EPA
EPA/63 5/R-l 6/3 50Fb
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)
December 2016
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC
-------
DISCLAIMER
This document has been reviewed in accordance with U.S. Environmental Protection
Agency policy and approved for publication. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
11
-------
CONTENTS
LIST OF TABLES vi
LIST OF FIGURES xi
LIST OF ABBREVIATIONS xiii
APPENDIX A. CRITICAL REVIEW OF EPIDEMIOLOGIC EVIDENCE A-l
A.l. BACKGROUND A-l
A.2. INDIVIDUAL STUDIES A-2
A.2.1. Hogstedt (1988), Hogstedt et al. (1986) A-2
A.2.2. Gardner et al. (1989) A-5
A.2.3. Kiesselbach et al. (1990) A-6
A.2.4. Greenberg et al. (1990) A-7
A.2.5. Steenland et al. (1991) A-9
A.2.6. Teta etal. (1993) A-11
A.2.7. Benson and Teta (1993) A-13
A.2.8. Stayner etal. (1993) A-14
A.2.9. Wong and Trent (1993) A-16
A.2.10. Bisanti et al. (1993) A-17
A.2.11. Hagmar etal. (1995) and Hagmar etal. (1991) A-18
A.2.12. Norman et al. (1995) A-20
A.2.13. Swaen et al. (1996) A-20
A.2.14. Olsen et al. (1997) A-21
A.2.15. Steenland et al. (2004) A-23
A.2.16. Steenland et al. (2003) A-25
A.2.17. Kardos etal. (2003) A-26
A.2.18. Tompa et al. (1999) A-27
A.2.19. Coggon et al. (2004) A-27
A.2.20. Swaen etal. (2009) and Valdez-Flores et al. (2010) A-27
A.2.21. Mikoczy etal. (2011) A-35
A.3. SUMMARY A-36
A.4. CONCLUSIONS A-56
APPENDIX B. REFERENCES FOR FIGURE 3-3 B-l
APPENDIX C. GENOTOXICITY AND MUTAGENICITY OF ETHYLENE OXIDE C-l
C.l. ADDUCTS C-2
C. 1.1. DNA Adducts C-2
C.1.2. EtO-He mo glob in Adducts C-10
C.2. GENE MUTATIONS C-l 1
C.2.1. Bacterial Systems C-ll
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.l. ENDOGENOUS PRODUCTION OF ETHYLENE AND EtO C-28
C.8. CONCLUSIONS C-33
iii
-------
APPENDIX D. REANALYSES OF ETHYLENE OXIDE EXPOSURE-RESPONSE
DATA D-l
D.l. BREAST CANCER INCIDENCE BASED ON THE SUBCOHORT WITH
INTERVIEWS D-4
D.l.l. Exposure Distribution among EtO-Exposed Women in Breast Cancer
Incidence Subcohort with Interviews (n = 5,139) D-4
D. 1.2. Lag Selection for the Breast Cancer Incidence Data D-5
D.l.3. Modeling of Breast Cancer Incidence Data Using a Variety of Models D-6
D.1.4. Risk Assessment for Breast Cancer Incidence Using the Cubic Spline
Curve Log RR Model D-20
D.1.5. Supplemental Results: Results for Cumulative Exposure and Log
Cumulative Exposure Cox Regression Models with Different Lag
Times D-21
D.1.6. Sensitivity of Unit Risk Estimates to Change in Lag Period D-22
D.1.7. Sensitivity of Unit Risk Estimates to Value of Knot D-23
D.1.8. Sensitivity of Unit Risk Estimates to Exclusion ofCovariates D-24
D.1.9. Analysis of Age Interaction for the Exposure Terms in the Two-piece
Linear Spline Model D-24
D.l. 10. Sensitivity of Unit Risk Estimates to Upper-Bound Estimation
Approach—Wald vs. Profile Likelihood D-25
D. 1.11. Sensitivity of Occupational Extra Risk Estimates to Change in Lag
Period D-25
D.2. BREAST CANCER MORTALITY D-28
D.2.1. Exposure Distribution among Women and Breast Cancer Deaths in the
Cohort Mortality Study (n = 9,544) D-28
D.2.2. Modeling of Breast Cancer Mortality Data Using a Variety of Models D-29
D.3. LYMPHOID CANCER MORTALITY (SUBSET OF ALL
HEMATOPOIETIC CANCERS COMBINED) (n = 17,530) D-37
D.3.1. Exposure Distribution in Cohort and among Lymphoid Cases in the
Cohort Mortality Study D-37
D.3.2. Lag Selection for the Lymphoid Cancer Mortality Data D-38
D.3.3. Modeling of Lymphoid Cancer Mortality Data Using a Variety of
Models D-41
D.3.4. Supplemental Results: Results for Log Cumulative Exposure Cox
Regression Model with No Lag D-51
D.3.5. Sensitivity of (Incidence) Unit Risk Estimates to Change in Lag Period.... D-52
D.3.6. Sensitivity of (Incidence) Unit Risk Estimates to Value of Knot D-52
D.3.7. Analysis of Age Interaction for the Exposure Terms in the Two-Piece
Linear Spline Model D-53
D.3.8. Sensitivity of (Incidence) Unit Risk Estimates to Upper-Bound
Estimation Approach—Wald vs. Profile Likelihood D-54
D.3.9. Sensitivity of Occupational Extra Risk Estimates to Change in Lag
Period D-54
D.4. HEMATOPOIETIC CANCER MORTALITY (ALL HEMATOPOIETIC
CANCERS COMBINED) (n = 17,530) D-57
iv
-------
D.4.1. Exposure Distribution in Cohort and among All
(Lympho)hematopoietic Cases in the Cohort Mortality Study D-57
D.4.2. Modeling of the Hematopoietic Cancer Mortality Data Using a Variety
of Models D-58
D.5. FURTHER CHARACTERIZATION OF THENIOSH COHORT D-64
D.5.1. Further Characterization of the Exposure Distributions and Other
Characteristics in the Full Cohort D-64
D.5.2. Further Characterization of the Exposure Distributions and Other
Characteristics in the Risk Sets D-69
D.6. POSSIBLE INFLUENCE OF THE HEALTHY WORKER SURVIVOR
EFFECT D-73
D.7. POSSIBLE INFLUENCE OF EXPO SURE MISMEASUREMENT D-74
APPENDIX E. LIFE-TABLE ANALYSIS E-1
APPENDIX F. EQUATIONS USED FOR WEIGHTED LINEAR REGRESSION OF
CATEGORICAL RESULTS 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 II-l
II. 1. SAB PANEL COMMENTS H-1
11.2. PUBLIC COMMENTS 11-30
APPENDIX I. EPA RESPONSES TO SAB COMMENTS ON 2014 EXTERNAL
REVIEW DRAFT 1-1
APPENDIX J. SUMMARY OF MAJOR NEW STUDIES SINCE THE LITERATURE
CUTOFF DATE J-l
J.l. SYSTEMATIC LITERATURE SEARCH J-l
J.2. REVIEWS OF MAJOR NEW STUDIES IDENTIFIED IN THE 2013
LITERATURE SEARCH J-5
J.2.1. Kiranetal. (2010) J-9
J.2.2. Mikoczy et al. (2011) J-10
J.3. REVIEWS OF MAJOR STUDIES IDENTIFIED BETWEEN THE 2013
LITERATURE SEARCH AND THE 2014 SAB REVIEW DRAFT J-16
J.3.1. Valdez-Flores and Sielken(2013) J-16
J.3.2. Parsons et al. (2013) [and Nagy et al. (2013)] J-16
J.4. REVIEW OF MAJOR STUDIES IDENTIFIED IN THE 2016 LITERATURE
SEARCH J-19
J.4.1. Zhang etal. (2015a) and Zhang etal. (2015b) J-19
APPENDIX K. SUMMARY OF PUBLIC COMMENTS RECEIVED ON THE JULY
2013 PUBLIC COMMENT DRAFT AND EPA RESPONSES K-l
REFERENCES R-l
v
-------
LIST OF TABLES
Table A-1. Estimated 8-hour time-weighted average ethylene oxide exposure, Plant 3 A-4
Table A-2. Cox regression results for hematopoietic cancer mortality (15-year lag) in
males A-24
Table A-3. Cox regression results for lymphoid cell line tumors (15-year lag) in males A-24
Table A-4. Exposure assessment matrix from Swaen et al. (2009)—8-hour TWA
exposures inppm A-29
Table A-5. Epidemiological studies of ethylene oxide and human cancer A-37
Table C-l. Levels of endogenous (background) N7-HEG adducts in unexposed human
and rodent tissues C-31
Table D-l. Distribution of cases in Cox regression for breast cancer morbidity analysis
after using a 15-year lag D-5
Table D-2. Minus 2 x log-likelihood results and AICs for different models and different
exposure lag times D-7
Table D-3. Categorical analysis of breast cancer incidence by deciles (exposures lagged
15 years) D-8
Table D-4. Fit of two-piece log-linear model to breast cancer incidence data, Cox
regression D-l 4
Table D-5. Fit of log-linear model to breast cancer incidence data, Cox regression (RR =
e(P x exposure)^ D-l5
Table D-6. Fit of the square root transformation log RR model to breast cancer incidence
data, Cox regression (RR = e^3 x sqrt(exP°sure))) D-16
Table D-7. Fit of the log-transform model to breast cancer incidence data, Cox
regression (RR = e^3 x ^exposure))) D-17
Table D-8. Change in -2 log-likelihood for log RR models for breast cancer incidence,
with addition of exposure term(s) D-17
Table D-9. Model fit statistics for linear RR models, breast cancer incidence D-19
Table D-10. Model coefficients for linear RR models, breast cancer incidence D-19
Table D-11. Comparison of some log-linear model results with different lag periods;
cumulative exposure inppm x days D-22
vi
-------
Table D-12. Comparison of unit risk estimates from two-piece linear spline models with
different lag periods; cumulative exposure in ppm x days, knot at 5,750 ppm x
days D-22
Table D-13. Comparison of unit risk estimates from two-piece linear spline models with
different knot; cumulative exposure in ppm x days, with lag of 15 years D-23
Table D-14. Comparison of unit risk estimates from two-piece linear spline models with
exclusion of nonexposure covariates; cumulative exposure in ppm x days with 15-
year lag, knot at 5,750 ppm x days D-24
Table D-15. Evaluation of age interaction for the exposure terms in the two-piece linear
spline model with knot at 5,750 ppm x days; cumulative exposure in ppm x days,
with lag of 15 years D-25
Table D-16. Comparison of unit risk estimates for breast cancer incidence from two-
piece linear spline model using Wald-based and profile-likelihood-based upper-
bound estimates on the 1st spline piece D-25
Table D-17. Parameter estimates for the two-piece linear spline model with the knot at
5,750 ppm x days for different lag periods; cumulative exposure in ppm x days D-26
Table D-18. Comparison of breast cancer incidence extra risk estimates from two-piece
linear spline models with different lag periods; cumulative exposure in ppm x
days, knot at 5,750 ppm x days D-27
Table D-19. Distribution of cases in Cox regression analysis of breast cancer mortality
after using a 20-year lag D-29
Table D-20. Categorical output breast cancer mortality, 20-year lag D-33
Table D-21. Two-piece log-linear spline, breast cancer mortality, 20-year lag, knot at
700 ppm-days D-33
Table D-22. Log-linear model, breast cancer mortality, 20-year lag D-34
Table D-23. Log-transform log RR model, breast cancer mortality, 20-year lag D-34
Table D-24. Two-piece log-linear spline model, breast cancer mortality, 20-year lag,
knot at 13,000 ppm-days D-35
Table D-25. Model results for breast cancer mortality, linear RR models D-36
Table D-26. Exposure categories and case distribution for lymphoid cancer mortality D-37
Table D-27. Minus 2 log-likelihood results and AICs for different models and different
exposure lag times D-39
Table D-28. Lymphoid cancer mortality results by sex D-41
vii
-------
Table D-29. Categorical results for lymphoid cancer mortality, men and women
combined D-44
Table D-30. Results of two-piece log-linear spline model for lymphoid cancer mortality,
men and women combined, knot at 100 ppm-days D-44
Table D-31. Results of the log-transform log RR model for lymphoid cancer mortality,
both sexes combined D-45
Table D-32. Results of the log-linear model for lymphoid cancer mortality, both sexes
combined D-45
Table D-33. Results of two-piece log-linear spline model for lymphoid cancer mortality,
men and women combined, knot at 1,600 ppm-days D-46
Table D-34. Distribution of cumulative exposures with a 15-year lag for the lymphoid
cancer deaths D-47
Table D-35. Model fit statistics and coefficients for log-linear RR model with square-
root of cumulative exposure, with a 15-year lag, lymphoid cancer mortality D-48
Table D-36. Model fit statistics and coefficients for linear RR models, lymphoid cancer
mortality D-49
Table D-37. Results for log cumulative exposure Cox regression model with no lag D-51
Table D-38. Comparison of unit risk estimates for lymphoid cancer incidence from two-
piece linear spline models with different lag periods; cumulative exposure in ppm
x days, knot at 1,600 ppm x days D-52
Table D-39. Comparison of unit risk estimates for lymphoid cancer incidence from two-
piece linear spline models with different knot; cumulative exposure in ppm x
days, with lag of 15 years D-53
Table D-40. Evaluation of age interaction for the exposure terms in the 2-piece linear
spline model with knot at 1,600 ppm x days; cumulative exposure in ppm x days,
with lag of 15 years D-53
Table D-41. Comparison of unit risk estimates for lymphoid cancer incidence from two-
piece linear spline model using Wald-based and profile-likelihood-based upper-
bound estimates on the 1st spline piece D-54
Table D-42. Parameter estimates for the two-piece linear spline model with the knot at
1,600 ppm x days for different lag periods; cumulative exposure in ppm x days D-55
Table D-43. Comparison of extra risk estimates for lymphoid cancer incidence from two-
piece linear spline models with different lag periods; cumulative exposure in ppm
x days, knot at 1,600 ppm x days D-56
viii
-------
Table D-44 Exposure categories and case distribution for hematopoietic cancer mortality ...D-58
Table D-45. All hematopoietic cancer mortality categorical results by sex D-58
Table D-46. Categorical results for all hematopoietic cancer mortality, men and women
combined, cumulative exposure with a 15-year lag D-61
Table D-47. Results of two-piece log-linear spline model for all hematopoietic cancer
mortality, men and women combined, cumulative exposure with a 15-year lag;
knot at 500 ppm-days D-61
Table D-48. Results of log-transform log RR model for all hematopoietic cancer
mortality, men and women combined, cumulative exposure with a 15-year lag D-62
Table D-49. Results of log-linear model for all hematopoietic cancer morality, men and
women combined, cumulative exposure with a 15-year lag D-62
Table D-50. Model fit statistics and coefficients for linear RR models, hematopoietic
cancer mortality D-63
Table D-51. Marginal summaries of workers' exposures, and years of entry to
employment and age at end of follow-up in full cohort D-64
Table D-52. Cumulative exposure to EtO by year of entry to employment in full cohort D-65
Table D-53. Cumulative exposure to EtO by duration of employment in full cohort D-65
Table D-54. Cumulative exposure to EtO in each of the risk categories in full cohort D-65
Table D-55. Sex distribution over time—case and control sexes by the year they entered
the workforce D-67
Table D-56. Year of entry to the EtO workforce D-68
Table D-57. Age of entry to the EtO workforce D-69
Table D-58. Duration of employment in the EtO workforce D-69
Table D-59. Year of departure/retirement from the EtO workforce D-69
Table D-60. Age of departure/retirement from the EtO workforce D-69
Table D-61. Summary of percentage of total case and control individual exposures in the
risk set worker histories that are excluded when the lag of 15 years is imposed D-73
Table E-l. Extra risk calculation for lymphoid cancer incidence from environmental
exposure to 0.00190 ppm (the LECoi) using the two-piece linear spline model
with knot at 1,600 ppm x days E-2
IX
-------
Table G-l. Analysis of grouped data, NTP (1987) mouse study; multistage model
parameters G-l
Table G-2. Analysis of grouped data from the Lynch et al. (1984a,c) study of male F344
rats; multistage model parameters G-l
Table G-3. Analysis of grouped data from the Garman et al. (1985) and Snellings etal.
(1984) reports onF344 rats; multistage model parameters G-2
Table G-4. Time-to-tumor analysis of individual animal data from the NTP (1987)
mouse study; multistage-Weibull model parameters G-2
Table 1-1. Number of cancer cases in the cohort attributable to EtO exposure, assuming
the selected models 1-22
Table 1-2. Plant 1, sterilizer volume and predicted EtO exposure levels by year 1-27
Table 1-3. Plant 5, sterilizer volume and predicted ETO exposure levels by year 1-27
Table J-1. Disposition of 56 new references identified as potentially relevant in 2013 J-2
Table J-2. Disposition of 17 new references identified as potentially relevant in 2016 J-5
Table J-3. New epidemiological studies of ethylene oxide and human cancer J-6
Table J-4. Comparison of Mikoczy etal. (2011) RR estimates with those obtained using
the selected models based on the NIOSH study J-12
Table J-5. Comparison of highest exposure levels estimated for the NIOSH cohort plants
with those in the Mikoczy et al. (2011) study plants J-14
Table J-6. Evaluation of reported measurements of GSH, GSSG, and HESG; averages
(SD) of poo led samples [(J,g/g tissue; Zhang et al. (2015a)] J-22
Table J-7. Evaluation of reported measurements of various DNA adducts; averages (SD)
[n = 5 analytical samples; Zhang et al. (2015b)] J-23
x
-------
LIST OF FIGURES
Figure D-l. Likelihoods vs. knots, two-piece log-linear spline model for breast cancer
incidence D-10
Figure D-2. Breast cancer incidence—two-piece log-linear spline model D-10
Figure D-3. Breast cancer incidence—log-linear (Cox regression) model D-ll
Figure D-4. Breast cancer incidence—effect on log-linear model of omitting highest
exposures D-ll
Figure D-5. Breast cancer incidence—log-linear model with log cumulative exposure D-12
Figure D-6. Breast cancer incidence—log-linear model with square root of cumulative
exposure D-l 3
Figure D-7. Likelihoods vs. knots, two-piece linear spline model, breast cancer incidence
(units are ppm-days, 15-year lag) D-l8
Figure D-8. Comparison ofWald and profile likelihood (one-sided) 95% upper-bound
estimates for 2-piece linear spline model D-20
Figure D-9. Likelihoods vs. knots for the two-piece log-linear model, breast cancer
mortality D-30
Figure D-10. Likelihoods vs. knots for the two-piece log-linear model, breast cancer
mortality, up to 15,000 ppm-days D-30
Figure D-ll. Dose-response models for breast cancer mortality D-31
Figure D-12. Breast cancer mortality—log-linear model with log cumulative exposure D-32
Figure D-13. Linear RR models for breast cancer mortality D-36
Figure D-14. AIC vs. knot for different lag periods for two-piece linear spline models D-40
Figure D-l5. Likelihoods vs. knots for two-piece log-linear model, lymphoid cancer
mortality D-42
Figure D-16. Exposure-response models for lymphoid cancer mortality D-43
Figure D-17. Comparison ofWald and profile likelihood (one-sided) 95% upper-bound
estimates for two-piece linear spline model D-50
Figure D-18. Linear RR models for lymphoid cancer D-51
xi
-------
Figure D-19. Likelihood vs. knots for two-piece log-linear model, all hematopoietic
cancer D-60
Figure D-20. Exposure-response models for hematopoietic cancer mortality D-60
Figure D-21. Linear RR models for hematopoietic cancer mortality D-64
Figure D-22. Estimated annual exposures experienced by cases and controls in the M
cohort while working—medians and interquartile ranges D-66
Figure D-23. Estimated annual exposures experienced by cases and controls in the M
cohort while working—means and 95th percentiles D-67
Figure D-24. Sex ratios for currently working populations D-68
Figure D-25. Box plots of both unlagged and 15-year lagged cumulative total exposures,
peak exposures, and exposure durations for risk sets D-71
Figure D-26. Lymphoid cancer case exposures compared to corresponding risk set
control mean exposures for cumulative total exposures, peak exposures, and
exposure durations both unlagged and with a 15-year lag D-73
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) H-15
Figure H-2. Induction of recessive lethal mutations by EtO in Drosophila (wild-type) H-17
xii
-------
LIST OF ABBREVIATIONS
8-OHdG
8-hydroxy-2'-deoxyguanosine
ACC
American Chemistry Council
ADAF
age-dependent adjustment factor
AIC
Akaike information criterion
AIDS
acquired immune deficiency syndrome
AP
apurinic/apyrimidinic
ARASP
Advancing Risk Assessment Science and Policy
BEIR
Committee on the Biological Effects of Ionizing Radiation
CI
confidence interval
DF
degrees of freedom
ECoi
effective concentration (modeled) resulting in 1% extra risk
EOSA
Ethylene Oxide Sterilization Association
EPA
U.S. Environmental Protection Agency
ERR
excess relative risk
EtO
ethylene oxide
FRG
Federal Republic of Germany
GC
gas chromatography
GSH
reduced glutathione
GSSG
oxidized glutathione
GST
glutathione S-transferase
HESG
2-hydroxyethylated glutathione
HEVal
N-(2-hydroxytheyl)valine
HOEtNU
N-(2-hy droxy ethyl)-N-nitros ourea
HPLC
high-performance liquid chromatography
i.p.
intraperitoneal
IARC
International Agency for Research on Cancer
ICD
International Classification of Diseases
IRIS
Integrated Risk Information System
IRR
incidence rate ratio
LH
lymphohematopoietic
LL
log likelihood
LECoi
95% (one-sided) lower confidence limit on the EC0i
LLOQ
lower limit of quantification
MF
mutant fraction
MLE
maximum likelihood estimate
MNBC
micronucleus frequencies in binucleated cells
MOA
mode of action
MS
mass spectrometry
Nl-HEA
N1 -(2-hy droxy ethyl)adenine
N3-HEA
N3-(2-hydroxyethyl)adenine
N3-HEC
N3-(2-hydroxyethyl)cytosine
N3-HET
N3-(2-hydroxyethyl)thymine
N3-HEU
N3-(2-hydroxyethyl)uracil
xiii
-------
N6-HEA N6-(2-hydroxyethyl)adenine
N7-HEG N7-(2-hydroxyethyl)guanine
NHL non-Hodgkin lymphoma
NIOSH National Institute for Occupational Safety and Health
NRC National Research Council
NS not specified
NTP National Toxicology Program
06-HEG 06-(2-hydroxyethyl)guanine
OPP Office of Pesticide Programs
OR odds ratio
PBPK physiologically based pharmacokinetic
POD point of departure
RR relative rate (i.e., rate ratio), or more generally, relative risk
SAB Science Advisory Board
SCE sister chromatid exchange
SD standard deviation
SE standard error
SEER Surveillance, Epidemiology, and End Results
SIR standardized incidence ratio
SMR standard mortality ratio
SN substitution nucleophilicity
TLC thin-layer chromatography
TWA time-weighted average
UCC Union Carbide Corporation
UCL upper confidence limit
xiv
-------
APPENDIX A. CRITICAL REVIEW OF EPIDEMIOLOGIC EVIDENCE
[EDITORIAL NOTE: The responses to the 2007 external peer-review and public comments can
be found in Appendix H, the responses to the 2013 public comments are in Appendix K, and the
responses to the 2015 SAB comments are in Appendix I.]
A.l. BACKGROUND
Prompted by studies indicating that ethylene oxide (EtO) is a mutagen and that exposure
toEtO produces increased chromosomal aberrations in human lymphocytes (Ehrenberg and
Gustafsson, 1970; Ehrenberg and Hallstrorq 1967; Rapoport, 1948), Hogstedt and colleagues
studied three small, independent cohorts of workers from Sweden. Reports on two of these
cohorts (Hogstedt et al., 1984; Hogstedt eta!, 1979b; Hogstedt et al., 1979a) were reviewed in
the earlier health assessment document (U.S. EPA, 1985). These two small cohorts plus a third
group of EtO-exposed workers from a third independent plant in Sweden were then combined
and studied as one cohort (Hogstedt, 1988; Hogstedt et a!, 1986). A review of this reconstituted
cohort study and subsequent independent studies is presented in Section A.2.
Shortly after the third Hogstedt study was completed, another independent study of
EtO-exposed employees was published (Gardner et al„ 1989) on a cohort of workers from four
companies and eight hospitals in Great Britain, and it was followed by a third independent study
on a cohort of exposed workers in eight chemical plants from the Federal Republic of Germany
(Kiesselbach etal., 1990). A follow-up study of the Gardner etal. (1989) cohort was conducted
by Coggon et al. (2004).
Greenberg etal. (1990) was the first in a series of studies of workers exposed to EtO at
two chemical manufacturing facilities in the Kanawha Valley (South Charleston, WV). The
workers at these two facilities were studied later by Teta et al. (1993), Benson and Teta (1993),
Teta et al. (1999), and Swaen et al. (2009) and became the basis for several quantitative risk
assessment analyses (Valdez-Flores et al., 2010; EPIC, 2001; Teta et al., 1999).
Another independent study of EtO-exposed workers in 14 sterilizing plants from across
the United States was completed by the National Institute for Occupational Safety and Health
(NIOSH) (Stayner etal., 1993; Steenland etal., 1991). The Stayner etal. (1993) paper presents
the exposure-response analysis performed by the NIOSH investigators. These same workers
were studied again from a different perspective by Wong and Trent (1993). The NIOSH
investigators then conducted a follow-up of the mortality study (Steenland et al., 2004) and a
breast cancer incidence study based in the same cohort (Steenland etal., 2003). The results of
the Steenland etal. (2004) and Steenland etal. (2003) analyses are the basis for the quantitative
A-l
-------
assessment in this document, for reasons explained in the review and summary sections of this
appendix as well as Section 4.1 of the main report.
Several additional studies of lesser importance have been done on EtO-exposed cohorts
of workers in Sweden (Hagmar etal., 1995; Hagmar et al.. 1991). Italy (Bisanti et al.. 1993).
Belgium (Swaen et al.. 1996) and western New York State (Norman et al.. 1995). and other parts
of the United States (Olsen et al.. 1997). These studies are discussed in the following review, but
they provide limited information to the overall discussion of whether EtO induces cancer in
humans.
The more important studies, which are discussed in detail in the summary, are those at
two facilities in the Kanawha Valley in West Virginia (Valdez-Flores et al., 2010; Swaen et al.,
2009; Teta etal., 1999; Benson and Teta, 1993; Teta et al„ 1993; Greenberg et al„ 1990) and at
14 sterilizing plants around the country (Steenland etal., 2004; Steenland et al„ 2003; Stayner et
al., 1993; Steenland et al„ 1991). These studies have sufficient follow-up to analyze latent
effects, and the cohorts appear to be large enough to test for small differences. In addition,
exposure estimates were derived for both cohorts, and attempts were made to assess
dose-response relationships.
More recently, a follow-up study of the Swedish cohort of Hagmar et al. (Hagmar et al.,
1995; Hagmar et al., 1991), which also had quantitative exposure estimates for the individual
workers, was published (Mikoczy etal., 2011). This follow-up study was published after the
general cutoff date for literature inclusion in this assessment and is reviewed in detail in Section
J.2.2 of Appendix J. However, because it is a follow-up of an earlier study and, with the
additional follow-up, provides important corroborating evidence, the study is also briefly
mentioned here.
A.2. INDIVIDUAL STUDIES
A.2.1. Hogstedt (1988), Hogstedt et al. (1986)
Hogstedt etal. (1986) combined workers from several cohorts for a total of 73 3 workers,
including 378 workers from two separate and independent occupational cohort mortality studies
by Hogstedt etal. (1979b) and Hogstedt etal. (1979a) and 355 employees from a third EtO
production plant who had not been previously examined. The combined cohort was followed
until the end of 1982. The first cohort comprised employees from a small technical factory in
Sweden where hospital equipment was sterilized with EtO. The second was from a production
facility where EtO was produced by the chlorohydrin method from 1940 to 1963. The third was
from a production facility where EtO was made by the direct oxidation method from 1963 to
1982.
A-2
-------
In the update of the 1986 occupational mortality report (Hogstedt, 1988). the cohort
inexplicably was reduced to 709 employees (539 men; 170 women). Follow-up for mortality
was extended to the end of 1985. The author reported that 33 deaths from cancer had occurred,
whereas only 20 were expected in the combined cohort. The excess was attributed primarily to
an increased risk of stomach cancer at one plant and an increase in blood and lymphatic
malignancies at all three. Seven deaths from leukemia occurred, whereas only 0.8 were expected
(standard mortality ratio [SMR] = 9.2). Ten deaths from stomach cancer occurred versus only
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 hypothesize 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, observed even at low-level and intermittent exposures to EtO,
should be "seriously considered by industry and regulating authorities."
In Plant 1 (Hogstedt et al., 1979a) in 1977, the average air concentrations of EtO 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. Exposure
levels were lower in the sterilization room.
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, average 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.
A-3
-------
Table A-l. Estimated 8-hour time-weighted
average ethylene oxide exposure, Plant 3
Group
1963-1976
1977-1982
A (n= 128)
5-8 ppm
1-2 ppm
B (n = 69)
3 ppm
1 ppm
C(n = 158)
1-3 ppm
0.4—1.6 ppm
Source: Hogstedtet al. f!986\
In the earlier studies (Hogstedt et al., 1979b; Houstedt etal., 1979a) of two of the plants
that contributed workers to this cohort, the authors note that there was exposure to ethylene
dichloride, ethylene chlorohydrin, ethylene, and small amounts of bis-(2-chloroethyl) ether, as
well as other chemicals in the respective plants. Although 170 women were present in the
workforce, no sex differences in risk were analyzed separately by the investigators. Of
16 patients with tumors in the two exposed cohorts, there were three cases of leukemia
(0.2 expected), six cases of alimentary tract cancer, and four cases of urogenital cancer. Of the
11 cancer cases in the full-time exposed cohort, 5.9 were expected (p < 0.05). This study was
criticized by Divine and Amanollahi (1986) for several reasons. First, they argued that the
study's strongest evidence in support of a carcinogenic claim for EtO was only a "single case of
leukemia" in subgroup C of Plant 3, where the workers had multiple chemical exposures;
however, there were no cases in subgroups A orB of Plant 3. Hogstedt etal. (1986) countered
that the expectation of leukemia in these two subgroups was 0.04 and 0.02, respectively, and that
the appearance of a case could only happen if EtO had "outstanding carcinogenic properties at
low levels." Divine and Amanollahi also pointed out that a study (Morgan etal., 1981) of a
cohort similar to that of Plant 3 found no leukemia cases or evidence of excessive mortality.
Hogstedt etal. (1986) replied that Morgan etal. (1981) stated in their paper that the statistical
power of their study to detect an increased risk of leukemia was not strong.
Divine and Amanollahi (1986) also stated that the exposures to EtO were higher in
Plants 1 and 2 than in Plant 3; therefore, combinations would "normally preclude comparisons
among the plants for similar causes of adverse health." This potential problem could be resolved
by structuring exposure gradients to analyze risk. Furthermore, they noted Plant 1 was a
nonproduction facility involved in sterilization of equipment. Plant 2 used the chlorohydrin
process for making EtO, and Plant 3 used the direct oxygenation process. Although these
conditions are obviously different, they "are grouped together as analogous." This criticism
would, in most instances, be valid only because the methods for producing EtO differ and there
were differing exposures to multiple chemicals.
A-4
-------
However, these concerns are not supported by the evidence. In all three plants the
leukemia risk was elevated, even if only slightly in Plant 3. This suggests that there may have
been a common exposure, possibly to EtO, endemic to all three plants that was responsible for
the measured excesses. Noteworthy is the elevated risk of leukemia seen in Plant 1 (3 observed
vs. 0.14 expected), where the exposures were almost exclusively to EtO in the sterilization of
equipment. The argument that Plant 1 leukemias form a "chance cluster," as Shore et al. (1993)
claim, and as such should be excluded from any analysis does not preclude the possibility that
these cases are in reality the result of exposure to EtO. Hogstedt (1988) argues that earlier
remarks by Ehrenberg and Gustafsson (1970) that EtO "constituted a potential cancer hazard" on
the basis of a considerable amount of evidence other than epidemiologic should have served as a
warning that the increased risk seen in Plant 1 was not necessarily a "chance cluster," and
because the chlorohydrin process was not used in Plant 1, it cannot be due to exposure to a
chemical in the chlorohydrin process.
A.2.2. Gardner etal. (1989)
Gardner et al. (1989) completed a cohort study of 2,876 men and women who had
potential exposure to EtO. The cohort was identified from employment records at four
companies that had produced or used EtO since the 1950s and from eight hospitals that have had
EtO clinical sterilizing units since the 1960s. The cohort was followed to December 31, 1987.
All but 1 of the 1,012 women and 394 of the men in the cohort worked at one of the hospitals.
The remaining woman and 1,470 men made up the portion of the cohort from the four
companies. By the end of the follow-up, 226 members (8% of the total cohort) had died versus
258.8 expected. Eighty-five cancer deaths were observed versus 76.64 expected.
No clear excess risk of leukemia (3 observed vs. 2.09 expected), stomach cancer
(5 observed vs. 5.95 expected), or breast cancer (4 observed vs. 5.91 expected) was present as of
the cutoff date. "Slight excesses" of deaths due to esophageal cancer (5 observed vs.
2.2 expected), lung cancer (29 observed vs. 24.55 expected), bladder cancer (4 observed vs.
2.04 expected), and non-Hodgkin lymphoma (NHL) (4 observed vs. 1.63 expected) were noted,
although an adjustment made to reflect local "variations in mortality" reduced the overall cancer
excess from 8 to only 3. According to the authors' published tabulations, all three leukemias
identified in this study fell into the longest latent category (20 years or longer), where only
0.35 were expected. All three were in the chemical plants. This finding initially would seem to
be consistent with laboratory animal evidence demonstrating excess risks of hematopoietic
cancer in animals exposed to EtO. But the authors note that because other known leukemogens
were present in the workplace, the excess could have been due to a confounding effect.
