 Kumar, R; Staffas, J; Forsti, A; et al. (1995) 32P-postlabelling method for the detection of 7-alkylguanine adducts
15 formed by the reaction of different 1,2-alkyl epoxides with DNA. Carcinogenesis 16(3):483-489.
16 Lambert, B; Andersson, B; Bastlova, T; et al. (1994) Mutations induced in the hypoxanthine phosphoribosyl
17 transferase gene by three urban air pollutants: acetaldehyde, benzo(a)pyrene diolepoxide, and ethylene oxide.
18 Environ Health Perspect Suppl 120:135-138.
19 Langholz, B; Richardson, DB. (2010) Fitting general relative risk models for survival time and matched case-control
20 analysis. Am JEpidemiol 171:377-383.
21 Leclercq, L; Laurent, C; De Pauw, E. (1997) High-performance liquid chromatography/electrospray mass
22 spectrometry for the analysis of modified bases in DNA: 7-(2-hydroxyethyl)guanine, the major ethylene oxide-DNA
23 adduct. Anal Chem 69(10): 1952-1955.
24 Leutbecher, M; Langhof, T; Peter, H; et al. (1992) Ethylene oxide: metabolism in human blood and its implication
25 to biological monitoring. Arch Toxicol Suppl 15:289.
26 Lewis, SE; Barnett, LB; Felton, C; et al. (1986) Dominant visible and electrophoretically expressed mutations
27 induced in male mice exposed to ethylene oxide by inhalation. Environ Mol Mutagen 8(6):867-872.
28 Li, F; Segal, A; Solomon, JJ. (1992) In vitro reaction of ethylene oxide with DNA and characterization of DNA
29 adducts. Chem Biol Interact 83(l):35-54.
30 Liou, SH; Lung, JC; Chen, YH; et al. (1999) Increased chromosome-type chromosome aberration frequencies as
31 biomarkers of cancer risk in a blackfoot endemic area. Cancer Res 59(7): 1481-1484.
32 Lorenti Garcia, C; Darroudi, F; Tates, AD; et al. (2001) Induction and persistence of micronuclei, sister-chromatid
33 exchanges and chromosomal aberrations in splenocytes and bone-marrow cells of rats exposed to ethylene oxide.
34 Mutat Res Genet Toxicol Environm Mutagen 492(l-2):59-67.
35 Lynch, DW; Lewis, TR; Moorman, WJ; et al. (1984b) Sister-chromatid exchanges and chromosome aberrations in
36 lymphocytes from monkeys exposed to ethylene oxide and propylene oxide by inhalation. Toxicol Appl Pharmacol
37 76:85-95.
3 8 Marsden, DA; Jones, DJ; Britton, RG; et al. (2009) Dose-response relationships for N7-(2-hydroxyethyl)guanine
39 induced by low-dose (14C)ethylene oxide: evidence for a novel mechanism of endogenous adduct formation.
40 Cancer Res 69(7):3052-3059.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 1-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 Marsden, DA; Jones, DJ; Lamb, JH; et al. (2007) Determination of endogenous and exogenously derived
2 N7-(2-hydroxyethyl)guanine adducts in ethylene oxide-treated rats. Chem Res Toxicol. 20(2):290-299.
3 Minino, AM; Arias, E; Kochanek, KD; Murphy, SL; Smith, BL. (2002). Deaths: Final Data for 2000. Hyattsville,
4 MD: National Center for Health Statistics, http://www.cdc.gov/nchs/data/nvsr/nvsr50/nvsr50 15.pdf
5 Nivard, MJ; Czene, K; Segerback, D; et al. (2003) Mutagenic activity of ethylene oxide and propylene oxide under
6 XPG proficient and deficient conditions in relation to N-7-(2-hydroxyalkyl)guanine levels in Drosophila. Mutat Res
7 529(l-2):95-107.
8 Ong, T; Bi, HK; Xing, S; et al. (1993) Induction of sister chromatid exchange in spleen and bone marrow cells of
9 rats exposed by inhalation to different dose rates of ethylene oxide. Environ MolMutagen 22(3): 147-151.
10 Otteneder, M; Lutz, WK. (1999) Correlation of DNA adduct levels with tumor incidence: carcinogenic potency of
11 DNA adducts [Review]. DNA Repair 424(l-2):237-247.
12 Pauwels, W; Veulemans, H. (1998) Comparison of ethylene, propylene and styrene 7,8-oxide in vitro adduct
13 formation onN-terminal valine in human haemoglobin and onN-7-guanine inhumanDNA. Mutat Res 418:21-33.
14 Pedersen-Bjergaard, J; Christiansen, DH; Desta, F; et al. (2006) Alternative genetic pathways and cooperating
15 genetic abnormalities in the pathogenesis of therapy-related myelodysplasia and acute myeloid leukemia [Review].
16 Leukemia 20:1943-1949.
17 Pero, RW; Widegren, B; Hogstedt, B; et al. 1981. In vivo and in vitro ethylene oxide exposure of human
18 lymphocytes assessed by chemical stimulation of unscheduled DNA synthesis. DNA Repair 83(2):271-289.
19 Ribeiro, LR; Rabello-Gay, MN; Salvadori, DMF; et al. (1987) Cytogenetic effects of inhaled ethylene oxide in
20 somatic and germ cells of mice. Arch Toxicol 59:332-335.
21 Ries, LAG; Melbert, D; Krapcho, M; Mariotto, A; Miller, BA; Feuer, EJ; Clegg, L; Horner, MJ; Howlader, N;
22 Eisner, MP. (2007). SEER (Surveillance Epidemiology and End Results) cancer statistics review, 19752004.
23 National Cancer Institute, U.S. Department of Health, Education, and Welfare, National Institutes of Health.
24 http://seer.cancer.gov/csr/1975 2004
25 Rossner, P; Boffetta, P; Ceppi, M; et al. (2005) Chromosomal aberrations in lymphocytes of healthy subjects and
26 risk of cancer. Environ Health Perspect 113(5):517-520.
27 Rusyn, I; Asakura, S; Li, Y; et al. (2005) Effects of ethylene oxide and ethylene inhalation on DNA adducts,
28 apurinic/apyrimidinic sites and expression of base excision DNA repair genes in rat brain, spleen, and liver. DNA
29 Repair 4:1099-1110.
3 0 SAB (U.S. EPA Science Advisory Board). (2007) Review of Office of Research and Development (ORD) draft
31 assessment entitled, "Evaluation of the carcinogenicity of ethylene oxide". Science Advisory Board, Washington,
32 DC. EPA-SAB-08-004; Available online at
33 http://yosemite.epa.gov/sab/sabproduct.nsf/368203f97al5308a852574ba005bbd01/5D661BC118B527A3852573
34 B80068C97B/$File/EPA-SAB-08-004-unsigned.pdf.
35 Saha, M; Abushamaa, A; Giese, RW. (1995) General method for determining ethylene oxide and related N7-guanine
36 DNA adducts by gas chromatography-electron capture mass spectrometry. J Chromatogr A 712(2):345-354.
37 Sarto, F; Cominato, I; Pinton, AM; et al. (1984b). Workers exposed to ethylene oxide have increased incidence of
38 sister chromatid exchange. IARC Sci Publ 59:413-419.
39 Segerback, D. (1983). Alkylation of DNA and hemoglobin in the mouse following exposure to ethene and ethene
40 oxide. Chem Biol Interact 45(2):139-151.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 1-5 DRAFT—DO NOT CITE OR QUOTE
-------
1 Segerback, D. (1990) Reaction products in hemoglobin and DNA after in vitro treatment with ethylene oxide and N-
2 (2-hydroxyethyl)-N-nitrosourea. Carcinogenesis 11(2):307-312.
3 Segerback, D. (1994) DNA alkylationby ethylene oxide and some mono-substituted epoxides [Review]. IARC Sci
4 Publ 125:37-47.
5 Shen, J; Kessler, W; Denk, B; et al. (1989) Metabolism and endogenous production of ethylene in rat and man.
6 Arch Toxicol Suppl 13:237-239.
7 Sielken, RL; Valdez-Flores, C. (2009a) Life-table calculations of excess risk for incidence versus mortality: ethylene
8 oxide case study. Regul Toxicol Pharmacol 55:82-89.
9 Sielken, RL; Valdez-Flores, C. (2009b) Calculating excess risk with age-dependent adjustment factors and
10 cumulative doses: ethylene oxide case study. Regul Toxicol Pharmacol 55:76-81.
11 Steenland, K; Deddens, J. (2004) A practical guide to dose-response analyses and risk assessment in occupational
12 epidemiology [Review]. Epidemiol 15:63-70.
13 Swaen, GM; Burns, C; Teta, JM; et al. (2009) Mortality study update of ethylene oxide workers in chemical
14 manufacturing: a 15 yearupdate. J Occup EnvironMed 51:714-723.
15 Swenberg, JA; Fryar-Tita, E; Jeong, YC; et al. (2008) Biomarkers in toxicology and risk assessment: informing
16 critical dose-response relationships. Chem Res Toxicol 21:253-265.
17 Swenberg, JA; Ham, A; Koc, H; et al. (2000) DNA adducts: effects of low exposure to ethylene oxide, vinyl
18 chloride and butadiene. DNA Repair 464:77-86.
19 Thiess, AM; Schwegler, H; Fleig, I; et al. (1981) Mutagenicity study of workers exposed to alkylene oxides
20 (ethylene oxide/propylene oxide) and derivatives. J Occup Med 23(5):343-347.
21 Tompa, A; Jakab, M; Biro, A; et al (2006) Chemical safety and health conditions among Hungarian hospital nurses.
22 Ann NY Acad Sci 1076:635-648.
23 Tompkins, EM; Jones, DJ; Lamb, JH; et al. (2008). Simultaneous detection of five different 2-hydroxyethyl-DNA
24 adducts formed by ethylene oxide exposure, using a high-performance liquid chromatography/electrospray
25 ionisation tandem mass spectrometry assay. Rapid Commun Mass Spectrom 22: 19-28.
26 Tompkins, EM; McLuckie, KI; Jones, DJ; et al. (2009) Mutagenicity of DNA adducts derived for ethylene oxide
27 exposure in the pSP189 shuttle vector replicated in human Ad293 cells. Mutat Res 678:129-137.
28 U.S. EPA (Environmental Protection Agency). (2006) Evaluation of the carcinogenicity of ethylene oxide: external
29 review draft. (EPA/635/R-06/003). Office of Research and Development, Washington, DC.
30 U.S. EPA (Environmental Protection Agency). (2012) Benchmark dose technical guidance. (EPA/100/R-12/001).
31 Risk Assessment Forum, Washington, DC. . Available online at
32 http://www.epa.gov/raf/publications/pdfs/benchmark_dose_guidance.pdf.
33 Uziel, M; Munro, NB; Katz, DS; etal. (1992) DNA adduct formation by 12 chemicals with populations potentially
34 suitable for molecular epidemiological studies [Review]. Mutat Res 277:35-90.
3 5 Valdez-Flores, C; Sielken, RL; Teta, MJ. (2010) Quantitative cancer risk assessment based on NIOSH and UCC
36 epidemiological data for workers exposed to ethylene oxide. Regul Toxicol Pharmacol 56(3):312-320.
37 van Delft, JH; van Winden, MJ; van den Ende, AM; et al. (1993) Determining N7-alkylguanine adducts by
3 8 immunochemical methods and HPLC with electrochemical detection: applications in animal studies and in
3 9 monitoring human exposure to alkylating agents. Environ Health Perspect 99:25-32.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 1-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 van Delft, JH; van Winden, MJ; Luiten-Schuite, A; et al. (1994) Comparison of various immunochemical assays for
2 the detection of ethylene oxide-DNA adducts with monoclonal antibodies against imidazole ring-opened
3 N7-(2-hydroxyethyl) guanosine: application in a biological monitoring study. Carcinogenesis 15(9):1867-1873.
4 van Sittert, NJ; de Jong, G. (1985) Biomonitoring of exposure to potential mutagens and carcinogens in industrial
5 populations. Food ChemToxicol 23:23-31.
6 van Wijingaarden, E; Hertz-Picciotto, I. (2004) A simple approach to performing quantitative cancer risk assessment
7 using published results from occupational epidemiology studies. Sci Total Environ 332(l-3):81-87.
8 Victorin, K; Stahlberg, M. (1988) A method for studying the mutagenicity of some gaseous compounds in
9 Salmonella typhimurium. Environ Mol Mutagen ll(l):65-77.
10 Vogel, EW; Barbin, A; Nivard, MJM; Stack, HF; Waters, MD; Lohman, PHM. (1998). Heritable and cancer risks of
11 exposures to anticancer drugs: inter-species comparisons of covalent deoxyribonucleic acid-binding agents. Mutat
12 Res 400: 509-540.
13 Vogel, EW; Natarajan, AT. (1995) DNA damage and repair in somatic and germ cells in vivo [Review]. Mutat Res
14 330(1-2):183-208.
15 Walker, VE; Fennell, TR; Upton, PB; et al. (1993) Molecular dosimetry of DNA and hemoglobin adducts in mice
16 and rats exposed to ethylene oxide. Environ Health Perspect 99:11-17.
17 Walker, VE; Wu, KY; Upton, PB; et al. (2000) Biomarkers of exposure and effect as indicators of potential
18 carcinogenic risk arising from in vivo metabolism of ethylene to ethylene oxide. Carcinogenesis 21(9): 1661-1669.
19 Warwick, GP. (1963) The mechanism of action of alkylating agents [Review]. Cancer Res 23:1315-1333.
20 Waters, MD; Stack, HF; Jackson, MA. (1999) Genetic toxicology data in the evaluation of potential human
21 environmental carcinogens [Review]. Mutat Res 437(l):21-49.
22 Wu, KY; Ranasinghe, A; Upton, PB; et al. (1999a) Molecular dosimetry of endogenous and ethylene oxide-induced
23 N7-(2-hydroxyethyl) guanine formation in tissues of rodents. Carcinogenesis 20(9): 1787-1792.
24 Wu, KY; Scheller, N; Ranasinghe, A; et al. (1999b) A gas chromatography/electron capture/negative chemical
25 ionization high-resolution mass spectrometry method for analysis of endogenous and exogenous
26 N7-(2-hydroxyethyl)guanine in rodents and its potential for human biological monitoring. Chem Res Toxicol
27 12(8):722-729.
28 Yager, JW; Benz, RD. (1982) Sister chromatid exchanges induced in rabbit lymphocytes by ethylene oxide after
29 inhalation exposure. Environ Mutagen 4(2): 121-134.
30 Yong, LC; Schulte, PA; Kao, CY; et al. (2007) DNA adducts in granulocytes of hospital workers exposed to
31 ethylene oxide. Am JIndMed50(4):293-302.
32 Zhao, C; Tyndyk, M; Eide, I; et al. (1999) Endogenous and background DNA adducts by methylating and
33 2-hydroxyethylating agents. Mutat Res 424(1-2): 117-125.
34 Zharlyganova, D; Harada, H; Harada, Y; et al. (2008) High frequency of AML1/RUNX1 point mutations in
3 5 radiation-associated myelodysplastic syndrome around Semipalatinsk nuclear test site. J Radiat Res (Tokyo)
36 49(5):549-555.
37
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 1-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX J.
2 SUMMARY OF MAJOR NEW STUDIES SINCE THE LITERATURE CUTOFF DATE
3 The cutoff date for literature inclusion into this carcinogenicity assessment was 30 June
4 2010. A systematic literature search was conducted for the time frame from January 2006 to
5 May 2013 to ensure that no major studies were missed from the time of the first external review
6 draft in 2006 until the cutoff date and to determine if any significant new studies had been
7 published since the cutoff date that might alter the findings of the assessment. No studies were
8 identified that would impact the assessment's major conclusions. Nonetheless, two new studies
9 of high pertinence to the assessment have been published since the cutoff date for literature
10 inclusion, and these studies are reviewed briefly in this Appendix for transparency and
11 completeness. The Appendix first provides a description of the systematic literature search that
12 was conducted to identify relevant new studies and then provides the reviews of the two major
13 new studies.
14
15 J.I. SYSTEMATIC LITERATURE SEARCH
16 A systematic literature search was conducted in May 2013, covering the time frame from
17 January 2006 to May 2013. The search was conducted using the LitSearch tool in EPA's HERO
18 database, and the following three literature databases were searched: PubMed, Web of Science,
19 and ToxNet. The search terms involved Ethylene Oxide AND (carcinogenicity OR cancer OR
20 mutagenicity OR mutation OR genotoxicity).
21 The search identified 372 references, of which 56 were determined to be potentially
22 relevant9. The disposition of the 56 potentially relevant references is summarized in Table J-l.
23 In brief, for the purposes of this carcinogenicity assessment, 26 references that were primarily
24 discussions of methods studies or exposure studies10 or were reviews or other secondary source
25 material were not generally considered further. The remaining 30 references were given further
26 consideration to see if they represented major new studies. No new studies were identified that
9In this first part of the screening, any references of potential relevance to the carcinogenicity assessment of ethylene
oxide were identified. References that pertained to other things and that were inadvertently captured in the literature
search were excluded. For example, in an alphabetical listing of the 372 references by first author, the first
reference is: Agarwal, A., Unfer, R. and Mallapragada, S. K. (2007), Investigation of in vitro biocompatibility of
novel pentablock copolymers for gene delivery. J. Biomed. Mater. Res., 81A: 24-39. This reference discusses some
copolymers of various chemicals, including poly(ethylene oxide), synthesized as vectors for gene delivery and tested
in some cancer cell lines; this reference was not relevant to the assessment and was excluded from further
consideration.
10This refers to general exposure studies; exposure studies related to any of the epidemiological studies of EtO
would be considered further.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-l DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
would impact the assessment's major conclusions. Two references were identified as highly
pertinent studies, and these are reviewed briefly in Section J.2 of this Appendix.
Table J-l. Disposition of 56 new references identified as potentially relevant
Category
Exposure studies
Methods studies
Reviews or other
secondary source
material
Cancer studies
References"
Davis et al. (2006)
Lin et al. (2007)
Tateo and Bononi (2006)
Ahn and Shin (2006)
Tretyakova et al. (2012)
Wuetal. (2011)
Brown etal. (2012)
Butterworth and Chapman (2007)
Chan etal. (2006)
Farmer and Singh (2008)
Hoenerhoffetal. (2009)
Jarabek et al. (2009)
Keshava et al. (2006a)
Keshava et al. (2006b)
(Manservigi et al., 2010)
McCarthy et al. (2009)
Mosavi-Jarrahi et al. (2009)
Okada etal. (2012)
Smith-Bindman(2012)
Snedeker (2006)
Steinhausen et al. (2012)
Weiderpass et al. (2011)
Won (20 10)
WHO, 2008 (same as IARC, 2008)
IARC (2008)
Kiran etal. (2010)
Mikoczyetal. (2011)
Swaen et al. (2009)
van Balen etal. (2006)
Fondelli et al. (2007)
Kim etal. (2011)
Disposition
Not considered further.
Not considered further.
Not considered further.
Already cited in the assessment.
Reviewed in Section J.2.
Already cited in the assessment.
Not considered further. Primarily a study of risks to
farmers. EtO left out of analysis because too few study
subjects were exposed to it. Subjects were part of the
EPILYMPH study analyzed by Kiran et al. (2010) (see
Section J.2. 1).
Not considered further. No EtO-specific results.
Not considered further. Case report study of 7 cases of
malignant lymphohematopoietic disorders found in 2
semiconductor plants. Various carcinogens suspected
of causing lymphohematopoietic cancers were
investigated; EtO not found in cases' processes.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
J-2 DRAFT—DO NOT CITE OR QUOTE
-------
Table J-l. Disposition of 56 new references identified as potentially relevant
(continued)
Category
Genotoxicity/
Mutagenicity studies
Other
References"
Dormer etal. (2010)
Godderis et al. (2006)
Hong et al. (2007)
Houle et al. (2006)
Marsden et al. (2007)
Marsden et al. (2009)
Tompkins et al. (2008)
Yong et al. (2007)
Tompa et al. (2006)
Tompkins et al. (2009)
Huang etal. (2011)
Lindberg etal. (2010)
Mazon et al. (2009)
Mazonetal. (2010)
Parsons etal. (2012)
Tompkins et al. (2006)
Sielken and Valdez-Flores (2009)
Sielken and Valdez Flores (2009)
Swenberg et al. (2008)
Valdez-Flores et al. (2010)
Haufroid et al. (2007)
Kensler etal. (2012)
Steenland etal. (2011)
Valdez-Flores et al. (201 1)
Swenberg etal. (2011)
Disposition
Already cited in the assessment.
Citations added to the assessment.
Not considered a major new study. Largely an
exposure study; examined use of urinary N7-HEG as a
biomarker of EtO exposure in EtO-exposed workers
and smokers in Taiwan.
Not considered further. This study examined utility of
a micronucleus assay for detecting genotoxic damage
in mouse alveolar Type II and Clara cells — EtO was
used as a test agent but at a high concentration (>3
times higher than the highest exposure concentration
used in the mouse cancer bioassay).
Not considered further. Focused on a specific repair
gene product in E. Coli.
Not considered further. Published abstracts, not full
papers.
Already cited in the assessment.
Citation added.
Not relevant; focused on chemoprevention.
Not considered further. Peer-reviewed publication of
analyses already in the assessment.
Not considered further. Quantitative risk assessment
for occupational exposures — issues pertaining to the
Valdez-Flores et al. risk assessment approach are
already addressed in the assessment in discussions of
the 2010 paper by the same authors (Valdez-Flores et
al., 2010).
Not considered further. Largely a review; focused on
implications of endogenous adducts for risk
assessment — this issue is already addressed in the
assessment (e.g., in the responses to SAB comments in
Appendix H).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
J-3 DRAFT—DO NOT CITE OR QUOTE
-------
1 Table J-l. Disposition of 56 new references identified as potentially relevant
2 (continued)
O
4 EtO: ethylene oxide.
5 N7-HEG: N7-(2-hydroxyethyl)guanine.
6
7 "Full citations for references cited elsewhere in the assessment are provided in the Reference section of the
8 assessment; full citations for references appearing only in this appendix are provided in Section J.3 of the appendix.
9
10
11 J.2. REVIEWS OF MAJOR NEW STUDIES PUBLISHED SINCE THE LITERATURE
12 CUTOFF DATE
13 As discussed in Section J. 1, a systematic literature search was conducted in part to
14 determine whether any significant new studies had been published since 30 June 2010, the cutoff
15 date for literature inclusion into this assessment. No new studies were identified that would
16 impact the assessment's major conclusions. Nonetheless, two studies of high pertinence to the
17 assessment have been published since the cutoff date for literature inclusion. The two studies are
18 epidemiology studies of occupational exposure to ethylene oxide (EtO). These studies are
19 reviewed briefly here for transparency and completeness, and key features of the studies are
20 summarized in Table J-2.
21
22 J.2.1. Kiran et al. (2010)
23 Kiran et al. (2010) investigated occupational exposure to EtO in a population-based case-
24 control study of lymphoma in six European countries (the "EPILYMPH study"). Cases
25 (n = 2,347) were consecutive adult patients with a first diagnosis of lymphoma, classified under
26 the 2001 WHO lymphoma classification system, in 1998-2004 at 22 centers in the six countries.
27 Controls from Germany and Italy were randomly selected from the general population, matched
28 on sex, 5-year age group, and residence area. Controls from the Czech Republic, France,
29 Ireland, and Spain were matched hospital controls with diagnoses other than cancer, infectious
30 diseases, and immunodeficient diseases (total controls = 2,463). Participation rates were 88% in
31 cases, 81% in hospital controls, and 52% in population controls. All study subjects were
32 interviewed in person using the same structured questionnaire, which included questions on
33 sociodemographic factors, lifestyle, health history, and complete work history (including all
34 full-time jobs held for >1 year). For each job, information was collected on type of industry,
35 tasks performed, machines used, and exposure to 35 specific agents (or groups of agents) of
36 interest, including EtO. Supplemental questionnaire modules for specific occupations were used
37 to get additional details on jobs and exposures of interest.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-4 DRAFT—DO NOT CITE OR QUOTE
-------
1 Exposure was evaluated in each center by specially trained industrial hygienists who
2 reviewed all the questionnaires and assessed frequency and intensity of exposure to each agent
3 on a 4-point scale (unexposed and low, medium and high exposures) as well as confidence in the
4 assessment (low, medium, or high). Most of the exposed cases and controls were classified with
5 medium or high confidence, although a greater proportion of cases than controls were thus
6 classified (23/31 versus 15/27). Because of the low prevalence (1.2%) of EtO exposure in the
7 study, the medium and high categories of exposure frequency and intensity were combined in the
8 statistical analyses. A cumulative exposure score for EtO was also developed for each study
9 subject, integrating duration, frequency, and intensity of exposure; these scores were then
10 categorized as above or below the median score among exposed subjects.
11 Risk was assessed for all lymphoma, B-cell lymphoma (which represented 80% of all the
12 lymphoma cases), and the most common subtypes of B-cell lymphoma. The OR was calculated
13 using unconditional logistic regression, adjusting for age, sex, and center. Including education,
14 farm work, and exposure to solvents in the model, reportedly did not change the risk estimates
15 (results not shown). Linear trends for the exposure metrics were calculated using the Wald test
16 for trend.
17 Because of the low prevalence of EtO exposure in the study (1.2%), the number of
18 exposed cases and controls was limited (31 and 27, respectively), so the study power was not
19 large, especially for analyses of lymphoma subtypes. Results for all lymphoma for ever exposed
20 and for the highest exposure category for each of the different exposure metrics are presented in
21 Table J-2. Increased risks were observed for ever exposed and for the highest exposure category
22 for each of the exposure metrics, and the OR for medium or high frequency of exposure was
23 statistically significant (4.3; 95% CI 1.4, 13.0). However, none of the trend tests was statistically
24 significant. The overall association appeared to be stronger using hospital controls; however,
25 when considering only subjects whose EtO exposures were assessed with medium or high
26 confidence, the increased ORs were similar using either hospital or population controls. Results
27 were similar when only B-cell lymphoma, which represented the majority of all lymphomas, was
28 evaluated. The strongest associations were observed for chronic lymphocytic leukemia, and
29 ^-values for trend were <0.051 for all the exposure metrics for that lymphoma subtype. The
30 investigators note that while random variation related to the low prevalence might account for
31 some positive results, their combined probability test (Fischer method) indicated that the chance
32 probability of an upward trend in chronic lymphocytic leukemia across the four metrics assumed
33 to be independent (confidence, frequency, intensity, and duration) was 0.003.
34 In conclusion, this study adds further support to the weight-of-evidence finding obtained
35 in Chapter 3 of strong, but less than conclusive, evidence of a causal association between EtO
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-5 DRAFT—DO NOT CITE OR QUOTE
-------
1 exposure and lymphohematopoietic cancers in humans. Because only categorical exposures
2 were assessed, no quantitative risk estimates can be derived from this study.
O
4 3.2.2. Mikoczyetal. (2011)
5 This study is an update of the Hagmar et al. (1991) and Hagmar et al. (1995) studies
6 discussed in Section 3.1 of the assessment and in Section A.2.11 of Appendix A. The first
7 update (Hagmar et al., 1995) had a median follow-up time of only 11.8 years; this update extends
8 the follow-up period through 2006, providing an additional 16 years of follow-up. The cohort
9 consists of 2,171 (1,309 females and 862 males11), employed for at least 1 year prior to 1986, at
10 two Swedish facilities that sterilized medical equipment using EtO (Plant A sterilization
11 operations ran from 1970 to 1994; Plant B sterilization operations ran from 1964 to 2002). Vital
12 status and emigration data at the end of follow-up were obtained from the Swedish population
13 registry; cause of death for 1972-2006 was obtained from Statistics Sweden; and malignant
14 tumor data for 1972-2006 were obtained from the Swedish Cancer Registry. At the end of
15 follow-up, the mean age of the cohort was 56 years and the cohort had contributed
16 58,305 person-years of risk; 171 cohort members had died (7.9%) and 126 (5.8%) had emigrated
17 and were of unknown vital status. Mean duration of employment in the cohort was 6.3 years.
18 In the original study (Hagmar et al., 1991), individual cumulative exposure estimates
19 were derived based on job-exposure matrices for each plant and exposure level estimates
20 determined up to 1986. While exposure levels were high in the early years of the operations
21 (e.g., peak levels of 75 ppm in 1964 in Plant B and exposure levels up to 40 ppm in 1970 in
22 Plant A), 8-hour TWA levels had decreased to below 1 ppm by 1985 (See Hagmar et al., 1991,
23 and Section A.2.11 of Appendix A for more details on the original exposure assessment). For
24 this update, worker histories for the 1,303 workers who were still employed at the two plants at
25 the end of the original study (1986) were extended up until the cessation of sterilization
26 operations in the plants, and exposure estimates for the follow-up period were determined from
27 yearly statutory industrial hygiene measurements of EtO from 1986 on. Because of the low
28 exposure levels after 1985, the impact of updating the cumulative exposure estimates was low
29 (the largest impact was reportedly on the 90th percentile, which changed from 1.17 to 1.29 ppm
30 x years). The mean and median cumulative exposures for the 2,020 cohort members for whom
31 job titles were available were 2.92 ppm x years and 0.13 ppm x years, respectively.
32 Standardized mortality and incidence ratios (SMRs and SIRs) were obtained by
33 comparing the number of deaths or incident cases observed to the number expected based on
"Without explanation, there is one additional male in the update; the 1991 and 1995 papers both reported
2,170 workers, including 861 males, in the cohort (Hagmar et al., 1995; 1991).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-6 DRAFT—DO NOT CITE OR QUOTE
-------
1 cause-, calendar year-, sex-, and 5-year age group-specific rates in the Swedish population
2 (external referents). For cancer incidence (but not mortality), internal analyses were also
3 conducted using Poisson regression analyses, adjusted for age group and calendar period, with
4 no induction (latency) period. In the internal analyses, incidence rate ratios were calculated by
5 comparing the incidence rates for the two highest cumulative exposure quartiles with that for the
6 50% of workers with cumulative exposures below the median of 0.13 ppm x years (internal
7 referents). Internal analyses are generally preferred over external analyses because the referents
8 are from the same cohort as the exposed subjects, potentially reducing confounding as well as the
9 healthy worker effect, which can mask an increase in risk; however, in this study, some of the
10 advantages of internal analyses may be mitigated by the absence of an unexposed referent group,
11 which could itself dampen relative risk estimates.
12 Results for cancer mortality and incidence for the cancer types of interest (i.e.,
13 lymphohematopoietic cancers and female breast cancer) are summarized in Table J-2. For
14 lymphohematopoietic cancers, nonsignificant increases in SMRs and SIRs were reported. For
15 the incidence data, the internal analysis shows no exposure-related association for
16 lymphohematopoietic cancers, although this analysis is relatively uninformative for these
17 cancers, given the small number of cases (five cases in each of the two highest exposure quartiles
18 and seven cases in the referent group of workers with cumulative exposures below the median),
19 the generally low estimated cumulative exposures, and the absence of an unexposed referent
20 group. It should also be noted that data were not reported or analyzed for the subgrouping of
21 "lymphoid" cancers.
22 For breast cancer mortality (results not shown), a "slight but nonsignificant decrease" in
23 the SMR was reported. With a 15-year induction period included, the SMR for breast cancer
24 was reportedly "somewhat increased." For workers with cumulative exposures above the
25 median, with a 15-year induction period, a "higher than expected" SMR, which was not
26 statistically significant, was reported.
27 For breast cancer incidence (41 incident cases), SIRs were nonsignificantly decreased,
28 both with and without a 15-year induction period. Internal analyses resulted in statistically
29 significant increases in the incidence rate ratios for the two highest cumulative exposure quartiles
30 as compared to the 50% of workers with cumulative exposures below the median (see Table J-2),
31 despite having a low-exposed rather than an unexposed referent group.
32 In conclusion, the nonsignificant increases in SMRs and SIRs for lymphohematopoietic
33 cancers reported in this study are consistent with an increase in lymphohematopoietic cancer risk
34 but, overall, the study is underpowered for the analysis of lymphohematopoietic cancers and
35 contributes little to the weight of evidence for these cancers. For breast cancer incidence,
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-7 DRAFT—DO NOT CITE OR QUOTE
-------
1 however, the statistically significant exposure-related increases in breast cancer incidence in
2 internal analyses add support to the weight-of-evidence finding obtained in Chapter 3 of strong,
3 but less than conclusive, evidence of a causal association between EtO exposure and female
4 breast cancer in humans. The cumulative exposure estimates for this study were very low
5 compared to those in other studies. For example, in the Swaen et al. (2009) study of the UCC
6 cohort of male EtO production workers, the average cumulative exposure was 67.16 ppm
7 x years. In the more comparable NIOSH cohort of sterilization workers, cumulative exposures at
8 the end of follow-up for the full cohort, which included workers with <1 year of employment,
9 had a mean of 27 ppm x years and median of 6 ppm x years (see Appendix D, Section D.I), and
10 in particular, the mean cumulative exposure at the end of follow-up in the breast cancer
11 incidence study cohort, which only included workers with >1 year of employment, was 37.0 ppm
12 x years. Yet, the breast cancer incidence RRs for the categorical exposure groups reported in
13 Steenland et al. (2003) for the NIOSH breast cancer incidence study were lower than those
14 observed in the Mikoczy et al. (2011) study.
15 Thus, if unit risk estimates were derived based on the Mikoczy et al. (2011) study, they
16 would be higher than the estimates calculated from the NIOSH study. However, no such
17 estimates were derived based on the Mikoczy et al. (2011) study because, in comparison to the
18 NIOSH study, the Mikoczy et al. (2011) study had limitations that would have made the
19 estimates more uncertain than those based on the NIOSH study. In particular, there was greater
20 uncertainty about the exposure estimates [e.g., measurement data were available only from 1973
21 for one plant and 1975 for the other; for earlier exposures, estimates were constructed taking into
22 account information on changes in production methods and environmental controls, subjective
23 memories, and time trends (Hagmar et al., 1991), but this is a less sophisticated approach than
24 that of the NIOSH exposure assessment, which used a detailed, validated regression model];
25 there were many fewer breast cancer cases (41 incident cases vs. 319 cases in NIOSH's full
26 incidence study cohort and 233 in the subcohort with interviews); and there was no information
27 on potential breast cancer risk factors, as was available for the NIOSH subcohort.