A-5
-------
The hospitals began using EtO during or after 1962, whereas all of the chemical
companies had handled EtO from or before 1960. In the hospitals there was occasional exposure
to formaldehyde and carbon tetrachloride but few other confounding agents. On the other hand,
the chemical workers were exposed to a wide range of compounds including chlorohydrin,
propylene oxide, styrene, and benzene. The earliest industrial hygiene surveys in 1977 indicated
that the TWA average exposures were less than 5 ppm in almost all jobs and less than 1 ppm in
many. No industrial hygiene data were available for any of the facilities prior to 1977, although
it is stated that peaks of exposure up to several hundred ppm occurred as a result of operating
difficulties in the chemical plants and during loading and unloading of sterilizers in the hospitals.
An odor threshold of 700 ppm was reported by both manufacturers and hospitals, according to
the authors. The authors assumed that past exposures were somewhat higher without knowing
precisely what they were. An attempt was made to classify exposures into a finite number of
subjectively derived categories (definite, possible, continual, intermittent, and unknown). This
exercise produced no discernable trends in risk of exposure to EtO. However, the exposure
status classification scheme was so vague as to be useless for determining risk by gradient of
exposure to EtO.
It is of interest that all three of the leukemia deaths entailed exposure to EtO, with very
little or no exposure to benzene, according to the authors. The findings are not inconsistent with
those ofHogstedt (1988) and Hogstedt etal. (1986). The possibility of a confounding effect
from substances other than benzene in these chemical workers cannot be entirely ruled out.
Other cancers were slightly in excess, but overall there was little increased mortality from cancer
in this cohort. It is possible that if very low levels of exposure to EtO had prevailed throughout
the history of these hospitals and plants, the periods of observation necessary to observe an effect
may not have been long enough.
A follow-up study of this cohort conducted by Coggon et al. (2004) is discussed below.
A.2.3. Kiesselbachetal. (1990)
Kiesselbach etal. (1990) carried out an occupational cohort mortality study of 2,658 men
from eight chemical plants in the Federal Republic of Germany (FRG) that were involved in the
production of EtO. The method of production is not stated. At least some of the plants that were
part of an earlier study by Thiess etal. (1982) were included. Each subject had to have been
exposed to EtO for at least 1 year sometime between 1928 and 1981 before person-years at risk
could start to accumulate. Most exposures occurred after 1950. By December 31, 1982, the
closing date of the study, 268 men had died (about 10% of the total cohort), 68 from malignant
neoplasms. The overall SMR for all causes was 0.87, and for total cancer, the SMR was 0.97,
A-6
-------
based on FRG rates. The authors reported that this deficit in total mortality indicates a
healthy-worker effect.
The only remarkable findings here are slightly increased risks of death from stomach
cancer (14 observed vs. 10.15 expected, SMR= 1.4), cancer of the esophagus (3 observed vs.
1.5 expected, SMR = 2), and cancer of the lung (23 observed vs. 19.86 expected, SMR= 1.2).
Although the authors claimed that they looked at latency, only stomach cancer and total
mortality has a latency analysis included. This was accomplished by not counting the first
10 years of follow-up in the parameter "years since first exposure." This study is limited by the
lack of further latency analyses at other cancer sites. The risk of stomach cancer shows only a
slight nonsignificant trend upward with increasing latency. Only two leukemias were recorded
versus 2.35 expected.
This is a largely unremarkable study, with few findings of any significance. No actual
exposure estimates are available. The categories of exposure that the authors constructed are
"weak," "medium," and "strong." It is not known whether any of these categories is based on
actual measurements. No explanation of how they were derived is provided except that the
authors claim that the information is available on 67.2% of the members of the cohort. If the
information was based on job categories, it should be kept in mind that exposures in jobs that
were classified the same from one plant to the next may have produced entirely different
exposures toEtO. The tabular data regarding these exposure categories shows that only 2.4% of
all members of the cohort were considered "strongly" exposed to EtO. Although 71.6% were
classified as "weak," the remaining 26% were considered as having "medium" exposure to EtO.
This is largely a study in progress, and further follow-up will be needed before any
definite trends or conclusions can be drawn. The authors reported that only a median 15.5 years
of follow-up had passed by the end of the cutoff" date, whereas the median length of exposure
was 9.6 years. Before any conclusions can be made from this study, several additional years of
follow-up would be needed with better characterization of exposure.
A.2.4. Greenberg etal. (1990)
Greenberg etal. (1990) retrospectively studied the mortality experience of 2,174 men
who were assigned to operations that used or produced EtO in either of two Union Carbide
Corporation (UCC) chemical plants in West Virginia. In 1970 and 1971, EtO production at the
two plants was phased out, but EtO was still used in the plants to produce other chemicals.
SMRs were calculated in comparison with the general U.S. population and the regional
population. Results based on regional population death rates were found to be similar to those
based on the U.S. general population. Follow-up began either on January 1, 1940, if exposure to
EtO began sooner, or on the date when exposure began, if it occurred after January 1, 1940.
A-7
-------
Follow-up ended on December 31, 1978. Note that this cohort is thus a mixture of a prevalent
cohort and an incident cohort, and the prevalent part of the cohort may be especially vulnerable
to bias from the healthy worker survivor effect. The healthy worker survivor effect might have
occurred if workers who were employed before 1940 and who were of greater susceptibility
preferentially developed a disease of interest prior to 1940 and were no longer employed when
cohort enumeration began. It appears that the chemical facilities began operating in 1925, so the
maximum latency for the development of a disease of interest between the time of first exposure
and cohort enumeration was 15 years; however, these early (pre-1940) hires would also have had
the highest EtO exposures (Swaen et al., 2009) and may thus have had short latency periods as
well. The healthy worker survivor effect bias can also dampen exposure-response relationships
(Applebaum et al., 2007). According to Greenberg et al. (1990), slightly over 10% of the cohort
consisted of prevalent hires (223 of 2,174). This is not a large proportion, but as noted above,
these early hires would also have had the highest exposures (Swaen et al., 2009). It is unknown
how many workers employed before 1940 were no longer employed when cohort enumeration
began. Two years of pre-1940 exposure were reportedly taken into account when categorizing
the cohort into groups with >2 years exposure in the different potential exposure categories (see
below); however, it is unclear how pre-1940 years of exposure were treated in other analyses, for
example, the analyses based on duration of exposure (although presumably they were taken into
account for those analyses as well).
Total deaths equaled 297 versus 375.9 expected (SMR = 0.79,p < 0.05). Only 60 total
cancer deaths were observed versus 74.6 expected (SMR= 0.81). These deficits in mortality
suggest a manifestation of the healthy-worker effect. In spite of this, nonsignificant elevated
risks of cancer of the liver, unspecified and primary, (3 observed vs. 1.8 expected, SMR = 1.7),
pancreas (7 observed vs. 4.1 expected, SMR= 1.7), and leukemia and aleukemia (7 observed vs.
3.0 expected, SMR = 2.3) were noted.
The authors also reported that in 1976 (3 years before the end of follow-up), an industrial
hygiene survey found that 8-hour TWA EtO levels averaged less than 1 ppm, although levels as
high as 66 ppm 8-hour TWA had been observed. In maintenance workers, levels averaged
between 1 and 5 ppm 8-hour TWA. Because of the lack of information about exposures before
1976 (e.g., when EtO was in production), the authors developed a qualitative exposure
categorization scheme with three categories of exposure (low, intermediate, and high) on the
basis of the potential for exposure in each department. The number of workers in each exposure
category was not reported; however, it appears from Teta etal. (1993) (see below) that only
425 workers were assigned to EtO production departments, which were apparently the only
departments with high potential exposure. No significant findings of a dose-response
relationship were discernable.
A-8
-------
Except for two cases of leukemia, all the workers who died of pancreatic cancer or
leukemia began their work—and hence exposure to EtO—many years before their deaths. The
leukemia and pancreatic cancer deaths were concentrated in the chlorohydrin production
department. Four of the seven workers who died of leukemia had been assigned to the
chlorohydrin department; only 0.8 deaths (SMR = 5.0) would have been expected in this
department of only 278 workers. Six of the workers who died of pancreatic cancer were
assigned to the chlorohydrin department, whereas only 0.98 deaths would have been expected to
occur (SMR = 6.1). All seven workers who died of leukemia, including the four in the
chlorohydrin department, were listed by the authors as having only low potential exposure to
EtO. In contrast, among workers ever assigned to a department in the high exposure category,
no leukemia deaths and only one pancreatic cancer death occurred.
The authors hypothesized that the excesses in leukemia and pancreatic cancers were
associated with production of ethylene chlorohydrin or propylene chlorohydrin, or both, in the
chlorohydrin department. Some later follow-up studies (described below) were done of the
cohort excluding the chlorohydrin production workers (Teta et al., 1993) and of the chlorohydrin
production workers alone (Benson and Teta, 1993) to further examine this hypothesis.
A.2.5. Steenland etal. (1991)
In an industry-wide analysis by NIOSH, Steenland etal. (1991) studied EtO exposure in
18,254 workers identified from personnel files of 14 plants that had used EtO for sterilization of
medical equipment, treating spices, or testing sterilizers. Each of the 14 plants (from 75 facilities
surveyed) that were considered eligible for inclusion in the study had at least 400 person-years at
risk prior to 1978. Within each eligible facility, at least 3 months of exposure to EtO qualified an
employee for inclusion in the cohort. Employees, including all salaried workers, who were
"judged never to have been exposed to EtO" on the basis of industrial hygiene surveys were
excluded. Follow-up ended December 31, 1987. The cohort averaged 16 years of latency.
Approximately 86% achieved the 9-year latent point, but only 8% reached the 20-year latency
category. The average year of first exposure was 1970, and the average length of exposure was
4.9 years. The workers' average age at entry was not provided, nor was an age breakdown.
Nearly 55% of the cohort were women.
Some 1,137 workers (6.4%) were found to be deceased at the end of the study period,
upon which the underlying cause of death was determined for all but 450. If a member was
determined to be alive as of January 1, 1979, but not after and no death record was found in the
National Death Index through December 31, 1987, then that member was assumed to be alive for
the purposes of the life-table analysis and person-years were accumulated until the cutoff date.
Altogether, 4.5% of the cohort fell into this category. This procedure would tend to increase the
A-9
-------
expected deaths and, as a consequence, potentially bias the risk ratio downward if a sizable
number of deaths to such persons during this period remained undiscovered to the researchers.
In the total cohort no significantly increased risks of death from any site-specific cancer
were noted. Analyses by job categories and by duration of exposure indicated no excess risks of
cancer when compared with the rate in the general population. However, there was an increased
trend in the risk of hematopoietic cancers, all sites, with increasing lengths of time since first
exposure. After 20 years latency, the SMR was 1.76, based on 13 cases. The test for trend was
significant at p = 0,03, For men (45%), without regard for latency, the SMR for hematopoietic
cancer was a significant 1.55 (p < 0.05), based on 27 cases. Among men with long latency
(greater than 20 years) and the longest duration of exposure (greater than 7 years) the SMR for
hematopoietic cancers was 2.63, based on 7 deaths (p < 0.05).
The authors pointed out that the SMR for leukemia among men was 3.45, based on
5 deaths (p< 0.05), for deaths in the latter period of 1985 to 1987. For kidney cancer, the SMR
was 3.27, based on six deaths (p < 0.05), after 20-years latency. The authors also reported on a
significant excess risk (p < 0.05) of lymphosarcoma-reticulosarcoma in men (SMR = 2.6), based
on seven deaths. Women had a lower nonsignificant rate. The risk of breast cancer was also
nonsignificant (SMR = 0.85 based on 42 cases). The authors hypothesized that men were more
heavily exposed to EtO than were women because "men have historically predominated in jobs
with higher levels of exposure." However, the lack of an association between EtO exposure and
lymphohematopoietic cancer in females was also observed in the exposure-response analyses of
this cohort, including in the highest exposure category, performed by Stayner et al. (1993) and
discussed below.
Industrial hygiene surveys indicated that sterilizer operators were exposed to an average
personal 8-hour TWA EtO level of 4.3 ppm, whereas all other workers averaged only 2 ppm,
based on 8-hour samples during the period 1976 to 1985. These latter employees primarily
worked in production and maintenance, in the warehouse, and in the laboratory. This was during
a time when engineering controls were being installed to reduce worker's exposure to EtO;
earlier exposures may have been somewhat higher. The authors reported that no evidence of
confounding exposure to other occupational carcinogens was documented.
The authors concluded that there was a trend toward an increased risk of death from
hematopoietic cancer with increasing lengths of time since the first exposure to EtO. This trend
might have been enhanced if the authors had added additional potential deaths identified from
the 820 (4.5%) "untraceable" members of the cohort from 1979 to 1987. The authors felt that
their results were not conclusive for the relatively rare cancers of a priori interest, based on the
limited number of cases and the short follow-up. The cohort averaged 16 years of latency and
86%) had at least 9 years, but only 8%> reached the 20-year latent category.
A-10
-------
Exposure-response analyses were conducted by Stavner etal. (1993) and are discussed
below. More recently, a follow-up mortality study (Steenland et al.. 2004) and abreast cancer
incidence study (Steenland et al.. 2003) of this cohort were conducted; these are also discussed
below.
A.2.6. Teta et al. (1993)
In a follow-up analysis of the cohort of 2,174 male UCC workers studied by Greenberg et
al. (1990), Teta and her colleagues excluded the 278 workers in the chlorohydrin unit in which
Greenberg and colleagues found a high risk of leukemia and pancreatic cancer, thereby removing
the potential confounding of the chlorohydrin production process. The 1,896 men in the
remaining cohort were followed for an additional 10 years, through all of 1988. (Among the
278 men who were excluded because they had worked in the chlorohydrin unit, 49 had also been
assigned to EtO production departments, which were considered high potential EtO exposure
departments, according to Greenberg etal. (1990). Data were reportedly examined with and
without the inclusion of these 49 workers with overlapping assignments; however, the results of
these analyses are not My presented.) According to Benson and Teta (1993). 112 of the
278 excluded workers were employed before 1940, reducing the prevalent part of the remaining
cohort to 111 of 1,896 workers, or just under 6%. (It is unclear how pre-1940 years of exposure
were treated in the analyses based on duration of exposure, although presumably they were taken
into account.) The update did not include additional work histories for the study subjects. Teta
et al. (1993) note that duration of assignment to an EtO production unit was not affected by the
update because EtO was no longer in production at the two plants; however, assignment to
EtO-using departments might have been affected, and according to Greenberg etal. (1990). some
of these departments had medium EtO exposure potential.
Teta et al. (1993) reported that the average duration of exposure was more than 5 years
and the average follow-up was 27 years. Furthermore, at least 10 years had elapsed since first
exposure for all the workers. The reanalysis demonstrated no increased risk of overall cancer, or
of leukemia, NHL, or cancers of the brain, pancreas, or stomach. The SMRfor total deaths,
based on comparison with mortality from the general population, was 0.79 (p < 0.01;
observed = 431). The SMR for total cancer was 0.86 (observed = 110). No site-specific cancers
were significantly elevated. Although the authors concluded that this study did not indicate any
significant trends of increasing site-specific cancer risk with increasing duration of potential
exposure to EtO, there appeared to be a nonsignificant increasing trend for leukemia and
aleukemia (p = 0.28, based on five cases) as well as stomach cancer (p = 0.13; eight cases).
According to Greenberg etal. (1990). 8-hour TWA EtO levels averaged less than 1 ppm,
based on the 1976 monitoring (after EtO production at the plants had ceased), although levels as
A-ll
-------
high as 66 ppm 8-hour TWA were reported. Teta et al. (1993) estimated that in the 1960s,
exposure in the units producing EtO by direct oxidation ranged from 3 to 20 ppm 8-hour TWA,
with peaks of several hundred ppm. These estimates were based on an industrial hygiene survey
conducted at another UCC facility in Texas that used the same direct oxidation process as the
two plants in West Virginia from which the UCC EtO cohort was taken. Ethylene oxide was
also produced via the chlorohydrin process in a closed building during the years 1925 to 1957.
Levels of exposure to EtO would have been higher than in the direct oxidation production
process because of start-up difficulties, fewer engineering controls, less complex equipment, and
the enclosed building. Employee nausea, dizziness, and vomiting were documented in the
medical department in 1949. These acute effects occur in humans at exposures of several
hundred ppm, according to the authors.
During the time periods under investigation, the estimated exposure ranges for
departments using or producing EtO were >14 ppm from 1925 to 1939; 14 ppm from 1940 to
1956; 5-10 ppm from 1957 to 1973; and <1 ppm from 1974 to 1988, with frequent peaks of
several hundred ppm in the earliest period and some peaks of similar intensity in the 1940s to
mid-1950s. In the absence of monitoring data prior to 1976, these estimates cannot be
confirmed. Furthermore, workers were eliminated from the analysis if they had worked in the
chlorohydrin unit because it was assumed that the increased risks of leukemia and pancreatic
cancer were possibly due to exposure to something in the chlorohydrin process, as conjectured
by Greenberg etal. (1990). However, even when the potential confounding influence of the
chlorohydrin process is removed, there remains the suggestion of a trend of an increasing risk of
leukemia and aleukemia with increasing duration of exposure to EtO in the remaining cohort
members (p = 0.28, based on 5 cases).
The authors indicated that their findings do not confirm the findings in laboratory animal
studies and are not consistent with the earliest results reported among EtO workers. They also
noted that they did not observe any significant trend of increasing risks of stomach cancer
(n = 8), leukemia (// = 5) or cancers of the pancreas or brain and nervous system with increasing
duration of exposure. No lagged exposure or latency analyses were conducted in this study.
In a later analysis, Teta etal. (1999) fitted Poisson regression dose-response models to
the UCC data (Teta etal., 1993) and to the NIOSHdata (Steenland etal., 1991). They reported
that latency and lagging of dose did not appreciably affect the fitted models. Because Teta et al.
(1999) did not present risk ratios for the categories used to model the dose-response
relationships, the only comparison that could be made between the UCC and NIOSH data is
based on the fitted models. The results of these models are almost identical for leukemia, but,
for the lymphoid category, the risk according to the fitted model for the UCC data decreased as a
function of dose, whereas the risk for the modeled NIOSH data increased as a function of dose.
A-12
-------
However, the models are based on small numbers of cases (16 [5 UCC, 11 NIOSH] for
leukemia; 22 [3 UCC, 19 NIOSH] for lymphoid cancers), and no statistics are provided to assess
model goodness-of-fit or to compare across models. This analysis is superseded by the more
recent analysis by the same authors (Valdez-F lores et al.. 2010) of the results of more recent
follow-up studies of these two cohorts [see discussion of the Swaen et al. (2009) study below],
A.2.7. Benson and Teta (1993)
In a companion mortality study (Benson and Teta, 1993), the remaining 278 employees
who were identified by Greenberg etal. (1990) as having worked at some time in the
chlorohydrin unit and who were not included in the cohort of Teta etal. (1993) were followed to
the end of 1988. Note that the prevalent part (i.e., those workers first employed before the cohort
enumeration date of 1 January 1940) of this reduced cohort is 112 of the 278 workers, or 40%,
and therefore, the potential for bias from a healthy worker survivor effect, as discussed for the
Greenberg etal. (1990) study above (see Section A. 2.4), may be more pronounced in this study
of the chlorohydrin unit workers. It is unknown how many chlorohydrin unit workers employed
before 1940 were no longer employed when cohort enumeration began.
Altogether, 40 cancer deaths occurred versus 30.8 expected (SMR = 1.3) in the subcohort
of chlorohydrin workers. In Greenberg etal. (1990), significant elevated risks of pancreatic
cancer and leukemia and aleukemia occurred in only those workers assigned to the chlorohydrin
process. Benson and Teta (1993) noted a significantly increased risk of pancreatic cancer
(SMR = 4.9, eight observed deaths, p < 0.05) in the same group and a significantly increased risk
of cancer in the enlarged category of lymphohematopoietic cancer (SMR = 2.9, eight observed
deaths, p < 0.05), which included leukemia and aleukemia, after an additional 10 years of
follow-up.
The authors concluded that these cancers were likely work-related and some exposure in
the chlorohydrin unit, possibly to the chemical ethylene dichloride, was probably the cause.
They pointed out that Greenberg etal. (1990) found that the chlorohydrin unit was likely to be a
low-EtO exposure area in the West Virginia plants. The other possibility was bis-chloroethyl
ether, which the authors pointed out is rated by the International Agency for Research on Cancer
(IARC) as a group 3 ("not classifiable as to its carcinogenicity to humans") chemical.
Circumstantial evidence seems to support the authors' contention that ethylene dichloride is the
cause: IARC designated ethylene dichloride as a group 2B chemical ("possibly carcinogenic to
humans"), exposure was likely heavier throughout the history of the facility, and plant medical
records documented many accidental overexposures occurring to the workers who died of
pancreatic cancer prior to diagnosis. However, this conclusion is disputed by Olsen et al. (1997)
whose analysis is discussed later.
A-13
-------
A.2.8. Stayner et al. (1993)
Stayner etal. (1993) provide an exposure-response analysis for the cohort study of EtO
workers described by Steenland etal. (1991). Nothing was modified concerning the follow-up,
cohort size, vital status, or cutoff date of the study. The exposure assessment and verification
procedures were presented in Greife et al. (1988) and Hornung etal. (1994). In brief, a
regression model was developed, allowing the estimation of exposure levels for time periods,
facilities, and operations for which industrial hygiene data were unavailable. The data for the
model consisted of 2,700 individual time-weighted exposure values for workers' personal
breathing zones, acquired from 18 facilities between 1976 and 1985. These data were divided
into two sets, one for developing the regression model and the second (from six randomly
selected plants) for testing it. Job titles were grouped into eight categories with similar potential
for EtO exposure, and arithmetic mean exposure levels by facility, year, and exposure category
were calculated from the data used for model development. The arithmetic means were
logarithmically transformed, and weighted linear regression models were fitted. Seven out of
23 independent variables tested for inclusion in the model were found to be significant predictors
(p <0,10) of EtO exposure and were included in the final model (exposure category [job], type
of product sterilized, sterilizer size, engineering controls [rear exhaust, aeration], days since
product sterilization, and calendar year). This model predicted 85% of the variation in average
EtO exposure levels in the test data. The model was also evaluated against estimates for the test
data derived by a panel of 11 industrial hygienists familiar with EtO levels in the sterilization
industry and provided with the values for the independent variables used in the model
corresponding to the arithmetic means from the test data. The overall mean of the modeled
estimates was not highly biased nor biased in one direction when compared to the overall mean
exposure estimates of the individual industrial hygiene experts. Using the test data as the
standard, the model estimates showed less bias (average difference) than 9 of the 11 industrial
hygienists and more precision (standard deviation of the differences) than all 11. Similarly, the
model outperformed the panel in terms of both bias and precision when the panel results were
averaged.
Average exposure levels, including early historical exposure levels, for the exposure
categories in the study plants were estimated using this industrial hygiene-based regression
model. Then, the cumulative exposure for each worker was estimated by calculating the product
of the average exposure in each job the worker held by the time spent in that job and then
summing these over all the jobs held by that worker. This value became the cumulative
exposure index for that employee and reflected the working lifetime total exposure to EtO.
(Details about the exposure estimates for the cohort are presented in Section D.5 of
Appendix D.)
A-14
-------
Stayner etal. (1993) generated SMRs based on standard life-table analysis. The three
categories of cumulative exposure were less than 1,200 ppm-days, 1,200 to 8,500 ppm-days, and
greater than 8,500 ppm-days. Additionally, the Cox proportional hazards model was used to
model the exposure-response relationship between EtO and various cancer types, using
cumulative exposure as a continuous variable.
Stayner etal. (1993) noted a marginally significant increase in the risk of hematopoietic
cancers, with an increase in cumulative exposure by both the life-table analysis as well as the
Cox model, although the magnitude of the increased risk was not substantial. At the highest
level—greater than 8,500 ppm-days of exposure—the SMR was a nonsignificant 1.24, based on
13 cases. However, 12 of these cases were in males, whereas only 6.12 were expected. Thus, in
this highest exposure category, a statistically significant (p < 0,05) SMR of 1.96 in males was
produced. This dichotomy produced a deficit in females (1 observed vs. 4.5 expected, p < 0.05).
The Cox analysis produced a significantly positive trend with respect to lymphoid cell
tumors (combination of lymphocytic leukemia and NHL) when EtO exposures were lagged
5 years. The authors stated that these data provide some support for the hypothesis that exposure
to EtO increases the risk of mortality from lymphatic and hematopoietic neoplasms. They
pointed out, however, that their data do not provide evidence for a positive association between
exposure to EtO and cancer of the stomach, brain, pancreas, or kidney or leukemia as a group.
Breast cancer was not analyzed in this report.
This cohort was not updated with vital status information on the "untraceables" (4.5%),
and cause of death information was not provided on deaths with unknown causes; thus, the
cohort lacks a complete follow-up, and therefore, the risk estimates may be understated. Another
potential limiting factor is the information regarding industrial hygiene measurements of EtO
that were completed in the plants. According to the authors, the median length of exposure to
EtO of the cohort was 2.2 years and the median exposure was 3.2 ppm. It may be unreasonable
to expect any findings of increased significant risks because follow-up was too short to allow the
accumulation of mortality experience (average follow-up =16 years; only 8% of cohort had
>20 years follow-up).
The authors also remind us that there is a lack of evidence for an exposure-response
relationship among females or for a sex-specific carcinogenic effect of EtO in either laboratory
animals or humans. In fact, the mortality rate from hematopoietic cancers among the women in
this cohort was lower than that of the general U.S. population. Therefore, the contrast seen here
is unusual.
The positive findings are somewhat affected by the presence in the cohort of one heavily
exposed case (although the authors saw no reason to exclude it from the analysis), and there is a
lack of definite evidence for an effect on leukemia as a group. Despite these limitations, the
A-15
-------
authors believe that their data provide support for the hypothesis that exposure to EtO increases
the risk of mortality from hematopoietic neoplasms.
This analysis is superseded by the more recent analysis by the same authors of the results
of a more recent follow-up study of this cohort [see discussion of the Steenland et al. (2004)
study below],
A.2.9. Wong and Trent (1993)
This study is a reanalysis of the same cohort that was studied by Stayner etal. (1993) and
Steenland et al. (1991), with some differences. The cohort was incremented without explanation
by 474 to a total of 18,728 employees and followed one more year, to the end of December 1988.
This change in the cohort resulted in the addition of 176 observed deaths and 392.2 expected
deaths. The finding of more than twice as many expected deaths as observed deaths is baffling.
A reduced total mortality of this magnitude suggests that many deaths may have been
overlooked, resulting in a further reduction of the overall SMR to a significant deficit of 0.73.
Sixty additional cancer deaths were added versus 65.9 expected, for an SMR = 0.9, based on 403
total cancer deaths observed versus 446.2 expected.
The authors reported no significant increase in mortality at the cancer sites found to be of
most interest in previous studies (i.e., stomach, leukemia, pancreas, brain, and breast). They also
reported the lack of a dose-response relationship and correlation with duration of employment or
latency. They did report a statistically significant increased risk of NHL among men
(SMR = 2.47; observed = 16, expected = 6.47; p < 0.05) that was not dose related and a
nonsignificant deficit of NHL among women (SMR = 0.32; observed = 2, expected = 6.27). The
authors concluded that the increase in men was not related to exposure to EtO but could in fact
have been related to the presence of acquired immune deficiency syndrome (AIDS) in the male
population. When this explanation was offered in a letter to the editor (Wong. 1991) regarding
the excess of NHL reported in Steenland etal. (1991). it was dismissed by Steenland and Stayner
(1993) as pure speculation. Steenland and Stayner (1993) responded that most of the NHL
deaths occurred prior to the AIDS epidemic, which began in the early 1980s. They also
indicated that there was no reason to suspect that these working populations would be at a higher
risk for AIDS than was the general population, the comparison group.
Wong and Trent (1993) also reported a slightly increased risk of cancer in other
lymphatic tissue (14 observed vs. 11.39 expected). In men, the risk was nonsignificantly higher
(11 observed vs. 5.78 expected). Forty-three lymphopoietic cancers were observed versus
42 expected. In men, the risk was higher (32 observed vs. 22.22 expected). Fourteen leukemia
deaths were noted versus 16.2 expected. The authors did not derive individual exposure
A-16
-------
estimates for exposure-response analysis as Stavner et al. (1993) did. Rather, they used duration
of employment as a surrogate for exposure.
This study has many of the same limitations as the Stavner et al. (1993) study. The
authors assumed that those individuals with an unknown vital status as of the cutoff date were
alive for the purposes of the analysis, and they were unable to obtain cause-of-death information
on 5% of the known deaths.
The differences between this cohort study and that of Stavner et al. (1993) are in the
methods of analysis. Stavner et al. (1993) used the 9th revision of the International Classification
of Diseases (ICD) to develop their site-specific cancer categories for comparison with expected
cancer mortality, whereas Wong and Trent (1993) used the 8th revision. This could account for
some of the differences in the observed numbers of site-specific cancers, because minor
differences in the coding of underlying cause of death could lead to a shifting of some unique
causes from one site-specific category to another. Furthermore, Wong and Trent (1993) did not
analyze separately the category "lymphoid" neoplasms, which includes lymphocytic leukemia
and NHL, whereas Stavner et al. (1993) did. Stavner etal. (1993) further developed cumulative
exposure information using exposure estimates, whereas Wong and Trent (1993) used length of
employment as their surrogate for exposure but did not code detailed employment histories.
Because Wong and Trent (1993) made no effort to quantify the exposures, as was the
case in Stavner etal. (1993), this study is less useful in determining a exposure-response
relationship. Furthermore, the assumption that a member of the cohort should be considered
alive if a death indication could not be found will potentially tend to bias risk ratios downward if
in fact, a large portion of this group is deceased. In this study all untraceable persons were
considered alive at the end of the follow-up; therefore, the impact of the additional person-years
of risk cannot be gauged.
A.2.10. Bisanti etal. (1993)
These authors reported on a cohort mortality study of 1,971 male chemical workers
licensed to handle EtO by the Italian government, whom they followed retrospectively from
1940 to 1984. Altogether, 76 deaths had occurred in this group by the end of the study period,
whereas 98.8 were expected. Of those, 43 were due to cancer versus 33 expected. The cause of
one death remained unknown, and 16 workers were lost to follow-up. A group of
637 individuals from this cohort was licensed to handle only EtO; the remaining 1,334 had
licenses valid for handling other toxic gases as well. Date of licensing for handling EtO became
the initiating point of exposure to EtO, although it is likely that some of these workers had been
exposed previously to EtO. The regional population ofLombardia was used as the reference
group from which comparison death rates were obtained.
A-17
-------
Although there were excess risks from almost all cancers, one of the greatest SMRs was
in the category known as "all hematopoietic cancers," where 6 observed deaths occurred when
only 2.4 were expected (SMR = 2.5). In the subgroup "lymphosarcoma, reticulosarcoma" there
were 4 observed deaths whereas only 0.6 were expected (SMR = 6.1, p <0 .05); the remaining
2 were leukemias. The authors note that five hematopoietic cancers occurred in the subgroup of
workers who were licensed to handle only EtO but no other chemicals versus only
0.7 hematopoietic cancers expected (SMR = 7.1, p<0 .05). These deaths occurred within
10 years from date of licensing (latent period), which is consistent with the shorter latent period
anticipated for this kind of cancer. According to the authors, all workers began their
employment in this industry when the levels of EtO were high, although no actual measurements
were available. The fact that this subgroup of workers was licensed only for handling EtO
reduces the likelihood of a confounding chemical influence.
The authors concluded that the excess risk of cancer of the lymphatic and hematopoietic
tissues in these particular EtO cohort members support the suggested hypothesis of a higher risk
of cancer found in earlier studies, but they added that the lack of exposure information on the
other industrial chemicals in the group that had a license for handling other toxic chemicals made
their findings inconclusive.
This study was of a healthy young cohort, and most person-years were contributed in the
latter years of observation. Many years of follow-up may be necessary in order to fully verily
any trend of excess risks for the site-specific cancers of interest and to measure latent effects.
Furthermore, the unusual deficit of total deaths versus expected contrasted with an excess of
cancer deaths versus expected raises a question about the potential for selection bias when the
members of this cohort were chosen for inclusion. Also, one of the study's major limitations is
the lack of exposure data.
A.2.11. Hagmar etal. (1995) and Hagmar et al. (1991)
Cancer incidence was studied in a cohort of 2,170 EtO-exposed workers from two plants
in Sweden that produced disposable medical equipment. To fit the definition for inclusion, the
subjects, 1,309 women and 861 men, had to have been employed for a minimum of 12 months
and some part of that employment had to have been during the period 1970-1985 in the case of
one plant and 1965-1985 in the case of the other. The risk ratios were not dichotomized by sex.
No records of anyone who left employment or died before January 1, 1972 in one plant and
January 1, 1975 in the other were included. Expected incidence rates were generated from the
Southern Swedish Regional Tumor Registries.
Because of a short follow-up period and the relative young age of the cohort, little
morbidity had occurred by the end of the cutoff" date of December 31, 1990. Altogether,
A-18
-------
40 cancers occurred, compared with 46.3 expected. After 10 years latency, 22 cases of cancers
were diagnosed versus 22.6 expected. However, 6 lymphohematopoietic cancers were observed
versus 3.37 expected, and when latency is considered, this figure falls to 3 versus 1.51 expected.
The authors pointed out that for leukemia the standard incidence ratio (SIR) is a nonsignificant
7.14, based on 2 cases in 930 subjects having at least 0.14 ppm-years of cumulative exposure to
EtO and a minimum of 10 years latency. The authors believed that the results provided some
minor evidence to support an association between exposure to EtO and an increased risk of
leukemia. However, for breast cancer, no increase in the risk was apparent for the total cohort
(SIR = 0.46; 5 cases). Even in the 10-years or more latency period, the risk was less than
expected (SIR = 0.36; 2 cases).
The authors made a reasonably good attempt to determine exposure levels during the
periods of employment in both plants for six job categories. Sterilizers in the years 1970-1972
were exposed to an average 40 ppm in both plants. These levels gradually dropped to 0.75 ppm
by 1985-1986. Packers and developmental engineers were the next highest exposed employees,
with levels in 1970-1972 of 20 to 35 ppm and by 1985-1986 of less than 0.2 ppm. During the
period 1964-1966 in the older plant, EtO levels averaged 75 ppm in sterilizers and 50 ppm in
packers. Peak exposures were estimated to have ranged from 500 to 1,000 ppm during the
unloading of autoclaves up to 1973. The levels gradually dropped to less than 0.2 ppm in both
plants by 1985-1986 in all job categories (developmental engineers, laboratory technicians,
repair men, store workers, controllers, foremen, and others) except sterilizers.
These exposure estimates were verified by measurement of hydroxy ethyl adducts to
N-terminal valine in hemoglobin in a sample of subjects from both plants. The adduct levels
reflect the average exposure during the few months prior to the measurement of EtO. The results
of this comparison were close except for sterilizers, whose air monitoring measurements were
2 to 3 times higher.
The authors pointed out two limitations in their study: a minority of subjects had a high
exposure to EtO, and the follow-up (median 11.8 years) resulted in relatively few person-years at
risk and was insufficient to assess the influence of a biologically relevant induction latency
period. Although this study has good exposure information and the authors used this information
to develop an exposure index per employee, they did not evaluate dose-response relationships
that might have been present, nor did they follow the cohort long enough to evaluate morbidity.
The strength of this study is the development of the cumulative exposure index as well as the
absence of any potential confounding produced by the chlorohydrin process, which was a
problem in workers who produced and manufactured EtO in other studies.
A follow-up study of this cohort conducted by Mikoczy et al. (2011) is discussed in
Section A.2.21 below.
A-19
-------
A.2.12. Norman etal. (1995)
These authors conducted a mortality/incidence study in a cohort of 1,132 workers, mainly
women (82%), who were exposed to EtO at some time during the period July 1, 1974, through
September 30, 1980. Follow-up was until December 31, 1987. Ethylene oxide was used at the
study plant to sterilize medical equipment and supplies that were assembled and packaged there.
This plant was selected for the study because in an earlier small study at this plant (Stolley etal.,
1984) there was an indication that in a sample of workers the average number of sister chromatid
exchanges was elevated over that of a control group selected from the nearby community.
Cancer morbidity was measured by comparing cancers occurring in this cohort with those
predicted from the National Cancer Institute's Surveillance, Epidemiology, and End Results
(SEER) Program for the period 1981-1985 and with average annual cancer incidence rates for
western New York for 1979-1984. Observed cancers were compared to expected cancers using
this method.
Only 28 cancer diagnoses were reported in the cohort; 12 were for breast cancers. Breast
cancer was the only cancer site in this study where the risk was significantly elevated, based on
the SEER rates (SIR = 2.55, p < 0.05). No significant excesses were seen at other cancer sites of
interest: leukemia (1 observed, 0.54 expected), brain (0 observed, 0.49 expected), pancreas
(2 observed, 0.51 expected) and stomach (0 observed, 0.42 expected). The authors offered no
explanation except chance as to why the risk of breast cancer was elevated in these workers.
In 1980, three 2-hour samples from the plant provided 8-hour TWA exposures to
sterilizer operators that ranged from 50 to 200 ppm. Corrective action reduced the levels to 5 to
20 ppm.
This study has little power to detect any significant risk of cancer at other sites because
morbidity was small, chiefly as a consequence of the short follow-up period. The mean number
of years from the beginning of follow-up to the end of the study was 11.4 years. In fact, the
authors stated that breast cancer was the only cancer site for which there was adequate power to
detect an increased relative risk. Additional weaknesses in this study include no historic
exposure information and too short a period of employment in some cases (<1 month) to result in
breast cancer. The authors maintained that their study was inconclusive.
A.2.13. Swaen et al. (1996)
A significant cluster of lOHodgkin lymphoma cases in the active white male workforce
of an unidentified large chemical manufacturing plant in Belgium led to a nested
case-control study by Swaen et al. (1996) to determine which, if any, chemical agents within the
plant may have led to the increase. By comparison with regional cancer incidence rates, the SIR
for this disease was 4.97 (95% confidence interval [CI] = 2.38-9.15) over a 23-year period, from
A-20
-------
1966 to 1992. This suggested that an occupational exposure may have produced the significant
excess risk of Hodgkin lymphoma seen in these workers.
The investigators randomly selected 200 individuals from a computerized sampling frame
of all men ever employed at the facility. From this list of 200, workers who were actively
employed at the time of diagnosis of each case were chosen as controls. No age matching was
done because the authors stated that age-specific incidence rates for Hodgkin lymphoma in the
United States were relatively flat for men between ages 18 and 65. The investigators felt that a
control could serve for more than one case.
Verification of the 10 cases revealed that 1 case was, in reality, a large-cell anaplastic
lymphoma. Two others could not be confirmed as Hodgkin lymphoma due to the lack of tissue.
The remaining seven were confirmed as Hodgkin lymphoma. In the ensuing case-control
analysis, significant odds ratios (ORs) for Hodgkin lymphoma were observed for five chemicals,
ammonia (6 cases, OR= 5.6), benzene (5 cases, OR= 11), EtO (3 cases, OR = 8.5), NaOH
(5 cases, OR= 8), and oleum (3 cases, OR= 6.9), based on the number of cases and controls
known to be exposed to the chemicals in question. This does not mean they were exposed only
to the chemical in question.
The availability of exposure information made it possible to calculate cumulative
exposure to the cases and controls of two chemicals, benzene and EtO. The cumulative exposure
for benzene-exposed cases was 397.4 ppm-months versus an expected 99.7 ppm-months for the
matched controls. This difference in cumulative exposures was not statistically significant;
although, the authors noted that one case had an exceptionally high cumulative benzene
exposure. Only a few studies have suggested that exposure to benzene could be related to an
increase in the risk of Hodgkin lymphoma. The cumulative total exposure to EtO for the cases
was 500.2 ppm-months versus 60.2 for the matched controls, which was statistically significant,
the significance being due to one extreme case.
This study is limited because the authors enumerated only cases among active employees
of the workforce; therefore, the distinct possibility exists that they could have missed potential
cases in the inactive workers. It is possible that latent Hodgkin lymphoma cases could have been
identified in the controls after the controls left active employment. However, given the many
different possible exposures to the chemicals produced in the workplaces of these employees, it
would be difficult to argue that either EtO or benzene could be considered solely responsible for
the excess risk of Hodgkin lymphoma in this working group.
A.2.14. Olsenetal. (1997)
Olsen et al. (1997) studied 1,361 male employees of four plants in Texas, Michigan, and
Louisiana who were employed a minimum of 1 month sometime during the period 1940 through
A-21
-------
1992 in the ethylene chlorohydrin and propylene chlorohydrin process areas. These areas were
located within the EtO and propylene oxide production plants. Some 300 deaths had occurred by
December 31, 1992.
Plant A in Texas produced EtO beginning in 1941 and ceased production in 1967.
Bis-chloroethyl ether, a byproduct of EtO continued to be produced at this plant until 1973. The
plant was demolished in 1974. Plant B, which was nearby, manufactured EtO from 1951 to 1971
and then again from 1975 until 1980. This plant continues to produce propylene oxide. The
Louisiana plant produced EtO and propylene oxide through the propylene chlorohydrin process
from 1959 until 1970, when it was converted to propylene oxide production. The Michigan plant
produced ethylene chlorohydrin and subsequently EtO beginning in 1936 and continuing into the
1950s. This plant produced propylene chlorohydrin and propylene oxide up to 1974.
The authors suggested that exposure to EtO was possible at the plants studied in this
report but that exposure was unlikely in the 278 chlorohydrin unit workers who were excluded
from the cohort studied by Teta et al. (1993). Unfortunately, no actual airborne measurements
were reported by Olsen etal. (1997), and thus only length of employment could be used as a
surrogate for exposure.
The SMR for all causes was 0.89 (300 observed). For total cancer the SMRwas 0.94
(75 observed, 79.7 expected). There were 10 lymphohematopoietic cancers versus 7.7 expected
(SMR =1.3). No significantly increased risks of any examined site-specific cancer (pancreatic,
lymphopoietic, hematopoietic, and leukemia) were noted even after a 25-year induction latency
period, although the SMR increased to 1.44 for lymphopoietic and hematopoietic cancer. When
only the ethylene chlorohydrin process was examined after 25 years latency, the SMR increased
to 1.94, based on six observed deaths.
The authors concluded that there was a weak, nonsignificant, positive association with
duration of employment for lymphopoietic and hematopoietic cancer with Poisson regression
modeling. They stated that the results of their study provide some assurance that this cohort of
ethylene chlorohydrin and propylene chlorohydrin workers has not experienced a significant
increased risk for pancreatic cancer and lymphopoietic and hematopoietic cancer. They believed
that this study contradicted the conclusions of Benson and Teta (1993) that ethylene dichloride,
perhaps in combination with chlorinated hydrocarbons, appeared to be the causal agent in the
increased risk of pancreatic cancer and hematopoietic cancer seen in that study. They pointed
out that ethylene dichloride is readily metabolized and rapidly eliminated from the body after
gavage or inhalation administration; therefore, they questioned whether experimental gavage
studies (NCI, 1978) are appropriate for studying the effects of ethylene dichloride in humans.
One study (Maltoni et al., 1980) found no evidence of tumor production in rats and mice
chronically exposed to ethylene dichloride vapor concentrations up to 150 ppm for 7 hours a day.
A-22
-------
Also, because this chemical is a precursor in the production of vinyl chloride monomer, the
authors wondered why an increase in these two site-specific cancers had not shown up in studies
of vinyl chloride workers. However, they believe that an additional 5 to 10 years of follow-up of
this cohort would be necessary to confirm the lack of risk for the two types of cancer described
above.
A.2.15. Steenland etal. (2004)
In an update of the earlier mortality studies of the same NIOSH cohort of workers
exposed to EtO described by Steenland etal. (1991) and Stayner et al. (1993), an additional
11 years of follow-up were added. This increased the number of deceased to 2,852. Work
history data were originally gathered in the mid-1980s. Approximately 25% of the cohort
continued working into the 1990s. Work histories on these individuals were extended to the last
date employed. It was assumed that these employees continued in the job they last held in the
1980s. Little difference was noted when cumulative exposure was calculated with and without
the extended work histories, chiefly because the exposure levels after the mid-1980s were very
low (see Section A.2.8 for a discussion of the NIOSH exposure assessment and Section D.5 of
Appendix D for further characterization of the NIOSH cohort). Again, no excess risk of
hematopoietic cancer was noted based on external rates. However, as in the earlier paper,
exposure-response analyses reported positive trends for hematopoietic cancers limited to males
(p = 0,02 for the log of cumulative exposure with a 15-year lag) using internal comparisons and
Cox regression analysis.1 (See Table A-2forthe categorical exposure results.)
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.
1 Valdez-Flores et al. C201CT) 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. (200^ did not estimate the lag using the likelihood; instead, they treated the
lagged exposure as an alternate exposure metric.
A-23
-------
Table A-2. Cox regression results for hematopoietic cancer
mortality (15-year lag) in males
Cumulative exposure (ppm-days)
Odds ratio (95% CI)
0
1
>0-1,199
1.23 (0.32-4.73)
1,200-3,679
2.52 (0.69-9.22)
3,680-13,499
3.13 (0.95-10.37)
13,500+
3.42 (1.09-10.73)
Source: Steenland et al. (2004Y
Table A-3. Cox regression results for lymphoid cell line
tumors (15-year lag) in males
Cumulative exposure (ppm-days)
Odds ratio (95% CI)
0
1
>0-1,199
0.9 (0.16-5.24)
1,200-3,679
2.89 (0.65-12.86)
3,680-13,499
2.74 (0.65-11.55)
13,500+
3.76 (1.03-13.64)
Source: Steenland et al. C2004\
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
in female exposure estimates to expect to be able to see a similar trend if it existed. In later
analyses conducted by Steenland and presented in Appendix D, the difference between the male
and female results was found not to be statistically significant, and the same pattern of
lymphohematopoietic cancer results observed for males by Steenland et al. (2004) was observed
for the males and females combined (i.e., statistically significant positive trends for both
hematopoietic and lymphoid cancers using log cumulative exposure and a 15-year lag).
A-24
-------
This study also reports a significant excess risk of breast cancer in the highest
cumulative-exposure quartile, with a 20-year lag (SMR = 2.07, 95% CI 1.1-3.54, n = 13) in
female employees. The results using internal Cox regression analyses with a 20-year lag time
produced an OR= 3.13 (95% CI 1.42-6.92) in the highest cumulative-exposure quartile. The
log of cumulative exposure with a 20-year lag was found to be the best model (p = 0.01) for the
analyses of breast cancer. As for hematopoietic cancer in males, cumulative exposure
untransformed showed a weaker trend (p = 0.16). Abreast cancer incidence study of this cohort
is discussed in Steenland etal. (2003).
A.2.16. Steenland etal. (2003)
In a companion study on breast cancer incidence in women employees of the same cohort
discussed in Steenland etal. (2004), the authors elaborated on the breast cancer findings in a
subgroup of 7,576 women from the cohort (76% of the original cohort). They had to be
employed at least 1 year and exposed while employed in commercial sterilization facilities. The
average length of exposure was 10.7 years. Breast cancer incidence analyses were based on
319 cases identified via interview, death certificates, and cancer registries in the full cohort,
including 20 in situ carcinomas. Interviews on 5,139 women (68% of the study cohort) were
obtained (next-of-kin interviews were sought for the 18% of the cohort who were deceased);
22% could not be located. Using external referent rates (SEER), the SIR was 0.87 for the entire
cohort based on a 15-year lag time. When in situ cases were excluded, the overall SIR increased
to 0.94. In the top quintile of cumulative exposure, with a 15-year lag time, the SIR was 1.27
(95% CI 0,94-1,69, n = 48), A significant positive linear trend of increasing risk with increasing
cumulative exposure was noted (p = 0.002) with a 15-year lag time. Breast cancer incidence was
believed to be underascertained owing to incomplete response and a lack of coverage by regional
cancer registries (68% were contacted directly and 50% worked in areas with cancer registries).
An internal nested case-control analysis, which is less subject to concerns about
underascertainment, produced a significant positive exposure-response with the log of
cumulative exposure and a 15-year lag time (p = 0.05). The top quintile was significant with an
OR of 1.74 (CI 1.16-2.65) based on all 319 cases (the entire cohort).
The authors also conducted separate analyses using the subcohort with interviews, for
which there was complete case ascertainment and additional information on potential
confounders. In the subcohort with interview data, the odds ratio for the top quintile equaled
1.87 (CI 1.12-3.1), based on 233 cases in the 5,139 women and controlled for with respect to
parity and breast cancer in a first-degree relative. Information on other risk factors was also
collected (e.g., body mass index, SES, diet, age at menopause, age at menarche, breast cancer in
A-25
-------
a first-degree relative, and parity), but only parity and breast cancer in a first-degree relative
were significant in the model. Continuous cumulative exposure, as well as the log cumulative
exposure, lagged 15 years, produced ^-values for the regression coefficient of 0.02 and 0.03,
respectively, for the Cox regression model, taking into account age, race, year of birth, parity,
and breast cancer in a first-degree relative.
The authors concluded that their data suggest that exposure to EtO is associated with
breast cancer, but because of inconsistencies in exposure-response trends and possible biases due
to nonresponse and incomplete cancer ascertainment, the case for breast cancer is not conclusive.
However, monotonically increasing trends in categorical exposure-response relationships are not
always the norm owing to lack of precision in the estimates of exposure. Furthermore, positive
trends were observed in both the full cohort and the subcohort with interviews, lessening
concerns about nonresponse bias and case underascertainment.
A.2.17. Kardos et al. (2003)
These authors reported on a study completed earlier by Muller and Bertok (1995) of
cancer among 299 female workers who were employed from 1976 to 1993 in a pediatric ward at
the county hospital in Eger, Hungary, where EtO gas sterilizers were used. Their observation
period for cancer was begun in 1987 on the assumption that cancer deaths before 1987 were not
due to EtO, based on a paper by Lucas and Teta (1996). Information about the Muller and
Bertok (1995) study is unavailable because the paper is in Hungarian and no translated copy is
available. Kardos and his colleagues evaluated mortality among these women and found a
statistically significant excess of total cancer deaths (n= 11) in the period from 1987 to 1999
when compared with expected deaths generated from three different comparison populations
(Hungary, n = 4.38; Heves County, n = 4.03; and city of Eger, n = 4.28). The SMRs are all
significant at thep < 0.01 level. Site-specific rates were not calculated. Among the 11 deaths
were 3 breast cancer deaths and 1 lymphoid leukemia death. The authors claim that their results
confirm "predictions of an increased cancer risk for the Eger hospital staff." They suggest an
etiological role for EtO in the excess risk. The observation of 3 breast cancer deaths, with at
most 4.4 (with Hungarian national rates as the referent) total cancer deaths expected, is indicative
of an increased risk of breast cancer.2
2Hungarian age-standardized female cancer mortality rates reported by the International Agency for Research on
Cancer (http://eu-cancer.iarc.fr/EUCAN/Countrv.aspx7ISOCountrv Cd=348) suggest that the ratio of breast cancer
deaths to total cancer deaths in Hungarian females is about 0.14 (23.5/100,000 breast cancer mortality rate versus
163.6/100,000 total cancer mortality rate). 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. (20031 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; nonetheless, the large difference between the ratios
(0.68 for the study versus 0.14 for the general population) indicates an increased risk of breast cancer in the study.
A-26
-------
A.2.18. Tompa etal. (1999)
The authors reported a cluster of eight breast cancer cases and eight other malignant
tumor cases that developed over a period of 12 years in 98 nurses who worked in a hospital in
the city ofEger, Hungary, and were exposed toEtO. These nurses were exposed for 5 to
15 years in a unit using gas sterilizer equipment. The authors report that EtO concentrations
were 5 to 150 mg/m3. The authors state that the high breast cancer incidence in the hospital in
Eger indicates a combined effect of exposure to EtO and naturally occurring radioactive tap
water, possibly due to the presence of radon. This case report study is discussed further in the
genotoxicity section.
A.2.19. Coggon et al. (2004)
Descriptive information about this cohort is available from the earlier study by Gardner et
al. (1989). In this update, the 1,864 men and 1,012 women described in the Gardner etal. (1989)
study were followed to December 31, 2000. This added 13 more years of follow-up resulting in
565 observed deaths versus 607.6 expected. For total cancer, the observed number of deaths
equaled 188 versus 184.2 expected. For NHL, 7 deaths were observed versus 4.8 expected. For
leukemia, 5 deaths were observed versus 4.6 expected. All 5 leukemia deaths fell into the subset
with definite or continual exposure to EtO, where only 2.6 were expected. In fact, the total
number of deaths classified to the lymphohematopoietic cancer category was 17 with 12.9
expected. This increased risk was not significant. When definite exposure was established, the
authors found that the risk of lymphatic and hematopoietic cancer was increased with 9 observed
deaths versus 4.9 expected. Deaths from leukemia were also increased in chemical workers with
4 leukemia deaths versus 1.7 expected. No increase was seen in the risk of hematopoietic cancer
in the hospital sterilizing unit workers, who are mostly female. Another finding of little
significance was that of cancer of the breast. Only 11 deaths were recorded in this cohort up to
the cutoff date versus 13.1 expected. Because there were no female workers in the chemical
industry, the results on breast cancer reflect only work in hospital sterilizing units. The
researchers concluded that the risk of cancer must be low at the levels sustained by workers in
Great Britain over the last 10 or 20 years.
A.2.20. Swaen et al. (2009) and Valdez-Flores etal. (2010)
Swaen et al. (2009) redefined and updated the cohort of 1,896 male UCC workers studied
by Teta et al. (1993), which was itself a follow-up of the 2,174 UCC workers originally studied
by Greenberg etal. (1990), excluding the 278 chlorohydrin unit workers because of potential
confounding. (However, confounding by chlorohydrin production has not been established, and
49 of those excluded workers were also employed in EtO production and thus had high potential
A-27
-------
EtO exposures.) Specifically, Swaen et al. (2009) extended the cohort enumeration period from
the end of 1978 to the end of 1988 (workers hired after 1988 were not added to the cohort
because they were considered to have no appreciable EtO exposure), identifying 167 additional
workers, and conducted mortality follow-up of the resulting cohort of 2,063 male workers
through 2003. Work histories were also extended through 1988; exposures after 1988 were
considered negligible compared to earlier exposure levels. Swaen et al. (2009) used an exposure
assessment reportedly based on the qualitative categorizations of potential for EtO exposure in
the different departments developed by Greenberg et al. (1990) and time-period exposure
estimates from Teta et al. (1993). The exposure assessment matrix for the exposure estimates of
Swaen et al. (2009) is presented in Table A-4 below. Cumulative exposures for the individual
workers were estimated by multiplying the time (in months) a worker was assigned to a
department by the estimated exposure level for the department and summing across the
assignments.
The exposure assessment used in this study was relatively crude, based on just a small
number of department-specific and time-period-specific categories, and with exposure estimates
for only a few of the categories derived from actual measurements. For the 1974-1988 time
period, based on measurements from environmental monitoring conducted in the (West Virginia)
plants since 1976, exposure estimates of 1 ppm and 0.3 ppm were chosen for the high- and
low-exposure-potential departments, respectively, and the average of 0.65 ppm was taken for the
medium-exposure-potential departments. For the 1957-1973 time period, exposure estimates
were based on measurements from an air-sampling survey of three EtO direct-oxidation
production units in aUCC plant in Texas in the early 1960s (during this 1957-1973 time period,
direct oxidation was the only method used for EtO production at the West Virginia plants as
well). The majority of the 8-hour TWA results in these units were between 3 and 20 ppm, with
levels between 5 and 10 ppm for operators. Because the West Virginia plants and equipment
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.
A-28
-------
Table A-4. Exposure assessment matrix from Swaen et al. (2009)—8-hour
TWA exposures in ppm
Time period
Exposure potential category
Low
(most EtO user departments)
Medium
(some EtO user departments)
High
(EtO production departments)
1925-1939
17
28
70
1940-1956
7
14
21
1957-1973
5
7.5
10
1974-1988
0.3
0.65
1
Source: Swaen et al. (2009Y
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 (one-half an order of
magnitude) was used between the low- and high-exposure-potential departments to obtain a
highly uncertain exposure estimate of 70 ppm for the high-exposure-potential departments.
Swaen et al. (2009) suggest that despite the high exposure estimates for the 1925-1939 time
period, the contribution of this time period to cumulative exposure estimates is limited because
only 98 workers (4.8% of the cohort) had employment histories before 1940. It appears, then,
that pre-1940 employment histories may have been missing for 13 of the workers, because
excluding the 112 pre-1940 chlorohydrin production workers (Benson and Teta, 1993) from the
original 223 pre-1940 workers (Greenberg et al., 1990) leaves 111 pre-1940 workers in the
cohort.
A-29
-------
At the end of the 2003 follow-up, 1,048 of the 2,063 workers had died and 23 were lost to
follow-up. In comparison with general population U.S. mortality rates, the all-cause mortality
SMR was 0.85 (95% CI = 0.80, 0.90) and the cancer SMR was 0.95 (95% CI = 0.84, 1.06).
None of the SMRs for specific cancer types showed any statistically significant increases. In
analyses stratified by hire date [pre-(inclusive) or post-1956], the SMR for leukemia was
elevated but not statistically significant (1.51; 95% CI 0.69, 2.87) in the early-hire group, based
on nine deaths. In analyses stratified by duration of employment, no trends were apparent for
any of the lymphohematopoietic cancers, although in the 9+ years of employment subgroup, the
SMR for NHL was nonsignificantly increased (1.49; 95% CI 0.48, 3.48), based on 5 deaths. In
SMR analyses stratified by cumulative exposure, no trends were apparent for any of the
lymphohematopoietic cancers, and there were no notable elevations for the highest cumulative
exposure category. Note that only 27 lymphohematopoietic cancer deaths (including
12 leukemias and 11 NHLs) were observed in the cohort.
Internal Cox proportional hazards modeling was also done for some disease categories
(all-cause mortality, leukemia mortality, and lymphoid cancer [NHL, lymphocytic leukemia, and
myeloma] mortality [17 deaths]), using cumulative exposure as the exposure metric. Year of
birth and year of hire were included as covariates in the Cox regression model, and the time
variable was presumably follow-up time. Year of hire was reportedly included to adjust for
potential cohort effects; however, adjusting for year of birth should have already adjusted for
cohort effects, and it is unclear whether year of hire was a statistically significant factor in the
regression. Furthermore, because year at hire is likely correlated with exposure (without being
correlated with disease trends over time, which would have been controlled for by year of birth),
including it in the regression model could overadjust and attenuate the observed exposure-related
effects. These internal analyses showed no evidence of an exposure-response relationship,
although, again, these analyses rely on small numbers of cases and a crude exposure assessment,
with a high potential for exposure misclassification.
Swaen et al. (2009) note that one of the strengths of their study is the long average
follow-up time of the workers. These authors further note that, because the UCC cohort is a
much older population (50% deceased) than the NIOSH cohort (Steenland et al., 2004), the
number of expected deaths is less than 3 times larger for the NIOSH cohort even though the
sample size is almost 9 times larger. However, the long follow-up and aged cohort might be a
limitation, as well. Because the follow-up is extended well beyond the time period of
nonnegligible exposures (pre-1989) for workers still employed and, especially, beyond the
highest exposures (e.g., pre-1940 or pre-1956), the follow-up is likely observing workers at the
high tail end of the distribution of latency times for EtO-associated lymphohematopoietic
cancers. In other words, workers that were at risk of developing lymphohematopoietic cancer as
A-30
-------
a result of their EtO exposures would likely have developed the disease earlier. Meanwhile,
having an older cohort means that the background rates of lymphohematopoietic cancers are
higher, and thus, relative risks may be attenuated. Such attenuation was observed even in the
younger NIOSH cohort between the 1987 follow-up (Steenland etal., 1991) and the 1998
follow-up (Steenland etaL 2004). when the follow-up was extended well beyond the period of
significant EtO exposures (exposure levels were considered very low by the mid-1980s).
Swaen et al. (2009) also note that their estimate of the average cumulative exposure for
the UCC cohort was more than twice the average cumulative exposure estimate for the NIOSH
cohort. However, there are substantial uncertainties in the exposure assessment, especially for
the early years when the highest exposures occurred. And despite the reported strengths of the
Swaen et al. (2009) study in terms of follow-up, cohort age, and high exposures, a limitation of
the study is the small cohort size. Based on data presented by Greenberg etal. (1990) and
Benson and Teta (1993), it appears that fewer than 900 workers were hired before 1956 (1,104 of
the original cohort were hired before 1960 and 233 of those were then excluded because they
worked in the chlorohydrin unit) and would have been potentially exposed to the higher pre-1956
exposures levels. Moreover, according to Teta et al. (1993), only 376 workers were assigned to
EtO production departments (but not the chlorohydrin unit), and these were the only departments
with high exposure potential (see Table A-4). In the lull cohort of 2,063 men, only 27
lympho hematopoietic (17 lymphoid) cancers were observed.
In alternate analyses of the UCC data, Valdez-Flores etal. (2010) fitted Cox proportional
hazards models and conducted categorical exposure-response analyses using a larger set of
cancer endpoints. These investigators also performed the same analyses using the data from the
last follow-up of the NIOSH cohort (Steenland etal., 2004) and from the two cohorts combined,
analyzing the sexes both separately and together. Valdez-Flores et al. (2010) reported that they
found no evidence of exposure-response relationships for cumulative exposure with either the
Cox model or categorical analyses for all of the cohort/endpoint data sets examined (endpoints
included all lymphohematopoietic cancers, lymphoid cancers, and female breast cancer, the latter
in the NIOSH cohort only). These investigators suggest that a review of the data from the
NIOSH and UCC studies supports combining them, but it should be recognized that the exposure
assessment conducted for the UCC cohort is much cruder (see above), especially for the highest
exposures, than the NIOSH exposure assessment (which was based on a validated regression
model; see A.2.8 above); thus, the results of exposure-response analyses of the combined cohort
data are considered to have greater uncertainty than those from analyses of the NIOSH cohort
alone, despite the additional cases contributed by the UCC cohort (e.g., the UCC cohort
contributes 17 cases of lympho id cancer to the 53 from the NIOSH cohort; however, as discussed
A-31
-------
above, some of these UCC cases occur in older workers, with longer postexposure follow-up,
and thus, may reflect background disease more than exposure-related disease).
Notable differences between the Steenland etal. (2004) and the Valdez-Flores etal.
(2010) analyses exist. A major difference is that Valdez-Flores et al. (2010) used only
cumulative exposure in the Cox regression model, so they considered only a sublinear
exposure-response relationship, whereas Steenland et al. (2004) also used log cumulative
exposure, which provides a supralinear exposure-response relationship model structure [e.g., see
Figure 4-1, illustrating the difference between the cumulative exposure and log cumulative
exposure Cox regression models (RR = e^ exP°sure) for the lymphoid cancers from Steenland et
al. (2004)1. Valdez-Flores et al. (2010) objected to the log cumulative exposure model for a
number of reasons, the primary one being that the use of log cumulative exposure forces the
exposure-response relationship to be supralinear regardless of the observed data. This is correct
but no different from the use of cumulative exposure imposing a sublinear exposure-response
relationship. Moreover, Steenland etal. (2004) used log cumulative exposure specifically when
the cumulative exposure Cox regression model did not yield a statistically significant fit to the
exposure-response data and the categorical analyses suggested increases in risk that were more
consistent with an underlying supralinear exposure-response relationship. With log cumulative
exposure, Steenland etal. (2004) observed statistically significant fits to the exposure-response
data for all lymphohematopoietic cancers in males, lymphoid cancers in males, and breast cancer
in females, none of which yielded statistically significant fits with the cumulative exposure
(sublinear exposure-response) model, supporting the apparent supralinearity of the data.3
Another key difference between the Steenland et al. (2004) and the Valdez-Flores etal.
(2010) analyses is that Valdez-Flores et al. (2010) present results only for unlagged analyses.
Valdez-Flores etal. (2010) state that their Cox regression results with different lag times were
similar to the unlagged results. Because the Valdez-Flores etal. (2010) categorical results are
for unlagged analyses, however, their referent groups are different from those used by Steenland
et al. (2004). Valdez-Flores et al. (2010) used the lowest exposure quintile (providing there were
sufficient data) as the referent group, whereas Steenland et al. (2004) used the no-exposure
(lagged-out) group as the referent. Because the NIOSH cohort data have an underlying
supralinear exposure-response relationship, the increased risk in the lowest exposure group is
already notably elevated and using the lowest exposure quintile as a referent group would
attenuate the relative risk. Nonetheless, Valdez-Flores et al. (2010) observed statistically
significant increases in response rates in the highest exposure quintile relative to the lowest
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).
A-32
-------
exposure quintile for lymphohematopoietic and lymphoid cancers in males in the NIOSH cohort,
consistent with the categorical results of Steenland etal. (2004). as well as a statistically
significant increase in the highest exposure quintile for lymphoid cancers in males and females
combined in the NIOSH cohort, consistent with the results in Appendix D.4
Although Valdez-Flores et al. (2010) found no statistically significant exposure-response
relationships for any of the cohort/endpoint data sets that they analyzed using the cumulative
exposure Cox regression model, these investigators derived risk estimates from the positive
relationships for the purposes of comparing those estimates with the U.S. Environmental
Protection Agency's (EPA's) 2006 draft risk estimates (U.S. EPA, 2006a). Valdez-Flores et al.
(2010) report that their estimate of the exposure level associated with 10 6 risk of
lymphohematopoietic cancer based on the male NIOSH cohort data is 1,500 times larger than the
EPA's 2006 draft estimate (their exposure level estimate based on the NIOSH and UCC male
and female data combined was a further 3 times higher). Most of the difference in magnitude
between the Valdez-Flores etal. (2010) and the EPA 2006 draft estimates is attributable to the
difference in the models used. The Valdez-Flores et al. (2010) estimate is based on the sublinear
Cox regression model, which the EPA rejected as not providing a good representation of the
low-exposure data (the EPA's 2006 draft risk estimate is based on a linear model). In addition,
Valdez-Flores etal. (2010)used maximum likelihood estimates, while the EPA uses upper
bounds on risk (or lower bounds on exposure). Valdez-Flores et al. (2010) also modeled down to
10 6 risk, whereas the EPA modeled to 10 2 risk and used the LECoi as a point of departure
(POD) for linear low-dose extrapolation. Valdez-Flores etal. (2010) suggest that PODs should
be within the range of observed exposures, and they chose a 10 6 risk level because the
corresponding exposure level was in the range of the observed occupational exposures
(converted to equivalent environmental exposures). The intention of the EPA's 2005 Guidelines
for Carcinogen Risk Assessment (U.S. EPA, 2005a), however, is for the POD to be (or more
specifically, to correspond to a response level) at the low end of the observable range of
responses (i.e., a response level that might reasonably be observed to have statistical significance
with respect to background responses). The underlying assumption in this approach is that one
can have relative confidence in an exposure-response model in the observable range, but there is
less confidence in any empirical exposure-response model for much lower exposures. The
estimates also differ because Valdez-Flores et al. (2010) truncated their life-table analysis at
70 years, while the EPA uses a cutoff" of 85 years.
4In Steenland's analyses of the NIOSH cohort data for both se?es combined, presented in Appendix D, the
categorical results for all lymphohematopoietic cancers were also statistically significantly increased.