28
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-8 DRAFT—DO NOT CITE OR QUOTE
-------
Table J-2. New epidemiological studies of ethylene oxide and human cancer
Population/
industry
Population-based
case-control study
involving
22 centers in
6 European
countries (Czech
Republic, France,
Germany, Italy,
Ireland, Spain)
[EPILYMPH
study]
Kiranetal. (2010)
Number of
subjects
2,347 cases
(1,3 14 male,
1,033 female);
2,463 controls
(1,321 male,
1,142 female),
matched on
sex, age
group, and
residence area
Extent of exposure to
ethylene oxide
1.2% of study population
defined as ever-exposed
(3 1 cases, 27 controls)
Health outcomes
All lymphoma
(# cases/* controls) OR (95% CI)
Unexposed (2,3 16/2,436) 1.0
[referent category]
Ever exposed (31/27) 1.3 (0.7, 2.1)
Confidence in exposure classification
low (8/12) 0.8(0.3,1.9)
med or high (23/15) 1.6(0.8,3.1)
p-trend = 0.242
Exposure frequency (no. working hr)
1-5% (16/23) 0.8(0.4,1.4)
>5%(15/4) 4.3(1.4,13.0)
^-trend = 0.107
Exposure intensity (ppm)
<0.5 (15/19) 0.9 (0.4, 1.7)
>0.5(16/8) 2.2(0.9,5.1)
^-trend = 0.197
Duration (years)
<10 (18/16) 1.2(0.6,2.4)
>10 (13/11) 1.3(0.6,3.0)
^-trend = 0.441
Cumulative exposure score
median (18/11) 1.8(0.8,3.9)
p-tiend = 0.246
Other chemicals to
which subjects were
potentially exposed
Would vary by
individual participant
since not industry-
based study; however,
inclusion of farm
work and
occupational
exposure to solvents
in the regression
model did not affect
the risk estimates
Limitations
Low exposure prevalence
in study population, so
small numbers of exposed
cases and controls
Lymphoma subtype
analyses, in particular,
limited by small numbers
Participation rate only
52% in population
controls, but the positive
association was observed
across centers with
different control types
§•
§
I
^
'TS
o
o
a
I
a,
8"
O 5
O I
^ x
31
-------
to
OJ
Table J-2. New epidemiological studies of ethylene oxide and human cancer (continued)
Population/
industry
Two plants that
produced
disposable medical
equipment, Sweden
Mikoczy et al.
(2011)
Same cohort as
Hagmar et al.
(1995);Hagmaret
al. (1991), followed
an additional
16 years
Number of
subjects
2,171
(862 men,
1,309 women)
Extent of exposure to
ethylene oxide
Exposure levels were up
to 75 ppmin 1964 in
Plant B and up to 40 ppm
in 1970 in Plant A.
By 1985, levels had
dropped to below 1 ppm.
For the 2,020 cohort
members for whom job
titles were available, the
median was 0.13 ppm
x years; the
75th percentile was
0.22 ppm x years; and
the 90th percentile was
1.29 ppm x years.
Health outcomes
Lymphohematopoietic cancers:
Mortality (results not shown):
Nonsignificant increases of deaths from
leukemia and lymphoma were reported;
with a 15-yr induction period, these
increases were lowered; with a 15-yr
induction period and restriction to workers
with cumulative exposure estimates above
the median; nonsignificant increases in
leukemia deaths were reported
Incidence:
Cancer (ICD-7) [cases! SIR (95% CD
All lymphohematopoietic
(200-209) [18] 1.25 (0.74, 1.98)
NHL (200+202) [9] 1.44 (0.66, 2.73)
Leukemia (204-205) [5] 1.40 (0.45, 3.26)
Internal analysis of lymphohematopoietic
cancers:
Cum exp gp
ppm x years [cases! IIR (95% CD
0-0.13 [7] 1.00
0.14-0.21 [5] 1.17(0.36,3.78)
>0.22 [5] 0.92 (0.28, 3.05)
Other chemicals to
which subjects were
potentially exposed
Fluorochlorocarbons,
methyl formate
(1:1 mixture with
EtO)
Limitations
Still a youthful cohort
(mean age 56 years), with
small numbers of events
for the study of the
incidence and mortality of
specific cancer types —
203 total cancer cases
(9.4%) and 171 total
deaths (7.9%)
Estimated cumulative
exposures were generally
low
There was no unexposed
referent group; internal
analyses involved
comparison of responses
in the top quartiles of
cumulative exposure to
those in the lower 50% of
cumulative exposures
§•
§
I
^
'TS
o
a
I
a,
8"
O 5
1
3
o
H
W
o
c
o
H
W
-------
to
OJ
Table J-2. New epidemiological studies of ethylene oxide and human cancer (continued)
Population/
industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
(continued from previous page)
Female breast cancer:
mortality (results not shown):
Slight but nonsignificant decrease in the
SMR was reported. With a 15-yr
induction period included, the SMR for
breast cancer was "somewhat increased."
For workers with cumulative exposures
above the median, with a 15-yr induction
period, a "higher than expected" SMR,
which was not statistically significant, was
reported.
Incidence:
41 female breast cancer cases vs.
50.9 expected (ICD-7 170);
SIR = 0.81 (95% CI = 0.58, 1.09)
Internal analysis:
Cum exp gp
ppm x Vrs [cases! IIR (95% CI)
0-0.13 [10] 1.00
0.14-0.21 [14] 2.76(1.20,6.33)
>0.22[17] 3.55(1.58,7.93)
Other chemicals to
which subjects were
potentially exposed
Limitations
§•
§
I
^
'TS
1
O
a
I
a,
O 5
3
o
H
W
1
o
c
o
H
W
-------
1 J.3. REFERENCES
2 Full citations for references cited elsewhere in the assessment are provided in the
3 Reference section of the assessment; full citations for references appearing only in this appendix
4 are provided below.
5
6 Ahn, H; Shin, H. (2006) Determination of ethylene oxide-hemoglobin adduct by silylation and gas chromatography-
7 electron impact-mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 843(2): 202-208.
8 Brown, T; Rushton, L; the British Occupational Cancer Burden Study Group. (2012) Occupational cancer in Britain
9 Haematopoietic malignancies: leukaemia, multiple myeloma, non-Hodgkins lymphoma. Br J Cancer 107: S41-S48.
10 Butterworth, BE; Chapman, JR. (2007) Exposure of hematopoietic stem cells to ethylene oxide during processing
11 represents a potential carcinogenic risk for transplant recipients. Regul Toxicol Pharmacol 49(3): 149-153.
12 Chan, C; Shie, R; Chang, T. (2006) Workers' exposures and potential health risks to air toxics in a petrochemical
13 complex assessed by improved methodology. Int Arch Occup Environ Health 79(2): 135-142.
14 Davis, FG; Erdal, S; Williams, L; Bigner, D. (2006) Work exposures to animal neurocarcinogens. Int J Occup
15 Environ Health 12(1): 16-23.
16 Farmer, PB; Singh, R. (2008) Use of DNA adducts to identify human health risk from exposure to hazardous
17 environmental pollutants: the increasing role of mass spectrometry in assessing biologically effective doses of
18 genotoxic carcinogens [Review]. Mutat Res 659(1-2): 68-76.
19 Fondelli, MC; Costantini, AS; Ercolanelli, M; et al. (2007) Exposure to carcinogens and mortality in a cohort of
20 restoration workers of water-damaged library materials following the River Arno flooding in Florence, 4 November
21 1966. MedLav 98(5): 422-431.
22 Grosse, Y; Baan, R; Straif, K; et al. (2007) Carcinogenicity of 1,3-butadiene, ethylene oxide, vinyl chloride, vinyl
23 fluoride, and vinyl bromide. Lancet Oncol 8(8): 679-680.
24 Hoenerhoff, MJ; Hong, HH; Ton, TV; et al. (2009) A review of the molecular mechanisms of chemically induced
25 neoplasia in rat and mouse models in National Toxicology Program bioassays and their relevance to human cancer
26 [Review]. Toxicol Pathol 37(7): 835-848.
27 Huang, CC; Wu, CF; Shin, WC; et al. (2011) Comparative analysis of urinary N7-(2-hydroxyethyl)guanine for
28 ethylene oxide- and non-exposed workers. Toxicol Lett 202(3): 237-243.
29 Jarabek, AM; Pottenger, LH; Andrews, LS; et al. (2009) Creating context for the use of DNA adduct data in cancer
30 risk assessment: I. Data organization [Review]. Crit Rev Toxicol 39(8): 659-678.
31 Kensler, TW; Ng, D; Carmella, SG; et al. (2012) Modulation of the metabolism of airborne pollutants by
32 glucoraphanin-rich and sulforaphane-rich broccoli sprout beverages in Qidong, China. Carcinogenesis 33(1):
33 101-107.
34 Keshava, N; Jinot, J; Sonawane, B. (2006a) An evaluation of mutagenic mode of action for carcinogenicity:
35 ethylene oxide. EnvironMolMutagen47(6): 442 [abstract].
36 Keshava, N; Woodall, GM; Rice, S; et al. (2006b) An evaluation of the mutagenic mode of action for four
37 environmental carcinogens. Toxicol Sci 90(1-S).
3 8 Kim, EA; Lee, HE; Ryu, HW; et al. (2011) Cases series of malignant lymphohematopoietic disorder in korean
39 semiconductor industry. Saf Health Work 2(2): 122-134.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-12 DRAFT—DO NOT CITE OR QUOTE
-------
1 Kiran, S; Cocco, P; Mannetje, A; et al. (2010) Occupational exposure to ethylene oxide and risk of lymphoma.
2 Epidemiology 21(6):905-910.
3 Lin, JS; Chuang, KT; Huang, MS; Wei, KM. (2007) Emission of ethylene oxide during frying of foods in soybean
4 oil. Food Chem Toxicol 45(4): 568-574.
5 Lindberg, HK; Falck, GC; Catalan, J; et al. (2010) Micronucleus assay for mouse alveolar Type II and Clara cells.
6 Environ Mol Mutagen 51(2): 164-172.
7 Manservigi, M; Tibaldi, E; Soffritti, M. (2010) Toxic and carcinogenic effects of ethylene and its ethylene oxyde
8 metabolite. Eur J Oncol 15(1): 5-23.
9 Mazon, G; Philippin, G; Cadet, J; et al. (2009) The alkyltransferase-like ybaZ gene product enhances nucleotide
10 excision repair of O(6)-alkylguanine adducts in E. coli. DNA Repair 8(6): 697-703.
11 Mazon, G; Philippin, G; Cadet, J; et al. (2010) Alkyltransferase-like protein (eATL) prevents mismatch repair-
12 mediated toxicity induced by O6-alkylguanine adducts in Escherichia coli. PNAS 107(42): 18050-18055.
13 McCarthy, MC; O'Brien, TE; Charrier, JG; Hafner, HR. (2009) Characterization of the chronic risk and hazard of
14 hazardous air pollutants in the United States using ambient monitoring data. Environ Health Perspect 117(5):
15 790-796.
16 Mikoczy, Z; Tinnerberg, H; Bjork, J; Albin, M. (2011) Cancer incidence and mortality in Swedish sterilant workers
17 exposed to ethylene oxide: updated cohort study findings 1972-2006. Int J Environ Res Public Health
18 8(6):2009-2019.
19 Mosavi-Jarrahi, AM; Mohagheghi, MA; Kalaghchi, B; et al. (2009) Estimating the incidence of leukemia
20 attributable to occupational exposure in Iran. Asian Pac J Cancer Prev 10(1): 67-70.
21 Okada, Y; Nakagoshi, A; Tsurukawa, M; et al. (2012) Environmental risk assessment and concentration trend of
22 atmospheric volatile organic compounds in Hyogo Prefecture, Japan. Environ Sci Pollut Res Int 19(1): 201-213.
23 Parsons, BL; Manjanatha, MG; Myers, MB; et al. (2012) Induction of ell and K-Ras mutation in lung DNA of Big
24 Blue mice exposed to ethylene oxide by inhalation. Environ Mol Mutagen 53: S62 [abstract].
25 Smith-Bindman, R. (2012) Environmental causes of breast cancer and radiation from medical imaging: findings
26 from the Institute of Medicine report. Arch Intern Med 172(13): 1023-1027.
27 Snedeker, SM. (2006) Chemical exposures in the workplace: effect on breast cancer risk among women [Review].
28 AAOHN 54(6): 270-279.
29 Steenland, K; Seals, R; Klein, M; et al. (2011) Risk estimation with epidemiologic data when response attenuates at
30 high-exposure levels. Environ Health Perspect 119(6): 831-837.
31 Steinhausen, M; Van Gelder, R; Gabriel, S. (2012) Work-related exposure to carcinogenic, mutagenic and
32 reprotoxic substances in Germany - Part 2: Substances with exposure-risk relationships according to BekGS 910.
33 Gefahrstoffe - Reinhaltung der Luft (Air Quality Control) 72(9): 347-358.
34 Swenberg, JA; Lu, K; Moeller, BC; et al. (2011) Endogenous versus exogenous DNA adducts: their role in
35 carcinogenesis, epidemiology, and risk assessment [Review]. Toxicol Sci 120 Suppl 1: S130-S145.
3 6 Tateo, F; Bononi, M. (2006) Determination of ethylene chlorohydrin as marker of spices fumigation with ethylene
37 oxide. J Food Compos Anal 19(1): 83-87.
38 Tompkins, EM; Jones, DJL; McLuckie, KIE; et al. (2006) Weak mutagenicity of DNA adducts derived from
39 ethylene oxide exposure. Mutagenesis 21(4): 292 [abstract].
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-13 DRAFT—DO NOT CITE OR QUOTE
-------
1 Tretyakova, N; Goggin, M; Sangaraju, D; Janis, G. (2012) Quantitation of DNA adducts by stable Isotope dilution
2 mass spectrometry. Chem Res Toxicol 25(10): 2007-2035.
3 Valdez-Flores, C; Sielken, RL; Teta, MJ. (2011) Quantitative cancer risk assessment for ethylene oxide inhalation in
4 occupational settings. Arch Toxicol 85(10): 1189-1193.
5 van Balen, E; Font, R; Cavalle, N; et al. (2006) Exposure to non-arsenic pesticides is associated with lymphoma
6 among farmers in Spain. Occup Environ Med 63(10): 663-668.
7 Weiderpass, E; Meo, M; Vainio, H. (2011) Risk factors for breast cancer, including occupational exposures. Saf
8 Health Work 2(1): 1-8.
9 Won, JU. (2010) Health effects of chemicals used in hospitals among healthcare workers. J Korean Med Assoc
10 53(6): 474-482.
11 Wu, KY; Chiang, SY; Shih, WC; et al. (2011) The application of mass spectrometry in molecular dosimetry:
12 ethylene oxide as an example. Mass Spectrom Rev 30(5):733-756.
13
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 J-14 DRAFT—DO NOT CITE OR QUOTE
-------
1 APPENDIX K.
2 DOCUMENTATION OF IMPLEMENTATIONS OF THE
3 2011 NATIONAL RESEARCH COUNCIL RECOMMENDATIONS
4 Background: On December 23, 2011, The Consolidated Appropriations Act, 2012, was
5 signed into law.12 The report language included direction to EPA for the Integrated Risk
6 Information System (IRIS) Program related to recommendations provided by the National
7 Research Council (NRC) in its review of EPA's draft IRIS assessment of formaldehyde.13 The
8 report language included the following:
9
10
11 The Agency shall incorporate, as appropriate, based on chemical-specific data sets
12 and biological effects, the recommendations of Chapter 7 of the National
13 Research Council' s Review of the Environmental Protection Agency's Draft IRIS
14 Assessment of Formaldehyde into the IRIS process.. .For draft assessments
15 released in fiscal year 2012, the Agency shall include documentation describing
16 how the Chapter 7 recommendations of the National Academy of Sciences (NAS)
17 have been implemented or addressed, including an explanation for why certain
18 recommendations were not incorporated.
19
20
21 The NRC's recommendations, provided in Chapter 7 of the review report, offered
22 suggestions to EPA for improving the development of IRIS assessments. Consistent with the
23 direction provided by Congress, documentation of how the recommendations from Chapter 7 of
24 the NRC report have been implemented in this assessment is provided in the tables below.
25 Where necessary, the documentation includes an explanation for why certain recommendations
26 were not incorporated.
27 The IRIS Program's implementation of the NRC recommendations is following a phased
28 approach that is consistent with the NRC's "Roadmap for Revision" as described in Chapter 7 of
29 the formaldehyde review report. The NRC stated that, "the committee recognizes that the
30 changes suggested would involve a multiyear process and extensive effort by the staff at the
31 National Center for Environmental Assessment and input and review by the EPA Science
32 Advisory Board and others."
33 The IRIS ethylene oxide carcinogenicity assessment is in Phase 1 of implementation,
34 which focuses on a subset of the short-term recommendations, such as editing and streamlining
12Pub. L. No. 112-74, Consolidated Appropriations Act, 2012.
13National Research Council, (2011). Review of the Environmental Protection Agency's Draft IRIS Assessment of
Formaldehyde.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 K-1 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
documents, increasing transparency and clarity, and using more tables, figures, and appendices to
present information and data in assessments. Phase 1 also focuses on assessments near the end
of the development process and close to final posting. Chemical assessments in Phase 2 of the
implementation will address all of the short-term recommendations from Table K-l. Chemical
assessments in Phase 3 of implementation will incorporate the longer-term recommendations
made by the NRC as outlined below in Table K-2. On May 16, 2012, EPA announced14 that as a
part of a review of the IRIS Program's assessment development process, the NRC will also
review current methods for weight-of-evidence analyses and recommend approaches for
weighing scientific evidence for chemical hazard identification. This effort is included in
Phase 3 of EPA's implementation plan.
Table K-l. The EPA's implementation of the National Research Council's
recommendations in the ethylene oxide (EtO) carcinogenicity assessment
NRC recommendations that the EPA is
implementing in the short term
Implementation in the EtO carcinogenicity
assessment
General recommendations for completing the IRIS formaldehyde assessment that the EPA will adopt
for all IRIS assessments (p. 152)
1. To enhance the clarity of the document, the
draft IRIS assessment needs rigorous editing to
reduce the volume of text substantially and
address redundancies and inconsistencies.
Long descriptions of particular studies should
be replaced with informative evidence tables.
When study details are appropriate, they could
be provided in appendices.
Partially Implemented. EtO is a post-peer review,
Phase 1 chemical; as such, implementation has
focused on a subset of the short-term
recommendations, such as editing and streamlining,
increasing transparency and clarity, and using more
tables, figures, and appendices to present information
and data. For example, the main body of the EtO
Assessment contains only abbreviated study
summaries and study summary tables. The longer
descriptions of the epidemiology studies and the
genotoxicity studies are contained in appendices
(Appendices A and C, respectively), along with a
detailed summary table of the epidemiology studies
in Appendix A. The main text of the hazard
identification chapter (see Chapter 3) is
comparatively streamlined.
16
14EPA Announces NAS' Review of IRIS Assessment Development Process (www.epa.gov/iris).
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 K-2 DRAFT—DO NOT CITE OR QUOTE
-------
Table K-l. The EPA's implementation of the National Research Council's
recommendations in the ethylene oxide (EtO) carcinogenicity assessment
(continued)
NRC recommendations that the EPA is
implementing in the short term
Implementation in the EtO carcinogenicity
assessment
Chapter 1 needs to be expanded to describe
more fully the methods of the assessment,
including a description of search strategies used
to identify studies with the exclusion and
inclusion criteria articulated and a better
description of the outcomes of the searches and
clear descriptions of the weight-of-evidence
approaches used for the various noncancer
outcomes. The committee emphasizes that it is
not recommending the addition of long
descriptions of the EPA guidelines to the
introduction, but rather clear concise statements
of criteria used to exclude, include, and
advance studies for derivation of the reference
concentrations (RfCs) and unit risk estimates.
Partially Implemented. The EPA's literature search
strategy is described briefly in Chapter 2 of the EtO
Assessment. To update the Assessment, a systematic
literature search was done covering the time span
from 2006 (the year of the 1st external review draft)
to May 2013; this search is described in detail in
Appendix J.
In addition, the text has been expanded to include
more description of the considerations made in
evaluating the epidemiology studies (p. 3-1) and in
selecting the study that formed the basis for the
quantitative cancer risk estimates (p. 4-1-4-3).
3. Standardized evidence tables for all health
outcomes need to be developed. If there were
appropriates tables, long text descriptions of
studies could be moved to an appendix or
deleted.
Partially Implemented. This assessment was
largely finalized before the release of the NRC
recommendations; thus, the tables herein may not be
consistent with current standardizations. However,
the EtO Assessment contains a detailed summary
table of the epidemiology studies in Appendix A (see
Table A-5) along with the longer text study
descriptions. Less detailed tables of the results are
presented in the main text (see Tables 3-1 and 3-2).
All critical studies need to be thoroughly
evaluated with standardized approaches that are
clearly formulated and based on the type of
research: for example, observational
epidemiologic or animal bioassays. The
findings of the reviews might be presented in
tables to ensure transparency.
Partially Implemented. All critical studies were
thoroughly evaluated in Chapter 3 and Appendix A.
As discussed above, the text has been expanded to
include more description of the considerations made
in evaluating the epidemiology studies (see p. 3-1),
and the epidemiology studies are summarized in a
detailed table in Appendix A (see Table A-5).
Standardized approaches for evaluating studies are
under development as part of Phases 2 and 3.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
K-3 DRAFT—DO NOT CITE OR QUOTE
-------
Table K-l. The EPA's implementation of the National Research Council's
recommendations in the ethylene oxide (EtO) carcinogenicity assessment
(continued)
NRC recommendations that the EPA is
implementing in the short term
Implementation in the EtO carcinogenicity
assessment
5. The rationales for the selection of the studies
that are advanced for consideration in
calculating the RfCs and unit risks need to be
expanded. All candidate RfCs should be
evaluated together with the aid of graphic
displays that incorporate selected information
on attributes relevant to the database.
Implemented. As discussed above, the text has been
expanded to include more description of the
considerations made in selecting the study that
formed the basis for the quantitative cancer risk
estimates (see p. 4-1-4-3). The selection
considerations are also summarized in a table (see
Table 4-1). The EtO Assessment is a carcinogenicity
assessment; thus, no RfCs or reference doses (RfDs)
are derived.
Strengthened, more integrative and more
transparent discussions of weight of evidence
are needed. The discussions would benefit
from more rigorous and systematic coverage of
the various determinants of weight of evidence,
such as consistency.
Implemented. The weight-of-evidence discussion in
the EtO Assessment has been substantially enhanced
(see Section 3.5.1), and two tables have been added
addressing consistency in the epidemiology study
results (see Table 3-1 for lymphohematopoietic
cancer and Table 3-2 for breast cancer).
General Guidance for the Overall Process (seep. 164)
1. Elaborate an overall, documented, and
quality-controlled process for IRIS
assessments.
Ensure standardization of review and
evaluation approaches among contributors and
teams of contributors; for example, include
standard approaches for reviews of various
types of studies to ensure uniformity.
9. Assess disciplinary structure of teams needed
to conduct the assessments.
Partially Implemented. A team approach was used
for the development of the EtO Assessment to help
ensure that the necessary disciplinary expertise was
available for assessment development and review.
Because EtO is a post-peer review, phase one
chemical, the EtO team did not have access to the
"overall, documented, and quality-controlled
process" that is now being developed in response to
the NRC recommendations.
Evidence Identification: Literature Collection and Collation Phase (seep. 164)
10. Select outcomes on the basis of available
evidence and understanding of mode of action.
Implemented. The EtO Assessment has detailed
discussions of genotoxicity (see Section 3.3.3 and
Appendix C) and mode of action (see Section 3.4),
and EPA concludes that the weight of evidence
supports a mutagenic mode of action for EtO
carcinogenicity. The cancer outcomes selected are
consistent with that mode-of-action finding as well as
the available hazard evidence.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
K-4 DRAFT—DO NOT CITE OR QUOTE
-------
Table K-l. The EPA's implementation of the National Research Council's
recommendations in the ethylene oxide (EtO) carcinogenicity assessment
(continued)
NRC recommendations that the EPA is
implementing in the short term
1 1 . Establish standard protocols for evidence
identification.
12. Develop a template for description of the
search approach.
13. Use a database, such as the Health and
Environmental Research Online (HERO)
database, to capture study information and
relevant quantitative data.
Implementation in the EtO carcinogenicity
assessment
Partially Implemented. This is being implemented
by the IRIS program as part of Phase 2. The EPA's
literature search strategy is described briefly in
Chapter 2 of the EtO Assessment. More details of
the original search are no longer available for this
Assessment, which was largely finalized before the
release of the NRC recommendations. To update the
Assessment, a systematic literature search was done
covering the time span from 2006 (the year of the 1st
external review draft) to May 2013; this search is
described in detail in Appendix J.
This is being implemented by the IRIS program as
part of Phase 2.
Implemented. HERO links were incorporated for all
citations.
Evidence Evaluation: Hazard Identification and Dose-Response Modeling (see p. 165)
14. Standardize the presentation of reviewed
studies in tabular or graphic form to capture
the key dimensions of study characteristics,
weight of evidence, and utility as a basis for
deriving reference values and unit risks.
15. Develop templates for evidence tables, forest
plots, or other displays.
16. Establish protocols for review of major types
of studies, such as epidemiologic and
bioassay.
Partially Implemented. This Assessment was
largely finalized before the release of the NRC
recommendations; thus, the tables herein may not be
consistent with current standardizations. The EtO
Assessment does include a detailed summary table of
key characteristics of the epidemiology studies (see
Table A-5 of Appendix A) and a table summarizing
the considerations made in selecting the study that
formed the basis for the quantitative cancer risk
estimates (see Table 4-1).
This is being implemented by the IRIS program as
part of Phase 2.
Partially Implemented. This is being implemented
by the IRIS program as part of Phase 2. The study
review process was not revised for this assessment
because EtO is a Phase 1 chemical. However, this
assessment was developed using standard protocols
for evidence evaluation that are provided in existing
EPA guidance.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
K-5 DRAFT—DO NOT CITE OR QUOTE
-------
Table K-l. The EPA's implementation of the National Research Council's
recommendations in the ethylene oxide (EtO) carcinogenicity assessment
(continued)
NRC recommendations that the EPA is
implementing in the short term
Implementation in the EtO carcinogenicity
assessment
Selection of Studies for Derivation of Reference Values and Unit Risks (seep. 165)
17. Establish clear guidelines for study selection.
a. Balance strengths and weaknesses.
b. Weigh human vs. experimental evidence.
c. Determine whether combining estimates
among studies is warranted.
Partially Implemented. As discussed above, the
text has been expanded to include more description of
the considerations made in selecting the study that
formed the basis for the quantitative cancer risk
estimates (see p. 4-1-4-3). The selection
considerations are also summarized in a table (see
Table 4-1). Consideration was given to combining
data from the Union Carbide Cohort (UCC) and
NIOSH cohort studies, and discussion is provided for
why the UCC data were ultimately not used (see
Section 4.1). The EtO Assessment is a
carcinogenicity assessment; thus, no RfCs or RfDs
are derived.
Calculation of Reference Values and Unit Risks (see pp. 165-166)
18. Describe and justify assumptions and models
used. This step includes review of dosimetry
models and the implications of the models for
uncertainty factors; determination of
appropriate points of departure (such as
benchmark dose, no-observed-adverse-effect
level, and lowest observed-adverse-effect
level), and assessment of the analyses that
underlie the points of departure.
19. Provide explanation of the risk-estimation
modeling processes (for example, a statistical
or biologic model fit to the data) that are used
to develop a unit risk estimate.
Implemented. The EtO Assessment has a detailed
discussion of model selection for the epidemiological
data sets (see Section 4.1) and the laboratory animal
data sets (see Section 4.2), including a discussion of
cross-species scaling (see Section 4.2.2). The EtO
Assessment is a carcinogenicity assessment; thus, no
reference values are derived.
Implemented. The EtO Assessment has a detailed
discussion of model selection for the epidemiological
data sets (see Section 4.1) and the laboratory animal
data sets (see Section 4.2).
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
K-6 DRAFT—DO NOT CITE OR QUOTE
-------
Table K-l. The EPA's implementation of the National Research Council's
recommendations in the ethylene oxide (EtO) carcinogenicity assessment
(continued)
NRC recommendations that the EPA is
implementing in the short term
Implementation in the EtO carcinogenicity
assessment
20. Provide adequate documentation for
conclusions and estimation of reference values
and unit risks. As noted by the committee
throughout the present report, sufficient
support for conclusions in the formaldehyde
draft IRIS assessment is often lacking. Given
that the development of specific IRIS
assessments and their conclusions are of
interest to many stakeholders, it is important
that they provide sufficient references and
supporting documentation for their
conclusions. Detailed appendixes, which
might be made available only electronically,
should be provided, when appropriate.
Implemented. The EtO Assessment includes, as an
appendix (see Appendix D), more detailed fit
statistics and modeling results for the epidemiological
cancer data sets. Appendix G provides results of the
laboratory animal tumor modeling. The EtO
Assessment is a carcinogenicity assessment; thus, no
reference values are derived.
1
2
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
K-7 DRAFT—DO NOT CITE OR QUOTE
-------
1
2
3
Table K-2. National Research Council recommendations that the EPA is generally
implementing in the long term
NRC recommendations that the EPA is implementing
in the long term
Implementation in the EtO Carcinogenicity
Assessment
Weight-of-Evidence Evaluation: Synthesis of Evidence
for Hazard Identification (seep. 165)
1. Review use of existing weight-of-evidence
guidelines.
2. Standardize approach to using weight-of-evidence
guidelines.
3. Conduct agency workshops on approaches to
implementing weight-of-evidence guidelines.
4. Develop uniform language to describe strength of
evidence on noncancer effects.
5. Expand and harmonize the approach for
characterizing uncertainty and variability.
6. To the extent possible, unify consideration of
outcomes around common modes of action rather
than considering multiple outcomes separately.
As indicated above, Phase 3 of EPA's
implementation plan will incorporate the
longer-term recommendations made by the
NRC. On May 16, 2012, EPA announced that
as a part of a review of the IRIS Program's
assessment development process, the NRC will
also review current methods for weight-of-
evidence analyses and recommend approaches
for weighing scientific evidence for chemical
hazard identification. In addition, EPA will
hold a workshop on August 26, 2013, on issues
related to weight of evidence to inform future
assessments.
Calculation of Reference Values and Unit Risks (see
pp. 165-166)
1. Assess the sensitivity of derived estimates to model
assumptions and end points selected. This step
should include appropriate tabular and graphic
displays to illustrate the range of the estimates and
the effect of uncertainty factors on the estimates.
Implemented. The EtO Assessment is a
carcinogenicity assessment; thus, no reference
values are derived. Chapter 4 presents
derivations of unit risk estimates for multiple
data sets, species, and models. Many of these
derivations are summarized in tables and
figures; for example, for the breast cancer
incidence subcohort, Figure 4-5 depicts the
range of relative risk estimates for different
exposure-response models considered and
Table 4-13 summarizes unit risk estimates
derived from different models.
7/2013
This document is a draft for review purposes only and does not constitute Agency policy.
K-8 DRAFT—DO NOT CITE OR QUOTE
-------
REFERENCES
This reference list includes all the references cited in the document except for
Appendix B, which is a reference list pertaining to Figure 3-3, and Appendix J, which includes
some more recent references. References added after the 2007 external peer review are also
listed separately in Appendix I. References identified in a May 2013 literature search but
appearing after the 30 June 2010 cutoff date for literature inclusion into this carcinogenicity
assessment are cited and discussed in Appendix J.
Abeles. FB: Heggestad. HE. (1973). Ethylene: An urban air pollutant. J Air Waste Manag Assoc 23: 517-521.
http://dx.doi.org/10.1080/00022470.1973.10469798
Adam. B: Bardos. H: Adany. R. (2005). Increased genotoxic susceptibility of breast epithelial cells to ethylene
oxide. MutatRes 585: 120-126. http://dx.doi.0rg/10.1016/i.mrgentox.2005.04.009
Agurell E: Cederberg. H: Ehrenberg. L: Lindahl-Kiessling. K: Rannug. U: Tornqvist M. (1991). Genotoxic effects
of ethylene oxide and propylene oxide: a comparative study. Mutat Res 250: 229-237.
Ahn. HS: Shin. HS. (2006). Determination of ethylene oxide-hemoglobin adduct by silylation and gas
chromatography-electron impact-mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 843: 202-
208. http://dx.doi.0rg/10.1016/i.ichromb.2006.06.003
Appelgren. LE: Eneroth. G: Grant. C: Lanstrom. LE: Tenghagen. K. (1978). Testing of ethylene oxide for
mutagenicity using the micronucleus test in mice and rats. Acta Pharmacol Toxicol 43: 69-71.
http://dx.doi.org/10.1111/i. 1600-0773.1978.tb02235.x
Applebaum. KM: Mallov. EJ: Eisen. EA. (2007). Reducing healthy worker survivor bias by restricting date of hire
ina cohort study of Vermont granite workers. Occup EnvironMed 64: 681-687.
http://dx.doi.org/10.1136/oem.2006.031369
Armstrong. B. (1990). The effects of measurement errors on relative risk regressions. Am J Epidemiol 132:1176-
1184.
Armstrong. B. (1998). Effect of measurement error on epidemiological studies of environmental and occupational
exposures [Review]. Occup EnvironMed 55: 651-656. http://dx.doi.org/10.1136/oem.55.10.651
Arrighi. HM: Hertz-Picciotto. I. (1994). The evolving concept of the healthy worker survivor effect [Review].
Epidemiology 5: 189-196.
Bastlova. T: Andersson. B: Lambert. B: Kolman. A. (1993). Molecular analysis of ethylene oxide-induced mutations
at the HPRT locus in human diploid fibroblasts. Mutat Res 287: 283-292.
BEIR (Committee on the Biological Effects of Ionizing Radiation). (1988). Health risks of radon and other
internally deposited alpha-emitters. Washington, DC: National Academy Press.
Benson. LO: Teta. MJ. (1993). Mortality due to pancreatic and lymphopoietic cancers in chlorohydrin production
workers. Br J IndMed 50: 710-716. http://dx.doi.0rg/10.1136/oem.50.8.710
Beranek. DT. (1990). Distribution of methyl and ethyl adducts following alkylation with monofunctional alkylating
agents [Review]. MutatRes 231: 11-30.