A-33
-------
A further reason for differences between the risk estimates of Valdez-Flores et al. (2010)
and the EPA's 2006 draft result is that Valdez-Flores etal. (2010) estimated mortality risks,
while the EPA estimates incidence risks. In a separate publication, Sielken and Valdez-Flores
(2009a) disagree with the assumption of similar exposure-response relationships for
lymphohematopoietic cancer incidence and mortality used by the EPA in deriving incidence
estimates and assert that the methods used by the EPA in calculating these estimates were
inappropriate. Sielken and Valdez-Flores (2009a) suggest that, except at high exposure levels,
the exposure-response data on all lymphohematopoietic cancers in males in the NIOSH cohort
are consistent with decreases in survival time as an explanation for the apparent increases in
mortality. For two of the four exposure groups, however, the best fitting survival times were
0 years, which seems improbable. Moreover, Sielken and Valdez-Flores (2009a) have not
established that the excess mortality is due to decreased survival time; the data are also
consistent with increased mortality resulting from increased incidence. Furthermore, the rodent
bioassays show that EtO is a complete carcinogen (see Section 3.2), and the mechanistic data
demonstrate that EtO is mutagenic (see Section 3.3.3), with sufficient evidence for a mutagenic
mode of action (see Section 3.4). Thus, EtO can be expected to act as an initiator in
carcinogenesis, and, consequently, be capable of inducing exposure-related increases in
incidence. As for the methods used by the EPA in calculating the incidence estimates, the EPA
used adjustments to the life-table analysis where warranted (U.S. EPA 2006a). The EPA did not
adjust the all-cause mortality rates in the lymphohematopoietic cancer analyses, because "the
lymphohematopoietic cancer incidence rates are small when compared with the all-cause
mortality rates" (U.S. EPA 2006a); Section 4.1.1.3 (actually, the differential rates obtained by
subtracting the mortality rates from the incidence rates) and, thus, the impact of taking into
account lymphohematopoietic cancer incidence when calculating interval "survival" is
negligible, as confirmed by Sielken and Valdez-Flores' own calculations, presented in their
Table 2 where the "multiplier" = 1 (Sielken and Valdez-Flores, 2009a). On the other hand, for
the breast cancer incidence analyses, where incidence rates (and the differentials between
incidence and mortality rates) are higher, the EPA adjusted the all-cause mortality rates to take
into account breast cancer incidence, effectively redefining interval "survival" (and thus the
resulting population at risk) as surviving the interval without developing an incident case of
breast cancer IU.S. EPA (2006a); Section 4.1.2.3], Therefore, the concerns raised by Sielken
and Valdez-Flores (2009a) about using life-table analyses to derive incidence estimates do not
apply to the EPA's calculations.
Finally, the risk estimates of Valdez-Flores etal. (2010) and the EPA's 2006 draft also
differ because Valdez-Flores etal. (2010), based on analyses in a separate publication by Sielken
and Valdez-Flores (2009b), misinterpreted the application of the age-dependent adjustment
A-34
-------
factors (ADAFs) such that, even though they purported to apply the factors, this application had
no impact on the risk estimate. The ADAFs are default adjustment factors intended to be applied
directly to the unit risk estimates (i.e., risk per unit constant exposure, or "slope factors") in
conjunction with age-specific exposure level estimates (U.S. EPA. 2005b). For the purposes of
applying the ADAFs, the unit risk estimate is parsed, as a proportion of an assumed 70-year
lifespan, across age groups with different adjustment factors and/or exposure levels. The
ADAFs were not designed to be applied in life-table analyses, as was done by Sielken and
Valdez-Flores (2009b). In addition, the use of the 15-year lag in exposure in the life-table
analyses does not mean that there is no risk from exposures before age 15 years, as intimated by
Sielken and Valdez-Flores (2009b). Indeed, those exposures do not increase risk for cancer
occurring before 15 years of age; however, they do contribute to lifetime risk. The assumption
of increased early-life susceptibility that underlies the application of the ADAFs is that early-life
exposure increases the lifetime risk of cancer, not just the risk of cancer in early life, so it is
inappropriate to apply the ADAFs only to the age-specific hazard rates, as was done by Sielken
and Valdez-Flores (2009b). One might conceivably incorporate the ADAFs into the life-table
analysis by weighting the age-specific exposures before they are aggregated into the cumulative
exposure, but such an integrated approach does not allow for the risks associated with less-than-
lifetime exposure scenarios to be calculated without redoing the life-table analysis each time.
A.2.21. Mikoczy etal. (2011)
Mikoczy etal. (2011) report the results of a follow-up study of the Swedish sterilizer
worker cohort investigated by Hagmar et al. (Hagmar et al.. 1995; Hagmar etal.. 1991).5 This
update extends the follow-up period through 2006, providing an additional 16 years of follow-up
(see Section J.2.2 of Appendix J for more details and discussion of this study).
For lymphohematopoietic cancers, nonsignificant increases in SMRs and SIRs were
reported. For the incidence data, the internal analysis shows no exposure-related association,
although this analysis is relatively uninformative for these cancers, given the small number of
cases (five cases in each of the two highest exposure quartiles and seven cases in the referent
group of workers with cumulative exposures below the median), the generally low estimated
cumulative exposures, and the absence of an unexposed referent group.
For breast cancer mortality (results not shown), a "slight but nonsignificant decrease" in
the SMRwas reported. With a 15-year induction period included, the SMRfor breast cancer
was reportedly "somewhat increased." For workers with cumulative exposures above the
5 This follow-up study was published after the general cutoff date for literature inclusion in this assessment and is
reviewed in detail in Section J.2.2 of Appendix J. However, as it is a follow-up of an earlier study and as, with the
additional follow-up, it provides important corroborating evidence, the study is also briefly mentioned here.
A-35
-------
median, with a 15-year induction period, a "higher than expected" SMR, which was not
statistically significant, was reported.
For breast cancer incidence (41 cases), SIRs were nonsignificantly decreased, both with
and without a 15-year induction period. Internal analyses resulted in statistically significant
increases in the incidence rate ratios for the two highest cumulative exposure quartiles as
compared to the 50% of workers with cumulative exposures below the median (see Table J-3 in
Appendix J), despite having a low-exposed rather than an unexposed referent group.
In conclusion, the EPA found that the nonsignificant increases in SMRs and SIRs for
lymphohematopoietic cancers reported in this study are consistent with an increase in
lymphohematopoietic cancer risk, but overall, the study is underpowered for the analysis of
lymphohematopoietic cancers and contributes little to the weight of evidence for these cancers.
For breast cancer incidence, however, the statistically significant exposure-related increases in
breast cancer incidence in internal analyses add support to the findings of increased risk of
female breast cancer observed in the studies of NIOSH (Steenland etal., 2004; Steenland et al.,
2003), Norman etal. (1995), and Kardos et al. (2003).
A.3. SUMMARY
The initial human studies by Hogstedt and colleagues (Hogstedt, 1988; Hogstedt et al.,
1986; Hogstedt et al., 1979b; Hogstedt etal., 1979a), in which positive findings of leukemia and
blood-related cancers suggested a causal effect, have been followed by studies that either do not
indicate any increased risks of cancer or else suggest a dose-related increased risk of cancer at
certain sites, chiefly cancers of the lymphohematopoietic system including leukemia,
lymphosarcoma, reticulosarcoma, and NHL. More recently, an association with breast cancer
has also been suggested. However, the overall epidemiological evidence is not conclusive
because of inadequacies and limitations in the epidemiological database. The main effects and
limitations in the epidemiological studies of EtO are presented in Table A-5.
A-36
-------
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
Sterilizers,
production workers,
Sweden
Hoestedt(,1988,l:
709
(539 men,
170 women)
Plant 1: mean < 20 ppm
Plant 2: mean 6-28 ppm in
early years, less than 6 ppm
later
Plant 3: mean 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 nitrate
No personalexposure
information from which to
estimate dose
No latency analysis
Mhed exposure to other
chemicals
Hoestedt etal.
(1986)
Sterilizing workers
in 8 hospitals and
users in 4
companies, Great
Britain
Gardner et al.
2,876
(1,864 men,
1,012
women)
After 1977, means < 5 ppm
In earlier years, means
likely higher, and peak
exposures above the odor
threshold of 700 ppm were
reported.
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 esophagealcancer 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
Mhed exposure to several
other chemicals
(1989)
-------
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
Coeeon et al. (2004)
Same cohort
followed
additional
13 years
Same
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)
11 breast cancers vs. 13.1
expected
Same
Same, also no latency
evaluation
Update of Gardner
et al. (1989)
-------
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
Production workers
(methods
unspecified) from 8
chemical plants in
former West
Germany
Kiesselbach et al.
2,658 men
No exposure information
available
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 esophagealcancer deaths
vs. 1.5 expected
23 lung cancer deaths vs.
19.9 expected
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
Thiess et al. (19821
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
(1990)
-------
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
>
o
Population/
industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
limitations
Production workers
and users at
2 chemical plants in
West Virginia
Greenbera 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 (ICD NS)
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
-------
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
Same cohort as
Greenbere et al.
(1990) minus all
chlorohydrin-
exposed employees,
followed an
additional 10 years
Teta et al. CI9931
1,896 men
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
5 leukemia and aleukemia
deaths vs. 4.7 expected (ICD
NS)
2 lymphosarcoma and
reticulosarcoma vs.2.03
expected
7 lymphohematopoietic
cancers vs. 11.8 expected
Trend of increasing risk of
leukemia and aleukemia
death with increasing
duration of exposure
Same (except for chemicals specific
to the chlorohydrin process)
Same
Only the
chlorohydrin-
exposed employees
from Greenbere et
al. fl99CT) cohort,
followed an
additional 10 years
278 men
Reported to be low
exposure to EtO in the
chlorohydrin process
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
Same
Same, also very small
cohort
Benson and Teta
(1993)
1 lymphosarcoma and
reticulo sarcoma vs. 0.50
expected
8 pancreatic cancer deaths
vs. 1.63 expected (p < 0.05)
-------
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
>
to
Population/
industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
limitations
Same cohort as for
Teta et al. (1993)
2,063 men
Individual exposure
estimates derived from an
exposure matrix based on
potential EtO exposure
categorizations developed
bv Greenbere et al. CI 9901
and time-period exposure
estimates developed bv Teta
et al. (19931 which relied
11 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
Same
Same
Crude exposure
assessment, especially for
the early time periods
Small cohort; thus, small
numbers of specific
cancers even though long
follow-up time
followed an
additional 15 years
plus cohort
enumeration
extended to end of
1988 (an additional
10 years), adding
167 workers
Swaen et al. (2009)
on measurements taken at
other facilities and rough
estimates for the time
periods before 1974
-------
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
Sterilizers of
medical equipment
and spices; and
manufacturers and
testers of medical
sterilization
equipment, in
14 plants in the
United States
Steenland et al.
18,254
(45% male,
55% female)
1938-1976 (estimated): 16
ppmfor 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
36 lymphohematopoietic
cancer deaths vs. 33.8
expected (ICD NS)
13 leukemia andaleukemia
deaths vs. 13.5 expected
8 lymphosarcoma and
reticulosarcoma deaths vs.
5.3 expected
After 20+ years latency,
SMR = 1.76 for
lymphohematopoietic
cancer; significant trend with
increasing latency (p < 0.03)
Significantly increasing
lymphohematopoietic cancer
and "lymphoid" cancer
(ICD-9 200, 202, 204) rrsks
with cumulative exposure
(Cox regression model)
No identified exposures to other
chemicals
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
(1991); Stavneret
al. (1993)
-------
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
Same cohort as
Steenland et al.
18,728
(45% male,
55% female)
Same as Steenland et al.
(1991) and Stavner et al.
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)
No identifiable exposures to other
chemicals
All of the limitations of
Steenland et al. (1991)
(1991) andStavner
(1993)
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 Stavner et
al. (1993) on the basis of
individual cumulative
exposure data but only on
duration of employment
et al. (1993) plus
474 additional
members, followed
1 more yr
Wone and Trent
(1993)
-------
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
Steenland et al.
(•20041
Update of Steenland
et al. (19911 and
Stavneret al. (19931
18,254
(45% male,
55% female)
Same as Steenland et al.
(19911. 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)
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 (p < 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)
No identified exposures to other
chemicals
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
-------
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
Women employees
from Steenland et
al. (2004) employed
in commercial
sterilization
facilities for at least
1 yr
Steenland et al.
7,576 women
Same as in Steenland et al.
(•20041
Minimum of 1 yr
(20031
SIR = 0.87
319 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)
Same as in Steenland et al. (20041
Stavneret al. (1993)
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
-------
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
>
-j
Population/
industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
limitations
Chemical workers
licensed to handle
EtO and other toxic
chemicals, Italy
Bisanti et al. CI9931
1,971 men
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
43 total cancer deaths vs. 33
expected
6 lymphohematopoietic
cancer deaths vs. 2.4
expected (ICD-9 200-208)
4 lymphosarcoma 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
Toxic gases, dimethyl sulphate,
methylene chloride, carbon
disulphide, phosgene, chlorine,
alkalic cyanides, sulfur dioxide,
anhydrous ammonia, hydrocyanic
acid
Lack of exposure data
Insufficient follow-up for
this young cohort
Potential selection bias
Possible earlier exposure
than date of licensing
would indicate
-------
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
>
00
Population/
industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
limitations
Two plants that
produced disposable
medical
equipment, Sweden
Haemar et al.
CI9951: Haemar et
al. (19911
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
2 cases
5 breast cancer cases vs. 10.8
expected (ICD-7 170)
Fluorochlorocarbons, methyl formate
(1:1 mixture with EtO)
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
-------
Table A-5. Epidemiological studies of ethylene oxide and human cancer (continued)
>
vo
Population/
industry
Number of
subjects
Extent of exposure to
ethylene oxide
Health outcomes
Other chemicals to which subjects
were potentially exposed
limitations
Mikoczv et al.
(2011)
2,171
(862 men,
1,309
women)
Exposure levels as for
Haemar et al. (1995).
18 lymphohematopoietic
cancer cases vs. 14.4
expected (ICD-7 200-209)
9 NHL cases vs. 6.25
expected (ICD-7 200 + 202)
5 leukemia cases vs. 3.58
expected (ICD-7 204-205)
41 breast cancer cases vs.
50.9 expected (ICD-7 170)
In internal Poisson
regression analyses of breast
cancer, IRR = 2.76 (p <
0.05) in the 3rd exposure
quartile and 3.55 (p < 0.05)
in the highest exposure
quartile, both compared to
the 50% of workers with
cumulative exposures below
the median
Fluorochlorocarbons, methyl formate
(1:1 mixture with EtO)
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 cancer 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
Update of Haemar
et al. CI9951 and
Haemar et al.
(1991)
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
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 (p < 0.05)
2 pancreatic cancer cases vs.
0.51 expected
No otherchemical 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
-------
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
Nested case-control
study; cases and
controls from a
large chemical
production plant,
Belgium
Swaen et al. (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,
NaOH, 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
Four EtO
production plants in
3 states using the
chlorohydrin
process (both
ethylene and
propylene)
Ols en et al. (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 ofEtO 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
-------
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
Female workers
from pediatric clinic
of hospital in Eger,
Hungary
Kardos et al. (2003)
299 female
employees
EtO sterilizing units with
unknown elevated
concentrations
11 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
No identifiable exposures to other
chemicals
Underlying cause of death
provided on all 11 cases
but no expected deaths
available by cause
Possible exposure to radon,
which is common in the
region
ICD NS: ICD codes not specified.
-------
Exposure information, where available, indicates that levels of EtO probably were not
high in these study cohorts. If a causal relationship exists between exposure to EtO and cancer,
the reported EtO levels may have been too low to produce a significant finding. Exposures in the
earlier years (prior to 1970) in most of the companies, hospitals, and other facilities where EtO
was made or used are believed to have been in the range of 20 ppm, with excursions many times
higher, although few actual measurements are available during this period. (One exception is the
environmental study by Jovner (1964). who sampled airborne levels of EtO from 1960 to 1962 in
a Texas City facility owned by Union Carbide.)
Almost all actual measurements of EtO were taken in the 1970s and 1980s at most plants
and facilities in the United States and Europe, and levels had generally fallen to 5 ppm and
below. Some plants may have never sustained high levels of airborne EtO. Assuming that there
is a true risk of cancer associated with exposure to EtO, then the risk is not evident at the levels
that existed in these plants except under certain conditions, possibly due to a lack of sensitivity in
the available studies to detect associated cancers at low exposures.
The best evidence of an exposure-response relationship for lymphohematopoietic cancers
comes from the large, diverse NIOSH study of sterilizer workers (Steenland etal., 2004; Stayner
et al., 1993; Steenland etal., 1991). This study estimated cumulative exposure (i.e., total lifetime
occupational exposure to EtO) in every member of the cohort. The investigators estimated
exposures from the best available data on airborne levels of EtO throughout the history of the
plants and used a regression model to estimate exposures for jobs/time periods where no
measurements were available. This regression model predicted 85% of the variation in average
EtO exposure levels. An added advantage to this study, besides its diversity, size, and
comprehensive exposure assessment, is the absence of other known confounding exposures in
the plants, especially benzene.
In the follow-up of the NIOSH cohort, as in the earlier study, Steenland etal. (2004)
observed no overall excess of hematopoietic cancers (ICD-9 codes 200-208). In internal
analyses, however, they found a significant positive trend (p = 0.02) for hematopoietic cancers
for males only, using log cumulative exposure and a 15-year lag, based on 37 male cases. In the
Cox regression analysis using categorical cumulative exposure and a 15-year lag, a positive trend
was observed and the OR in the highest exposure quartile was statistically significant
(OR = 3.42; 95% CI 1.09-10.73). Similar results were obtained for the "lymphoid" category
(lymphocytic leukemia, NHL, and myeloma). No evidence of a relationship between EtO
exposure and hematopoietic cancers in females in this cohort was observed. In later analyses
conducted by Steenland and presented in Appendix D, the difference between the male and
female results was found not to be statistically significant, and the same pattern of
lymphohematopoietic cancer results observed for males by Steenland et al. (2004) was observed
A-52
-------
for the males and females combined (i.e., statistically significant positive trends for both
hematopoietic [// = 74] and lymphoid [// = 53] cancers using log cumulative exposure and a
15-year lag, as well as statistically significant ORs in the highest exposure quartile for both
hematopoietic and lymphoid cancers).
In the analysis by Swaen et al. (2009) of male UCC workers, the authors discussed the
development of the exposure assessment matrix used in combination with worker histories to
estimate cumulative exposures for each worker in West Virginia UCC cohort. The exposure
matrix was based on the qualitative categorization of potential EtO exposure in the different
departments developed by Greenberg etal. (1990) and the time-period exposure estimates from
Teta et al. (1993). Eight-hour TWA concentrations (ppm) were estimated over four time periods
(1925-1939, 1940-1956, 1957-1973, and 1974-1978) at the two facilities for three
exposure-potential categories (high-, medium-, and low-exposure departments). Average
exposures in the latter time period (1974-1978) were based on industrial hygiene monitoring
conducted at the locations where the study subjects worked. Estimates for the earlier time
periods were inferred from data on airborne exposure levels in "similar" manufacturing
operations during the time periods of interest. The estimates for the 1957-1973 time period were
inferred from measurements reported for the EtO production facility at Texas City studied by
Joyner (1964), and the estimates for the 1940-1956 time period were inferred from "rough"
estimates of exposure reported for the Swedish company described by Hogstedt et al. (1979a).
Exposures for the 1925-1939 time period were assumed to be greater than for the later time
periods, but the exposure estimates for this period are largely guesses.
This relatively crude exposure assessment formed the basis of the exposure-response
analyses of the UCC study described in Swaen et al. (2009). Swaen et al. (2009) conducted
SMR analyses for the UCC workers stratified into those hired before and after December 31,
1956; for three subgroups of employment duration; and for three subgroups of cumulative
exposure. These investigators also conducted Cox proportional hazards modeling for leukemia
mortality and lymphoid malignancy mortality. No statistically significant excesses in cancer risk
or positive trends were reported. Despite the long follow-up of the UCC cohort, its usefulness is
limited by its small size (e.g., a total of 27 lymphohematopoietic cancer deaths were observed).
Valdez-Flores etal. (2010)used the same exposure assessment to conduct further
exposure-response modeling of the UCC data. These authors used the Cox proportional hazards
model to model various cancer endpoints, using the UCC data, the NIOSH data (Steenland et al.,
2004), or the combined data from both cohorts. Using cumulative exposure as a continuous
variable, no statistically significant positive trends were observed from any of the analyses.
Unlike Steenland etal. (2004), Valdez-Flores et al. (2010) rejected the log cumulative exposure
model. Using cumulative exposure as a categorical variable, statistically significant increased
A-53
-------
risks in the highest exposure quintile were reported for all lymphohematopoietic cancers and for
lymphoid cancers in the NIOSH male workers, consistent with results reported by Steenland et
al. (2004). Statistically significant increased risks in the highest exposure quintile were also
reported for NHL in the NIOSH male workers and for lymphoid cancers and NHL in both sexes
combined in the NIOSH cohort.
The many different analyses of the UCC data are weakened by the reliance on the crude
exposure assessment. The NIOSH investigators, on the other hand, based their exposure
estimates on a comprehensive, validated regression model. Furthermore, the NIOSH cohort was
a much larger, more diversified group of workers who were exposed to fewer potential
confounders.
One other study that provides cumulative exposure estimates is the incidence study by
Hagmar and colleagues (Hagmar et al.. 1995; Hagmar et al.. 1991). The short follow-up period
and relative youthfulness of the cohort produced little morbidity by the end of the study,
although some support for an excess risk of leukemia and lymphohematopoietic cancer had
appeared. More recently, a follow-up of this cohort by Mikoczy etal. (2011) observed
nonsignificant increases in SMRs and SIRs for lymphohematopoietic cancers, consistent with an
increase in lymphohematopoietic cancer risk; however, overall, the study is still underpowered
for the analysis of lymphohematopoietic cancers (n = 18) and contributes little to the weight of
evidence for these cancers.
In a separate analysis of the NIOSH cohort by Wong and Trent (1993), duration of
exposure to EtO was used as a surrogate for exposure. These authors did not find any positive
exposure-response relationships. They did observe an elevated significant risk of "NHL" in
males (SMR = 2.47, p < 0.05), based on 16 deaths, which was not dose related or time related.
However, a deficit in females remained.
Increases in the risk of hematopoietic cancers are also suggested in several other studies
(Coggon et al„ 2004; Olsen et al„ 1997; Swaen et al„ 1996; Norman et al„ 1995; Bisanti et al„
1993; Gardner etal., 1989). However, in all these studies the deaths were few and the risk ratios
were mostly nonsignificant except at higher estimated exposures or after long observation
periods. The findings were not robust, and there were potentially confounding influences, such
as exposure to benzene and/or chlorohydrin derivatives.
In those plants with no detectable risks (Norman etal., 1995; Kiesselbach etal., 1990),
the cohorts were generally relatively youthful or had not been followed for a sufficient number
of years to observe any effects from exposure to EtO. In the study by Olsen et al. (1997),
although a slight increase in the risk of cancer of the lymphopoietic and hematopoietic system
was evident, the authors stated that their study provided some assurance that working in the
chlorohydrin process had not produced significantly increased risks for pancreatic cancer or
A-54
-------
lymphopoietic or hematopoietic cancer, thus contradicting the findings of Benson and Teta
(1993). This study lacks any measurement of airborne exposure to any of the chemicals
mentioned and the authors indicated that an additional 5 to 10 years of follow-up would be
needed to confirm the lack of a risk for the cancers described in their study.
Although the largest database pertaining to the cancer risks from EtO exposure is for
lymphohematopoietic cancers, described above, more recent evidence suggests that exposure to
EtO also increases the risk of breast cancer. The study by Norman et al. (1995) of women who
sterilized medical equipment observed a significant twofold elevated risk of breast cancer, based
on 12 cases. A study by Tompa et al. (1999) reported on a cluster of breast cancers occurring in
Hungarian hospital workers exposed to EtO. In another Hungarian study of female hospital
workers by Kardos et al. (2003), three breast cancers were noted out of 11 deaths reported by the
authors. Although expected breast cancer deaths were not reported, the total expected deaths
calculated was just slightly more than four, making this a significant finding for cancer in this
small cohort. The most recent follow-up (Mikoczy et al., 2011) of the Swedish cohort of
sterilizer workers originally studied by Hagmar et al. (Haemar et al., 1995; Haemar et al., 1991)
reported that the overall SMR and SIR for breast cancer were nonsignificantly decreased.
However, in internal exposure-response analyses, statistically significant increases were
observed in the incidence rate ratios in the highest two cumulative exposure quartiles compared
to the workers with cumulative exposures below the median.
The most compelling evidence on breast cancer comes from the NIOSH cohort. In the
latest update of this cohort (Steenland et al., 2004), no overall excess of breast cancer mortality
was observed in the female workers; however, a statistically significant SMR of 2.07 was
observed in the highest cumulative exposure quartile, with a 20-year lag. In internal Cox
regression analyses, a positive exposure-response (p = 0.01) was observed for log cumulative
exposure with a 20-year lag, based on 103 cases. Similar evidence of an excess risk of breast
cancer was reported in a breast cancer incidence study of a subgroup of 7,576 female workers
from the NIOSH cohort who were exposed for 1 year or longer (Steenland etal., 2003). A
significant (p = 0,002) linear trend in SIR was observed across cumulative exposure quintiles,
with a 15-year lag. In internal Cox regression analyses, there was a significant regression
coefficient with log cumulative exposure and a 15-year lag, based on 319 cases. Using
categorical cumulative exposure, the OR of 1.74 was statistically significant in the highest
exposure quintile. In a subcohort of 5,139 women with interviews, similar results were obtained
based on 233 cases, and the models for this subcohort were also able to take information on other
potential risk factors for breast cancer into account. Additionally, the coefficient for continuous
cumulative exposure was also significant (p = 0.02), with a 15-year lag.
A-55
-------
Two other studies with female employees in the defined cohorts reported no increased
risks of breast cancer due to exposure to EtO (Coggon et al.. 2004; Hagmar et al.. 1995;
Hogstedt 1988). However, these studies have much lower statistical power than the NIOSH
studies, as evidenced by the much lower numbers of breast cancer cases that they report. The
largest number of cases in these other studies is 11 cases in the Coggon etal. (2004) study.
Furthermore, none of these other studies conducted internal (or external) exposure-response
analyses, which are the analyses that provided the strongest evidence in the NIOSH studies and
the Mikoczy et al. (2011) study.
Although the strongest evidence of a cancer risk is with cancer of the hematopoietic
system and female breast cancer, there are indications that the risk of stomach cancer may have
been elevated in some studies (Teta et al., 1993; Kiesselbach et al., 1990; Hogstedt et al., 1986;
Hogstedt etal., 1979b); however, this increased risk attained significance only in the study by
Hogstedt etal. (1979b), with 9 observed versus 1.27 expected. Shore et al. (1993) reported that
this excess may have been because early workers at this plant "tasted" the chemical reaction
product to assess the result of the EtO synthesis. This reaction mix would have also contained
ethylene dichloride, a suspected carcinogen, and other chemicals. This increased risk of stomach
cancer was not supported by analyses of intensity or duration of exposure in the remaining
studies, except that Benson and Teta (1993) suggested that exposure to this chemical increased
the risk of pancreatic cancer and perhaps hematopoietic cancer but not stomach cancer.
A significant risk of pancreatic cancer first reported by Morgan etal. (1981) was also
reported by Greenberg etal. (1990) in their cohort of chemical workers, but only in those
workers assigned to the ethylene chlorohydrin production process, where the authors reported
that exposure to EtO was low. Benson and Teta (1993) attributed the increase in pancreatic
cancer seen in Greenberg et al. (1990) to exposure to ethylene dichloride in the chlorohydrin
process. However, Olsen et al. (1997) refuted this finding in their study. The pancreatic cancers
from the study by Morgan et al. (1981) also occurred in workers in a chlorohydrin process of
EtO production. The possibility that exposure to a byproduct chemical such as ethylene
dichloride may have produced the elevated risks of pancreatic cancer seen in these workers
cannot be ruled out.
A.4. CONCLUSIONS
Although several human studies have indicated the possibility of a carcinogenic effect
from exposure to EtO, especially for lymphohematopoietic cancers and female breast cancer, the
total weight of the epidemiologic evidence is not sufficient to support a causative determination.
The causality factors of temporality, coherence, and biological plausibility are satisfied. There is
also evidence of consistency in the human studies. When combined under the rubric
A-56
-------
"lymphohematopoietic cancers," this loosely defined combination of blood malignancies
produces a slightly elevated risk of cancer in most studies but not in all. Similarly, for breast
cancer, increased risks are observed in the most of the studies with females, except for two with
just a small number of cases. In addition, there is evidence of a biological gradient in the
significant exposure-response relationships seen in the large, high-quality Steenland et al. (2004)
study and in the Steenland etal. (2003) breast cancer incidence study and the Mikoczy et al.
(2011) breast cancer incidence results.
For lymphohematopoietic cancer, the best evidence of a carcinogenic effect produced by
exposure to EtO is found in the NIOSH cohort of workers exposed to EtO in 14 sterilizer plants
around the country (Steenland et al., 2004; Stayner etal., 1993; Steenland etal., 1991). A
positive trend in the risk of lymphohematopoietic and "lymphoid" neoplasms with increasing log
cumulative exposure to EtO with a 15-year lag is evident. But there are some limitations to
concluding that this is a causal relationship at this time. For example, there was alack of
dose-response relationship in females, although, as presented in Appendix D, later calculations
show that the difference in response between females and males is not statistically significant
and that significant increases are also observed with both sexes combined.
An elevated risk of lymphohematopoietic cancers from exposure to EtO is also apparent
in several other studies. In some of these studies, confounding exposure to other chemicals
produced in the chlorohydrin process concurrent with EtO may have been partially responsible
for the excess risks. In other studies, where the chlorohydrin process was not present, there are
no known confounding influences that would produce a positive risk of lymphohematopoietic
cancer. Overall, the evidence on lymphohematopoietic cancers in humans is considered to be
strong but not sufficient to support a causal association.
For breast cancer, the best evidence is again found in the NIOSH studies (Steenland et al„
2004; Steenland etal., 2003) discussed earlier, with some corroborating support from the
Norman et al. (1995), Kardos et al. (2003), and Mikoczy etal. (2011) studies of breast cancer in
women exposed to EtO. The risk of breast cancer was analyzed in two other studies (Coggon et
al., 2004; Hogstedt, 1988), and no increase in the risk of breast cancer was found; however, these
studies had far fewer cases to analyze, did not have individual exposure estimates, and relied on
external comparisons. The NIOSH studies (Steenland etal., 2004; Steenland etal., 2003), on the
other hand, used the largest cohort of women potentially exposed to EtO and clearly show
significantly increased risks of breast cancer incidence and mortality, based on internal
exposure-response analyses. The authors suggest that the case is not conclusive of a causal
association "due to inconsistencies in exposure-response trends and possible biases due to
nonresponse and an incomplete cancer ascertainment." While these are not decisive
limitations—exposure-response relationships are often not strictly monotonically increasing
A-57
-------
across finely dissected exposure categories, and the consistency of results between the lull cohort
(less nonresponse bias) and the subcohort with interviews (lull case ascertainment) alleviates
some of the concerns about those potential biases—the evidence for a causal association between
breast cancer and EtO exposure is less than conclusive at this time.
See Section 3.5 for a more detailed and comprehensive weight-of-evidence discussion.
A-58
-------
APPENDIX B. REFERENCES FOR FIGURE 3-3
The references in this list correspond to the additional data that were added to Figure 3-3
since the I ARC (1994b) genetic toxicity profile was published. See the Figure 3-3 legend for
details.
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. Mutat Res 328:31-47.
Hengstler, JG; Fuchs, J; Gebhard, S; et al. (1994) Glycolaldehyde causes DNA-protein
crosslinks: a new aspect of ethylene oxide genotoxicity. Mutat Res 304(2):229-234.
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.
Major, J; Jakab, MG; Tompa, A. (1999) The frequency of induced premature centromere
division in human populations occupationally exposed to genotoxic chemicals. Mutat
Res 445(2):241-249.
Nygren, J; Cedervall, B; Eriksson, S; et al. (1994) Induction of DNA strand breaks by ethylene
oxide in human diploid fibroblasts. Environ Mol Mutagen 24(3):161-167.
Oesch, F; Hengstler, JG; Arand, M; etal. (1995) Detection of primary DNA damage:
applicability to biomonitoring of genotoxic occupational exposure and in clinical therapy.
Pharmacogenetics 5 Spec No:S118-S122.
Ribeiro, LR; Salvadori, DM; Rios, AC; et al. (1994) Biological monitoring of workers
occupationally exposed to ethylene oxide. Mutat Res 313:81-87.
Sisk, SC; Pluta, LJ; Meyer, KG; et al. (1997) Assessment of the in vivo mutagenicity of ethylene
oxide in the tissues of B6C3F1 lacl transgenic mice following inhalation exposure.
Mutat Res 391(3):153-164.
Swenberg, JA; Ham, A; Koc, H; et al. (2000) DNA adducts: effects of low exposure to ethylene
oxide, vinyl chloride and butadiene. DNA Repair 464:77-86.
Tates, AD; vanDam, FJ; Natarajan, AT; et al. (1999) Measurement of HPRT mutations in splenic
lymphocytes and haemoglobin adducts in erythrocytes of Lewis rats exposed to ethylene
oxide. DNA Repair 431(2):397-415.
van Sittert, NJ; Boogaard, PJ; Natarajan, AT; et al. (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.
B-l
-------
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(2):124-135.
Vogel, EW; Nivard, MJ. (1998) Genotoxic effects of inhaled ethylene oxide, propylene oxide
and butylene oxide on germ cells: sensitivity of genetic endpoints in relation to dose and
repair status. Mutat Res 405(2):259-271.
Walker, VE; Sisk, SC; Upton, PB; etal. (1997) In vivo mutagenicity of ethylene oxide at the hprt
locus in T- lymphocytes of B6C3F1 lacl transgenic mice following inhalation exposure.
Mutat Res 3 92(3):211-222.
Walker, VE; Wu, KY; Upton, PB; et al. (2000) Biomarkers of exposure and effect as indicators
of potential carcinogenic risk arising from in vivo metabolism of ethylene to ethylene
oxide. Carcinogenesis 21 (9):1661—1669.
B-2
-------
APPENDIX C. GENOTOXICITY AND MUTAGENICITY OF ETHYLENE
OXIDE
A summary of the available genotoxicity and mutagenicity data for ethylene oxide (EtO)
is presented in Chapter 3 (see Section 3.3.3). This appendix provides further details on the
available genotoxicity and mutagenicity data and on some of the studies that are briefly
mentioned in Chapter 3. The genotoxic potential of EtO is a key component of the assessment of
its carcinogenicity. The relationship between genotoxicity/mutagenicity and carcinogenicity is
based on the observations that genetic alterations are observed in almost all cancers and that
many of these alterations have been shown to play an important role in carcinogenesis. Exposure
to EtO has been found to result in a number of genotoxic effects in laboratory animal studies and
in studies of humans exposed in occupational settings. In particular, EtO has been shown to alter
or damage genetic material in such a manner that the genetic alterations are transmissible during
cell division. Evidence of genotoxicity/mutagenicity provides strong mechanistic support for
potential carcinogenicity in humans (Waters etal., 1999).