Bisanti. L: Magginj M: Raschetti R: Alegianj SS: Ippolito. FM: Caffari. B: Segnaa N: Ponti A. (1993). Cancer
mortality in ethylene oxide workers. Br J Ind Med 50: 317-324.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-1 DRAFT—DO NOT CITE OR QUOTE
-------
Boffetta. P: vanderHel O: Norppa. H: Fabianova. E: Fucic. A: Gundy. S: Lazutka. J: Cebulska-Wasilewska. A:
Puskailerova. D: Znaor. A: Kelecsenyi. Z: Kurtinaitis. J: Rachtan. J: Forni. A: Vermeulen. R: Bonassi. S. (2007).
Chromosomal aberrations and cancer risk: results of a cohort study from Central Europe. Am J Epidemiol 165: 36-
43. http://dx.doi.org/10.1093/aje/kwj367
Bolt HM: Leutbecher. M: Golka. K. (1997). A note on the physiological background of the ethylene oxide adduct 7-
(2-hydroxyethyl) guanine in DNA from human blood [Letter]. Arch Toxicol 71: 719-721.
http://dx.doi.org/10.1007/s002040050451
Bolt HM: Peter. H: Post U. (1988). Analysis of macromolecular ethylene oxide adducts [Review]. Int Arch Occup
Environ Health 60: 141-144.
Bonassi. S: Znaor. A: Ceppi. M: Lando. C: Chang. WP: Holland. N: Kirsch-Volders. M: Zeiger. E: Ban. S: Barale.
R: Bigatti. MP: Bolognesi. C: Cebulska-Wasilewska. A: Fabianova. E: Fucic. A: Hagmar. L: Joksic. G: Martelli. A:
Migliore. L: Mirkova. E: Scarfi. MR: Zijno. A: Norppa. H: Fenech. M. (2007). An increased micro nucleus
frequency in peripheral blood lymphocytes predicts the risk of cancer in humans. Carcinogenesis 28: 625-631.
http://dx.doi.org/10.1093/carcin/bgll77
Boogaard. PJ. (2002). Use of haemoglobin adducts in exposure monitoring and risk assessment [Review]. J
Chromatogr B Analyt Technol Biomed Life Sci 778: 309-322. http://dx.doi.org/10.1016/S0378-4347(01)00445-5
Boysen. G: Pachkowski. BF: Nakamura. J: Swenberg. JA. (2009). The formation and biological significance of N7-
guanine adducts [Review]. MutatRes 678: 76-94. http://dx.doi.0rg/10.1016/j.mrgentox.2009.05.006
Britton. DW: Tornqvist. M: van Sittert NJ: Watson. WP: Wraith. MJ: Wright AS. (1991). Immunochemical and
GC/MS analysis of protein adducts: dosimetry studies with ethylene oxide [Review]. Prog Clin Biol Res 372: 99-
106.
Brown. CD: Bahman. A: Turner. MJ: Fennell. TR. (1998). Ethylene oxide dosimetry in the mouse. Toxicol Appl
Pharmacol 148: 215-222. http://dx.doi.org/10.1006/taap.1997.8349
Brown. T: Rushton. L: Group. tBOCBS. (2012). Occupational cancer in Britain Haematopoietic malignancies:
leukaemia, multiple myeloma, non-Hodgkins lymphoma. Br J Cancer 107: S41-S48.
http://dx.doi.org/10.1038/bic.2012.117
Butterworth. BE: Chapman. JR. (2007). Exposure of hematopoietic stem cells to ethylene oxide during processing
represents a potential carcinogenic risk for transplant recipients. Regul Toxicol Pharmacol 49: 149-153.
http://dx.doi.0rg/10.1016/i.vrtph.2007.07.004
Chaa C: Shie. R: Chang. T. (2006). Workers' exposures and potential health risks to air toxics in a petrochemical
complex assessed by improved methodology. Int Arch Occup Environ Health 79: 135-142.
http://dx.doi.org/10.1007/s00420-005-0028-9
Chandra. GR: Spencer. M. (1963). A micro apparatus for absorption of ethylene and its use in determination of
ethylene in exhaled gases from human subjects. Biochim Biophys Acta 69: 423-425.
Clare. MG: Dean. BJ: De Jong. G: Van Sittert. NJ. (1985). Chromosome analysis of lymphocytes from workers at
an ethylene oxide plant. DNA Repair 156: 109-116.
Coggon. D: Harris. EC: Poole. J: Palmer. KT. (2004). Mortality of workers exposed to ethylene oxide: extended
follow up of a British cohort. Occup Environ Med 61: 358-362.
Csanady. GA: Denk. B: Putz. C: Kreuzer. PE: Kessler. W: Baur. C: Gargas. ML: Filser. JG. (2000). A physiological
toxicokinetic model for exogenous and endogenous ethylene and ethylene oxide in rat, mouse, and human:
formation of 2-hydroxyethyl adducts with hemoglobin and DNA. Toxicol Appl Pharmacol 165: 1-26.
http://dx.doi.org/10.1006/taap.2000.8918
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-2 DRAFT—DO NOT CITE OR QUOTE
-------
Cushnir. JR: Lamb. JH: Parry. A: Farmer. PB. (1991). Tandem mass spectrometric approaches for determining
exposure to alkylating agents. In IK O'Neill; J Chen; H Bartsch (Eds.), Relevance to human cancer of N-nitroso
compounds, tobacco, and mycotoxins (pp. 107-112). Lyon, France: International Agency for Research on Cancer.
Davis. FG; Erdal S; Williams. L; Bigner. D. (2006). Work exposures to animal neurocarcinogens. Int J Occup
Environ Health 12: 16-23.
de Serres. FJ; Brockman. HE. (1995). Ethylene oxide: induction of specific-locus mutations in the ad-3 region of
heterokaryon 12 of Neurospora crassa and implications for genetic risk assessment of human exposure in the
workplace. MutatRes 328: 31-47.
Deddens. JA; Hornung. R. (1994). Quantitative examples of continuous exposure measurement errors that bias risk
estimates away from the null. In CM Smith; DC Christiani; KT Kelsey (Eds.), Chemical risk assessment of
occupational health: current applications, limitations, and future prospects (pp. 77-85). Westport, CT: Auburn
House.
Dellarco. VL; Generoso. WM; Sega. GA; Fowle. JR. Ill; Jacobson-Kram. D. (1990). Review of the mutagenicity of
ethylene oxide [Review]. EnvironMol Mutagen 16: 85-103.
Divine. BJ; Amanollahi. KS. (1986). Ethylene oxide and cancer [Letter]. JAMA 256: 1726-1727.
Dormer. EM; Wong. BA; James. RA; Preston. RJ. (2010). Reciprocal translocations in somatic and germ cells of
mice chronically exposed by inhalation to ethylene oxide: implications for risk assessment. Mutagenesis 25: 49-55.
http://dx.doi.org/10.1093/mutage/gep042
Ehrenberg. L; Gustafsson. A. (1970). Chemical mutagens: Their uses and hazards in medicine and technology [A
report of February 1959 to the National Board of Health]. Lund, Sweden: Ehrenberg, L; Gustafsson, A.
Ehrenberg. L; Hallstrom. T. (1967). Haematologic studies on persons occupationally exposed to ethylene oxide. In
Radiosterilization of medical products (pp. 327-334). (SM 92/26; STI/PUB/157). Vienna: International Atomic
Energy Agency. http://www-pub.iaea.org/books/IAEABooks/2114/Radiosterilization-of-Medical-Products-
Budapest-5-9-June-1967
Ehrenberg. L; Hussain. S. (1981). Genetic toxicity of some important epoxides [Review]. DNA Repair 86: 1-113.
Ehrenberg. L; Osterman-Golkar. S; Segerback. D; Svensson. K; Calleman. CJ. (1977). Evaluation of genetic risks of
alkylating agents. III. Alkylation of haemoglobin after metabolic conversion of ethene to ethene oxide in vivo. Mutat
Res 45: 175-184.
Eide. I: Zhao. C: Kumar. R: Hemminki. K: Wu. K: Swenberg. JA. (1999). Comparison of (32)P-postlabeling and
high-resolution GC/MS in quantifying N7-(2-Hydroxyethyl)guanine adducts. ChemRes Toxicol 12: 979-984.
EPIC (Ethylene Oxide Industry Council). (2001). Toxicological review of ethylene oxide in support of summary
information on the integrated risk information system. Arlington, VA.
Farmer. PB; Bailey. E; Naylor. S; Anderson. D; Brooks. A; Cushnir. J; Lamb. JH; Sepal O; Tang. YS. (1993).
Identification of endogenous electrophiles by means of mass spectrometric determination of protein and DNA
adducts [Review]. Environ Health Perspect 99: 19-24.
Farmer. PB; Shuker. DE. (1999). What is the significance of increases in background levels of carcinogen-derived
protein and DNA adducts? Some considerations for incremental risk assessment [Review]. Mutat Res 424: 275-286.
Farmer. PB; Singh. R. (2008). Use of DNA adducts to identify human health risk from exposure to hazardous
environmental pollutants: the increasing role of mass spectrometry in assessing biologically effective doses of
genotoxic carcinogens [Review]. MutatRes 659: 68-76. http://dx.doi.0rg/10.1016/i.mrrev.2008.03.006
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-3 DRAFT—DO NOT CITE OR QUOTE
-------
Farooqi Z: Tornqvist M: Ehrenberg. L: Nataraian. AT. (1993). Genotoxic effects of ethylene oxide and propylene
oxide in mouse bone marrow cells. Mutat Res 288: 223-228.
Fennell TR: MacNeela. JP: Morris. RW: Watson. M: Thompson. CL: Bell DA. (2000). Hemoglobin adducts from
acrylonitrile and ethylene oxide in cigarette smokers: effects of glutathione S-transferase Tl-null and Ml-null
genotypes. Cancer Epidemiol Biomarkers Prev 9: 705-712.
Filser. JG: Denk. B: Tornqvist. M: Kessler. W: Ehrenberg. L. (1992). Pharmacokinetics of ethylene in man: body
burden with ethylene oxide and hydroxyethylation of hemoglobin due to endogenous and environmental ethylene.
ArchToxicol66: 157-163.
Fondelli. MC: Costantinj AS: Ercolanelli M: Pizzo. AM: Maltoni. SA: Quinn. MM. (2007). Exposure to
carcinogens and mortality in a cohort of restoration workers of water-damaged library materials following the River
Arno flooding in Florence, 4 November 1966. Med Lav 98: 422-431.
Post. U: Marczynskj B: Kasemann. R: Peter. H. (1989). Determination of 7-(2-hydroxyethyl)guanine with gas
chromatography/mass spectrometry as a parameter for genotoxicity of ethylene oxide. Arch Toxicol Suppl 13: 250-
253.
Fuchs. J: Wullenweber. U: Hengstler. JG: Bienfait. HG: Hiltl. G: Oesch. F. (1994). Genotoxic risk for humans due
to work place exposure to ethylene oxide: remarkable individual differences in susceptibility. Arch Toxicol 68: 343-
348.
Galloway. SM: Berry. PK: Nichols. WW: Wolman. SR: Soper. KA: Stollev. PD: Archer. P. (1986). Chromosome
aberrations in individuals occupationally exposed to ethylene oxide, and in a large control population. Mutat Res
Genet Toxicol 170: 55-74. http://dx.doi.org/10.1016/0165-1218(86')90082-0
Gardner. MJ: Coggon. D: Pannett. B: Harris. EC. (1989). Workers exposed to ethylene oxide: a follow up study.
Occup Environ Med 46: 860-665.
Carman. RH: Snellings. WM: Maronpot RR. (1985). Brain tumors in F344 rats associated with chronic inhalation
exposure to ethylene oxide. Neurotoxicology 6: 117-137.
Garry. VF: Hozier. J: Jacobs. D: Wade. RU Gray. DG. (1979). Ethylene oxide: evidence of human chromosomal
effects. Environ Mutagen 1: 375-382. http://dx.doi.org/10.1002/em.2860010410
Generoso. WM: Cain. KT: Cornett CV: Cacheiro. NLA: Hughes. LA. (1990). Concentration-response curves for
ethylene-oxide-induced heritable translocations and dominant lethal mutations. Environ Mol Mutagen 16: 126-131.
Generoso. WM: Caia KT: Hughes. LA: Sega. GA: Bradea PW: Gosslee. DG: Shelby. MD. (1986). Ethylene oxide
dose and dose-rate effects in the mouse dominant-lethal test. Environ Mol Mutagen 8: 1-7.
Generoso. WM: Rutledge. JC: Cain. KT: Hughes. LA: Downing. DJ. (1988). Mutagen-induced fetal anomalies and
death following treatment of females within hours after mating. DNA Repair 199: 175-181.
htto://dx.doi.org/10.1016/0165-1161(88)90245-2
Godderis. L: Aka. P: Matecuca. R: Kirsch-Volders. M: Lisoa D: Veulemans. H. (2006). Dose-dependent influence
of genetic polymorphisms on DNA damage induced by styrene oxide, ethylene oxide and gamma-radiation.
Toxicology 219: 220-229. http://dx.doi.0rg/10.1016/i.tox.2005.ll.021
Golberg. L. (1986). Chemical and physical properties. In Hazard assessment of ethylene oxide. Boca Raton, FL:
CRC Press.
Greenberg. HL: Ott MG: Shore. RE. (1990). Men assigned to ethylene oxide production or other ethylene oxide
related chemical manufacturing: A mortality study. Br J Ind Med 47: 221-230.
http://dx.doi.0rg/10.1136/oem.47.4.221
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-4 DRAFT—DO NOT CITE OR QUOTE
-------
Greife. AL: Hornung. RW: Stayner. LG: Steenland. KN. (1988). Development of a model for use in estimating
exposure to ethylene oxide in a retrospective cohort mortality study. Scand J Work Environ Health 1: 29-30.
Grosse. Y: Baaa R: Straif. K: Secretan. B: El Ghissassi F: Bouvard. V: Altieri. A: Cogliano. V. (2007).
Carcinogenicity of 1,3-butadiene, ethylene oxide, vinyl chloride, vinyl fluoride, and vinyl bromide. Lancet Oncol 8:
679-680. http://dx.doi.org/10.1016/81470-2045(07)70235-8
Gupta. RC: Lutz. WK. (1999). Background DNA damage for endogenous and unavoidable exogenous carcinogens:
a basis for spontaneous cancer incidence? DNA Repair 424: 1-8.
Hagmar. L: Mikoczy. Z: Welinder. H. (1995). Cancer incidence in Swedish sterilant workers exposed to ethylene
oxide. Occup EnvironMed 52: 154-156. http://dx.doi.0rg/10.1136/oem.52.3.154
Hagmar. L: Stromberg. U: Bonassi. S: Hansteen. IL: Knudsen. LE: Lindholm. C: Norppa. H. (2004). Impact of
types of lymphocyte chromosomal aberrations on human cancer risk: results from Nordic and Italian cohorts. Cancer
Res 64: 2258-2263. http://dx.doi.org/10.1158/0008-5472.CAN-03-3360
Hagmar. L: Welinder. H: Linden. K: Attewell R: Osterman-Golkar. S: Tornqvist M. (1991). An epidemiological
study of cancer risk among workers exposed to ethylene oxide using hemoglobin adducts to validate environmental
exposure assessments. Int Arch Occup Environ Health 63: 271-277'.
Hairier. E: Langhof. T: Dannappel. D: Leutbecher. M: Schroder. K: Goergens. HW: Muller. A: Bolt. HM. (1993).
Polymorphism of glutathione conjugation of methyl bromide, ethylene oxide and dichloromethane in human blood:
Influence on the induction of sister chromatid exchanges (SCE) in lymphocytes. Arch Toxicol 67: 173-178.
Haufroid. V: Merz. B: Hofmann. A: Tschopp. A: Lison. D: Hotz. P. (2007). Exposure to ethylene oxide in hospitals:
biological monitoring and influence of glutathione S-transferase and epoxide hydrolase polymorphisms. Cancer
Epidemiol Biomarkers Prev 16: 796-802. http://dx.doi.org/10.1158/1055-9965.EPI-06-0915
Hengstler. JG: Fuchs. J: Gebhard. S: Oesch. F. (1994). Glycolaldehyde causes DNA-protein crosslinks: a new aspect
of ethylene oxide genotoxicity. MutatRes 304: 229-234.
Hill. AB. (1965). The environment and disease: Association or causation? Proc R Soc Med 58: 295-300.
Hoenerhoff. MJ: Hong. HH: Ton. TV: Lahousse. SA: Sills. RC. (2009). A review of the molecular mechanisms of
chemically induced neoplasia in rat and mouse models in National Toxicology Program bioassays and their
relevance to human cancer [Review]. Toxicol Pathol 37: 835-848. http://dx.doi.org/10.1177/0192623309351726
Hogstedt. B: Gullberg. B: Hedner. K: Kolnig. AM: Mitelman. F: Skerfving. S: Widegrea B. (1983). Chromosome
aberrations and micronuclei in bone marrow cells and peripheral blood lymphocytes in humans exposed to ethylene
oxide. Hereditas 98: 105-113.
Hogstedt. C. (1988). Epidemiologic studies on ethylene oxide and cancer: An updating. In Methods for detecting
DNA damaging agents in humans: Applications in cancer epidemiology and prevention. Lyon, France: International
Agency for Research on Cancer.
Hogstedt. C: Aringer. L: Gustavssoa A. (1984). Etylenoxid och cancer - litteraturoversikt och uppfoljning av tva
epidemiologiska studier [Ethylene oxide and cancer - a literature review and follow-up of two epidemiological
studies] [Review]. Arbete ochHalsa41: 49.