Since the first report of EtO's role in inducing sex-linked recessive lethals in Drosophila
(Rapoport, 1948), numerous papers have been published on the mutagenicity of EtO in
biological systems, spanning a whole range of assay systems, from bacteriophage to higher
plants and animals (see Figure 3-3 in Chapter 3). EtO, being a mono-functional alkylating
agent, is DNA-reactive, capable of forming DNA adducts and inducing mutations at both the
chromosome and gene levels under appropriate conditions, as evidenced in numerous in vitro
and in vivo studies reviewed elsewhere (IARC, 2008; Kolman etal., 2002; Bolt, 2000; Natarajan
et al., 1995; Vogel and Natarajan, 1995; Dellarco etal., 1990; Kolman etal., 1986). In
prokaryotes (bacteria) and lower eukaryotes (yeasts and fungi), EtO induces DNA damage and
gene mutations and conversions. In mammalian cells, EtO induces DNA adducts, unscheduled
DNA synthesis, gene mutations, sister chromatid exchanges (SCEs), micronuclei, and
chromosomal aberrations (IARC, 2008; Bolt, 2000; Natarajan etal., 1995; Preston et al„ 1995;
Dellarco et al„ 1990; Walker etal., 1990; Ehrenberg and Hussain, 1981). The results of in vivo
studies on the genotoxicity of EtO following ingestion, inhalation, or injection have also been
consistently positive (IARC, 2008, 1994b). Furthermore, in vivo exposure to EtO-induced gene
mutations in the Hprt locus in mouse and rat splenic T-lymphocytes and SCEs in lymphocytes
from rabbits, rats, and monkeys, in bone marrow cells from mice and rats, and in rat spleen.
Increases in the frequency of gene mutation in the lung and bone marrow (LacI locus) (Recio et
al., 2004; Sisk etal., 1997) and in the Hprt locus in T-lymphocytes (Walker et al., 1997) in
transgenic mice exposed to EtO via inhalation have been observed at concentrations similar to
those in carcinogenesis bioassays (NTP, 1987). Furthermore, the frequency of Kras mutations
C-l
-------
was increased in lung and Harderian gland tumors from EtO-exposed mice compared with
spontaneous tumors from control mice, and the spectrum of Kras mutations in lung tumors
arising from EtO exposure was dramatically different from that found in spontaneous tumors
(Hong etal., 2007; NTP. 1987). Likewise, Hras and Trp53 mutations were more frequently
induced in mammary carcinomas from EtO-exposed mice, were more frequently concurrent, and
expressed different mutation profiles than mammary carcinomas from control mice (Houle et al..
2006; NTP. 1987). EtO has also induced heritable mutations or effects in germ cells in rodents
(Generoso et al., 1990; Lewis et al., 1986). In addition, significant increases in the frequency of
SCEs and chromosomal aberrations in peripheral blood lymphocytes have been consistently
reported in workers exposed to concentrations of EtO of greater than 5 ppm (TWA) RARC
(2008), and references therein]. Thus, there is consistent evidence from in vitro studies and in
vivo studies of laboratory animals and occupationally exposed humans that EtO interacts with
the genome. Based on these observations, exposure to EtO is considered to cause cancer through
a mutagenic mode of action (see Chapter 3, Section 3.4).
The following sections provide further details on different genotoxicity test results on the
mutagenic potential of EtO.
C.l. ADDUCTS
C.1.1. DNA Adducts
Covalent bonding of a chemical (direct-acting) or its electrophilic intermediates or
metabolites (indirect-acting chemicals following metabolic activation) with the nucleophilic sites
in DNA results in the formation of "DNA adducts," which represent the biologically effective
dose of the chemical agent in question. Alkylating agents, such as EtO, are direct-acting
chemical agents that can transfer alkyl groups (e.g., ethyl groups) to nucleophilic sites in DNA,
alkylating the nucleotide bases. Alkylating agents are classified as SNl-type or SN2-type
depending on the substitution nucleophilicity (Sn). The SNl-type chemicals follow first-order
kinetics (e.g., ethylnitrosourea and methylnitrosourea), while the SN2-type agents exhibit an
intermediate transition state (e.g., EtO and methyl methanesulfonate). EtO is a direct-acting Sn2
(substitution-nucleophilic-bimolecular)-type alkylating agent that forms adducts with cellular
macromolecules such as proteins (e.g., hemoglobin) and DNA. The reactivity of an alkylating
agent can be estimated by its Swain-Scott substrate constant (s1-value), which ranges from 0 to 1
(Warwick, 1963). Alkylating agents such as EtO and methyl methanesulfonate, which have high
.s-values (0.96 and >0.83, respectively), target the nucleophilic centers of ring nitrogens (e.g., N7
of guanine and N3 of adenine) in DNA, while agents such as ethylnitrosourea with a low .s-
values (0.26) target the less nucleophilic centers such as O6 of guanine. EtO has a high substrate
constant favoring efficient alkylation at N7 of guanine (Beranek, 1990; Golbere„ 1986; Warwick,
C-2
-------
1963). Due to the high nucleophilicity and steric availability of the N7 of guanine, EtO
predominantly forms the N7-(2-hydroxyethyl)guanine (N7-HEG) adduct, although minor 2-
hydroxyethyl adducts such as those forming at the O6 of guanine (06-HEG), the N1 (Nl-HEA),
N3 (N3-HEA), and N6 of adenine (N6-HEA), and the N3 of cytosine (N3-HEC), uracil (N3-
HEU) and thymine (N3-HET) are found in some instances (Segerback, 1994).6
Several methods have been developed since 1988 to detect EtO-induced DNA adducts in
vitro and in vivo. However, sensitivity and specificity of these methods have been a concern.
These methods include immunochemical assays, fluorescence techniques, high-performance
liquid chromatography (HPLC), gas chromatography/mass spectrometry (GC/MS),
32P-postlabeling and electrochemical detection, with varying sensitivities for detection of
EtO-DNA adducts (Marsden et al., 2009; Huang et al., 2008; Tompkins etal., 2008; Marsden et
al., 2007; Bolt etal., 1997; Leclercq et al., 1997; Kumar et al., 1995; Sahaetal., 1995; van Delft
et al., 1994; van Delft etal., 1993; Uziel etal., 1992; Bolt et al„ 1988). In the following
paragraphs, a brief summary of available methods is provided to aid in the discussion of the
DNA adduct data.
van Delft et al. (1993) developed monoclonal antibodies against the imidazole ring of
N7-alkyldeoxyguanosine, with the limits of detection being 5-10, 1-2, and 20 adducts per
106 nucleotides when used in the direct and competitive enzyme-linked immunosorbent assay
and in immunofluorescence microscopy, respectively. Later, the same authors developed an
immunoslot-blot assay with increased sensitivity that detected 0.34 N7-HEG adducts per 106
nucleotides (van Delft et al„ 1994). Kumar et al. (1995) developed a 32P-postlabeling method
using thin-layer chromatography (TLC) and HPLC, which detected 0.1-1.0 ftnol 7-alkylguanine
adducts in rats exposed to different alkenes. Despite occasional inefficient labeling and poor
recovery of adduct due to depurination, this method has potential for use in measuring human
exposure to alkenes or their corresponding epoxides, as well as the endogenously formed
7-alkylguanine adducts.
Bolt etal. (1997) developed a HPLC method involving derivatization with phenylglyoxal
and fluorescence detection, using 7-methylguanine as an internal standard, for measuring the
physiological background of the N7-HEG adduct in DNA isolated from human blood. Using
this method, the authors were able to detect N7-HEG levels in five individuals ranging between
2.1 and 5.8 pmol/mg DNA (mean 3.2). Furthermore, Leclercq etal. (1997) developed a method
based on DNA neutral thermal hydrolysis, adduct micro-concentration, and HPLC coupled to
6For simplicity, this assessment generally uses the nomenclature and abbreviations for the nucleobase adducts; these
are the same adducts encompassed in the larger deo?yribonucleoside adductforms. Thus, for example, N3-HEA is
used synonymously to refer to both the N3-(2-hydro?y ethy l)adenine and the N3-(2-hy droxy ethy l)-2' -
deo?yadenosine(TNT3-HEdA) adducts.
C-3
-------
single-ion monitoring electrospray mass spectrometry which has a detection limit of 1 fmol,
reportedly allowing the detection of approximately 3 adducts/108 normal nucleotides. Using this
method, Leclercq et al. (1997) detected a dose-response relationship for N7-HEG after exposing
calf thymus DNA and blood samples to various doses of EtO. Marsden et al. (2007) used a
highly sensitive LC-MS/MS assay with selected reaction monitoring that offers a limit of
detection of 0.1 fmol of N7-HEGto establish background levels of N7-HEG (1.1-3.5
adducts/108 nucleotides) in rat tissue. Huang et al. (2008) developed an isotope-dilution online
solid-phase extraction and liquid chromatography coupled with tandem mass spectrometry
method with reportedly excellent accuracy, sensitivity, and specificity to analyze N7-HEG in
urine samples of nonsmokers. This method also demonstrated high-throughput capacity for
detecting EtO-DNA adducts and may be particularly useful for future molecular epidemiology
studies of individuals with low-dose EtO exposure. Tompkins etal. (2008) used a
high-performance liquid chromatography/electrospray ionization tandem mass spectrometry and
reported ~8 N7-HEG adducts/108 nucleotides in the livers of control rats. This method was also
capable of detecting the less prevalent but potentially more biologically significant Nl-HEA,
06-HEG, N6-HEA, and N3-HEU adducts. However, these minor adducts were below the level
of detection in control rat tissue DNA.
Overall, the sensitivity of EtO adduct detection depends on the method used for analysis.
Hence, use of appropriate methods is important when analyzing for these adducts and will be
highlighted in the following discussion.
C.l.1.1. Detection of EtO Adducts in In Vitro and In Vivo Systems
Numerous studies have been conducted to investigate the formation of DNA adducts
following EtO exposure in a wide range of experimental models, including cell-free systems,
bacteria, fungi, Drosophila, and other laboratory animals, as well as in exposed human subjects.
The following discussion is a review of the available studies of exposure to EtO and DNA adduct
formation in in vitro systems, laboratory animals, and humans (Bovsen etal.. 2009; Pauwels and
Veulemans. 1998; Bolt et al.. 1988; van Sittert and de Jong. 1985).
C.l.1.2. In Vitro DNA Binding Studies
The capacity of EtO to bind to DNA and form DNA adducts has been documented in a
few in vitro studies. Segerback (1990) showed that [14C]-labeled EtO reacted in vitro with calf
thymus DNA to produce N7-HEG adduct as the predominant adduct, with relatively low
amounts of Oe-HEG and N3-HEA adducts. The levels of N3-HEA and Oe-HEG are 4.4 and
0.5%, respectively, of the N7-HEG levels. Thus, the ratio ofN7-HEG, N3-HEA, and Oe-HEG
produced in vitro was 200:8.8:1, respectively. In the same study, the in vitro reaction products of
C-4
-------
radiolabeled N-(2-hydroxyethyl)-N-nitrosourea (HOEtNU) with calf thymus DNA exhibited a
higher relative amount of 06-HEG, which was 63% of the N7-HEG formed. The difference in
reactivity towards the N7 and O6 positions in guanine by these two alkylating agents was
explained by the difference in their 5-values. EtO, with an 5-value of 0.9, has a greater relative
preference for reacting with N rather than O atoms than does HOEtNU, with an 5-value of 0.2.
In another study, Li et al. (1992) observed that EtO in aqueous solution incubated with
calf thymus DNA in vitro for 10 hours produced several 2-hydroxyethyl DNA adducts whose
relative yields (nmol/mg DNA) were in the descending order: N7-HEG (330) > N3-HEA (39) >
Nl-HEA (28), N6-HEA (6.2) > N3-HEC (3.1) > N3-HET (2.0) > N3-HEU (0.8). This in vitro
study did not detect the 06-HEG adduct.
More recently, Tompkins et al. (2009) treated pSP189 shuttle vector plasmid to a range of
EtO concentrations in water and reported that, of the five 2-hydroxyethyl DNA adducts
measurable using their LC-MS/MS analytical method, only the N7-HEG adduct was detectable
at EtO concentrations up to 2,000 |iM,7 At the 10 mM concentration, the level of N7-HEG
adducts was about 19 times higher than that of Nl-HEA adducts and about 1,000 times higher
than that of 06-HEG adducts. At 30 mM, N3-HEU adducts were detectable, but this adduct was
not quantifiable due to the lack of a suitable internal standard. Detection of the N3-HEU adduct
implies that the N3-HEC adduct is also formed, as the former is the hydrolytic deamination
product of the latter (Tompkins etal., 2009). No results for the N6-HEA adduct were reported.
(N3-HEA, N3-HEC, and N3-HET adducts are not measurable by their method.)
C.l.1.3. In Vivo Studies—Laboratory Animals
Several studies evaluated N7-HEG levels following one or a range of doses with repeated
exposures of EtO given by inhalation or intraperitoneal injection in laboratory animals.
Segerback (1983) showed that in male CBA mice exposed by inhalation to [14C]-labeled EtO
N7-HEG adducts are formed in spleen, testes, and liver with half-lives of 24, 20, and 12 hours,
respectively.
Walker et al. (1990) conducted a time-course study to investigate the formation and
persistence of N7-HEG adducts in various tissues (e.g., brain, kidney, liver, spleen, lung, and
kidney) of male Fischer 344 rats exposed to one high dose of 300 ppm EtO by inhalation for
4 consecutive weeks (6 hours/day, 5 days/week) and sacrificed 1-10 days after the end of
exposure. The N7-HEG adduct was detectable in both target (brain, spleen, and white blood
cells) and nontarget (kidney, liver, lung, and testis) tissues with maximum levels (1.5 times
7The 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 06-HEG; andO.082/106 nucleotides for
N3-HEU (Tompkins et al., 2009).
C-5
-------
control levels) seen in brain compared to other tissues 1 day after exposure. The similarities in
N7-HEG levels in various tissues are possibly due to efficient pulmonary uptake of EtO and
rapid distribution by the circulatory system. The N7-HEG adduct levels increased linearly for 3-
5 days followed by a slow removal from DNA with an apparent half-life of 7 days, suggesting
that the adduct was probably removed by spontaneous depurination. The calculated in vivo
half-life for N7-HEG formed by EtO confirms the persistence of this adduct and is consistent
with another study in rats exposed to another alkylating agent, N-nitrosomethyl-(2-
hydroxyethyl)amine (Koepke et al.. 1988). Walker et al. (1990) suggested that the similarity in
N7-HEG formation in the target as well as nontarget tissues could also be due to factors such as
cell replication, location of the adducts in the genome, and tissue susceptibility genes, which
might be critical determinants quantitatively affecting tissue-specific and/or dose-response
relationships.
Using fluorescence-coupled HPLC, Walker et al. (1992) measured N7-HEG levels in
DNA of target and nontarget tissues from male B6C3Fi mice and F344 rats exposed to 0, 3, 10,
33, 100, or 300 (rats only) ppm EtO by inhalation for 4 weeks (6 hours/day, 5 days/week).
Another group of mice was exposed to 100 ppm EtO for 1, 3, 7, 14, or 28 days (5 days/week).
The authors reported linear dose-response relationships for N7-HEG in rat tissues following EtO
exposures between 10 and 100 ppm, with the slope increasing for exposures above 100 ppm. In
mice, only exposures to 100 ppm EtO resulted in significant increase in N7-HEG levels. Walker
et al. (1992) observed N7-HEG adduct levels of 2-6 pmols/mg DNA in control mice and rats,
while in mice exposed to 100 ppm EtO, N7-HEG levels ranged from 17.5 ±3.0 (testis) to
32.9 ± 1.9 (lung) pmol/mg DNA after 4 weeks of exposure. Rats and mice concurrently exposed
to 100 ppm EtO for 4 weeks showed two- to threefold lower N7-HEG levels in all tissues of
mice compared to rats, suggesting species differences in the susceptibility to EtO-induced
genotoxicity. The half-life of N7-HEG in mouse kidney DNA was 6.9 days, and in rat brain and
lung 5.4-5.8 days. The half-lives of N7-HEG adducts in DNA from other tissues of mouse and
rat were 1.0-2.3 days and 2.9-4.8 days, respectively. The authors suggested that the slow linear
removal of N7-HEG adducts from the DNA was mainly due to chemical depurination, while the
rapid removal was due to loss by depurination and DNA repair. Rats exposed to 300 ppm EtO
showed 06-HEG adducts at a steady-state concentration of ~1 pmol/mg DNA. Based on the
results from rats and mice, the authors suggested that DNA repair was saturated at the
concentration of EtO used in the time-course studies and that repeated exposures to lower
concentrations of EtO should lead to species- and tissue-specific differences in the levels of
N7-HEG (Walker et al. 1992).
Wu et al. (1999a) analyzed DNA from liver, brain, lung, and spleen of B6C3Fi mice and
F344 rats for N7-HEG adducts after exposure to EtO (0, 3, 10, 33, or 100 ppm) for 4 weeks
C-6
-------
(6 h/day, 5 days/week). The authors observed tissue- and species-specific dose-response
relationships of N7-HEG adducts in the EtO-exposed animals. Mice showed linear
dose-response relationships for N7-HEG adducts in liver, brain, and spleen at exposures between
3 and 100 ppm, and sub linear responses in lung between 33 and 100 ppmEtO exposure. Rats
showed linear increases in adduct levels in liver and spleen DNA between 3 and 100 ppm EtO,
and sub linear responses in the brain and lung between 33 and 100 ppmEtO exposure. Overall,
rats and mice exposed to 3 ppm EtO showed 5.3- to 12.5- and 1.3- to 2.5-fold higher N7-HEG
adducts, respectively, compared to the corresponding unexposed control animals. Thus, results
from this study suggest species differences, with rats being more susceptible to adduct formation
than mice, at lower levels of EtO exposure. This study also showed a clear difference in
N7-HEG levels between unexposed and exposed mice at these lower exposure levels, unlike the
study of Walker et al. (1992) discussed above. This difference is possibly due to the use of a
highly sensitive gas chromatography high-resolution mass spectrometry assay in the Wu et al.
(1999a) study.
van Sittert et al. (2000) exposed Lewis rats to 50, 100, and 200 ppm EtO by inhalation
(4 weeks, 5 days/week, 6 h/day) and measured N7-HEG adducts 5, 21, 35, and 49 days after
cessation of exposure. The authors used mass spectrometry following neutral thermal hydrolysis
of DNA to release the N7-HEG adducts and observed a clear exposure-response relationship
across the control and EtO-exposed rats. The mean levels of liver N7-HEG immediately after
cessation of exposure to 50, 100, and 200 ppm were estimated by extrapolation to be 310, 558,
and 1,202 adducts/108 nucleotides, respectively, while the mean level in control rats was
2.6 adducts/108 nucleotides. By 49 days postexposure, N7-HEG adducts had returned to near
background levels. The N7-HEG levels in liver DNA showed a linear response between 0 and
200 ppm EtO, suggesting that detoxification and DNA repair processes were not saturated up to
the highest exposure level tested. The authors observed statistically significant linear
relationships between mean N7-HEG levels at "day 0" postexposure and (1) Hprt mutant
frequencies at expression times of 21/22 and 49/50 days postexposure, (2) SCEs at 5 days
postexposure, or (3) high-frequency cells measured 5 days postexposure. The authors also
observed that SCEs and high-frequency cells continued to be present at 21-days postexposure
and significantly correlated with N7-HEG adducts at that time. However, induction of
micronuclei, chromosome breaks, or translocations did not show a dose-response relationship.
Nivard et al. (2003) showed that in male Drosophila, EtO exposure (2-1,000 ppm) by
inhalation for 24 hours induced a linear dose-response relationship for N7-HEG adduct
formation (0.15 to 105.4 adducts/106 nucleotides) over the entire dose range, as detected by
32P-postlabeling assay. The N7-HEG adducts were undetectable in controls (i.e., below the
detection limit of 1 adduct/108 nucleotides).
C-7
-------
A study by Rusvn et al. (2005) tested the hypothesis that EtO exposure results in an
accumulation of apurinic/apyrimidinic (AP) sites in DNA and induces changes in expression of
genes involved in DNA base excision repair (BER). The authors exposed male F344 rats by
inhalation to 100 ppm EtO or ethylene (40 or 3,000 ppm) for 1, 3, or 20 days (6 h/day,
5 days/week) and sacrificed them 2, 6, 24, or 72 hours after a single-day exposure. Brain and
spleen were considered as target sites for EtO-induced carcinogenesis, and liver as a nontarget
organ. Rusvn et al. (2005) observed a time-dependent increase in N7-HEG in brain and spleen
(target organs) and liver (nontarget organ) and in N-(2-hydroxyethyl)valine (HEVal) adducts in
hemoglobin. However, they could not detect any increase in AP sites in control or EtO-exposed
rats for any given duration or dose of exposure. Rats exposed to EtO for 1 day showed a
threefold to sevenfold decrease in expression of the DNA repair enzyme 3-methyladenine-DNA
glycosylase in the brain and spleen, while rats exposed to EtO for 20 days showed increased
expression of hepatic 8-oxoguanine DNA glycosylase, 3-methyladenine-DNA glycosylase, AP
endonuclease, polymerase beta, and alkylguanine methyltransferase by 20-100%. Levels of
brain AP endonuclease and polymerase beta were increased by <20% only in rats exposed to
3,000 ppm ethylene for 20 days. Results from this study suggest that EtO-induced DNA damage
is repaired without accumulation of AP sites or involvement of the BER pathway in target
organs. The authors concluded that accumulation of AP sites is not likely a primary mechanism
for mutagenicity and carcinogenicity of EtO, and further suggested that minor DNA adducts such
as 06-HEG or Nl-HEA are likely to be involved in mutagenicity. In fact, in a previous study
from the same group (Walker et al„ 1992), steady-state concentrations of Oe-HEG were reported
after 4 weeks of exposure with 300 ppm EtO, a finding which warrants further investigation.8
Marsden et al. (2007) have shown that intraperitoneal administration of a single or three
daily doses of EtO (0.01-1.0 mg/kg) induced dose-related increases in N7-HEG adduct levels in
male F344 rats, except at the lowest dose (0.01 mg/kg), where N7-HEG levels were similar to
endogenous levels detected in control animals. Further, they observed that N7-HEG adducts did
not accumulate in rats given three daily doses of EtO.
8 In a study published after the cutoff date for literature inclusion and described in more detail in Section J.4.1 of
Appendix J, Zhang et al. (2015b) exposed male B6C3Fi mice to 0, 100, or 200 ppm EtO for 6 hours/day,
5 days/week, for 12 weeks and examined the lungs for DNA adducts using more sensitive techniques than those
used by Walker etal. (19921. The Zhang et al. (2015b') study supports the identification of the 06-HEG adduct as a
direct product of EtO reactivity and adds coherence to the available database by observing an exposure-related
increase in lung 06-HEG levels at lower concentrations than previously evaluated (i.e., 100-200 ppm vs. 300 ppm),
quantification in another rodent species (i.e., mice vs. rats), and even detection in the majority of unexposed lung
samples (3/5), suggesting that endogenous EtO may be responsible for a low background level of this potentially
mutagenic DNA adduct. Significant increases in other potentially mutagenic purine adducts (e.g., Nl-HEA and N6-
HEA) were also observed.
C-8
-------
More recently, using a dual-isotope approach combining HPLC-accelerated mass
spectrometry with LC-MS/MS analysis, Marsden et al. (2009) observed linear dose-response
relationships for [14C]N7-HEG adducts (0.002 to 4 adducts/108 nucleotides) in spleen, liver, and
stomach DNA of F344 rats after exposure to low, occupationally relevant concentrations of
[14C]EtO (0, 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, and 0.1 mg/kg daily for 3 consecutive
days, with the rats killed 4 h after the last exposure). These results suggest that by using a highly
sensitive assay, it is possible to measure the N7-HEG adducts resulting from low EtO exposures
above the background adduct levels.
Otteneder and Lutz (1999) reviewed the quantitative relationship between DNA adduct
levels and tumor incidence in rodents that received repeated administration of EtO. The authors
observed a correlation with tumor incidence when the DNA adduct levels measured at a given
dose were normalized to the TD50 dose (the dose which results in 50% tumor incidence in a
two-year study). The calculated adduct level in mice associated with the hepatocellular TD50
was 812 N7-HEG adducts/108 normal nucleotides.
C.l.1.4. In Vivo Studies—Humans
A few studies have examined the effect of EtO exposure on humans, particularly in
occupational settings, and these have been comprehensively reviewed by Kolman etal. (2002).
In that review, the authors examined the use of hemoglobin and DNA adducts as biomarkers of
EtO exposure and the roles of genetic polymorphisms and confounding factors. Kolman etal.
(2002) also described the genotoxic effects of EtO in mammalian cells and summarized the
genotoxic and carcinogenic effects of EtO in humans. Some of the relevant studies in humans
are briefly discussed below.
An immunoslot blot assay was used to analyze N7-HEG levels in white blood cell DNA
from individuals exposed to EtO (2-5 ppm) and from controls (van Delft et al.. 1994). The
authors reported 0.1 and 0.065 N7-HEG adducts/106 nucleotides, respectively, in EtO-exposed
individuals (n = 42) and controls (// = 29) by this method. However, these differences were not
statistically significant.
In a study involving 58 sterilizer operators exposed to low and high levels of EtO (<32
and >32 ppm-hour, respectively) and 6 nonexposed controls from different hospitals, Yong etal.
(2007) examined N7-HEG adducts in granulocyte DNA. During the 4-month study, the
cumulative exposure to EtO (ppm-hour) was estimated before the blood sample collection. After
adjusting for cigarette smoking and other potential confounders, the mean N7-HEG adduct levels
in the nonexposed, low-, and high-exposure groups were 3.8, 16.3, and
20.3 adducts/107 nucleotides, respectively, with considerable interindividual variation (range:
1.6-241.3 adducts/107 nucleotides). However, these differences in mean adduct level were not
C-9
-------
statistically significant. The large variability across workers may reflect differences in their
recent exposure patterns because granulocytes have a lifespan of less than a day. Also, the study
did not find a significant correlation between the levels of N7-HEG adducts and HEVal adducts.
Mayer etal. (1991) observed an apparent suppression of DNA repair capacity in
EtO-exposed individuals as measured by the DNA repair index, that is, the ratio of unscheduled
DN A synthesis and N-acetoxy-2-acetylaminofluorene-DNA binding, (p < 0.01). In this study,
34 sterilization unit workers of a large university hospital and 23 controls working in the
university library were used. Overall, this study demonstrates significant correlations between
EtO-induced hemoglobin adduct levels and SCEs and the number of high frequency cells, at low
levels of EtO exposure (<1 ppm), independent of smoking history.
C.l.1.5. DNA Adducts—Summary
In summary, EtO predominantly forms N7-HEG adducts. Minor adducts are 06-HEG
adducts and reaction products with Nl, N3, and N6 of adenine and with N3 of cytosine, uracil
and thymine in vitro. However, the minor adducts are not observed to the same extent in vivo,
which may reflect a limitation in the sensitivity of the adduct assays available to date. Repeated
inhalation exposure of EtO induces N7-HEG adducts in both target organs (brain, spleen, and
white blood cells) and nontarget organs (kidney, liver, and lung) in rodents, with an apparent
half-life of 3-6 days in rats and 1-3 days in mice (Walker et al.. 1992). The dose-response
relationship of N7-HEG and EtO exposure is influenced by the analytical method used, which
also affects the background (endogenous) levels of adducts observed in unexposed rodents.
Steady-state levels of 06-HEG adducts (1 pmol/mg DNA) are detected in rats exposed by
inhalation to high doses of EtO (300 ppm) which are -250-300 times lower than the N7-HEG
levels (Walker etal.. 1992). Although N7-HEG adducts are likely to be removed by
depurination forming apurinic/apyrimidinic (AP) sites, Rusvn etal. (2005) showed that DNA
damage induced by exposure to EtO is repaired without accumulation of AP sites and without
affecting base excision repair (BER) in target organs of Fischer rats. Only two studies are
available on EtO-induced DNA adducts in human populations. Although higher levels of
N7-HEG DNA adducts were observed in human white blood cells (van Delft et al.. 1994) and
granulocytes (Yong et al.. 2007) of exposed cases compared to controls, these differences were
not statistically significant, possibly due to high interindividual variability.
C.1.2. EtO-Hemoglobin Adducts
Several studies have shown that EtO-induced hemoglobin adducts (e.g., HEVal) are good
biomarkers of exposure for this compound in human studies and that predicted hemoglobin
adduct levels resulting from exposure to ethylene or EtO are in agreement with measured values
C-10
-------
(Boogaard. 2002; Yong etal.. 2001; Fennell et al.. 2000; Tates et al.. 1999; Walker et al.. 1992;
Britton et al.. 1991). Csanadv et al. (2000) found a good agreement between the predicted and
measured hemoglobin adduct levels in humans. However, in rodents, hemoglobin adducts were
under-predicted by a factor of 2 to 3, while DNA adduct levels were comparable, suggesting
inconsistencies between the two biomarkers. Walker et al. (1993) also observed that the
relationships between HEVal and N7-HEG concentrations varied with length of exposure,
interval since exposure, species, and tissue, which may be due to differences in formation,
persistence, repair, and chemical depurination of the DNA adduct. Thus, Walker etal. (1993)
suggested that HEVal adducts do not provide accurate prediction of DNA adducts in specific
tissues of humans under actual exposure conditions. In summary, HEVal adducts do not appear
to be predictable markers for DNA adducts.
C.2. GENE MUTATIONS
EtO has consistently yielded positive results, at both the gene and chromosome levels, in
a broad range of in vitro and in vivo mutational assays, including those performed in bacteria,
fungi, yeast, insects, plants, Drosophila, and rodents, in both repair-deficient and proficient
organisms, and in mammalian cell cultures, including cells from humans [reviewed in (IARC.
2008; Kolman et al.. 2002; Bolt. 2000; Nataraian et al.. 1995; Vogel and Nataraiaa 1995; IARC.
1994b; Dellarco etal.. 1990)1. The results of in vivo studies on the mutagenicity of EtO have
also been consistently positive following ingestion, inhalation, or injection [e.g., Tates et al.
(1999)1. Increases in the frequency of gene mutations in the lung (LacI locus) (Sisk et al.. 1997).
in T-lymphocytes (Hprt locus) (Walker etal.. 1997). and bone marrow and testes in B6C3Fi
Lac I transgenic mice (Recio etal.. 2004) have been observed in mice exposed to EtO via
inhalation at concentrations similar to those used in the carcinogenesis bioassays (NTP. 1987).
clearly documenting that EtO is a DNA-reactive mutagenic agent. Furthermore, occupational
studies provide evidence for the genotoxic potential of EtO.
C.2.1. Bacterial Systems
Studies have been conducted to investigate the ability of EtO to induce gene mutations in
bacterial systems. Victorin and Stahlberg (1988) treated Salmonella typhimurium strain TA100
with EtO at concentrations of 1-200 ppm for 6 hours and demonstrated that EtO was mutagenic
in this system. In another study, Agurell et al. (1991) compared EtO and propylene oxide (two
alkylating agents) for genotoxic effectiveness in various test systems. The abilities of the two
compounds to induce point mutations in S. typhimurium strains TA 100 and TA1535 were
approximately equal. EtO induced a dose-dependent increase in the number of revertants in both
tester strains. No toxic effects were observed under the conditions tested.
C-ll
-------
In contrast, Agurell etal. (1991) found EtO to be 5-10 times more effective than
propylene oxide with respect to gene conversion and reverse mutation in the Saccharomyces
cerevisiae D7 and S. cerevisiae RSI 12 strains. The greater effectiveness of EtO over propylene
oxide in inducing these types of mutations was probably due to the difference in these
compounds' abilities to cause strand breaks via alkylation of DNA-phosphate groups.
Mutagenicity studies of EtO have also been conducted using different Escherichia coli
strains. Kolman (1985) investigated the influence of the uvrB and umuC genes on the induction
of LacI- mutants and nonsense mutants by EtO in the LacI gene of is. coli and found that uvrB
gene mutation was associated with higher mutation frequencies whereas umuC mutation did not
significantly affect the induction of LacI mutations. Thus, mutations induced by EtO were
enhanced by a lack of excision repair but not influenced by changes in error-prone repair. In
another study by the same group of authors (Kolman and Naslund, 1987), the mutagenicity of
EtO in E. coli B strains with different repair capacities was investigated. Deficiencies in
excision repair (uvrA,polA) led to considerable increases in mutation frequency compared to the
wild-type strain and strains deficient in error-prone repair (recA, lexA).
The induction of specific-locus mutations in the adenine-3 (ad-3) region of a
two-component heterokaryon (H-12) of Neurospora crassa by EtO was studied by de Serres and
Brockman (1995). The objective of this study was to compare EtO's mutational spectrum for
induced specific-locus mutations with those of other chemical mutagens. Conidial suspensions
were treated with five different concentrations of EtO (0.1-0.35%) for 3 hours. The results from
these experiments showed (1) the dose-response curve for EtO-induced specific-locus mutations
in the ad-3 region was linear, with an estimated slope of 1.49 ± 0.07, and (2) the maximum
forward-mutation frequency was between 10 and 100 ad-3 mutations per 106 survivors. The
overall data demonstrate that EtO-induced ad-3 mutations were the result of a high percentage
(96.9%) of gene/point mutations at the ad-3A and ad-3B loci.
C.2.2. Mammalian Systems
EtO has yielded positive results in virtually all in vitro mammalian cell culture systems
tested, including human cells (IARC. 2008; Kolman et al.. 2002; Bolt 2000; Prestoa 1999;
Nataraian et al.. 1995; Vogel and Nataraiaa 1995; IARC. 1994b; Dellarco etal.. 1990). Only
select in vitro studies of human cells will be reviewed here. For reviews of other in vitro studies
using mammalian cell cultures, see the aforementioned references.