Hogstedt. C: Aringer. L: Gustavssoa A. (1986). Epidemiologic support for ethylene oxide as a cancer-causing
agent. JAMA 255: 1575-1578.
Hogstedt. C: Malmqvist N: Wadman. B. (1979a). Leukemia in workers exposed to ethylene oxide. JAMA 241:
1132-1133. http://dx.doi.org/10.1001/iama.1979.03290370036024
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-5 DRAFT—DO NOT CITE OR QUOTE
-------
Hogstedt C: Rohlen. O: Berndtsson. BS: Axelson. O: Ehrenberg. L. (1979b). A cohort study of mortality and cancer
incidence in ethylene oxide production workers. Occup Environ Med 36: 276-280.
http://dx.doi.0rg/10.1136/oem.36.4.276
Hong. HH: Houle. CD: Ton. TV: Sills. RC. (2007). K-ras mutations in lung tumors and tumors from other organs
are consistent with a common mechanism of ethylene oxide tumorigenesis in the B6C3F1 mouse. Toxicol Pathol 35:
81-85. http://dx.doi.org/10.1080/01926230601063839
Hornung. RW: Greife. AL: Stayner. LT: Steenland. NK: Herrick. RF: Elliott. LJ: Ringenburg. VL: Morawetz. J.
(1994). Statistical model for prediction of retrospective exposure to ethylene oxide in an occupational mortality
study. Am J Ind Med 25: 825-836.
Houle. CD: Ton. TV: Clayton. N: Huff. J: Hong. HH: Sills. RC. (2006). Frequent p53 and H-ras mutations in
benzene- and ethylene oxide-induced mammary gland carcinomas fromB6C3Fl mice. Toxicol Pathol 34: 752-762.
http://dx.doi.org/10.1080/01926230600935912
Huang. CC Shih. WC Wu. CF: Chea MF: Chea YL: Lin. YH: Wu. KY. (2008). Rapid and sensitive on-line liquid
chromatographic/tandem mass spectrometric determination of an ethylene oxide-DNA adduct, N7-(2-
hydroxyethyl)guanine, in urine of nonsmokers. Rapid CommunMass Spectrom22: 706-710.
http://dx.doi.org/10.1002/rcm.3414
Huang. CC: Wu. CF: Shih. WC: Chea MF: Chen. CY: Chien. YC: Liou. SH: Chiang. SY: Wu. KY. (20ID.
Comparative analysis of urinary N7-(2-hydroxyethyl)guanine for ethylene oxide- and non-exposed workers. Toxicol
Lett 202: 237-243. http://dx.doi.0rg/10.1016/i.toxlet.2011.02.009
IARC (International Agency for Research on Cancer). (1994a). Ethylene. In IARC monographs on the evaluation of
carcinogenic risks to humans Some industrial chemicals (pp. 45-71). Lyon, France.
http://monographs.iarc.fr/ENG/Monographs/vol60/volume60.pdf
IARC (International Agency for Research on Cancer). (1994b). Ethylene oxide. In IARC monographs on the
evaluation of carcinogenic risks to humans Some industrial chemicals (pp. 73-159). Lyon, France.
http://monographs.iarc.fr/ENG/Monographs/vol60/volume60.pdf
IARC (International Agency for Research on Cancer). (2008). 1,3-butadiene, ethylene oxide and vinyl halides (vinyl
fluoride, vinyl chloride and vinyl bromide). In IARC monographs on the evaluation of carcinogenic risks to humans
(pp. 3-471). Lyon, France. http://monographs.iarc.fr/ENG/Monographs/vol97/mono97.pdf
Jarabek. AM: Pottenger. LH: Andrews. LS: Casciano. D: Embry. MR: Kim. JH: Preston. RJ: Reddv. MV: Schoeny.
R: Shuker. D: Skare. J: Swenberg. J: Williams. GM: Zeiger. E. (2009). Creating context for the use of DNA adduct
data in cancer risk assessment: I. Data organization [Review]. Crit Rev Toxicol 39: 659-678.
http://dx.doi.org/10.1080/10408440903164155
Jenssea D: Ramel. C. (1980). The micronucleus test as part of a short-term mutagenicity test program for the
predicition of carcinogenicity evaluated by 143 agents tested. MutatRes75: 191-202.
http://dx.doi.org/10.1016/0165-1110(80)90014-7
Joyner. RE. (1964). Chronic toxicity of ethylene oxide. Arch Environ Occup Health 8: 700-710.
Kardos. L: Szeles. G: Gombkoto. G: Szeremi. M: Tompa. A: Adany. R. (2003). Cancer deaths among hospital staff
potentially exposed to ethylene oxide: an epidemiological analysis. Environ Mol Mutagen 42: 59-60.
http://dx.doi.org/10.1002/em. 10167
Kelsev. KT: Wiencke. JK: Eisen. EA: Lynch. DW: Lewis. TR: Little. JB. (1988). Persistently elevated sister
chromatid exchanges in ethylene oxide-exposed primates: The role of a subpopulation of high frequency cells.
Cancer Res 48: 5045-5050.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-6 DRAFT—DO NOT CITE OR QUOTE
-------
Kensler. TW: Ng. D: Carmella. SG: Chen. M: Jacobson. LP: Mufioz. A: Egner. PA: Chen. JG: Oian. GS: Chen. TY:
Fahey. JW: Talalay. P: Groopman. JD: Yuaa JM: Hecht. SS. (2012). Modulation of the metabolism of airborne
pollutants by glucoraphanin-rich and sulforaphane-rich broccoli sprout beverages in Qidong, China. Carcinogenesis
33: 101-107. http://dx.doi.org/10.1093/carcina)gr229
Keshava. N: Jinot J: Sonawane. B. (2006a). An evaluation of mutagenic mode of action for carcinogenicity:
Ethylene oxide. Environ Mol Mutagen 47: 442-442.
Keshava. N: Woodall. GM: Rice. S: Sonawane. B: Cote. I. (2006b). An evaluation of the mutagenic mode of action
for four environmental carcinogens [Abstract]. Toxicol Sci 90: 334-335.
Kiesselbach. N: Ulm. K: Lange. HJ: Korallus. U. (1990). A multicentre mortality study of workers exposed to
ethylene oxide. Br J Ind Med 47: 182-188.
Kim EA: Lee. HE: Ryu. HW: Park. SH: Kang. SK. (2011). Cases series of malignant lymphohematopoietic disorder
in Korean semiconductor industry. Saf Health Work 2: 122-134. http://dx.doi.Org/10.5491/SHAW.2011.2.2.122
Kiran. S: Cocco. P: Mannetje. A: Satta. G: D'Andrea. I: Becker. N: de Sanjose. S: Foretova. L: Staines. A: Kleefeld.
S: Maynadie. M: Nieters. A: Brennaa P: Boffetta. P. (2010). Occupational exposure to ethylene oxide and risk of
lymphoma. Epidemiology 21: 905-910. http://dx.doi.org/10.1097/EDE.Ob013e3181f4ccOf
Kirman. CR: Sweeney. LM: Teta. MJ: Sielken. RL: Valdez-Flores. C: Albertini. RJ: Gargas. ML. (2004).
Addressing nonlinearity in the exposure-response relationship for a genotoxic carcinogen: cancer potency estimates
for ethylene oxide. Risk Anal 24: 1165-1183. http://dx.doi.org/10.1111/i.0272-4332.2004.00517.x
Krigerman. AD: Erexson. GL: Phelps. ME: Wilmer. JL. (1983). Sister-chromatid exchange induction in peripheral
blood lymphocytes of rats exposed to ethylene oxide by inhalation. Mutat Res Lett 120: 37-44.
http://dx.doi.org/10.1016/0165-7992(83)90071-4
Koepke. SR: Kroeger-Koepke. MB: Bosan. W: Thomas. BJ: Alvord. WG: Michejda. CJ. (1988). Alkylation of DNA
in rats by N-nitrosomethyl-(2-hydroxyethyl)amine: dose response and persistence of the alkylated lesions in vivo.
Cancer Res 48: 1537-1542.
Kolmaa A. (1985). Effect of deficiency in excision repair and umuC function on the mutagenicity with ethylene
oxide in the lacl gene of E. coli. Mutat Res 146: 43-46.
Kolmaa A: Chovanec. M. (2000). Combined effects of gamma-radiation and ethylene oxide in human diploid
fibroblasts. Mutagenesis 15: 99-104. http://dx.doi.0rg/10.1093/mutage/15.2.99
Kolmaa A: Chovanec. M: Osterman-Golkar. S. (2002). Genotoxic effects of ethylene oxide, propylene oxide and
epichlorohydrin in humans: update review (1990-2001) [Review]. DNA Repair 512: 173-194.
Kolmaa A: Naslund. M. (1987). Mutagenicity testing of ethylene oxide in Escherichia coli strains with different
repair capacities. Environ Mol Mutagen 10:311-315.
Kolmaa A: Naslund. M: Calleman. CJ. (1986). Genotoxic effects of ethylene oxide and their relevance to human
cancer [Review]. Carcinogenesis 7: 1245-1250.
Kumar. R: Staffas. J: Forsti. A: Hemminki. K. (1995). 32P-postlabelling method for the detection of 7-alkylguanine
adducts formed by the reaction of different 1,2-alkyl epoxides with DNA. Carcinogenesis 16: 483-489.
Lambert. B: Andersson. B: Bastlova. T: Hou. SM: Hellgren. D: Kolmaa A. (1994). Mutations induced in the
hypoxanthine phosphoribosyl transferase gene by three urban air pollutants: Acetaldehyde, benzo[a]pyrene
diolepoxide, and ethylene oxide. Environ Health Perspect Suppl 102: 135-138.
Langholz. B: Richardson. DB. (2010). Fitting general relative risk models for survival time and matched case-
control analysis. Am J Epidemiol 171: 377-383. http://dx.doi.org/10.1093/aie/kwp403
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-7 DRAFT—DO NOT CITE OR QUOTE
-------
Laurent. C: Frederic. J: Leonard. AY. (1984). Sister chromatid exchange frequency in workers exposed to high
levels of ethylene oxide, in a hospital sterilization service. Int Arch Occup Environ Health 54: 33-43.
Leclercq. L: Laurent. C: De Pauw. E. (1997). High-performance liquid chromatography/electrospray mass
spectrometry for the analysis of modified bases inDNA: 7-(2-hydroxyethyl)guanine, the major ethylene oxide-DNA
adduct. Anal Chem69: 1952-1955.
Lerda. D: Rizzj R. (1992). Cytogenetic study of persons occupationally exposed to ethylene oxide. Mutat Res 281:
31-37.
Leutbecher. M: Langhof. T: Peter. H: Post. U. (1992). Ethylene oxide: metabolism in human blood and its
implication to biological monitoring. ArchToxicol 15: 289.
Lewis. SE: Barnett LB: Felton. C: Johnson. FM: Skow. LC: Cacheiro. N: Shelby. MD. (1986). Dominant visible
and electrophoretically expressed mutations induced in male mice exposed to ethylene oxide by inhalation. Environ
Mol Mutagen 8: 867-872.
Li. F: Segal. A: Solomon. JJ. (1992). In vitro reaction of ethylene oxide with DNA and characterization of DNA
adducts. ChemBiol Interact 83: 35-54.
Lin. JS: Chuang. KT: Huang. MS: Wei. KM. (2007). Emission of ethylene oxide during frying of foods in soybean
oil. Food Chem Toxicol 45: 568-574. http://dx.doi.0rg/10.1016/i.fct.2006.10.002
Lindberg. HK: Falck. GC: Catalan. J: Santonen. T: Norppa. H. (2010). Micronucleus assay for mouse alveolar Type
II and Clara cells. Environ Mol Mutagen 51: 164-172. http://dx.doi.org/10.1002/em.20520
Liou. SH: Lung. JC: Chen. YH: Yang. T: Hsieh. LL: Chea CJ: Wu. TN. (1999). Increased chromosome-type
chromosome aberration frequencies as biomarkers of cancer risk in a blackfoot endemic area. Cancer Res 59: 1481-
1484.
Lorenti Garcia. C: Darroudi. F: Tates. AD: Natarajan. AT. (2001). Induction and persistence of micronuclei, sister-
chromatid exchanges and chromosomal aberrations in splenocytes and bone-marrow cells of rats exposed to
ethylene oxide. Mutat Res Genet Toxicol Environ Mutagen 492: 59-67. http://dx.doi.org/10.1016/S1383-
5718(01)00149-8
Lucas. LJ: Teta. MJ. (1996). Breast cancer and ethylene oxide exposure [Letter]. Int J Epidemiol 25: 685-686.
Lynch. DW: Lewis. TR: Moorman. WJ: Burg. JR: Groth. DH: Khaa A: Ackerman. LJ: Cockrell. BY. (1984a).
Carcinogenic and toxicologic effects of inhaled ethylene oxide and propylene oxide in F344 rats. Toxicol Appl
Pharmacol 76: 69-84.
Lynch. DW: Lewis. TR: Moorman. WJ: Burg. JR: Gulati DK: Kaur. P: Sabharwal PS. (1984b). Sister-chromatid
exchanges and chromosome aberrations in lymphocytes from monkeys exposed to ethylene oxide and propylene
oxide by inhalation. Toxicol Appl Pharmacol 76: 85-95.
Lynch. DW: Lewis. TR: Moormaa WJ: Burg. JR: Lai. JB: Setzer. JV: Groth. DH: Gulati. DK: Zavos. PM:
Sabharwal. PS: Ackerman. LJ: Cockrell. BY: Sprinz. H. (1984c). Effects on monkeys and rats of long-term
inhalation exposure to ethylene oxide: Majorfindings of the NIOSH study. In Inhospital ethylene oxide sterilization:
Current issues in ethylene oxide toxicity and occupational exposure (pp. 7-10). (AAMI Technology Assessment
Report No. 8-84). Arlington, VA: Association for the Advancement of Medical Instrumentation.
Major. J: Jakab. MG: Tompa. A. (1996). Genotoxicological investigation of hospital nurses occupationally exposed
to ethylene-oxide: I Chromosome aberrations, sister-chromatid exchanges, cell cycle kinetics, and UV-induced DNA
synthesis in peripheral blood lymphocytes. Environ Mol Mutagen 27: 84-92. http://dx.doi.org/10.1002/(SICI) 1098-
2280(1996)27:2<:84::AID-EM2>:3.0.CO:2-E
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-8 DRAFT—DO NOT CITE OR QUOTE
-------
Major. J: Jakab. MG: Tompa. A. (1999). The frequency of induced premature centromere division in human
populations occupationally exposed to genotoxic chemicals. Mutat Res Genet Toxicol Environ Mutagen 445: 241-
249. http://dx.doi.org/10.1016/S1383-5718(99)00129-l
Major. J: Jakab. MG: Tompa. A. (2001). Genotoxicological investigation of hospital nurses occupationally exposed
to ethylene oxide. II. HPRT mutation frequencies. Central Eur J Occup Env Med 7: 195-208.
Maltoni. C: Valgimigri. L: Scarnato. C. (1980). Long-term carcinogenic bioassays on ethylene dichloride
administered by inhalation to rats and mice. In B Ames; P Infante; R Reitz (Eds.), Ethylene dichloride: A potential
health risk? (pp. 3-29). Cold Spring Harbor, NY: Cold Spring Harbor Laboratory.
Manservigi M; Tibaldi. E; Soffritti. M. (2010). Toxic and carcinogenic effects of ethylene and its ethylene oxyde
metabolite. Eur J Oncol 15: 5-23.
Marsden. DA; Jones. DJ; Britton. RG; Ognibene. T; Ubick. E; Johnson. GE; Farmer. PB; Brown. K. (2009). Dose-
response relationships for N7-(2-hydroxyethyl)guanine induced by low-dose [14C]ethylene oxide: evidence for a
novel mechanism of endogenous adduct formation. Cancer Res 69: 3052-3059. http://dx.doi.org/10.1158/0008-
5472.CAN-08-4233
Marsdea DA: Jones. DJ: Lamb. JH: Tompkins. EM: Farmer. PB: Brown. K. (2007). Determination of endogenous
and exogenously derived N7-(2-hydroxyethyl)guanine adducts in ethylene oxide-treated rats. Chem Res Toxicol 20:
290-299. http://dx.doi.org/10.1021/tx600264t
Mayer. J; Warburton. D; Jeffrey. AM; Pero. R; Walles. S; Andrews. L; Toor. M; Latriano. L; Wazneh. L; Tang. D;
Tsai. WY; Kuroda. M; Perera. F. (1991). Biologic markers in ethylene oxide exposed workers and controls. Mutat
Res 248: 163-176. http://dx.doi.org/10.1016/0027-5107r9D90098-9
Mazon. G: Philippia G: Cadet. J: Gasparutto. D: Fuchs. RP. (2009). The alkyltransferase-like ybaZ gene product
enhances nucleotide excision repair of O(6)-alkylguanine adducts in E. coli. DNA Repair 8: 697-703.
http://dx.doi.0rg/10.1016/i.dnarep.2009.01.022
Mazon. G; Philippia G; Cadet. J; Gasparutto. D; Modesti M; Fuchs. RP. (2010). Alkyltransferase-like protein
(eATL) prevents mismatch repair-mediated toxicity induced by O6-alkylguanine adducts in Escherichia coli. PNAS
107: 18050-18055. http://dx.doi.org/10.1073/pnas.1008635107
McCarthy. MC; O'Brien. TE; Charrier. JG; Hafner. HR. (2009). Characterization of the chronic risk and hazard of
hazardous air pollutants in the United States using ambient monitoring data. Environ Health Perspect 117: 790-796.
http://dx.doi.org/10.1289/ehp.11861
Mikoczy. Z: Tinnerberg. H: Bjork. J: Albin. M. (2011). Cancer incidence and mortality in Swedish sterilant workers
exposed to ethylene oxide: updated cohort study findings 1972-2006. Int J Environ Res Public Health 8: 2009-2019.
http://dx.doi.org/10.3390/ijerph8062009
Morgan. RW; Claxton. KW; Divine. BJ; Kaplan. SD; Harris. VB. (1981). Mortality among ethylene oxide workers.
J Occup Environ Med 23: 767-770.
Mosavi-Jarrahi. A: Mohagheghi. MA: Kalaghchi. B: Mousavi-Jarrahi. Y: Kolahi. AA: Noori. MK. (2009).
Estimating the incidence of leukemia attributable to occupational exposure in Iran. Asian Pac J Cancer Prev 10: 67-
70.
Muller. E; Bertok. A. (1995). Tumor cases in the personnel of Eger hospital: Hygienic conditions and consequences.
K6rhaz-esorvostechnika33: 17-22.
Natarajan. AT: Preston. RJ: Dellarco. V: Ehrenberg. L: Generoso. W: Lewis. S: Tates. AD. (1995). Ethylene oxide:
evaluation of genotoxicity data and an exploratory assessment of genetic risk [Review]. DNA Repair 330: 55-70.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-9 DRAFT—DO NOT CITE OR QUOTE
-------
NCI (National Cancer Institute). (1978). Bioassay of 1,2-dichloroethane for possible carcinogenicity. (CAS No. 107-
06-2 & NCI-CG-TR-55). Bethesda MD: National Institute of Health.
Nivard. MJ: Czene. K: Segerback. D: Vogel EW. (2003). Mutagenic activity of ethylene oxide and propylene oxide
under XPG proficient and deficient conditions in relation to N-7-(2-hydroxyalkyl)guanine levels in Drosophila.
Mutat Res 529: 95-107.
Norman. SA: Berlin. JA: Soper. KA: Middendorf. BF: Stollev. PP. (1995). Cancer incidence in a group of workers
potentially exposed to ethylene oxide. Int J Epidemiol 24: 276-284.
NRC (National Research Council). (2011). Review of the Environmental Protection Agency's draft IRIS assessment
of formaldehyde. Washington, DC: National Academies Press, http://www.nap.edu/catalog/13142.html
NTP (National Toxicology Program). (1987). Toxicology and carcinogenesis studies of ethylene oxide (CAS no 75-
21-8) inB6C3Fl mice (inhalation studies).
Nygren. J: Cedervall B: Eriksson. S: Dusinska. M: Kolman. A. (1994). Induction of DNA strand breaks by ethylene
oxide in human diploid fibroblasts. Environ Mol Mutagen 24: 161-167.
Oesch. F: Hengstler. JG: Arand. M: Fuchs. J. (1995). Detection of primary DNA damage: applicability to
biomonitoring of genotoxic occupational exposure and in clinical therapy. Pharmacogenetics 5 Spec No: SI 18-S122.
Okada. Y: Nakagoshi. A: Tsurukawa. M: Matsumura. C: Eiho. J: Nakano. T. (2012). Environmental risk assessment
and concentration trend of atmospheric volatile organic compounds in Hyogo Prefecture, Japan. Environ Sci Pollut
Res Int 19: 201-213. http://dx.doi.org/10.1007/sll356-011-0550-0
Olsen. GW: Lacy. SE: Bodner. KM: Chau. M: Arceneaux. TG: Cartmill JB: Ramlow. JM: Boswell JM. (1997).
Mortality from pancreatic and lymphopoietic cancer among workers in ethylene and propylene chlorohydrin
production. Occup Environ Med 54: 592-598. http://dx.doi.0rg/10.1136/oem.54.8.592
Ong. T: Bi. HK: Xing. S: Stewart. J: Moorman. W. (1993). Induction of sister chromatid exchange in spleen and
bone marrow cells of rats exposed by inhalation to different dose rates of ethylene oxide. Environ Mol Mutagen 22:
147-151.
Otteneder. M: Lutz. WK. (1999). Correlation of DNA adduct levels with tumor incidence: carcinogenic potency of
DNA adducts [Review]. DNA Repair 424: 237-247.
Parsons. BL: Manjanatha. MG: Myers. MB: McKim KL: Wang. Y: Gollapudi BB: Moore. N: Haber. LT: Moore.
MM. (2012). Induction of ell and K-Ras mutation in lung DNA of Big Blue mice exposed to ethylene oxide by
inhalation. Environ Mol Mutagen 53: S62-S62.
Pauwels. W: Veulemans. H. (1998). Comparison of ethylene, propylene and styrene 7,8-oxide in vitro adduct
formation on N-terminal valine in human haemoglobin and on N-7-guanine in human DNA. Mutat Res 418: 21-33.
Pero. RW: Widegren. B: Hogstedt. B: Mitelman. F. (1981). In vivo and in vitro ethylene oxide exposure of human
lymphocytes assessed by chemical stimulation of unscheduled DNA synthesis. DNA Repair 83: 271-289.
Prentice. RL. (1982). Covariate measurement errors and parameter estimation in a failure time regression model.
Biometrika69: 331-342. http://dx.doi.0rg/10.1093/biomet/69.2.331
Preston. PJ. (1999). Cytogenetic effects of ethylene oxide, with an emphasis on population monitoring [Review].
Crit Rev Toxicol 29: 263-282. http://dx.doi.org/10.1080/10408449991349212
Preston. PJ: Fennell TR: Leber. AP: Sielken. RL: Swenberg. JA. (1995). Reconsideration of the genetic risk
assessment for ethylene oxide exposures. Environ Mol Mutagen 26: 189-202.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-10 DRAFT—DO NOT CITE OR QUOTE
-------
Rapoport IA. (1948). The effect of ethylene oxide, glycide and glycol on genetic mutations. Dokl Biochem Biophys
60: 469-472.
Recio. L: Donner. M: Abernethy. D: Pluta. L: Steen. AM: Wong. BA: James. A: Preston. RJ. (2004). In vivo
mutagenicity and mutation spectrum in the bone marrow and testes of B6C3F1 lacl transgenic mice following
inhalation exposure to ethylene oxide. Mutagenesis 19: 215-222.
Ribeiro. LR: Rabello-Gav. MN: Salvadori. DM: Pereira. CA: Becak. W. (1987). Cytogenetic effects of inhaled
ethylene oxide in somatic and germ cells of mice. Arch Toxicol 59: 332-335.
Ribeiro. LR: Salvadori. DM: Rios. AC: Costa. SL: Tates. AD: Tornqvist M: Natarajaa AT. (1994). Biological
monitoring of workers occupationally exposed to ethylene oxide. Mutat Res 313: 81-87.
Robins. JM: Blevins. D: Ritter. G: Wulfsohn. M. (1992). G-estimation of the effect of prophylaxis therapy for
Pneumocystis carinii pneumonia on the survival of AIDS patients. Epidemiology 3: 319-336.
Rossner. P: Boffetta. P: Ceppi M: Bonassi S: Smerhovsky. Z: Landa. K: Juzova. D: Sram. RJ. (2005).
Chromosomal aberrations in lymphocytes of healthy subjects and risk of cancer. Environ Health Perspect 113: 517-
520. http://dx.doi.org/10.1289/ehp.6925
Rothman. K: Greenland. S: Lash. T. (2008). Modern epidemiology. Philadelphia, PA: Lippincott Williams &
Wilkins.
Rothman. KJ. (1986). Modern epidemiology. Boston, MA: Little Brown & Co.
Rothman. KJ: Greenland. S. (1998). Modern epidemiology (2nd ed.). Philadelphia, PA: Lippincott, Williams, &
Wilkins.
Rusvn. I: Asakura. S: Li. Y: Kosyk. O: Koc. H: Nakamura. J: Upton. PB: Swenberg. JA. (2005). Effects of ethylene
oxide and ethylene inhalation on DNA adducts, apurinic/apyrimidinic sites and expression of base excision DNA
repair genes in rat brain, spleen, and liver. DNA Repair 4: 1099-1110.
http://dx.doi.0rg/10.1016/i.dnarep.2005.05.009
SAB (Science Advisory Board). (2007). Review of Office of Research and Development (ORD) draft assessment
entitled "Evaluation of the carcinogenicity of ethylene oxide". Washington, DC: Science Advisory Board, U.S.
Environmental Protection Agency.
http://vosemite.epa.gov/sab/sabproduct.nsf/368203f97al5308a852574ba005bbd01/5D661BC118B527A3852573B8
0068C97B/$File/EP A-SAB-08-004-unsigned.pdf
Saha. M: Abushamaa. A: Giese. RW. (1995). General method for determining ethylene oxide and related N7-
guanine DNA adducts by gas chromatography-electron capture mass spectrometry. J Chromatogr A 712: 345-354.
Sarto. F: Clonfero. E: Bartolucci. GB: Franceschi. C: Chiricolo. M: Levis. AG. (1987). Sister chromatid exchanges
and DNA repair capability in sanitary workers exposed to ethylene oxide: Evaluation of the dose-effect relationship.
Am J Ind Med 12: 625-637. http://dx.doi.org/10.1002/aiim.4700120515
Sarto. F: Cominato. I: Pintoa AM: Brovedani. PG: Faccioli. CM: Bianchi. V: Levis. AG. (1984a). Cytogenetic
damage in workers exposed to ethylene oxide. Mutat Res 138: 185-195.
Sarto. F: Cominato. I: Pinton. AM: Brovedani. PG: Faccioli. CM: Bianchi. V: Levis. AG. (1984b). Workers exposed
to ethylene oxide have increased incidence of sister chromatid exchange. IARC Sci Publ413-419.
Sarto. F: Tomanin. R: Giacomelli. L: lannini. G: Cupiraggi AR. (1990). The micronucleus assay in human
exfoliated cells of the nose and mouth: application to occupational exposures to chromic acid and ethylene oxide.
Mutat Res 244: 345-351.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-11 DRAFT—DO NOT CITE OR QUOTE
-------
Sarto. F: Tornqvist MA: Tomanin. R: Bartolucci GB: Osterman-Golkar. SM: Ehrenberg. L. (1991). Studies of
biological and chemical monitoring of low-level exposure to ethylene oxide. Scand J Work Environ Health 17: 60-
64.
Schulte. PA: Boeniger. M: Walker. JT: Schober. SE: Pereira. MA: Gulati. DK: Wojciechowskj JP: Garza. A:
Froelich. R: Strauss. G. (1992). Biologic markers in hospital workers exposed to low levels of ethylene oxide. Mutat
Res 278: 237-251.
Segerback. D. (1983). Alkylation of DNA and hemoglobin in the mouse following exposure to ethene and ethene
oxide. ChemBiol Interact 45: 139-151.
Segerback. D. (1990). Reaction products in hemoglobin and DNA after in vitro treatment with ethylene oxide and
N-(2-hydroxyethyl)-N-nitrosourea. Carcinogenesis 11: 307-312.
Segerback. D. (1994). DNA alkylationby ethylene oxide and some mono-substituted epoxides [Review]. IARC Sci
Publ37-47.
Shen. J: Kessler. W: Denk. B: Filser. JG. (1989). Metabolism and endogenous production of ethylene in rat and
man. ArchToxicol Suppl 13: 237-239.
Shore. RE: Gardner. MJ: Pannett B. (1993). Ethylene oxide: an assessment of the epidemiological evidence on
carcinogenicity. Occup Environ Med 50: 971-997.
Sielken. RL: Valdez-Flores. C. (2009). Life-table calculations of excess risk for incidence versus mortality: ethylene
oxide case study. Regul Toxicol Pharmacol 55: 82-89. http://dx.doi.0rg/10.1016/i.vrtph.2009.06.003
Sielken. RL: Valdez Flores. C. (2009). Calculating excess risk with age-dependent adjustment factors and
cumulative doses: Ethylene oxide case study. Regul Toxicol Pharmacol 55: 76-81.
http://dx.doi.0rg/10.1016/i.vrtph.2009.06.004
Sisk. SC: Pluta. LJ: Meyer. KG: Wong. BC: Recio. L. (1997). Assessment of the in vivo mutagenicity of ethylene
oxide in the tissues of B6C3F1 lacl transgenic mice following inhalation exposure. Mutat Res 391: 153-164.
Smith-Bindman. R. (2012). Environmental causes of breast cancer and radiation from medical imaging: findings
from the Institute of Medicine report. Arch Intern Med 172: 1023-1027.
http://dx.doi.org/10.1001/archinternmed.2012.2329
Snedeker. SM. (2006). Chemical exposures in the workplace: effect on breast cancer risk among women [Review].
AAOHN J 54: 270-279; quiz 280-271.
Snellings. WM: Weil CS: Maronpot RR. (1984). A two-year inhalation study of the carcinogenic potential of
ethylene oxide in Fischer 344 rats. Toxicol Appl Pharmacol 75: 105-117.
Stayner. L: Steenland. K: Greife. A: Hornung. R: Haves. RB: Nowlin. S: Morawetz. J: Ringenburg. V: Elliot L:
Halperin. W. (1993). Exposure-response analysis of cancer mortality in a cohort of workers exposed to ethylene
oxide. AmJEpidemiol 138: 787-798.
Steenland. K: Deddens. J: Piacitelli. L. (2001). Risk assessment for 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
based on an epidemiologic study. Am J Epidemiol 154: 451-458.
Steenland. K: Deddens. J: Salvan. A: Stayner. L. (1996). Negative bias in exposure-response trends in occupational
studies: modeling the healthy workers survivor effect. Am J Epidemiol 143: 202-210.
Steenland. K: Deddens. J: Zhao. S. (2000). Biases in estimating the effect of cumulative exposure in log-linear
models when estimated exposure levels are assigned. Scand J Work Environ Health 26: 37-43.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-12 DRAFT—DO NOT CITE OR QUOTE
-------
Steenland. K: Deddens. JA. (1997). Increased precision using countermatching in nested case-control studies.
Epidemiology 8: 238-242.
Steenland. K: Deddens. JA. (2004). A practical guide to dose-response analyses and risk assessment in occupational
epidemiology [Review]. Epidemiology 15: 63-70. http://dx.doi.org/10.1097/01.ede.0000100287.45004.e7
Steenland. K: Seals. R: Klein. M: Jinot. J: Kahn. HP. (2011). Risk estimation with epidemiologic data when
response attenuates at high-exposure levels. Environ Health Perspect 119: 831-837.
http://dx.doi.org/10.1289/ehp.1002521
Steenland. K: Stayner. L. (1993). An epidemiological study of workers potentially exposed to ethylene oxide [letter]
[Letter]. Br J Ind Med 50: 1125.
Steenland. K: Stayner. L: Deddens. J. (2004). Mortality analyses in a cohort of 18 235 ethylene oxide exposed
workers: follow up extended from 1987 to 1998. Occup Environ Med 61: 2-7.
Steenland. K: Stayner. L: Greife. A: Harperin. W: Haves. R: Hornung. R: Nowlin. S. (1991). Mortality among
workers exposed to ethylene oxide. N Engl J Med 324: 1402-1407.
http://dx.doi.org/10.1056/NEJM199105163242004
Steenland. K: Whelan. E: Deddens. J: Stayner. L: Ward. E. (2003). Ethylene oxide and breast cancer incidence in a
cohort study of 7576 women (United States). Cancer Causes Control 14: 531-539.
Steinhausen. M: Van Gelder. R: Gabriel S. (2012). Arbeitsbedingte expositionen gegeniiber krebserzeugenden,
erbgutverandernden oder fortpflanzungsgefahrdenden substanzen in Deutschland teil 2: Stoffe mit ERB nach BekGS
910. GefahrstReinhaltLuft72: 347-358.
Stolley. PD: Soper. KA: Galloway. SM: Nichols. WW: Norman. SA: Wolman. SR. (1984). Sister-chromatid
exchanges in association with occupational exposure to ethylene oxide. Mutat Res 129: 89-102.
http://dx.doi.org/10.1016/0027-5107(84)90127-l
Swaen. GM: Burns. C: Teta. JM: Bodner. K: Keenaa D: Bodnar. CM. (2009). Mortality study update of ethylene
oxide workers in chemical manufacturing: a 15 year update. J Occup Environ Med 51: 714-723.
http://dx.doi.org/10.1097/JOM.Ob013e3181a2ca20
Swaen. GM: Slangen. JM: Ott MG: Kusters. E: Van Den Langenbergh. G: Arends. JW: Zober. A. (1996).
Investigation of a cluster often cases of Hodgkin's disease in an occupational setting. Int Arch Occup Environ
Health 68: 224-228.