C.2.2.1. In Vitro Studies
Single base pair deletion and base substitution (both transitions and transversions)
mutations were observed in the HPRT gene in human diploid fibroblasts exposed to EtO
C-12
-------
(Bastlova et al.. 1993). Sequence analysis revealed that EtO induces many different kinds of
HPRT mutations—several mutants had large HPRT gene deletions, a few mutants showed
deletion of the entire HPRT gene, and other mutants had a truncated HPRT gene; overall, as
many as 50% were large deletions. In another study by the same group of authors (Lambert et
al.. 1994). comparisons of the HPRT mutations in human diploid fibroblasts were made for three
urban air pollutants (acetaldehyde, benzo[a]pyrene, and EtO). Large genomic deletions in the
HPRT gene were observed for acetaldehyde and EtO, whereas benzo[a]pyrene induced point
mutations. The authors concluded that the HPRT locus could be a useful target for the study of
chemical-specific mutational events (Lambert etal., 1994).
The effect of EtO as a pretreatment or posttreatment to ionizing radiation was studied by
Kolman and Chovanec (2000). Human diploid VH-10 fibroblasts were either preexposed to
gamma rays (0.66 Gy/minute or 10 Gy/minute) and then treated with EtO (2.5 mMh) or
pretreated with EtO and then exposed to gamma rays. Cell killing/cytotoxicity, DNA
double-strand breakage, and mutagenicity were studied in both types of exposures. The results
of the study indicate that preexposure of the cells to gamma radiation (1 Gy) followed by
treatment with EtO (2.5 mMh) led to an additive interaction, irrespective of the dose rate. On the
other hand, pretreatment with EtO followed by gamma ray exposure resulted in an antagonistic
effect, which was most pronounced in the high-dose group (10 Gy/minute). In this group, the
mutant frequency was half that of the sum of the mutant frequencies after the individual
treatments. The authors suggest that one possible explanation for the difference in the results is
that DNA damage induced by preexposure to gamma radiation persisted into the EtO treatment
phase, and EtO might also prohibit DNA repair enzymes from operating; thus, both treatments
contributed to the mutant frequency. However, when cells were exposed to gamma radiation
following EtO treatment, the cells may have been able to repair, at least in part, the promutagenic
lesions induced by the gamma rays.
Tompkins et al. (2009) investigated the mutagenicity of EtO-derived DNAadducts in a
supF forward mutation assay. Aliquots of pSP189 plasmid containing the supl'' gene were
exposed to various concentrations of EtO in water to induce the formation of DNA adducts. The
plasmids were then transfected into human embryonic adenovirus-transformed kidney (Ad293)
cells and allowed to replicate to propagate any mutations. Replicated plasmids were isolated and
used to treat E. coli indicator bacteria under conditions in which only bacteria containing the
plasmid can grow; nonmutant colonies appear dark blue and mutant colonies appear white or
pale blue. Two studies were conducted: Study 1, in which the plasmid was incubated with EtO
concentrations ranging from 10 to 2,000 |iM at 22°C for 4 hours, and Study 2, in which the
plasmid was treated under "refined" conditions optimised to produce more of the minor
2-hydroxyethyl adducts, which involved incubation of the plasmid with EtO concentrations
C-13
-------
ranging from 10 to 100 mM at 37°C for 24 hours. For Study 1, Tompkins etal. (2009) reported
that N7-HEG was the only detectable adduct of the five they measured (before transfection; see
Section C.l.1.2 above), and there was no clear exposure-response relationship for the relative
mutation frequency. In Study 2, Nl-HEAand Oe-HEG adducts were also quantifiable, but at
lower levels than the N7-HEG adduct, and there was an apparent exposure-response relationship
for the relative mutation frequency for plasmids exposed to the 10 and 30 mM EtO
concentrations. Plasmids exposed to higher concentrations of EtO failed to produce any E. coli
colonies; this was attributed to excessive strand breaks in the plasmid DNA at those
concentrations. For the DNA damage induced by EtO-derived adducts, this limitation in the
assay imposes a short response range for the relative mutation frequency for the mutations
measured by the assay—the relative mutation frequency was 5.34 for plasmids exposed to
30 mM and no E. coli colonies were produced with plasmids exposed to the next highest EtO
concentration of 50 mM, due to excessive DNA strand breaks.
Tompkins etal. (2009) concluded that EtO is a relatively weak mutagen and that their
results suggest that a certain level of total DNA adducts or of specific promutagenic adducts
must be achieved before mutations become detectable above background levels. However,
several methodological issues raise concerns about the interpretation of the results. For example,
two solvent controls were used in the study—Solvent Control 1 was prepared in "a separate fume
hood to totally exclude any possibility of [EtO] contamination" and Solvent Control 2 was
prepared "alongside the [EtO] reactions." Solvent Control 1 was used as the referent group for
the relative mutation frequency determinations. In two replicates, Solvent Control 2 had a
relative mutation frequency of 3.0 and 2.6 compared to Solvent Control 1. If this difference
reflects a real difference between the two different solvent control preparations, it raises the
possibility that cross-contamination may have been a problem, and if any cross-contamination
also occurred across the different EtO concentrations, it could have dampened any
exposure-response relationship. In addition, if the "refined conditions" for plasmid treatment
used to produce more of the minor (more directly promutagenic) adducts in Study 2, which
included incubation at a temperature more comparable to mammalian body temperatures, had
also been used for Study 1, a different adduct profile, and different relative mutation frequencies,
might have resulted. The authors themselves acknowledged that "[in] order to categorically
determine whether a threshold exists for [EtO] in this system, a more detailed examination of the
dose-response relationship using the optimised reaction protocol and including more
concentrations around the mutagenic range is needed" (Tompkins et al., 2009). Moreover, there
is uncertainty about the generalizability of mutagenicity results from this in vitro experimental
system to the mutagenicity and genotoxicity induced by EtO exposure in vivo; for example,
C-14
-------
human embryonic adenovirus-transformed kidney cells were used for plasmid replication and
mutation production, but embryonic kidneys are not a known target for EtO carcinogenesis.
C.2.2.2. In Vivo Studies—Laboratory Animals
The results of in vivo studies on the mutagenicity of EtO following ingestion, inhalation,
or injection have also been consistently positive [e.g., Tates etal. (1999)1. For example,
increases in the frequency of gene mutations in T-lymphocytes (Hprt locus) (Walker etal., 1997)
and in bone marrow and testes (LacLlocus) (Recio etal., 2004)have been observed in transgenic
mice exposed to EtO via inhalation at concentrations similar to those in carcinogenesis bioassays
with this species (NTP, 1987). At somewhat higher concentrations than those used in the
carcinogenesis bioassays (200 ppm, but for only 4 weeks), increases in the frequency of gene
mutations have also been observed in the lung of transgenic mice (LacI locus) (Sisk etal., 1997)
and in T-lymphocytes of rats (Hprt locus) (van Sittert et al„ 2000; Tates et al., 1999). These and
other key in vivo studies are discussed in more detail below.
An approach for determining mutational spectra in exon 3 of the Hprt gene in splenic
T-lymphocytes ofB6C3Fi mice was developed by Walker and Skopek (1993). Mice (12 days
old) were given 2, 6, or 9 single intraperitoneal (i.p.) injections of 100 mg/kg EtO every other
day or 30, 60, 90, or 120 mg/kg of EtO for 5 consecutive days to achieve different cumulative
doses. In mice exposed every other day, cumulative doses of 200, 600, and 900 mg/kg produced
average mutant frequencies of 15 x 10 6, 45 x 10 6, and 73 x 10 6, respectively, 8 weeks after
dosing began. However, in mice exposed daily, cumulative doses of 150, 300, 450, and
600 mg/kg yielded average mutant frequencies of 4 x 10 6, 8 x 10 6, 11 x 10 6, and 16 x 10 6,
20 weeks after initiation of dosing. Hprt mutants obtained from mice exposed to 600 or
900 mg/kg EtO were isolated and analyzed for mutations, specifically in exon 3. DNA
sequencing showed 11 base-pair substitutions, including 4 A-to-T and 2 G-to-C transversions as
well as 3 A-to-T and 2 G-to-C transitions and seven +1 frameshift mutations in a run of six
cosecutive guanine bases. The results suggested both modified guanine and adenine bases being
involved in EtO-induced mutagenesis.
The same group of authors (Walker et al., 1997) studied the in vivo mutagenicity of EtO
at the Hprt locus of T-lymphocytes following inhalation exposure of male B6C3Fi LacI
transgenic mice. Big Blue mice at 6-8 and 8-10 weeks of age were exposed to 0, 50, 100, or
200 ppm EtO for 4 weeks (6 h/day, 5 days/week). T-cells were isolated from the thymus and
spleen and cultured in the presence of concanavalin A, IL-2, and 6-thioguanine. Mice were
sacrificed at 2 hours, 2 weeks, and 8 weeks after exposure to 200 ppm EtO to determine a time
course for the expression of Hprt-negative lymphocytes in the thymus. The results of this study
showed that following 2 hours of exposure, the Hprt mutant frequency in the thymic
C-15
-------
lymphocytes of the exposed mice was increased and reached an average maximum mutant
frequency of 7.5 ± 0.9 x 10 6 at 2 weeks postexposure when compared to 2.3 ± 0.8 x 10 6 in the
thymic lymphocytes of control mice. Dose-related increases in Hprt mutant frequency were
found in thymic lymphocytes from mice exposed to 100 and 200 ppm EtO. Furthermore, a
greater mutagenic efficiency (mutations per unit dose) was found at higher concentrations than at
lower concentrations of EtO in splenic T-cells. The average induced mutant frequencies in
splenic T-cells were 1.6, 4.6, and 11.9 x 10 6 following exposures to 50, 100, or 200 ppm EtO,
respectively. For the analysis of the LacI mutations, lymphocytes (both B-and T-cells) were
isolated from the spleen in the same animals. Two of three EtO-exposed mice at the 200 ppm
exposure level demonstrated an elevated LacI mutant frequency. The authors suggest that these
elevations were probably due to the in vivo replication of pre-existing mutants and not to the
induction of new mutations associated with EtO exposure. The results of this study indicate that
repeated inhalation exposures to high concentrations of EtO produce dose-related increases in
mutations at the Hprt locus of T-lymphocytes in male LacI transgenic mice.
LacI mutant frequencies as a result of exposure to EtO were further investigated bySisk
et al. (1997). Male transgenic LacI B6C3Fi mice (n = 15) were exposed to 0, 50, 100, or
200 ppm EtO for 4 weeks (6 hours/day, 5 days/week) and were sacrificed at 0, 2, or 8 weeks
after the last EtO exposure. To determine the LacI mutant frequency, the LacI transgene was
recovered from several tissues, including lung, spleen, germ cells, and bone marrow, selected
because they were the target sites for tumor formation (particularly lung tumors and lymphomas)
in chronic bioassays or germ cells. The results of this study indicate that the LacI mutant
frequency in the lung was significantly increased at 8-weeks postexposure to 200 ppm EtO. In
contrast, no significant increase in the LacI mutant frequencies was observed in the spleen, bone
marrow, or germ cells at either 2 or 8 weeks following exposure. These results suggest that a
4-week inhalation exposure to EtO is mutagenic in lung but not in other tissues examined under
similar conditions. The authors predict that the lack of mutagenic response in other tissues
examined is probably because of large deletions that were either not detected or recovered in the
current lambda-based shuttle vector systems. Based on the above study, the authors also suggest
that the primary mechanism of EtO-induced mutagenicity in vivo is likely through the induction
of deletions.
Tates etal. (1999) exposed rats to EtO via three routes: a single i.p. injection
(10-80 mg/kg), ingestion of drinking water (4 weeks at concentrations of 2, 5, and 10 mM), or
inhalation (50, 100, or 200 ppm for 4 weeks, 5 days/week, 6 hours/day). The goal of this study
was to measure the induction of Hprt mutations in splenic lymphocytes using a cloning assay.
Mutagenic effects of EtO following EtO administration via the three routes were compared in the
Hprt assay based on blood doses, which were determined from HEVal adduct levels in
C-16
-------
hemoglobin. Exposure to EtO via both injection and ingestion of drinking water led to a
statistically significant dose-dependent induction of mutations (up to 2.3- and 2.5-fold increases
in mutant frequency compared to background, respectively). Exposure via inhalation also caused
a statistically significant increase in mutant frequency, although to a lesser extent (up to 1.4-fold
over background). Plotting of the mutagenicity data for the three exposure routes against blood
doses as a common denominator indicated that, at equal blood doses, the order of increased
mutant frequency was i.p. injection > ingestion (drinking water) > inhalation. In the injection
experiments, there was evidence for a saturation of detoxification processes at the highest doses,
although such effects were not seen following subchronic administration. Taken together, the
mutagenicity data from this study provide consistent results, showing that exposure to EtO gives
rise to a linear dose-dependent increase in mutant frequency.
In a study by Recio et al. (2004), male Big Blue (LacI transgenic) B6C3Fi mice were
exposed to 0, 25, 50, 100, or 200 ppm EtO (6 hours per day, 5 days per week) for 12, 24, and
48 weeks. An unambiguous mutagenic response in the bone marrow was observed only after
48 weeks, with dose-related LacImutant frequencies of 7.3 x 10 5, 11.3 x 10 5, 9,3 x 10 5,
14.1 x 10 5, and 30.3 x 10 5. The mutagenic response in bone marrow is consistent with a linear
exposure-response relationship, contrary to the assertion by Recio et al. (2004) which appears to
be based on a misleading plotting scale. Mutant frequencies from testes (seminiferous tubules)
were significantly greater than in controls at 25, 50, and 100 ppm (48-week exposure). No
difference between the control and treated groups was observed in the LacI mutant frequency
after 48 weeks of 200 ppm EtO exposure. The authors suggest that this was probably due to
testicular toxicity. Furthermore, a mutation spectrum analysis of induced mutations in bone
marrow indicated a decrease in mutations at G:C base pairs and an increase at A:T base pairs,
exclusively in A:T to T:A transversions; however, the mutation spectrum from testes was similar
to that of the untreated animals. The difference in mutation spectrum between the two tissues
was probably due to differences in the repair of the DNA adducts formed.
Mutations in proto-oncogenes (Kras, Hras) and in the Trp53 tumor suppressor gene have
been studied in tumor tissues of several types from B6C3Fi mice exposed to EtO. Hong et al.
(2007) obtained tumor tissues from lung, Harderian gland, and uterus from a 2-year study (NTP,
1987) in which male and female mice were exposed to 0, 50, or 100 ppm EtO by inhalation
6 hours/day, 5 days/week and from control mice from other National Toxicology Program (NTP)
2-year bioassays. The authors analyzed the tissues for Kras mutations in codons 12, 13, and 61.
A high frequency of Kras mutations (23/23 examined, 100%) was observed in the lung
neoplasms in EtO-exposed mice compared to spontaneous lung neoplasms (27/108, 25%). The
lung neoplasms in EtO-exposed mice predominantly exhibited GGT-to-GTT mutations in codon
12 (21/23), a transversion that was rare in spontaneous lung tumors (1/108). A similar spectrum
C-17
-------
of Kras mutations was detected in the lung neoplasms in EtO-exposed mice regardless of
histological subtype (adenomas or carcinomas) or dose group. In the case of Harderian gland
neoplasms, a high frequency (18/21, 86%) of Kras mutations was detected in neoplasms from
EtO-exposed mice compared to spontaneous tumors (2/27, 7%). The predominant mutations in
the Harderian gland neoplasms in EtO-exposed mice consisted of GGC-to-CGC transversions at
codon 13 and GGT-to-TGT transversions atcodon 12, neither of which was observed in the
spontaneous tumors. When the six uterine neoplasms from EtO-exposed mice were examined
(there were no uterine tumors in the controls), the predominant mutation was a GGC-to-GGT
transition in codon 13 (5/6, 83%). Based on the above results, the authors propose that the
prominent targeting of guanine bases in the lung and Harderian gland neoplasms suggests that
the formation of N7-HEG adducts by EtO plays a role in the induction of these tumors. The
authors further propose that EtO can specifically target the Kras gene in multiple types of tissues
and that interaction with this gene is a critical component of EtO-induced tumorigenesis and is of
potential relevance to humans.
In an earlier study by the same group of authors (Houle et al.. 2006), mammary
carcinoma tissues from the same NTP study of mice exposed to EtO (0, 50, or 100 ppm)
mentioned above were examined for p53 protein expression and for Trp53 (exons 5-8) and Hras
(codon 61) mutations. The authors supplemented the number of spontaneous mammary
carcinomas with tissues from female control mice in other NTP studies from the same time
period. P53 protein expression was detected in 67% (8/12) of the mammary carcinomas in
EtO-exposed mice and 42% (8/19) of the spontaneous tumors; however, expression levels were
about sixfold higher in the tumors in the EtO-exposed mice than in the spontaneous tumors.
Trp53 mutations were observed in 67% (8/12) of the mammary carcinomas in EtO-exposed mice
and 58%) (7/12) of the spontaneous tumors. Hras mutations were detected in 33% (4/12) of the
mammary carcinomas in EtO-exposed mice and 26% (5/19) of the spontaneous tumors. While
the mutation levels for these two genes were not substantially elevated in the mammary
carcinomas in EtO-exposed mice compared to the spontaneous tumors, a shift in the mutational
spectrum was observed. Hras mutations in the tumors from EtO-exposed mice exhibited a
preference for A-to-G and A-to-T transversions, while spontaneous Hras mutations exhibited a
preference for C-to-A transversions. Trp53 mutations in the tumors from EtO-exposed mice
exhibited a base preference for guanine, while spontaneous Trp53 mutations exhibited a
preference for cytosine. In addition, concurrent Hras and Trp53 mutations were more common
in the tumors in EtO-exposed mice than in the spontaneous tumors. Based on the results of the
above two studies, it is suggested that the purine bases serve as primary targets for mutations
induced by EtO, while mutations of these genes involving cytosine appears to be a more
common spontaneous event.
C-18
-------
In vivo exposure to EtO also induced heritable mutations or effects in germ cells in
rodents (IARC, 1994b). EtO induces dominant lethal effects in mice and rats and heritable
translocations in mice (Generoso et al.. 1990; Lewis et al.. 1986). Generoso etal. (1986) and
Generoso et al. (1988) have reported that short bursts of EtO at high concentrations, such as
those that may occur in the workplace, lead to a possibly greater risk of germ cell damage than
cumulative, long-term exposure to lower levels.
Dominant-lethal mutations were investigated by Generoso et al. (1986) in two studies
(dose-response and dose-rate) in mice exposed to different doses of EtO. Dominant-lethal
responses were assessed based on matings involving sperm exposed as late spermatids and early
spermatozoa because these are the stages most sensitive to EtO exposure. In the dose-response
study, male mice were exposed by inhalation to 300 ppm, 400 ppm, or 500 ppm EtO, 6 hours per
day, for 4 consecutive days. A dose-related increase in dominant-lethal mutations was observed.
In the dose-rate study, mice were given a total exposure of 1,800 ppm x hours per day, also for 4
consecutive days, delivered either as 300 ppm in 6 hours, 600 ppm in 3 hours, or 1,200 ppm in
1.5 hours. Dominant-lethal responses increased with increasing concentration level, indicating a
dose-rate effect for the production of dominant-lethal mutations.
C.2.2.3. In Vivo Studies—Humans
Workers occupationally exposed to EtO have been studied using different physical and
biological measures (Tates etal.. 1991). Blood samples from 9 hospital workers and 15 factory
workers engaged in sterilization of medical equipment with EtO and from matched controls were
collected. Average exposure levels during 4 months (the lifespan of erythrocytes) prior to blood
sampling were estimated from levels of HEVal adducts in hemoglobin. The adduct levels were
significantly increased in hospital workers and factory workers exposed to a 40-hour
time-weighted average of 0.025 ppm and 5 ppm, respectively. Exposures were usually received
in bursts, with EtO concentrations in air ranging from 22 to 72 ppm in hospital workers and 14 to
400 ppm in factory workers. All blood samples were analyzed for HPRT mutant frequencies,
chromosomal aberrations, micronuclei, and SCEs. Mutant frequencies were significantly
increased in factory workers but not in hospital workers. The chromosomal aberration and SCE
results are discussed below in Sections C.3 and C.5, respectively.
The same authors (Tates etal.. 1995) conducted another study of workers in an EtO
production facility. HPRT mutations were measured in three exposed groups (one with high
acute exposures and two with low chronic exposures) and one unexposed group (seven workers
per group). Contrary to the earlier study, no significant differences in mutant frequencies were
observed among the groups; this discrepancy may be attributable to lower overall exposures in
C-19
-------
these workers than in the factory workers in the previous study and to the small number of
subjects per group.
Major et al. (2001) measured HPRT mutations in female nurses employed in hospitals in
Eger and Budapest, Hungary. This study examined a possible causal relationship between EtO
exposure and a cluster of cancers (mostly breast) in nurses exposed to EtO in the Eger hospital.
Controls were female hospital workers in the respective cities. The mean peak levels of EtO
were 5 mg/m3 (2.7 ppm) in Budapest and 10 mg/m3 (5.4 ppm) in Eger. HPRT variant
frequencies in both controls and EtO-exposed workers in the Eger hospital were higher than
either group in the Budapest hospital, but there was no significant increase among the
EtO-exposed workers in either hospital when compared with the respective controls.
C.2.3. Gene Mutations—Summary
In summary, there is sufficient evidence for mutagenicity of EtO in various organisms
(prokaryotes, eukaryotes, in vitro and in vivo in rodents and in vitro in human cells) tested in a
variety of mutational assays. In addition, increases in mutations in specific proto-oncogenes and
tumor suppressor genes in EtO-induced mouse tumors have been reported. Dominant-lethal
mutations have also been observed in several in vivo studies. Although data in humans are
limited, there is some evidence of increased frequencies of mutations from occupational studies.
C.3. CHROMOSOMAL ABERRATIONS
The induction and persistence of EtO-induced chromosomal alterations have been studied
both in in vitro and in vivo systems in rodent and monkey models (Lorenti Garcia et al., 2001;
Farooqi et al., 1993; Lynch et al., 1984; Kligerman et al., 1983). In addition, several studies
examined the association of chromosomal aberrations and EtO exposure in humans (WHO,
2003; Lerda and Rizzi, 1992; Galloway et al., 1986; Clare etal., 1985; Sarto et al„ 1984a;
Stolley et al., 1984; Pero et al„ 1981; Thiess etal., 1981). Chromosomal aberrations have been
linked to an increased risk of cancer in several large prospective studies [e.g., (Boffetta et al.,
2007; Rossner et al., 2005; Haemar et al., 2004; Liou et al., 1999)1. This section discusses key
studies on EtO and chromosomal aberrations.
Lorenti Garcia etal. (2001) studied the effect of EtO on the formation of chromosomal
aberrations in rat bone-marrow cells and splenocytes following in vivo exposure. Rats were
exposed to EtO either chronically by inhalation (50-200 ppm, 4 weeks, 5 days/week,
6 hours/day) or acutely by i.p. injection at doses of 50 or 100 mg/kg. Frequencies of both
spontaneous and EtO-induced chromosomal aberrations (and other endpoints, such as
micronucleus formation and SCEs, which are discussed in Section 3.3.3.3) were determined in
the splenocytes and bone-marrow cells following in vivo mitogen stimulation. No significant
C-20
-------
increase in chromosomal aberrations was observed from the chronic or acute exposures. In
another study, by Kligerman et al. (1983). no increase in chromosomal aberrations was observed
in peripheral blood lymphocytes from rats exposed to EtO by inhalation at concentrations of
either 50, 150, or 450 ppm, for 6 hours per day, for 1 and 3 days.
A study by Donner et al. (2010) in mice, however, showed clear, statistically significant
increases in chromosomal aberrations with longer durations of exposure (>12 weeks). Male
B6C3Fi mice were exposed by inhalation to 0, 25, 50, 100, or 200 ppm EtO, 5 days/week, 6
hours/day, for 6, 12, 24, or 48 weeks. The frequency of total chromosomal aberrations in
peripheral blood lymphocytes was statistically significantly increased after 12 weeks exposure to
100 or 200 ppm EtO. By 48 weeks, statistically significant increases were observed for all the
exposure groups. In addition, reciprocal translocation frequencies were statistically significantly
increased in spermatocytes for all the exposure groups at 48 weeks. Ribeiro etal. (1987)
similarly observed chromosomal aberrations in mouse bone marrow cells and spermatocytes
following 1-day and 2-week inhalation exposures to higher levels of EtO. Male Swiss Webster
mice were exposed to 0, 200, 400, or 600 ppm EtO for 6 hours in 1 day or to 0, 200, or 400 ppm
EtO for 6 hours/day, 5 days/week, for 2 weeks. Statistically significant increases in
chromosomal aberrations were observed in bone marrow cells and in spermatocytes following a
1-day exposure of 400 or 600 ppm EtO or a 2-week exposure of 200 or 400 ppm EtO.
Chromosomal aberrations in bone marrow cells were also reported in a study of acute EtO
exposure in mice (Farooqi et al., 1993). Female Swiss albino mice were administered single
doses of EtO in the range of 30-150 mg/kg by i.p. injection. A dose-related increase in
chromosomal aberrations in the bone marrow cells was observed.
Chromosomal aberrations induced by long-term exposures to inhaled EtO were also
investigated in the peripheral lymphocytes ofcynomolgus monkeys (Lynch etal., 1984). Groups
of 12 adult male monkeys were exposed at 0, 50, or 100 ppm EtO (7 hours/day, 5 days/week) for
2 years. Exposure to EtO at 100 ppm resulted in statistically significant increases in
chromosome-type aberrations in monkey lymphocytes, and exposure at both 50 and 100 ppm
resulted in statistically significant increases in chromatid-type aberrations and in chromosome-
and chromatid-type aberrations in combination. No differences in the number of gaps were
found.
Increases in chromosomal aberrations in peripheral blood lymphocytes have been
consistently reported in studies of workers exposed to high occupational concentrations of EtO
(>5 ppm, TWA). Effects observed at lower concentrations have been mixed (WHO, 2003).
Chromosomal aberrations that have been detected in the peripheral blood lymphocytes of
workers include breaks, gaps, and exchanges and supernumerary chromosomes (Lerda and Rizzi,
C-21
-------
1992; Galloway etal.. 1986; Clare et al.. 1985; Sarto et al.. 1984a; Pero et al.. 1981; Thiess etal..
19811
Clare etal. (1985) conducted chromosomal analyses of lymphocytes from 33 workers
employed in the manufacture of EtO. A slightly higher frequency of chromatid aberrations was
observed in workers exposed to EtO than in controls. Further, a positive correlation between
length of employment in the EtO-exposed group and the number of aberrations was observed. In
another study, Galloway etal. (1986) analyzed chromosomal aberration frequencies in
61 employees potentially exposed to EtO. Three work sites (I, n, and III) with different
historical ambient levels of EtO were chosen for the study. Blood samples were drawn over a
24-month period and aberrations were analyzed in 100 cells per sample after culture for
48-51 hours. At work sites I and II, no consistent differences in aberration frequencies were
found. However, at work site III, aberration frequencies in potentially exposed individuals were
significantly increased when compared with controls. A previous study by the same group
(Stolley etal., 1984) showed an association between SCEfrequency and EtO exposure. When
the aberrations were compared with the levels of SCEs, the authors found a weak overall
association. In addition, Lerda and Rizzi (1992) showed a significant increase in chromosomal
aberration frequencies in EtO-exposed individuals when compared with controls. Major et al.
(1996) studied hospital nurses exposed to low doses and high doses of EtO to identify changes in
structural and numerical chromosomal aberrations. Chromosomal aberrations were found to be
significantly elevated in both the low-dose and the high-dose exposure groups. Deletions and, to
a lesser extent, chromatid exchanges and dicentrics were detected in the low-dose exposure
group; however, in the high-dose group, in addition to the increased number of deletions, the
frequencies of dicentrics and rings showed a significant excess when compared with controls.
The authors suggest that a natural radioactivity from local tap water may have been a
confounding factor.
A study by Sarto et al. (1984a) showed significant increases in chromosomal aberrations
after exposure to EtO. Chromosomal aberrations were detected in the peripheral lymphocytes of
41 workers exposed to EtO in the sterilizing units of eight hospitals in the Venice region
compared to 41 age- and smoking-matched controls. In another study of 28 EtO-exposed
sterilizer workers and 20 unexposed controls, Hogstedt etal. (1983) reported a statistically
significant increase in total chromosomal aberrations and gaps, but not breaks, in the peripheral
blood lymphocytes of the exposed workers, adjusted for age, smoking, drug intake, and exposure
to ionizing radiation; no significant increases in chromosomal aberrations were observed in bone
marrow cells. Tates etal. (1991) reported a significant increase in chromosomal aberrations in
hospital workers and in factory workers (details of this study are provided in the section on gene
mutations above). Tompa et al. (2006) reported statistically significant increases in
C-22
-------
chromosomal aberrations and SCEs in 66 Hungarian hospital nurses exposed to sterilizing gases
in uncontrolled environments compared to 94 nonexposed controls; however, it is difficult to sort
out any effects ofEtO exposure from possible effects from smoking or exposure to ionizing
radiation, formaldehyde, or other possible sterilizing gases in this study.
In summary, the above data clearly indicate that EtO is genotoxic and can cause a variety
of chromosomal aberrations, including breaks, gaps and exchanges [reviewed in detail in Preston
(1999)1. Chromosomal aberrations have been observed in both in vitro and in vivo studies in
rodent models and mammalian cells. Increases in chromosomal aberrations in peripheral blood
lymphocytes have been consistently reported in studies of workers exposed to EtO.
C.4. MICRONUCLEUS FORMATION
Micronucleus formation also demonstrates the genotoxic effects of a chemical. When
appropriate methods are used to identify the origin of the micronucleus (kinetochore-positive or
kinetochore-negative), this assay can provide information about a chemical's mechanism of
action (e.g., if a chemical causes direct DNA damage resulting from strand breaks [clastogen] or
indirect numerical changes [aneugen] resulting from spindle disruption). An association between
increased micronucleus frequency and cancer risk has been reported in at least one large
prospective study (Bonassi et al.. 2007). Several in vitro and in vivo studies in both laboratory
animals (Lorenti Garcia et al.. 2001; Jenssen and Ramei 1980; Appelgren et al.. 1978) and
humans (Ribeiro et al.. 1994; Schulte etal.. 1992; Mayer et al.. 1991; Tates et al.. 1991; Sarto et
al.. 1990; Hogstedt etal.. 1983) have been conducted to explore the induction ofmicronuclei as a
result of exposure to EtO.
Lorenti Garcia etal. (2001) studied the effect ofEtO on the formation of micronuclei in
rat bone marrow cells and splenocytes following in vivo exposure. Rats were exposed to EtO
either subchronically by inhalation (50-200 ppm, 5 days/week, 6 hours/day, for 4 weeks) or
acutely by i.p. injection at doses of 50 or 100 mg/kg. Spontaneous and induced frequencies of
micronuclei were determined in the bone marrow cells (only for acute EtO exposure) and
splenocytes following in vitro mitogen stimulation. Following chronic exposure, no significant
increase in micronuclei was observed in rat splenocytes. Following acute exposure, micronuclei
increased significantly in rat bone marrow cells as well as splenocytes.
In the Hogstedt etal. (1983) study of 28 EtO-exposed sterilizer workers and
20 unexposed controls, a statistically significant increase in micronuclei was observed in bone
marrow cells (erythroblasts and polychromatic erythrocytes), but not in lymphocytes, in the
exposed workers, adjusted for age, smoking, drug intake, and exposure to ionizing radiation.
The frequency of micronuclei in peripheral blood cells was increased in workers exposed
to relatively high (3.7-60.4 mg/m3) levels ofEtO (Ribeiro etal.. 1994; Tates etal.. 1991).
C-23
-------
Schulte et al. (1992) did not observe increased micronuclei in the lymphocytes of hospital
workers with low levels of EtO exposure (up to 2.5 mg/m3 8-hour TWAs). Sarto et al. (1990)
studied micronucleus formation in human exfoliated cells of buccal and nasal cavities to monitor
the genotoxic risk in a group of workers (n = 9) chronically exposed to EtO (concentrations
lower than 0.38 ppm as time-weighted average). The mean frequencies of micronucleated
buccal cells were similar to control values. The frequency of nasal micronucleated cells was
higher than in controls (0.77 vs. 0.44); however, the difference was not statistically significant.
In another group of three subjects that were acutely exposed (concentration not provided) to EtO,
buccal cavity and nasal mucosa samples were taken 3, 9, or 16 days after acute exposure. The
frequencies of micronucleated buccal cells did not change, while the frequencies of
micronucleated nasal cells significantly increased.
Peripheral blood cells of 34 EtO-exposed workers at a sterilization plant and
23 unexposed controls were assessed for different biological markers, such as EtO-hemoglobin
adducts, SCEs, micronuclei, chromosomal aberrations, DNA single-strand breaks and an index
of DNA repair (Mayer et al.. 1991). Neither chromosomal aberrations nor micronuclei differed
significantly by exposure status, whether or not adjusted for smoking status.
In summary, increases in the frequency of micronuclei have been observed in in vivo
animal studies. The frequency of micronuclei in peripheral blood cells was also increased in
workers exposed to relatively high (3.7-60.4 mg/m3) levels of EtO (Ribeiro et al. 1994; Tates et
al., 1991). However, in the majority of human studies involving exposures at lower levels, no
effects on the frequency of micronuclei were observed. Apparent inconsistencies in the data
could reflect the influence of peak exposures, differences in exposure measurement errors,
duration of exposure, and/or smoking status.
C.5. SISTER CHROMATID EXCHANGES (SCES)
There is a significant body of evidence for the induction of SCEs as a result of exposure
to EtO. Studies have been conducted both in laboratory animals (Lorenti Garcia et al.. 2001;
Ong et al.. 1993; Kelsev etal.. 1988; Lynch et al.. 1984; Kligerman et al.. 1983; Yager and Benz.
1982) and in humans (Agurell etal.. 1991; Galloway etal.. 1986; Laurent et al.. 1984; Sarto et
al.. 1984a. b; Stollev et al.. 1984; Yager et al.. 1983; Garry et al.. 1979). In particular, several
occupational exposure studies have yielded positive results when EtO-exposed workers were
studied. The following is a summary of both the animal and human studies.
Inhalation studies with rats have shown that exposures to EtO at 50 ppm or more for
3 days result in an increase in SCEs in peripheral blood lymphocytes (Kligerman et al.. 1983).