Swenberg. JA: Fryar-Tita. E: Jeong. YC: Boysea G: Starr. T: Walker. VE: Albertini. RJ. (2008). Biomarkers in
toxicology and risk assessment: Informing critical dose-response relationships [Review]. Chem Res Toxicol 21:
253-265. http://dx.doi.org/10.1021/tx700408t
Swenberg. JA: Ham. A: Koc. H: Morinello. E: Ranasinghe. A: Tretvakova: Upton. PB: Wu. K. (2000). DNA
adducts: effects of low exposure to ethylene oxide, vinyl chloride and butadiene. DNA Repair 464: 77-86.
http://dx.doi.org/10.1016/S1383-5718(99)00168-0
Swenberg. JA: Lu. K: Moeller. BC: Gao. L: Upton. PB: Nakamura. J: Starr. TB. (2011). Endogenous versus
exogenous DNA adducts: their role in carcinogenesis, epidemiology, and risk assessment [Review]. Toxicol Sci 120
Suppl 1: S130-S145. http://dx.doi.org/10.1093/toxsci/kfq371
Tateo. F: Bononi. M. (2006). Determination of ethylene chlorohydrin as marker of spices fumigation with ethylene
oxide. J Food Compos Anal 19: 83-87. http://dx.doi.0rg/10.1016/i.jfca.2004.12.003
Tates. AD: Boogaard. PJ: Darroudi. F: Natarajan. AT: Caubo. ME: Van Sittert. NJ. (1995). Biological effect
monitoring in industrial workers following incidental exposure to high concentrations of ethylene oxide. DNA
Repair 329: 63-77.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-13 DRAFT—DO NOT CITE OR QUOTE
-------
Tales. AD: Grummt T: Tornqvist M: Farmer. PB: van Dam. FJ: vanMossel H: Schoemaker. HM: Osterman-
Golkar. S: Uebel. C: Tang. YS: Zwinderman. AH: Natarajan. AT: Ehrenberg. L. (1991). Biological and chemical
monitoring of occupational exposure to ethylene oxide. Mutat Res 250: 483-497. http://dx.doi.org/10.1016/0027-
5107(91190205-3
Tates. AD: Van Dam. FJ: Natarajan. AT: Van Teylingen. CMM: De Zwart. FA: Zwinderman. AH: Van Sittert NJ:
Nilsen. A: Nilsea OG: Zahlsea K: Magnusson. AL: Tornqvist. M. (1999). Measurement of HPRT mutations in
splenic lymphocytes and haemoglobin adducts in erythrocytes of Lewis rats exposed to ethylene oxide. DNA Repair
431:397-415.
Teta. MJ: Benson. LO: Vitale. JN. (1993). Mortality study of ethylene oxide workers in chemical manufacturing: a
10 year update. Br J Ind Med 50: 704-709.
Teta. MJ: Sielken. RU Valdez-Flores. C. (1999). Ethylene oxide cancer risk assessment based on epidemiological
data: application of revised regulatory guidelines. Risk Anal 19: 1135-1155.
Thier. R: Bolt. HM. (2000). Carcinogenicity and genotoxicity of ethylene oxide: new aspects and recent advances
[Review]. Crit Rev Toxicol 30: 595-608. http://dx.doi.org/10.1080/10408440008951121
Thiess. AM: Schwegler. H: Fleig. I: Stocker. WG. (1981). Mutagenicity study of workers exposed to alkylene
oxides (ethylene oxide/propylene oxide) and derivatives. J Occup Environ Med 23: 343-347.
Tompa. A: Jakab. M: Biro. A: Magyar. B: Fodor. Z: Klupp. T: Major. J. (2006). Chemical safety and health
conditions among Hungarian hospital nurses. Ann N Y Acad Sci 1076: 635-648.
http://dx.doi.org/10.1196/annals.1371.054
Tompa. A: Major. J: Jakab. MG. (1999). Is breast cancer cluster influenced by environmental and occupational
factors among hospital nurses in Hungary. Pathol Oncol Res 5: 117-121.
Tompkins. EM: Jones. DJL: Lamb. JH: Marsden. DA: Farmer. PB: Brown. K. (2008). Simultaneous detection of
five different 2-hydroxyethyl-DNA adducts formed by ethylene oxide exposure, using a high-performance liquid
chromatography/electrospray ionisation tandem mass spectrometry assay. Rapid Commun Mass Spectrom 22: 19-
28. http://dx.doi.org/10.1002/rcm.3328
Tompkins. EM: Jones. DJL: McLuckie. KIE: Farmer. PB: Brown. K. (2006). Weak mutagenicity of DNA adducts
derived from ethylene oxide exposure. Mutagenesis 21: 292-292.
Tompkins. EM: McLuckie. KI: Jones. DJ: Farmer. PB: Brown. K. (2009). Mutagenicity of DNA adducts derived
from ethylene oxide exposure in the pSP189 shuttle vector replicated in human Ad293 cells. Mutat Res 678: 129-
137. http://dx.doi.0rg/10.1016/i.mrgentox.2009.05.011
Tornqvist M. (1996). Ethylene oxide as a biological reactive intermediate of endogenous origin [Review]. Adv Exp
Med Biol 387: 275-283.
Tornqvist. MA: Almberg. JG: Bergmark. EN: Nilssoa S: Osterman-Golkar. SM. (1989). Ethylene oxide doses in
ethene-exposed fruit store workers. Scand J Work Environ Health 15: 436-438.
Tretvakova. N: Goggin. M: Sangaraju. D: Janis. G. (2012). Quantitation of DNA Adducts by Stable Isotope Dilution
Mass Spectrometry. ChemRes Toxicol 25: 2007-2035. http://dx.doi.org/10.1021/tx3002548
Tucker. JD: Xu. J: Stewart. J: Baciu. PC: Ong. TM. (1986). Detection of sister chromatid exchanges induced by
volatile genotoxicants. Teratog CarcinogMutagen6: 15-21. http://dx.doi.org/10.1002/tcm. 1770060103
U.S. EPA (U.S. Environmental Protection Agency). (1985). Performance evaluation of the Sciex TAGAR 600
mobile laboratory for ambient air monitoring Summary report [EPA Report].
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-14 DRAFT—DO NOT CITE OR QUOTE
-------
U.S. EPA (U.S. Environmental Protection Agency). (2005a). Guidelines for carcinogen risk assessment [EPA
Report]. (EPA/630/P-03/001F). Washington, DC: Risk Assessment Forum, http://www.epa.gov/cancerguidelines/
U.S. EPA (U.S. Environmental Protection Agency). (2005b). Supplemental guidance for assessing susceptibility
from early-life exposure to carcinogens. (EPA/630/R-03/003F). Washington, DC: U.S. Environmental Protection
Agency, Risk Assessment Forum. http://www.epa.gov/raf/publications/pdfs/childrens_supplement_fmal.pdf.
U.S. EPA (U.S. Environmental Protection Agency). (2006a). Evaluation of the carcinogenicity of ethylene oxide:
external review draft [EPA Report]. (EPA/635/R-06/003). Washington, DC: National Center for Environmental
Assessment, Office of Research and Development, http://cfpub.epa. gov/ncea/cfm/recordisplay.cfm?deid= 157664
U.S. EPA (U.S. Environmental Protection Agency). (2006b). Science policy council peer review handbook 3rd
edition [EPA Report]. (100B06002). Science Policy Council.
http://cfpub.epa. gov/ncea/cfm/recordisplav.cfm?deid= 157664
U.S. EPA (U.S. Environmental Protection Agency). (2012). Benchmark dose technical guidance. (EPA/100/R-
12/001). Washington, DC: Risk Assessment Forum.
http://www.epa.gov/raf/publications/pdfs/benc hmark_dose_guidance.pdf
Uziel. M: Munro. NB: Katz. PS: Vo-Dinh. T: Zeighami. EA: Waters. MD: Griffith. JD. (1992). DNA adduct
formation by 12 chemicals with populations potentially suitable for molecular epidemiological studies [Review].
MutatRes 277: 35-90. http://dx.doi.org/10.1016/0165-1110(92)90025-5
Valdez-Flores. C: Sielken. RL: Teta. MJ. (2010). Quantitative cancer risk assessment based on NIOSH and UCC
epidemiological data for workers exposed to ethylene oxide. Regul Toxicol Pharmacol 56: 312-320.
http://dx.doi.0rg/10.1016/i.vrtph.2009.10.001
Valdez-Flores. C: Sielkea RL: Teta. MJ. (2011). Quantitative cancer risk assessment for ethylene oxide inhalation
in occupational settings. Arch Toxicol 85: 1189-1193. http://dx.doi.org/10.1007/s00204-011-0669-2
vanBalen. E: Font. R: Cavalle. N: Font. L: Garcia-Villanueva. M: Benavente. Y: Brennan. P: de Sanjose. S. (2006).
Exposure to non-arsenic pesticides is associated with lymphoma among farmers in Spain. Occup Environ Med 63:
663-668. http://dx.doi.org/10.1136/oem.2005.024026
van Delft. JH: van Winden. MJ: Luiten-Schuite. A: Ribeiro. LR: Baan. RA. (1994). Comparison of various
immunochemical assays for the detection of ethylene oxide-DNA adducts with monoclonal antibodies against
imidazole ring-opened N7-(2-hydroxyethyl)guanosine: application in a biological monitoring study. Carcinogenesis
15: 1867-1873.
van Delft. JH: van Winden. MJ: van den Ende. AM: Baan. RA. (1993). Determining N7-alkylguanine adducts by
immunochemical methods and HPLC with electrochemical detection: applications in animal studies and in
monitoring human exposure to alkylating agents. Environ Health Perspect 99: 25-32.
van Sittert. NJ: Boogaard. PJ: Natarajan. AT: Tates. AD: Ehrenberg. LG: Tornqvist MA. (2000). Formation of
DNA adducts and induction of mutagenic effects in rats following 4 weeks inhalation exposure to ethylene oxide as
a basis for cancer risk assessment. Mutat Res-Fundam Mol Mech Mutagen 447: 27-48.
http://dx.doi.org/10.1016/80027-5107(99)00208-0
Van Sittert. NJ: de Jong. G. (1985). Biomonitoring of exposure to potential mutagens and carcinogens in industrial
populations. Food Chem Toxicol 23: 23-31.
van Wijngaarden. E: Hertz-Picciotto. I. (2004). A simple approach to performing quantitative cancer risk assessment
using published results from occupational epidemiology studies. Sci Total Environ 332: 81-87.
http://dx.doi.0rg/10.1016/i.scitotenv.2004.04.005
Victoria K: Stahlberg. M. (1988). A method for studying the mutagenicity of some gaseous compounds in
Salmonella typhimurium. Environ Mol Mutagen 11: 65-11.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-15 DRAFT—DO NOT CITE OR QUOTE
-------
Vogel EW: Natarajan. AT. (1995). DNA damage and repair in somatic and germ cells in vivo [Review]. Mutat Res
330: 183-208.
Vogel EW: Nivard. MJ. (1997). The response of germ cells to ethylene oxide, propylene oxide, propylene imine
and methyl methanesulfonate is a matter of cell stage-related DNA repair. Environ Mol Mutagen 29: 124-135.
Vogel EW: Nivard. MJ. (1998). Genotoxic effects of inhaled ethylene oxide, propylene oxide andbutylene oxide on
germ cells: sensitivity of genetic endpoints in relation to dose and repair status. Mutat Res 405: 259-271.
Walker. VE: Fennell. TR: Boucheron. JA: Fedtke. N: Ciroussel. F: Swenberg. JA. (1990). Macromolecular adducts
of ethylene oxide: a literature review and a time-course study on the formation of 7-(2-hydroxyethyl)guanine
following exposures of rats by inhalation [Review]. DNA Repair 233: 151-164.
Walker. VE: Fennell TR: Upton. PB: MacNeela. JP: Swenberg. JA. (1993). Molecular dosimetry of DNA and
hemoglobin adducts in mice and rats exposed to ethylene oxide. Environ Health Perspect 99: 11-17.
Walker. VE: Fennell TR: Upton. PB: Skopek. TR: Prevost V: Shuker. PEG: Swenberg. JA. (1992a). Molecular
dosimetry of ethylene oxide: formation and persistence of 7-(2-hydroxyethyl)guanine in DNA following repeated
exposures of rats and mice. Cancer Res 52: 4328-4334.
Walker. VE: MacNeela. JP: Swenberg. JA: Turner. MJ. Jr: Fennell. TR. (1992b). Molecular dosimetry of ethylene
oxide: formation and persistence of N-(2-hydroxyethyl)valine in hemoglobin following repeated exposures of rats
and mice. Cancer Res 52: 4320-4327.
Walker. VE: Sisk. SC: Upton. PB: Wong. BA: Recio. L. (1997). In vivo mutagenicity of ethylene oxide at the hprt
locus in T-lymphocytes of B6C3F1 lacl transgenic mice following inhalation exposure. Mutat Res 392: 211-222.
Walker. VE: Skopek. TR. (1993). A mouse model for the study of in vivo mutational spectra: sequence specificity
of ethylene oxide at the hprt locus. Mutat Res 288: 151-162. http://dx.doi.org/10.1016/0027-5107(93)90216-3
Walker. VE: Wu. KY: Upton. PB: Ranasinghe. A: Scheller. N: Cho. MH: Vergnes. JS: Skopek. TR: Swenberg. JA.
(2000). Biomarkers of exposure and effect as indicators of potential carcinogenic risk arising from in vivo
metabolism of ethylene to ethylene oxide. Carcinogenesis 21: 1661-1669.
Warwick. GP. (1963). The mechanism of action of alky lating agents [Review]. Cancer Res 23: 1315-1333.
Waters. MD: Stack. HF: Jackson. MA. (1999). Genetic toxicology data in the evaluation of potential human
environmental carcinogens [Review]. Mutat Res 437: 21-49. http://dx.doi.org/10.1016/S1383-5742(99)00037-X
Weiderpass. E: Meo. M: Vainio. H. (2011). Risk factors for breast cancer, including occupational exposures. Saf
Health Work 2: 1-8. http://dx.doi.Org/10.5491/SHAW.2011.2.l.l
WHO (World Health Organization). (2003). Concise International Chemical Assessment Document: Ethylene oxide.
Geneva: International Programme on Chemical Safety (IPCS).
http ://www. inchem.org/documents/cicads/cicad54. htm
Won. JU. k. (2010). Health Effects of Chemicals used in hospitals among Healthcare Workers. J Korean Am Med
Assoc 53: 474-482. http://dx.doi.0rg/10.5124/ikma.2010.53.6.474
Wong. O. (1991). Mortality among workers exposed to ethylene oxide [Letter]. N Engl J Med 325: 1254.
http://dx.doi.org/10.1056/NEJM199110243251716
Wong. O: Trent LS. (1993). An epidemiological study of workers potentially exposed to ethylene oxide. Occup
Environ Med 50: 308-316.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-16 DRAFT—DO NOT CITE OR QUOTE
-------
Wu. KY: Chiang. SY: Shih. WC: Huang. CC: Chen. MF: Swenberg. JA. (2011). The application of mass
spectrometry in molecular dosimetry: Ethylene oxide as an example. Mass Spectrom Rev 30: 733-756.
http://dx.doi.org/10.1002/mas.2Q299
Wu. KY: Ranasinghe. A: Upton. PB: Walker. V. (1999a). Molecular dosimetry of endogenous and ethylene oxide-
induced N7-(2-hydroxyethyl) guanine formation in tissues of rodents. Carcinogenesis 20: 1787-1792.
Wu. KY: Scheller. N: Ranasinghe. A: Yea TY: Sangaiah. R: Giese. R: Swenberg. JA. (1999b). A gas
chromatography/electron capture/negative chemical ionization high-resolution mass spectrometry method for
analysis of endogenous and exogenous N 7-(2-hydroxyethyl) guanine in rodents and its potential for human
biological monitoring. ChemRes Toxicol 12: 722-729. http://dx.doi.org/10.1021/tx990059n
Yager. JW: Benz. RD. (1982). Sister chromatid exchanges induced in rabbit lymphocytes by ethylene oxide after
inhalation exposure. Environ Mutagen 4: 121-134.
Yager. JW: Hines. CJ: Spear. RC. (1983). Exposure to ethylene oxide at work increases sister chromatid exchanges
in human peripheral lymphocytes. Science 219: 1221-1223. http://dx.doi.org/10.1126/science.6828851
Yong. LC: Schulte. PA: Kao. CY: Giese. RW: Boeniger. MF: Strauss. GH: Petersen. MR: Wiencke. JK. (2007).
DNA adducts in granulocytes of hospital workers exposed to ethylene oxide. Am J Ind Med 50: 293-302.
http://dx.doi.org/10.1002/aiim.20443
Yong. LC: Schulte. PA: Wiencke. JK: Boeniger. MF: Connally. LB: Walker. JT: Whelan. EA: Ward. EM. (2001).
Hemoglobin adducts and sister chromatid exchanges in hospital workers exposed to ethylene oxide: effects of
glutathione S-transferase Tl and Ml genotypes. Cancer Epidemiol Biomarkers Prev 10: 539-550.
Zhao. C: Kumar. R: Zahlsea K: Sundmark. HB: Hemminki. K: Eide. I. (1997). Persistence of 7-(2-
hydroxyethyl)guanine-DNA adducts in rats exposed to ethene by inhalation. Biomarkers 2: 355-359.
Zhao. C: Tyndyk. M: Eide. I: Hemminki. K. (1999). Endogenous and background DNA adducts by methylating and
2-hydroxyethylating agents. Mutat Res 424: 117-125.
This document is a draft for review purposes only and does not constitute Agency policy.
7/2013 R-17 DRAFT—DO NOT CITE OR QUOTE
------- |