Increased incidences of SCEs in the peripheral blood lymphocytes of monkeys exposed to EtO at
500 or 100 ppm were also reported by Lynch et al. (1984). A follow-up study in these same
C-24
-------
monkeys by Kelsev et al. (1988) indicated that the high SCE counts persisted for 6 years after
exposure.
Lorenti Garcia etal. (2001) studied the effect ofEtO on the persistence of SCEs in rat
bone marrow cells and splenocytes following in vivo exposure. Rats were exposed to EtO either
subchronically by inhalation (50-200 ppm, 5 days/week, 6 h/day, for 4 weeks) or acutely by
i.p. injection at dose levels of 50 or 100 mg/kg. Frequencies of SCEs were determined in the
bone marrow cells and splenocytes after in vitro mitogen stimulation. Following chronic
exposure, cytogenetic analyses were carried out at Days 5 and 21 in the splenocytes. In these
experiments, EtO was effective in inducing SCEs, and marked increases in cells with high
frequency SCEs were observed which persisted until Day 21 postexposure. Following acute
exposure, SCEs were increased significantly in rat bone marrow cells as well as splenocytes.
New Zealand white male rabbits (n = 4) were exposed in inhalation chambers to 0, 10,
50, and 250 ppm EtO for 6 hours a day, 5 days a week, for 12 weeks (Yager and Benz, 1982).
Peripheral blood samples were drawn in three regimes (before the start of exposure, at intervals
during exposure, and up to 15 weeks after the end of exposure) to measure SCE rates. No
change in SCE rates was observed from exposure to 10 ppm; however, an increase was seen after
exposure to 50 and 250 ppm. Above-baseline levels were observed even after 15 weeks
postexposure, although the levels were not as high as during exposure. These results indicate
that inhalation exposure to the EtO results in a dose-related increase in SCEs.
Lynch etal. (1984) investigated the effect of long-term exposures to inhaled EtO on SCE
rates in peripheral lymphocytes of monkeys. Groups of 12 adult male cynomolgus monkeys
were exposed at 0, 50, or 100 ppm EtO (7 hours/day, 5 days/week) for 2 years. Statistically
significant increases in SCE rates were observed in monkey lymphocytes in both exposure
groups. Both exposure groups had increased numbers of SCEs/metaphase as compared to
controls, and these numbers increased in a dose-dependent manner.
In an in vitro study of human cells, peripheral lymphocyte cultures were exposed to
methyl bromide, EtO, and propylene oxide, as well as diesel exhaust (Tucker et al., 1986). SCE
frequency was measured, and the frequency more than doubled in the cultures treated with EtO.
Agurell etal. (1991) also studied the effect ofEtO on SCEs in human peripheral blood
lymphocytes in vitro. An increase in SCE frequency was observed as a result of exposure
(0-20 mMh) to EtO. Similarly, Hallier etal. (1993) observed that the frequency of SCEs in
human peripheral blood lymphocytes exposed in vitro to EtO was higher in cells isolated from
individuals expressing low levels of glutathione S-transferase T1 than in cells from subjects
expressing higher levels of this enzyme.
Several studies of EtO-exposed workers have also reported an increased incidence of
SCEs in peripheral lymphocytes [e.g., Garry etal. (1979), Yager et al. (1983), Sarto et al.
C-25
-------
(1984a). Sarto et al. (1984b). Galloway etal. (1986). Schulte et al. (1992)1. although the
Hogstedt etal. (1983) study discussed in Sections C.3 and C.4 did not report significant increases
in SCEs in the lymphocytes of the exposed workers.
Garry et al. (1979) analyzed SCEs in lymphocytes cultured from EtO-exposed individuals
as well as comparable controls. Significant increases in SCEs were observed at 3 weeks and at
8 weeks following exposure. Although this study does not describe the exact exposure estimates,
EtO was recognized as a mutagenic or genotoxic agent. Laurent etal. (1984) studied SCE
frequency in workers exposed to high levels of EtO in a hospital sterilization service. Blood
samples were obtained retrospectively from a group of 25 subjects exposed to high levels of EtO
for a period of 2 years. A significant increase in SCEs was observed in the exposed group when
compared with the control group. The authors concluded that the effect of exposure to EtO was
sufficient to produce a cumulative and, in some cases, a persistent genetic change.
Peripheral blood lymphocytes of nurses exposed to low and high concentrations of EtO
were studied by Major et al. (1996). SCEs were slightly elevated in the low-exposure group but
were significantly increased in the high-exposure group. Similarly, several studies (Sarto et al.,
1991; Sarto et al., 1990; Sarto et al„ 1987; Sarto et al., 1984a, b) showed significant increases in
SCEs.
Tates etal. (1991) studied workers occupationally exposed to EtO using different
physical and biological measures. Blood samples from 9 hospital workers and 15 factory
workers engaged in sterilization of medical equipment with EtO and from matched controls were
collected. Exposures were usually received in bursts, with EtO concentrations in air ranging
from 22 to 72 ppm in hospital workers and 14 to 400 ppm in factory workers. The mean
frequency of SCEs was significantly elevated by 20% in hospital workers and by almost 100% in
factory workers. In contrast, no significant increase in SCEs was observed in lymphocytes of
workers who were accidentally exposed to high concentrations of EtO or of workers with low
exposure concentrations (Tates et al„ 1995).
Schulte et al. (1992) observed a statistically significant increase in SCEs in 43 workers
exposed to EtO in U.S. hospitals compared to 8 unexposed hospital workers. The frequency of
SCEs was also significantly associated with cumulative EtO exposure in a regression analysis
that controlled for various potential confounding factors, including smoking. A similar
relationship was not observed in 22 Mexican hospital workers. Schulte etal. (1992)
hypothesized that the difference may have been due to longer shipping times of the Mexican
specimens for the cytogenetic assays.
In summary, significant increases in the frequency of SCEs were observed in rats and in
monkeys both by inhalation and i.p. injection. In humans, multiple occupational studies have
reported positive responses, with significant increases in frequency of SCEs in peripheral blood
C-26
-------
lymphocytes having been observed among individuals exposed to higher levels ofEtO. In some
studies, increases in the frequency of SCEs have been observed to persist after exposure has
ceased. The results of studies of individual workers exposed to very low levels (<0.9 mg/m3) of
EtO have been mixed.
C.6. OTHER ENDPOINTS (GENETIC POLYMORPHISM, SUSCEPTIBILITY)
Dose-dependent effects of polymorphisms in the genes for epoxide hydrolase (EPHX1),
different subfamilies of glutathione-^'-transferase (GSTM1, GSTP1, GSTT1), and various DNA
repair enzymes (hOGGl, XRCC1, XRCC3) on EtO-induced genotoxicity were evaluated by
Godderis etal. (2006). Peripheral blood mononuclear cells from 20 individuals were exposed to
three doses ofEtO (0.45, 0.67, 0.9 mM), and genotoxicity was evaluated by measuring comet tail
length and micronucleus frequencies in binucleated cells (MNBC). A dose-dependent increase
in tail length (indicating DNA strand breaks) was observed in exposed individuals compared to
controls. No change in MNBC was observed. None of the epoxide hydrolase or
glutathione-^'-transferase polymorphisms had a significant influence on the tail length or MNBC
results for any EtO dose. Further analysis revealed a significant contribution of the hOGGl
(involved in base excision repair) and XRCC3 (involved in repair of cross-links and
chromosomal double-strand breaks) genotypes to the interindividual variability of EtO-induced
increases in tail length. Homozygous hOGGl326 wild-type cells showed significantly lower
effects ofEtO on tail length compared to the heterozygous cells. Also, significantly higher tail
lengths were found in EtO-exposed cells carrying at least one variant XRCC3241 Met allele. For
the latter effect, there was a significant interaction between the XRCC3241 polymorphism and
dose, signifying a greater impact of the polymorphism on DNA damage at higher doses.
In contrast to the findings of no significant effect of glutathione-^'-transferase
polymorphisms on DNA breaks and micronuclei production by Godderis etal. (2006). Hallier et
al. (1993) observed that the frequency of SCEs in human peripheral blood lymphocytes exposed
in vitro to EtO was higher in cells isolated from individuals expressing low levels of GSTT1 than
in cells from subjects expressing higher levels of this enzyme. Similarly, Yong etal. (2001)
measured approximately twofold greater EtO-hemoglobin adduct levels in occupationally
exposed persons with a (AS"/"/'/-null genotype than in those with positive genotypes.
In a study involving small numbers (// = 4-12 per group) of nonsmoking males and
females exposed to EtO through the sterilization of medical equipment, Fuchs et al. (1994)
reported 1.5-, 2.2-, and 1.5-fold increases in DNA single-strand breaks in peripheral blood
mononuclear cells obtained from individuals exposed to EtO concentrations of 0.1-0.49 mg/m3,
0.5-2.0 mg/m3, and >2 mg/m3, respectively. Fuchs et al. (1994) further noted that these
nonsmokers could be divided into two distinct susceptibility groups, with 67% of the subjects
C-27
-------
exhibiting approximately fivefold higher levels of DNA single-strand breaks in response to EtO
exposure than the remaining subjects, and that the bimodal nature of the differential
susceptibility suggested that the susceptibility was attributable to an unidentified polymorphism.
Primary and secondary cultures of lymphob lasts, breast epithelial cells, peripheral blood
lymphocytes, keratinocytes, and cervical epithelial cells were exposed to 0-100 mM EtO, and
DNA damage was measured using the comet assay (Adam et al.. 2005). A dose-dependent
increase in DNA damage was observed in all cell types without notable cytotoxicity. Breast
epithelial cells (26% increase in tail length) were more sensitive than keratinocytes (5% increase)
and cervical epithelial cells (5% increase) but less sensitive than lymphob lasts (51% increase)
and peripheral lymphocytes (71% increase) at the same dose of 20 mM.
C.7. ENDOGENOUS PRODUCTION OF ETHYLENE AND ETO
Ethylene, a biological precursor of EtO, is ubiquitous in the environment as an air
pollutant and is produced in plants, animals, and humans (Abeles and Heggestad. 1973).
Ethylene is generated in vivo endogenously during normal physiological processes such as (1)
oxidation of methionine, (2) oxidation of hemoglobin, (3) lipid peroxidation, and (4) metabolism
of intestinal bacteria [reviewed by (Bolt 2000; IARC. 1994a)1. Marsden et al. (2009) proposed
that oxidative stress can induce the endogenous formation of ethylene, which can in turn be
metabolized to EtO. Endogenous production of ethylene has been documented in laboratory
animals and in humans (Filser et al.. 1992; Shen et al.. 1989; Ehrenberg etal.. 1977; Chandra and
Spencer. 1963).
Shen et al. (1989) reported an endogenous production rate of 2.8 and 41 nmol/h ethylene
in Sprague-Dawley rats and humans, respectively, with similar thermodynamic partition
coefficients between the two species. Filser et al. (1992) reported a low degree of endogenous
production of ethylene (32 ±12 nmol/h) in healthy volunteers based on exhalation data. The
authors indicated that the endogenous levels of ethylene would account for ~66%> of the
background level of EtO-hemoglobin adducts (HEVal), while the remaining one-third (15 ppb) is
contributed by exogenous environmental ethylene exposure. Although the percentage of
endogenous ethylene converted to EtO is not known, Tornqvist etal. (1989) have shown that in
fruit-store workers exposed to 0.3 ppm ethylene, only 3%> is metabolized to EtO. Thus, the
amount of endogenous ethylene converted to EtO should be minimal. Furthermore, with
inadequate laboratory animal and human evidence available for ethylene as a carcinogen (IARC.
1994a). exogenous ethylene exposure may not produce enough EtO to contribute significantly to
carcinogenicity under standard bioassay conditions (Walker etal.. 2000).
Ethylene formed from endogenous sources is converted to EtO by cytochrome
P450-mediated metabolism (Tornqvist. 1996; IARC. 1994a). EtO formed from the endogenous
C-28
-------
conversion of ethylene leads to 2-hydroxyethylation of DNA and forms N7-HEG adducts,
contributing to the background levels of this adduct in unexposed humans and rodents. As
shown in Table C-l, improvements in analytical methodology have led to the detection and
quantification of background N7-HEG adducts in DNA of unexposed laboratory animals and
humans (Marsden et al.. 2009; Swenberg etal.. 2008; Tompkins et al.. 2008; Marsden et al..
2007; Swenberg et al.. 2000; van Sittert et al.. 2000; Walker et al.. 2000; Eide etal.. 1999;
Farmer and Shuker. 1999; Wu et al.. 1999b; Wu et al.. 1999a; Zhao etal.. 1999; Bolt etal.. 1997;
Zhao etal., 1997; Kumar etal., 1995; van Delft etal., 1994; Farmer et al., 1993; van Delft etal„
1993; Leutbecher et al„ 1992; Walker et al., 1992; Cushnir etal., 1991; Fostetal., 1989).
However, the levels of adducts detected in rodents and humans vary widely and appear to depend
on the type of the analytical method used. Even with the most advanced techniques (Tompkins
et al., 2008), minor DNA adducts such as 06-HEG and N3-HEA are below the level of detection.
Also, some researchers consistently demonstrate higher background levels of DNA adducts (Wu
et al., 1999a; Walker et al., 1992). However, the higher background levels in some of these
studies are possibly due to the methodology used, which may have caused an artifactual increase
in the adduct levels.
Using sensitive detection techniques and a dual-isotype labeling approach designed to
separately quantify both endogenous N7-HEG adducts and "exogenous" N7-HEG adducts
induced by EtO treatment in F344 rats, Marsden et al. (2009) reported detectable levels of
exogenous adducts in DNA of spleen and liver tissues at the lowest dose administered (0.0001
mg/kg injected i.p. daily for 3 days). The authors also reported statistically significant linear
dose-response relationships (p < 0.05) for exogenous adducts in all three tissues examined
(spleen, liver, and stomach), although the authors caution that some of the adduct levels induced
at low EtO concentrations are at or below the limit of accurate quantitation (0.2-0.5 adducts/1010
nucleotides). EtO doses of >0.001 and >0.05 mg/kg induced elevated endogenous N7-HEG
levels in the liver and spleen, respectively, with endogenous adduct levels increasing in an
apparent dose-responsive manner above >0.05 mg/kg in both tissues; endogenous adduct levels
in the stomach, however, remained unchanged at any dose. Note that the whole range of doses
studied by Marsden et al. (2009) lies well below the dose corresponding to the lowest LOAEL
from an EtO cancer bioassay. For example, an approximate calculation indicates that the low
exposure level of 10 ppm for 6 hours/day used in the Snellings etal. (1984) bioassay of F344 rats
is equivalent to a daily dose of about 1.7 mg/kg, which is over 10 times higher than the largest
daily dose of 0.1 mg/kg used by Marsden et al. (2009).9
9This calculation uses themean alveolar ventilation rate for rats of 52.9 mL/minute/100 g reported by Brown et al.
f!998\ 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
C-29
-------
In summary, endogenous ethylene and EtO production, which contribute to background
N7-HEG DNA adducts indicative of DNA damage, have been observed in unexposed rodents
and humans. Although a constant reduction in DNA damage in vivo is carried out by DNA
repair and DNA replicative synthesis, a certain steady-state background level of adducts is
measurable at all times. The quantitative relationships between the background DNA damage
and the spontaneous rates of mutation and cancer are not well established. Experimental
evidence is needed that can unequivocally measure artifact-free levels of background DNA
damage, including effects other than adducts, clearly establish mutagenic potency of such
background lesions, and demonstrate the organ- and cell-type-specific requirements for the
primary DNA damage to be expressed as heritable genetic changes (Gupta and Lutz. 1999).
Some investigators have posited that the high and variable background levels of
endogenous EtO-induced DNA damage in the body may overwhelm any contribution from
exogenous EtO exposure (Marsden et al„ 2009; SAB, 2007). It is true that the existence of these
high and variable background levels may make it hard to observe statistically significant
increases in risk from low levels of exogenous exposure. However, there is clear evidence of
carcinogenic hazard from the rodent bioassays and strong evidence from human studies (see
Chapter 3, Section 3.5), and the genotoxicity/mutagenicity of EtO (see Section 3.4) supports
low-dose linear extrapolation of risk estimates from those studies (U.S. EPA 2005a). In fact, as
discussed above, Marsden et al. (2009) reported increases in exogenous adducts in DNA of the
spleen and liver consistent with a linear dose-response relationship (p < 0.05), down to the
lowest dose administered (0.0001 mg/kg injected i.p. daily for 3 days, which is a very low dose
compared to the LOAELs in the carcinogenicity bioassays). Furthermore, while the
contributions to cancer risk from low exogenous EtO exposures may be relatively small
compared to those from endogenous EtO exposure, low levels of exogenous EtO may
nonetheless be responsible for levels of risk (above background risk) that exceed de minimis risk
(e.g., >10 6), This is not inconsistent with the much higher levels of background cancer risk, to
which endogenous EtO may contribute, for the two cancer types observed in the human studies:
lymphoid cancers, which have a background lifetime incidence risk on the order of 3%, and
breast cancer, which has a background lifetime incidence risk on the order of 15%.
equates to about 3.48 mg/kg. I ARC (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.
C-30
-------
Table C-l. Levels of endogenous (background) N7-HEG adducts in unexposed human and rodent tissues
Species
Tissue
Detection method
Adduct levels reported
Adducts/107
nucleotides3
Reference
Human
Lymphocytes
GC/MS
8.5 pmol/mg DNA
28
Fostetal. CI9891
Human
WBC
Immuno-slotblot
0.34 adducts/106 nucleotides
3.4
van Delft et al. (1994)
Human
Blood
HPLC-fluores cence
3.2 pmol/mg DNA
11
Bolt et al. (19971
Human
Lymphocytes
GC/MS
2-19 adducts per 107 nucleotides
2-19
Wu et al. (1999bl
Human
WBC
32P/TLC/HPLC
0.6 adducts/107 nucleotides
0.6
Zhao et al. (19991
Human
WBC
32P/TLC/HPLC
2.9 adducts/107 nucleotides
2.9
Zhao et al. (19991
Human
Lung
32P/TLC/HPLC
4.0 adducts/107 nucleotides
4.0
Zhao et al. (19991
Human
Granulocytes
GC-EC-MS
3.8 adducts/107 nucleotides
3.8
Yone et al. (20071
Rat
Lymphocytes
GC/MS
5.6 pmol/mg DNA
18.48
Fostetal. (19891
Mice/Rats
6 Control tissues'3
HPLC-fluores cence
2-6 pmol/mg DNA
6.6-19.8
Walker et al. (19921
Rat
Liver, kidney, spleen
3 2P/GC/MS
0.4 to 1.1 adducts/107 nucleotides
0.4-1.1
Eide et al. (19991
Mice/Rats
Spleen
GC/EC/N CI-HRM S
0.2 to 0.3 pmol/mmol guanine
0.5-0.8c
Wu et al. (1999al
Rat
Lymphocytes, liver, kidney
32P/TLC/HPLC
0.6 to 0.9 adducts/107 nucleotides
0.6-0.9
Zhao et al. (19991
Rat
Liver
GC/MS
2.6 adducts/108 nucleotides
0.26
van Sittert et al. (20001
Rat Study 1
Study 2
Heart, colon, liver
7 control tissues'1
LC-MS/MS
LC-MS/MS
0.27-2.38 adducts/108 nucleotides
1.06-3.52 adducts/108 nucleotides
0.027-0.238
0.106-0.352
Marsdenet al. (20071
Rat
Liver
HPLC/ESI TMS
8 adducts/108 normal nucleotides
0.8
Tompkins et al. (20081
Rat
Liver, spleen, stomach
HPLC/LC-M S/MS
233-373 adducts/1010 nucleotides
0.233-0.373
Marsdenet al. (20091
-------
Table C-l. Levels of endogenous (background) N7-HEG adducts in unexposed human and rodent tissues
(continued)
aAdduct 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;
LC-MS, liquid chromatography mass spectrometry; ESI TMS, electrospray ionization tandem mass spectrometry; GC/EC/NCI-HRMS, gas chromatography/electron
capture/negative chemical ionization high-resolution mass spectrometry; WBC, white blood cells.
bBrain, lung, spleen, kidney,liver, and testes.
cEstimated bv Marsden et al. f2007\
dLiver, heart, colon, lung, kidney, spleen, and stomach.
-------
C.8. CONCLUSIONS
The overall available data from in vitro studies, laboratory animal studies, and human
studies indicate that EtO is both a mutagen and a genotoxicant. In addition, increases in
mutations in specific oncogenes and tumor suppressor genes in EtO-induced mouse tumors have
been reported. Stable translocations seen in human leukemias may arise from similar DNA
adducts that produce chromosome breaks, micronuclei, SCEs, and even gene mutations observed
in peripheral lymphocytes. Dominant lethal mutations, heritable translocations, chromosomal
aberrations, DNA damage, and adduct formation in rodent sperm cells have been observed in a
number of studies involving the exposure of rats and mice to EtO. Based on the likely role of
DNA alkylation in producing the genotoxic effects in germ cells in laboratory animals exposed
to EtO, as well as the lack of qualitative differences in the metabolism of EtO between humans
and laboratory animals, EtO can also be considered a likely human germ cell mutagen (WHO,
2003). There is consistent evidence that EtO interacts with the genome of cells within the
circulatory system in occupationally exposed humans and overwhelming evidence of
carcinogenicity and genotoxicity in laboratory animals. Based on these considerations, there is a
strong weight of evidence suggesting that EtO would be carcinogenic to humans (see Chapter 3,
Section 3.4).
C-33
-------
APPENDIX D. REANALYSES OF ETHYLENE OXIDE EXPOSURE-RESPONSE
DATA
[EDITORIAL NOTE: This Appendix contains a revised version of the report submitted by
Dr. Kyle Steenland in 2010 summarizing the results of analyses he conducted under contract
to the EPA. The terminology originally used by Steenland to designate the different
exposure-response model forms has been changed to be consistent with that used in the rest
of this assessment (see end of Section 4.1, pages 4-5 to 4-6). Models that are linear in log
RR and which were previously referred to as "linear" models have been renamed "log-linear"
or "Cox regression" models (except in some cases where it is stated that they are log RR
models). Models of the form RR = 1 + p x exposure, which were previously referred to as
"excess relative risk" (ERR) models, have been renamed "linear" models in most cases. In
addition, section headings, figures, and tables were renumbered for the table of contents, and
some supplemental analyses performed by Steenland after the main report was submitted
have been added. Finally, after the SAB review of the 2014 draft assessment (U.S. EPA,
2014a, b), some additional analyses of the NIOSH data were conducted and the results of
these analyses have been included here, and some of the text has been edited for consistency
with the new analyses. The breast cancer incidence data are not publically available, and so,
were no longer accessible to Steenland, as he is no longer at NIOSH. Therefore, the revised
breast cancer incidence analyses, primarily categorical and linear RR analyses and the
reexamination of lag periods, have been performed by Dr. James Deddens of NIOSH.
Supplemental analyses of the lymphoid cancer mortality data were again conducted by
Steenland under contract to the EPA. No further analyses were done of the all-
lymphohematopoietic cancer or breast cancer mortality data because these analyses are not of
primary interest in this assessment but rather are provided for comparison with the lymphoid
cancer mortality and breast cancer incidence results. Some of the original pieces (e.g.,
figures and risk assessment sections) of the report submitted by Steenland in 2010 have been
deleted because revised versions are presented in Chapter 4. Finally, some minimal technical
editing was done.]
This report contains the results of reanalyses of the National Institute for Occupational
Safety and Health cohort of workers exposed to ethylene oxide (EtO) conducted for the
U.S. Environmental Protection Agency. The report begins with an overview of the modeling
strategy used, followed by the results of reanalyses of the breast cancer incidence, breast cancer
mortality, lymphoid cancer mortality, and finally, hematopoietic cancer mortality databases.
Various models were used for these reanalyses, as discussed in this report. The report concludes
D-l
-------
with the results of some sensitivity analyses and discussions of the possible influences of the
healthy worker survivor effect and exposure mismeasurement.
Introduction. Modeling strategy for ethylene oxide (EtO) risk assessment
The modeling strategy adopted here for EtO risk assessment relies principally on the
usual epidemiologic models in which the log of the rate ratio (RR) is some function of exposure,
in this case cumulative exposure with a lag to reflect a length of time that is likely necessary
before an exposure can result in (observable or fatal) cancer. We have relied primarily on Cox
regression as a flexible method of modeling the log RR; however, we have also included some
linear relative risk models. Cumulative exposure is typically the exposure metric of interest in
predicting chronic disease. Transformations (natural log and sqaure root) of cumulative
exposure have been used in some of the Cox regression models. As is commonly done, 1 ppm x
day was added to cumulative exposures in analyses using the log of cumulative exposure with a
lag, to avoid taking the log of 0.
We have also used two-piece spline models, in which log RR (in the log-linear model) or
RR(in the linear model) is a function of two lines that join continuously at a single point of
inflection, or knot. These two-piece spline models have been described as part of a general
description of exposure-response modeling by Steenland and Deddens (2004) and have been
used previously in risk assessment [e.g., see the risk assessment for dioxin by Steenland etal.
(2001)1. The two-piece log-linear model has the form log
RR = Po + Pi x cumexp + P2 x (max(0,cumexp-knot)), where cumexp is cumulative exposure, the
last term equals either 0 or cumexp-knot, whichever is greater, and the knot is the point of
inflection or point of change of slope for the two linear pieces. The slope of the last term is
Pi + P2 for cumulative exposure values above the knot.
Log RR models are not linear when the log RR function is transformed via
exponentiation back to a nonlogarithmic function, but they are nearly so in the low-dose region
of interest. The splines are linear using the linear RR model.
"Plateau-like" exposure-response curves, in which the exposure-response curve begins
steeply but is attenuated at higher exposure, have been seen for many occupational carcinogens.
This may occur for a variety of reasons, including depletion of susceptible subpopulations,
mismeasurement at high exposure resulting in attenuation, and the healthy worker survivor effect
(Stayner et al„ 1993). Attenuation of the exposure-response relationship occurs for the breast
cancer and (lympho)hematopoietic endpoints of interest for EtO. For these endpoints, a simple
linear model, where the log RR (for the log-linear model) or the RR increases linearly with
cumulative exposure, does not fit the data well, based on simple visual inspection of the
categorical data.
D-2
-------
Frequently, such plateau-like curves may be modeled by using the log of cumulative
exposure rather than cumulative exposure itself but this approach has the disadvantage that the
curve is usually highly supralinear at low doses. Two-piece spline models are particularly useful
in modeling exposure-response relationships in which the log RR or RR increases initially with
increasing exposure but then tends to increase less or plateau at high exposures. The two-piece
spline models lessen this supralinearity in the low-dose region (Steenland and Deddens. 2004).
The shape of the two-piece spline model, in particular the slope of the curve in the low-
dose region, depends on the choice of the point of inflection where the two splines are joined.
Here, we have chosen the point of inflection based on the best model likelihood, trying a range
of points of inflection (knots) across the range of exposure starting from 0 and incrementing by
100, 500, or 1,000 ppm-day intervals. The model likelihood often does not change much across
these different points of inflection, but it does change some and we have chosen the point of
inflection resulting in the best model likelihood. The model likelihood used to find the best fit in
all models used in this analysis is the usual partial likelihood (Langholz and Richardsoa 2010),
as used with the Cox models, which maximizes the probability, across all the cases, that a case
fails (the numerator) relative to its case-control risk set (which includes the case) (the
denominator) and has the form
L( P) = q> case (ZjP)/2j cases and controls Cpj (ZjjP),
where (p(Z;P) is some function of a vector of covariates Z and the parameters of interest p.
For example, for the linear RR model with only cumulative exposure in the model,
cp(Z;P) = 1 + zp, where z is cumulative exposure and P is the exposure-response coefficient of
interest. For the log RR (i.e., log-linear) model, (p(Z;P) = e(z|3).
In contrast to log-linear RR models, for which the Wald approach was used to estimate
confidence intervals, linear RR models may not have symmetrical confidence intervals
(Laneholz and Richardsoa 2010).10 For linear RR models, a profile likelihood approach was
used to derive confidence intervals. To obtain profile likelihood confidence intervals for the first
linear piece of the two-piece linear spline model, the sample space for the beta coefficient for the
first piece (betal) of the two-piece spline model was searched manually for a beta value that was
1.92 lower in likelihood value from the optimal model, on either side of the MLE
(< MLE, > MLE). The resulting profile likelihood confidence interval is a 90% interval. The
upper bound of this interval corresponds to a 95% one-sided upper confidence limit. The profile
10 This is because beta is constrained in nonlog-linear hazard functions because the hazard cannot be less thanO;
beta in log-linear models is unconstrained because ePx will never be less thanO.
D-3
-------
likelihood bounds are time-consuming to calculate, and lower bounds and the upper bounds on
beta2 were not uniformly reported.
D.l. BREAST CANCER INCIDENCE BASED ON THE SUBCOHORT WITH
INTERVIEWS
D.l.l. Exposure Distribution among EtO-Exposed Women in Breast Cancer Incidence
Subcohort with Interviews (n = 5,139)
The estimated daily exposure to EtO across different jobs and time periods ranged from
0.05 ppm to 77 ppm. Exposure intensities from this broad range were multiplied by the length of
time in different jobs to get estimates of cumulative exposure. The duration of exposure had a
mean of 10.8 years (std. dev. 9.1), and a median of 7.4 years. The range was from 1.00 to 50.3
years. The 25th percentile was 2.8 years and the 75th percentile was 17.6 years. Multiplying
exposure intensity and exposure duration results in a wide range of cumulative exposures.
Cumulative exposure at the end of follow-up, with no lag, had a mean of 13,524
ppm-days (37.0 ppm-years), with a standard deviation of 13,254 ppm-days. These data are
highly skewed, with a range from 5 to 253,848 ppm-days. The 25th percentile is 926 ppm-days,
while the 75th is 10,206 ppm-days. Log transformation of these data results in an approximately
normal distribution of the data.
As a caveat, it should be remembered that 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. For
example, standard life-table analyses calculate cumulative exposure at different times during
follow-up for each person. Subsequently, both person-time and disease events are put into
categories of cumulative exposure. A given person may pass through many such categories,
contributing person-time to each. Poisson regression, analogous to life-table analyses (and often
based directly on output from life table programs), similarly relies on person-time (and disease
occurrence) categorized by cumulative exposure. Both of these types of analyses are inherently
categorical.
In the analyses presented here, we have used Cox regression in which age is the time
variable. The basic approach is to compare each case to a set of 100 randomly chosen controls,
whose exposure is evaluated at the same age at which the case fails (gets disease or dies of
disease). Using 100 controls generally would be expected to give the same result as the full risk
set and shortens analysis time (Steenland and Deddens. 1997). Hence, cumulative exposure is
again time dependent. For the case who fails at an early age, the cumulative exposure of the case
and many of his or her controls at that same age may be low. For the case who fails late in life,
the cumulative exposure of the case and his or her controls will be higher. When cumulative
D-4
-------
exposure is lagged so that no exposure is counted until after a lag period (e.g., 15 years) is
fulfilled, many cases and their respective controls will be "lagged out" (i.e., will have no
cumulative exposure, if the case fails at an early age). Note that Cox regression uses individual
data, and there is no inherent categorization typical of life-table analyses and Poisson regression,
although categorical analyses can still be done in Cox regression and are often useful.
For these reasons, it is difficult to describe the cumulative exposure distribution of all
subjects in the Cox regression. Controls may appear more than once matched to different cases,
and their cumulative exposure will differ each time depending on the age of the case. However,
cases only appear once in the data and their exposure distribution can be easily presented. In our
situation, we have used Cox regression with a 15-year lag to analyze breast cancer incidence.
The exposure distribution of the cases, by deciles above the lagged out category, is shown in
Table D-1. Creating deciles such that cases are equally distributed is a good a priori way of
creating categories in which rate ratios will have approximate equal variance, a desirable feature.
The lagged out cases are women who got incident breast cancer within 15 years of first exposure.
Table D-l. Distribution of cases in Cox regression for breast cancer
morbidity analysis after using a 15-year lag
Cumulative exposure,
15-year lag
Mean cumulative exposure
(ppm-days)
Number of incident breast
cancer cases
0 (lagged out)
62
>0-<364 ppm-days
178
17
364-<854 ppm-days
524
17
854—<1,379 ppm-days
1,107
17
l,379-<2,207 ppm-days
1,767
17
2,207-<3,895 ppm-days
2,918
17
3,895-<5,542 ppm-days
4,638
17
5,542-<8,012 ppm-days
6,442
17
8,012-<14,551 ppm-days
10,447
17
14,551-<25,458 ppm-days
19,506
17
>25,458 ppm-days
44,778
17
D.1.2. Lag Selection for the Breast Cancer Incidence Data
After the SAB review of the 2014 draft assessment, the issue of lag selection was
revisited. Table D-2 provides -2 log-likelihood results comparing different models with
different lags. Table D-2 also presents the Akaike information criterion (AIC) values for the
same models, to facilitate comparison with the two-piece spline models, which include an extra
D-5
-------
parameter. [The knot is preselected and is not considered a parameter in these analyses,
consistent with the SAB's concept of parsimony (SAB. 2015)1.11
For four of the eight models, the lowest -2 log-likelihoods (and AICs) occur with a lag of
15 years, consistent with the lag used in the original published paper (Steenland etal., 2003).
For both the log-linear and linear log cumulative exposure models, the lowest -2 log-likelihoods
(and AICs) occur with no lag, which is not biologically likely. For both the log-linear and linear
two-piece spline models, the lowest -2 log-likelihoods (and AICs) occur with a lag of 20 years,
but the differences between the results for the 20-year lag and the 15-year lag is small, less than
1 AIC unit in each case. Thus, for consistency in comparisons and to optimize the best fitting lag
overall, a lag of 15 years was selected for analyzing the breast cancer incidence data. Selecting
the lag time based on the strongest associations is a common statistical approach (Checkoway et
al., 2004). A lag of 15 years is also biologically plausible for a solid tumor like breast cancer.
Sensitivity of the results to choice of lag time is examined in Sections D.1.5 and D.1.6 below.
D.1.3. Modeling of Breast Cancer Incidence Data Using a Variety of Models
D. 1.3.1. Cox Regression (Log RR) Models
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 (log-)linear model, a two-piece
(log-)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 used deciles, as indicated in Table D-3. 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-3 below.
11 "in some settings the principle ofparsimony may suggest that the most informative analysis will rely uponfixing
some parameters rather than estimating them from the data. The impact of the fhed parameter choices can be
evaluated in sensitivity analyses. In the draft assessment, fixing the knot when estimating linear spline model fits
from relative risk regressions is one such example" [page 12 of SAB f2015TI.
D-6
-------
Table D-2. Minus 2 x log-likelihood results and AICs for different models
and different exposure lag times
Minus twice LL
Lag
To get AIC
0.0
5.0
10.0
15.0
20.0
LOG-LINEAR EXPOSURE MODELS
CUMEXP
1,946.5
1,945.9
1,945.5
1,944.7
1,946.0
add 12
LCUMEXP
1,943.7
1,946.8
1,944.0
1,944.2
1,947.0
add 12
SQRT_CUMEXP
1,945.1
1,944.4
1,943.6
1,941.0
1,943.1
add 12
2-PIECE
1,943.8
1,943.1
1,943.5
1,940.5
1,939.6
add 14
knot3
500.0
6,250.0
500.0
5,500.0
5,500.0
LINEAR EXPOSURE MODELS
CUMEXP
1,946.1
1,945.2
1,944.5
1,942.5
1,944.7
add 12
LCUMEXP
1,943.3
1,947.3
1,944.4
1,944.8
1,947.3
add 12
SQRT_CUMEXP
1,944.9
1,944.3
1,943.4
1,940.5
1,942.6
add 12
2-PIECE
1,943.8
1,943.2
1,942.9
1,940.4
1,939.7
add 14
knot3
500.0
6,250.0
500.0
5,750.0
5,750.0
NULL
1,967.8
NONEXPOSURE
COVARIATES
1,948.9
AIC
Lag
0.0
5.0
10.0
15.0
20.0
LOG-LINEAR EXPOSURE MODELS
CUMEXP
1,958.5
1,957.9
1,957.5
1,956.7
1,958.0
LCUMEXP
1,955.7
1,958.8
1,956.0
1,956.2
1,959.0
SQRT_CUMEXP
1,957.1
1,956.4
1,955.6
1,953.0
1,955.1
2-PIECE
1,957.8
1,957.1
1,958.1
1,954.5
1,953.6
LINEAR EXPOSURE MODELS
CUMEXP
1,958.1
1,957.2
1,956.5
1,954.5
1,956.7
LCUMEXP
1,955.3
1,959.3
1,956.4
1,956.8
1,959.3
SQRT_CUMEXP
1,956.9
1,956.3
1,955.4
1,952.5
1,954.6
2-PIECE
1,957.8
1,957.2
1,956.9
1,954.4
1,953.7
3knots for two-piece spline models were obtained by doing a grid search by increments of 500 ppm x days and then
interpolating.
CUMEXP: cumulative exposure.
LCUMEXP: log (In) cumulative exposure.
SQRT_CUMEXP: square root of cumulative exposure.
AIC: Akaike information criterion
D-7
-------
Table D-3. Categorical analysis of breast cancer incidence by deciles
(exposures lagged 15 years)
Parameter
Estimate
SE
RR
Lower RR
Upper RR
CAT 1
-0.1171
0.29340
0.88953
0.50051
1.58091
CAT 2
-0.02152
0.29716
0.97871
0.54665
1.75228
CAT 3
0.1925
0.29767
1.21226
0.67642
2.17257
CAT 4
0.1438
0.29972
1.15471
0.64172
2.07776
CAT 5
-0.00308
0.29966
0.99692
0.55410
1.79364
CAT 6
0.4381
0.30283
1.54977
0.85605
2.80568
CAT 7
0.3955
0.30573
1.48513
0.81568
2.70401
CAT 8
0.2980
0.30652
1.34711
0.73874
2.45649
CAT 9
0.5583
0.31129
1.74774
0.94950
3.21703
CAT 10
0.7732
0.31304
2.16675
1.17311
4.00199
SE = standard error.
-2 Log-likelihood = 1,937.0; degrees of freedom = 15 (10 exposure terms, 5 covariates)
AIC = 1,967.0
We then fit a cubic spline (restricted at the ends to be linear) which presents a description
of 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 etal., 2003). Five knots for the cubic spline were chosen using every
other midpoint from the categorical analysis (598, 1,774, 4,647, 11,187, and 37,668 ppm-days)
(using Steenland's 2010 cutpoints, which were slightly different from those currently used).
We then ran a two-piece (log-)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-l shows the -2 log-likelihood graphed against the knots.
In this figure, the lower peak corresponds to the highest likelihood.12
Figures D-2 and D-3 show the results of the two-piece (log-)linear, the categorical, the
(log-)linear, and the cubic spline log RR models. In these figures, the categorical points are the
midpoints of the decile categories, with the final category assigned the final cutpoint plus 50%,
using Steenland's 2010 cutpoints, so the decile RR estimates differ somewhat from those
reported in the current assessment.
12Editorial 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 x2 with degrees of freedom equal to the difference in
degrees of freedom between the two (nested)models.
D-8
-------
It appears that the two-piece log-linear curve in Figure D-2 approximates the shape of the
exposure-response seen in the decile and cubic spline log RR analyses, better than the log-linear
curve in Figure D-3.
The log-linear curve appears to have a low slope versus the other models, suggesting
possible influential observations in the upper tail of exposure. To further explore this, we
excluded from the analysis increasing amounts of the upper tail of the data using the log-linear
model (i.e., via excluding the upper 1, 2.5, 5, 10, 15, 20, and 27% of exposure) based on the
exposure distribution of the cases (the last amount, 27%, corresponds to excluding subjects with
cumulative exposure above 6,000 ppm-days, which was close to the knot in the two-piece
log-linear model [5,800 ppm-days]). The ratios of the slope (coefficient) for the linear term (log
RR model) with these exclusions versus the slope for the linear term (log RR model) with no
exclusions were 1.5, 2.3, 3.2, 2.5, 3.1, 6.1, and 9.2, respectively. As expected, the slope
increases markedly as the data are restricted to the lower range of exposure. For example, a
modified log-linear curve after excluding the upper 5% of the data is seen in Figure D-4, along
with the lull log-linear curve from Figure D-3. Nonetheless, even the log-linear curve from these
truncated data has a markedly lower slope in the low-exposure region than the two-piece
log-linear (or spline) curves. For example, inspection shows that the RR for 6,000 ppm-days is
about 1.2 for the log-linear curve from the truncated data and 1.6 from the two-piece log-linear
model. Use of the log-linear curve based on truncated data has the disadvantage of having to
choose rather arbitrarily where to truncate the data. This disadvantage is avoided by using the
two-piece log-linear model.
A two-piece log-linear model, then, is preferred for estimating risk parsimoniously in the
low-exposure region. For comparison purposes, we also show the model using the logarithm of
exposure (see Figure D-5), which we have not used for risk assessment because it is supralinear
in the low-dose region.
We also fit a square-root transformation (square root of cumulative exposure, 15-year
lag) log RR model, which is shown in Figure D-6. This model also fits the breast cancer
morbidity well, and can be used for risk assessment, but with the disadvantage that it is not linear
or approximately linear in the low-dose region. For this reason, we prefer the two-piece
log-linear curve, with is approximately linear in the low-dose region (and strictly linear in the
linear RR models discussed below). Excess lifetime risk does not vary greatly among all these
models (see below), with the exception of the log RR model with a linear term for cumulative
exposure, which is below other excess risk estimates.
D-9
-------
Log RR Model
Knot Location, CUMEXP15
Figure D-l. Likelihoods vs. knots, two-piece log-linear spline model for
breast cancer incidence.
ODEBF15
Figure D-2. Breast cancer incidence—two-piece log-linear spline model.
Plot of the two-piece log-linear spline dose-response relationship overlaid with a plot of restricted
cubic (log RR) splines with continuous exposure. Dots represent the effect of exposure grouped in
deciles. Deciles were formed by allocating cases approximately equally in ten groups,above
lagged-out cases (using Steenland's 2010 cutpoints, so the decile RR estimates differ somewhat
from thosereported in the current assessment). The v-axis is rate ratio and the x-axis is cumulative
exposure lagged 15 years, in ppm-days.
D-10
-------
CEDiEIH?I5
Figure D-3. Breast cancer incidence—log-linear (Cox regression) model.
Plot of the log-linear dose-response relationship overlaid with a dose-response relationship generated using
restricted cubic (log RR) spline model with continuous exposure. Dots represent the effect of exposure
grouped in deciles. Deciles were formed by allocating cases approximately equally in ten groups, above
lagged-out cases (using Steenland's 2010 cutpoints, so the decile RR estimates differ somewhat from those
reported in the current assessment).
RR
2.6
2.4
2.2
2.0
1.8
1.6
1.4
1.2
1.0
0 5000 10000 15000 20000 25000
CUMEXP15
Figure D-4. Breast cancer incidence—effect on log-linear model of omitting
highest exposures.
Comparison of log-linear curve (log RR = p x cumexp) with all the data (lower blue curve) and the
log-linear curve (higher red curve) after excluding those in the top 5% of exposure (>27,500 ppm-days).
Comparing log linear models, model with higher slope omits highest 5% of exposure
^
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—r
D-ll
-------
Breast cancer morbidity log transformed
- 2 I og I i kel i hood i s 1944. 153
Ckt egor i cal anal yses over I ayed
CLfvGxFI 5
Figure D-5. Breast cancer incidence—log-linear model with log cumulative
exposure.
Plot of a logarithmic transformation log RR dose-response model [log RR = p x log(cumexp)] overlaid
with categorical RR results (deciles). Deciles were formed by allocating cases approximately equally in ten
groups, above lagged-out cases (using Steenland's 2010 cutpoints, so the decile RR estimates differ
somewhat from those reported in the current assessment).
D-12
-------
Breast cancer morbidity sqrt root transformed
-2 I og I i kel i hood i s 1941. 02S
C&t egor i cal anal yses over I ayed
0 5000 1OOOO 15000 20000 25000 30000 35000 40000
Figure D-6. Breast cancer incidence—log-linear model with square root of
cumulative exposure.
Plot of a square-root transformation log RR dose-response model overlaid with categorical RR results
(deciles). Deciles were formed by allocating cases approximately equally in ten groups, above lagged-out
cases (using Steenland's 2010 cutpoints, so the decile RR estimates differ somewhat from those reported in
the current assessment).
Tables D-4, D-5, D-6, and D-7 below present the model fit statistics for the two-piece
log-linear, the log-linear, the square root log RR model, and the log-transform log RR model
seen above. Table D-8 summarizes the goodness-of-fit data with regard to the exposure term.
Table D-8 shows that the addition of exposure terms to the various models results in similar
model fits. The exposure terms in the two-piece log-linear model improve model fit marginally
better than those in the other models except the square root log RR model, with which the
two-piece log-linear model is tied. If one adds a degree of freedom to the x2test for the two-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 two-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 two-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.
D-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 Section 4.1.2.3 of the assessment (with the exception of the cubic spline results, which
are not part of Section 4.1.2.3, Steenland's original risk assessment sections were deleted
because they were based on older mortality and disease rates than were the analyses presented in
the current Section 4.1.2.3). 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, presented in
Section D.1.4 below, provides a good comparison to other parametric models.
Table D-4. Fit of two-piece log-linear model to breast cancer incidence data,
Cox regression13
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
1,967.813
1,940.485
AIC
1,967.813
1,954.485
SBC
1,967.813
1,978.612
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
27.3281
7
0.0003
Score
29.0949
7
0.0001
Wald
28.4426
7
0.0002
Analysis of maximum likelihood estimates
Variable
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
LIN_0 (pi)
0.0000770
0.0000317
5.4642
0.0194
1.000
LIN_1
-0.0000724
0.0000334
4.1816
0.0409
1.000
DOB1
0.08770
0.21805
0.1618
0.6875
1.092
DOB2
0.41958
0.24430
2.9496
0.0859
1.521
DOB3
0.55168
0.29096
3.5950
0.0580
1.736
PARITY1
-0.23398
0.18793
1.5502
0.2131
0.791
FREL_BR_ CAN 1
0.47341
0.17934
6.9686
0.0083
1.605
Covariance linO andlinl: -1 x 10 9
13For environmental exposures, only exposures below the knot are of interest. Below the knot, RR = e®1 * e??50sure).
D-14
-------
Table D-5. Fit of log-linear model to breast cancer incidence data, Cox
regression (RR = e(PXexP°sure))
Model fit statistics
Criterion
Without covariates
With covariates
-2 LOG L
1,967.813
1,944.675
AIC
1,967.813
1,956.675
SBC
1,967.813
1,977.356
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
23.1374
6
0.0008
Score
25.8389
6
0.0002
Wald
25.3594
6
0.0003
Analysis of maximum likelihood estimates
Variable
Parameter estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
CUMEXP15 (P)
9.5482E-6
4.09902E-6
5.4261
0.0198
1.000
DOB1
0.13558
0.21676
0.3912
0.5316
1.145
DOB2
0.53147
0.23741
5.0116
0.0252
1.701
DOB3
0.74477
0.27425
7.3748
0.0066
2.106
PARITY
-0.23394
0.18882
1.5351
0.2154
0.791
FREL_BR_ CAN 1
0.46449
0.17928
6.7126
0.0096
1.591
D-15
-------
Table D-6. Fit of the square root transformation log RR model to breast
cancer incidence data, Cox regression (RR = e^Xsqrt(exposure®)
Model fit statistics
Crtierion
Without
covariates
With covariates
-2 LOG L
1,967.813
1,941.028
AIC
1,967.813
1,953.028
SBC
1,967.813
7,973.708
Testing global null hypothesis: BETA =0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
26.7851
6
0.0002
Score
28.9446
6
< .0001
Wald
28.5277
6
< .0001
Analysis of maximum likelihood estimates
Variable
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
DOB1
1
0.09778
0.21756
0.2020
0.6531
DOB2
1
0.43872
0.24177
3.2929
0.0696
DOB3
1
0.58623
0.28404
4.2596
0.0390
sqrtcumexpl5 (P)
1
0.00349
0.00118
8.7489
0.0031
PARITY1
1
-0.22539
0.18787
1.4393
0.2302
FREL_BR_ CAN 1
1
0.46937
0.17922
6.8589
0.0088
D-16
-------
Table D-7. Fit of the log-trans form model to breast cancer incidence data,
Cox regression (RR= e(Pxln(exP°sure)))
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
1,967.813
1,944.176
AIC
1,967.813
1,956.176
SBC
1,967.813
1,976.856
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
23.6371
6
0.0006
Score
24.0044
6
0.0005
Wald
23.5651
6
0.0006
Analysis of maximum likelihood estimates
Parameter
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
DOB1
1
0.08605
0.21943
0.1538
0.6949
1.090
DOB2
1
0.38780
0.25363
2.3378
0.1263
1.474
DOB3
1
0.47303
0.31528
2.2509
0.1335
1.051
LCUMEXP15 (P)
1
0.04949
0.02288
4.6787
0.0305
1.051
PARITY1
1
-0.25908
0.18638
1.9322
0.1645
0.772
FREL_BR_ CAN 1
1
0.47620
0.17923
7.0595
0.0079
1.610
Table D-8. Change in -2 log-likelihood for log RR models for breast cancer
incidence, with addition of exposure term(s)a
Log RR model
Change (x2)
DF
/j-val uc
Log transform
4.8
1
0.03
Linear
4.2
1
0.04
Categorical
12.0
10
0.29
Cubic spline
8.8
4
0.07
Two-piece linear
8.4
2
0.01
Square root
7.7
1
0.006
aAll models had 3 variables for date ofbirth, 1 for family history, and 1 for parity.
D. 1.3.2. Linear Relative Risk Models for Breast Cancer Incidence
We also ran linear relative risk models for breast cancer incidence, using the techniques
described by Lanuholz 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
D-17
-------
can be cumulative dose, the log of cumulative dose, a two-piece linear function of cumulative
dose, and so on.
To choose the knot for the linear two-piece spline model, Deddens examined inflection
points over the range 500 to 10,000 ppm-days by 500 ppm-day increments, and then interpolated
where appropriate.
Figure D-7 shows the likelihood profile for different possible knots for the two-piece
linear spline, with the search conducted by using increments of 500 ppm-days. The best knot
was 5,750 ppm-days, similar to the knot of 5,800 ppm-days for the two-piece log-linear model.
Table D-9 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 x2
does not attain significance at the 0.05 level. For the two-piece linear model, the model
likelihood is 1940.36 versus a likelihood of 1940.49 for the two-piece log-linear model. Among
the linear RR models, the two-piece spline model fits better than the other models, although not
significantly so. Table D-10 gives the exposure parameter values for the linear RR models.
fS
—'—r—
mm
¦
Figure D-7. Likelihoods vs. knots, two-piece linear spline model, breast
cancer incidence (units are ppm-days, 15-year lag).
D-18
-------
Table D-9. Model fit statistics for linear RR models, breast cancer
incidence"
Linear RR model
DF (full
model)b
-2 LL (full
model)
-2 LL (model
without
exposure)
-2 LL
(model
without
any
covariates)
/j-valuc
(full model)
/>-value
(for addition
of exposure
terms)0
CUMEXP15
6
1,942.526
1,948.935
1,967.813
0.0030
0.0113
Sqrt(CUMEXP 15)
6
1,940.501
1,948.935
1,967.813
0.0001
0.0037
Spline, knot = 5,750,
CUMEXP15
7
1,940.360
1,948.935
1,967.813
0.0003
0.0137
aFor the linear RR models, all nonexposure covariates were included multiplicatively.
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)
LL = log likelihood
Table D-10. Model coefficients for linear RR models, breast cancer
incidence
Linear RR model
Parameter(s)
SE
Profile likelihood 95%
(one-sided) upper bounds0
CUMEXP15
B = 2.2964 x 10~5
SE= 1.210 x 10~5
UB = 4.666 x 10~5
Sqrt(CUMEXP 15)
B = 5.531 x 10~3
SE = 2.585 x 10~3
UB = 0.01067
Spline, knot = 5,750,
CUMEXP15ab
B1 = 8.9782 x 10~5
B2 = -7.7859 x 10~5
SE1 = 5.378 x 10~5
SE2 = 5.930 x 10~5
UB1 = 1.837 x 10~4
UB2 = 4.309 x 10~6
SE = standard error.
aVarl = SE12 = 2.892 x 10 9: Var2 = SE22 = 3.516 x 10 9: Covariance = -3.11 x 10 9,
bFor estimating risks from occupational exposures (see Section 4.7), both pieces of the two-piece linear spline
model are used. For the maximum likelihood estimate, for exposures below the knot, RR = 1 + (B1 x exp); for
exposures above the knot, RR = 1 + (B1 x exp + B2 x [exp - knot]). For the (one-sided) 95% upper confidence
limit, the Wald approach is used as an approximation because it was not possible to obtain a formula for the profile
likelihood upper-bound estimates that could be used in the life-table analysis. Thus, for exposures below the knot,
RRu = 1 + ([B1+ 1.645 x SE1] x exp); for exposures above the knot, RRu = 1 + (B1 x exp + B2 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. As shown in Figure D-8, the difference between the Wald
upper-bound estimates and the profile likelihood upper-bound estimates is not large. In the range of occupational
exposures of interest (i.e., up to 12,775 ppm x days), the Wald RRu estimate is at most ~5% less than the profile
likelihood RRu estimate (at the extreme, i.e., at 12,775 ppm x days).
"Calculating the profile likelihood bounds is computationally difficult and estimating the lower bounds was not
pursued here.
D-19
-------
3.50
Figure D-8. Comparison ofWald and profile likelihood (one-sided) 95%
upper-bound estimates for 2-piece linear spline model.
D.1.4. Risk Assessment for Breast Cancer Incidence Using the Cubic Spline Curve Log RR
Model
Risk assessment using the spline curve is more difficult due to the semiparametric
complicated nature of the restricted cubic spline function. The cubic spline function for the
breast cancer incidence rate ratio is:
RR=exp((ns_0*cumexpl5) + ns_l*(((cumexpl5-598)**3)*(cumexpl5>= 598) -
((37668-598) /(37668-11187)) *(((cumexpl5-11187)**3) *(cumexpl5>= 11187)) +
((11187 -598)/(37668 - 11187)) *(((cumexpl5-37668 )**3) *(cumexpl5>= 37668))
) + ns_2*(((cumexpl5-1774)**3)*(cumexpl5>= 1774) - ((37668-1774) /(37668-
11187)) *(((cumexpl5-11187)**3) *(cumexpl5>= 11187)) + ((11187 -1774) /(37668
- 11187))*(((cumexpl5-37668 )**3) *(cumexpl5>= 37668)) ) + ns_3*(((cumexpl5-
4647)**3)*(cumexpl5>= 4647) - ((37668-4647) /(37668-11187)) *(((cumexpl5-
11187)**3) *(cumexpl5>= 11187)) + ((11187 -4647)/(37668 - 11187))
*(((cumexpl5-37668 )**3) *(cumexpl5>= 37668)) )).
D-20
-------
The coefficients ns_0, ns_l, ns_2, and ns_3 used in this function are 0.00008294999811,
-0.00000000000310, 0.00000000000425, and -0.00000000000114, respectively. The
expression "cumexpl5>=" is a logical statement whereby the term is 0 when "cumexp" is less
than the specified value.
Here we calculate only the ECoi (without the LECoi) for comparison with the
corresponding ECoi from the two-piece log-linear model. The point is to show that the cubic
spline log RR model and the two-piece log-linear spline give similar answers, not to propose the
cubic spline for use in risk assessment, given its relatively complicated formula above.
Calculation of the LECoi is also particularly complicated because to doit correctly one must use
not only the standard errors 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 two-piece log-linear model. [Note that although
these ECoi values are internally consistent for the comparison made here, they are not directly
comparable to values reported in Chapter 4 because the calculations presented here were made
using background mortality and incidence rates from 1997-2001 and were not updated for the
current assessment. Nonetheless, the difference between the value of 0.0152 ppm presented here
for the two-piece log-linear model and the value of 0.0155 ppm reported in Chapter 4 is
negligible.]
D.1.5. Supplemental Results: Results for Cumulative Exposure and Log Cumulative
Exposure Cox Regression Models with Different Lag Times
Sensitivity of the exposure parameter estimates to choice of exposure lag time (no lag,
5 years, 10 years, 15 years, and 20 years) for the log-linear cumulative exposure (standard Cox
regression) and log cumulative exposure models is summarized in Table D-11.
D-21
-------
Table D-ll. Comparison of some log-linear model results with
different lag periods; cumulative exposure in ppm x days
Likelihood
Exposure
Standard
ratio test
parameter
error
-2
p-value for
estimate (per
(per unit
Model
Lag (years)
log-likelihood
exposure term
unit exposure)
exposure)
Log-linear
cumulative
0
1,946.492
0.09
5.93879 x 10~6
3.52892 x 10~6
5
1,945.875
0.06
6.8565 x 10~6
3.59626 x 10~6
exposure
10
1,945.521
0.04
7.75726 x 10~6
3.80799 x 10~6
(standard Cox
model)
15
1,944.675
0.02
9.54826 x 10~6
4.09902 x 10~6
20
1,946.040
0.055
1.01 x 10~5
5.27041 x 10~6
0
1,943.662
0.02
0.09294
0.04097
Log-linear log
5
1,946.843
0.16
0.04458
0.03135
cumulative
10
1,944.040
0.03
0.05654
0.02594
exposure
15
1,944.176
0.03
0.04949
0.02288
20
1,947.020
0.17
0.02970
0.02151
D.1.6. Sensitivity of Unit Risk Estimates to Change in Lag Period
Sensitivity of the unit risk estimates to changes in exposure lag time for the two-piece
linear spline model with the knot at 5,750 ppm x days is summarized in Table D-12.
Table D-12. Comparison of unit risk estimates from two-piece linear
spline models with different lag periods; cumulative exposure in ppm x
days, knot at 5,750 ppm x days
Lag
(years)
-2
log-likelihood
Parameter
estimate for 1st
spline segment
(per ppm x day)
Profile likelihood 95%
one-sided
upper-hound estimate
for 1st spline segment
(per ppm x day)
ECoi
(PP"i)
LEG,,
(PP"i)
Unit risk
estimate
(per ppm)
0
1,944.5
5.9472 x 10~5
1.5063 x 10~4
0.0160
6.29 x 10~3
1.59
5
1,943.2
6.7229 x 10~5
1.5589 x 10~4
0.0153
6.59 x 10~3
1.52
10
1,943.9
4.5903 x 10~5
1.2655 x 10~4
0.0245
8.88 x 10 3
1.13
15
1,940.4
8.9782 x 10~5
1.8372 x 10~4
0.0138
6.75 x 10~3
1.48
20
1,939.6
1.3725 x 10~4
2.4727 x 10~4
0.0101
5.59 x 10 3
1.79
The sensitivity analysis for choice of lag reveals that the unit risk estimates for the
selected two-piece linear spline model with the knot at 5,750 ppm x days for different lag
periods (0, 5, 10, 15, and 20 years) ranged from about 35% less than (10-year lag) to about 21%
D-22
-------
greater than (20-year lag) the estimate for the selected model (15-year lag). Of these specific
models, the model with the 20-year lag was the best fitting model, based on log likelihood. The
models for lags of 0, 5, and 10 years had ^-values > 0.05 for inclusion of the exposure terms
(0.11, 0.057, and 0.080, respectively).
The optimal knot for the two-piece linear spline model with a 20-year lag was the same
as that for the model with a 15-year lag [i.e., 5,750 ppm x days (see Table D-2)]. For lags of 0
and 10 years, the optimal knot was in the vicinity of 500 ppm x days. Using this lower knot
would have yielded higher regression coefficients for the 1st spline segment and correspondingly
higher unit risk estimates. For the lag of 5 years, the optimal knot was slightly higher (6,250
ppm x days) than the knot for the selected model (5,750 ppm x days), which would have yielded
a slightly lower unit risk estimate than that presented in Table D-12. However, even with the
optimal knot, the models for lags of 0 and 5 years had ^-values > 0.05 for the exposure terms
(0.077 and 0.057, respectively), and the model with a lag of 10 years had p = 0.049.
D.1.7. Sensitivity of Unit RiskEstimates to Value of Knot
Sensitivity of the unit risk estimates to value of knot for the two-piece linear spline model
is summarized in Table D-13, with knots of 5,750 ± 1,000 ppm x days.
Table D-13. Comparison of unit risk estimates from two-piece linear
spline models with different knot; cumulative exposure in ppm x days,
with lag of 15 years
Knot
(ppm x
days)
-2
log-likelihood
Parameter
estimate for 1st
spline segment
(per ppm x day)
Profile likelihood 95%
one-sided
upper-bound estimate
for 1st spline segment
(per ppm x day)
ECoi
(ppm)
LEG,,
(PI*n)
Unit risk
estimate
(per ppm)
4,750
1,940.43
1.0008 x 10~4
2.0898 x 10~4
0.0124
5.93 x 10 3
1.69
5,750
1,940.36
8.9782 x 10~5
1.8372 x 10~4
0.0138
6.75 x 10 3
1.48
6,750
1,940.48
8.0280 x 10~5
1.6357 x 10~4
0.0154
7.58 x 10 3
1.32
The sensitivity analysis for knot selection in the two-piece linear spline model shows
very little difference in the unit risk estimates for knots 1,000 ppm x days below and above the
selected knot of 5,750 ppm x days. The unit risk estimates for these alternate knot values are
about 14% greater and 11% lower, respectively, than the unit risk estimate for the selected model
(with the knot at 5,750 ppm x days).
D-23
-------
D.1.8. Sensitivity of Unit Risk Estimates to Exclusion of Covariates
Sensitivity of the unit risk estimates to exclusion of (nonexposure) covariates (i.e.,
significant breast cancer risk factors) for the two-piece linear spline model is summarized in
Table D-14.
Table D-14. Comparison of unit risk estimates from two-piece linear spline
models with exclusion of nonexposure covariates; cumulative exposure in ppm
x days with 15-year lag, knot at 5,750 ppm x days
Excluded
covariates
-2
log-likelihood
Parameter
estimate for 1st
spline segment
(per ppm x day)
Profile likelihood 95%
one-sided
upper-bound estimate for
1st spline segment (per
ppm x day)
ECoi
(ppm)
LEG,,
(ppm)
Unit risk
estimate
(per ppm)
None
1,940.4
8.9782 x 1CT5
1.8372 x 10~4
0.0138
6.75 x 10 3
1.48
Parity
1,941.8
9.0441 x 10~5
1.8516 x 10~4
0.0137
6.70 x 10 3
1.49
Parity and breast
cancer in
1st-degree relative
1,948.0
8.7427 x 10~5
1.8109 x 10~4
0.0142
6.84 x 10 3
1.46
The sensitivity analysis for exclusion of covariates in the two-piece linear spline model
shows very little difference in the unit risk estimates. Excluding parity and both parity and
breast cancer in a first-degree relative would change the unit risk estimate by only about 1%
from the unit risk estimate derived for the selected model (i.e., with inclusion of both covariates).
D.1.9. Analysis of Age Interaction for the Exposure Terms in the Two-piece Linear Spline
Model
Table D-15 shows the ^-values for the inclusion of age interaction terms for the spline
exposure regression coefficients. The interaction terms have ^-values well above 0.05,
indicating that the exposure terms are independent of age (i.e., the proportional hazards
assumption is validated).
D-24
-------
Table D-15. Evaluation of age interaction for the exposure terms in the
two-piece linear spline model with knot at 5,750 ppm x days; cumulative
exposure in ppm x days, with lag of 15 years
Parameter
-2 log likelihood without
age interaction term
-2 log likelihood with
age interaction term
Difference in-2 log
likelihoods
/j-value for the inclusion
of age interaction term
Betal
1940.360
1940.167
0.193
0.66
Beta2
1940.360
1940.284
0.076
0.78
D.1.10. Sensitivity of Unit Risk Estimates to Upper-Bound Estimation Approach—Wald
vs. Profile Likelihood
Sensitivity of the unit risk estimates to the approach used to estimate the upper bound on
the first spline piece from the two-piece linear spline model is summarized in Table D-16.
According to Laneholz and Richardson (2010), the distribution of estimated parameters in
nonlog-linear models (hazard functions) is often not symmetrical (because beta is constrained so
that the hazard cannot be less than 0) and profile likelihood confidence intervals are
recommended as being more accurate than Wald-type intervals. The Wald-based result is 3%
lower than the profile-likelihood-based estimate.
Table D-16. Comparison of unit risk estimates for breast cancer incidence
from two-piece linear spline model using Wald-based and
profile-likelihood-based upper-bound estimates on the 1st spline piece
Estimation
approach
Betal
estimate
(per
ppm x day)
Wald SE1
estimate
(per ppm x day)
95% one-sided
upper-hound estimate
for 1st spline segment
(per ppm x day)
ECoi
(ppm)
LEG,,
(PI*n)
Unit risk
estimate
(per ppm)
Wald
8.98 x 10~5
5.38 x 10~5
1.78 x 10~4
0.0138
6.95 x 10 3
1.44
Profile likelihood
8.98 x 10~5
--
1.84 x 10~4
0.0138
6.75 x 10 3
1.48
D.l.ll. Sensitivity of Occupational Extra Risk Estimates to Change in Lag Period
In Section 4.7, extra risk estimates from the selected model are presented for some
occupational exposure scenarios of interest (3 5-year exposures to 8-hour TWAs ranging from 0.1
to 1 ppm between ages 20 and 55 years), because the scenarios include cumulative exposures
above the level at which the unit risk estimate is valid. Here, the sensitivity of the selected
model (i.e., the two-piece linear spline model with the knot at 5,750 ppm x days and a lag of
D-25
-------
15 years) to changes in lag is explored. Parameter estimates for the two-piece linear spline
model with the knot at 5,750 ppm x days and different lag periods (0, 5, 10, 15, and 20 years) are
presented in Table D-17. The Wald approach was used as an approximation to derive the
upper-bound estimates because it was not possible to obtain a formula for the profile likelihood
upper-bound estimates that could be used in the life-table analysis. As shown in Figure D-8, the
difference between the Wald upper-bound estimates and the profile likelihood upper-bound
estimates is not large (the Wald upper-bound RR estimates are about 4% lower than the profile
likelihood upper-bound RR estimates in the region of the second spline segment, and the
difference is even less than that in the region of the first spline segment). The equations for
deriving the MLE and upper-bound estimates across the range of exposures are presented in
footnote 2 of Table D-10. Sensitivity of the extra risk estimates for the occupational exposure
scenarios to changes in exposure lag time for the two-piece linear spline model with the knot at
5,750 ppm x days is summarized in Table D-18.
Table D-17. Parameter estimates for the two-piece linear spline model with
the knot at 5,750 ppm x days for different lag periods; cumulative exposure
in ppm x days
Lag (years)
Betal
(per ppm x day)
Beta2
(per ppm x day)
SE1
(per ppm x day)
SE2
(per ppm x day)
Covariance
[per (ppm x day)2]
0
5.947 x 10-5
-5.472 x 10-5
4.606 x 10-5
4.889 x 10-5
-2.23 x 10-9
5
6.723 x lO-5
-6.134 x lO-5
4.602 x lO-5
4.913 x lO-5
-2.23 x 10-9
10
4.590 x 10-5
-3.544 x 10-5
4.402 x 10-5
4.797 x 10-5
-2.07 x 10-9
15
8.978 x lO-5
-7.786 x lO-5
5.378 x lO-5
5.930 x lO-5
-3.11 x 10-9
20
1.373 x 10-4
-1.343 x 10-4
6.387 x 10-5
6.969 x 10-5
-4.36 x 10-9
D-26
-------
Table D-18. Comparison of breast cancer incidence extra risk estimates from two-piece linear spline models
with different lag periods; cumulative exposure in ppm x days, knot at 5,750 ppm x days
O
to
8-hour TWA
15-year lag
0-year lag
Ratio to 15-
year-lagged
estimates
5-year lag
Ratio to 15-
year-lagged
estimates
10-year lag
Ratio to 15-
year-lagged
estimates
20-year lag
Ratio to 15-
year-lagged
estimates
MLEs
0.1
0.0128
0.0106
0.83
0.0114
0.89
0.00728
0.57
0.017
1.33
0.2
0.0255
0.0211
0.83
0.0227
0.89
0.0145
0.57
0.0336
1.32
0.3
0.0379
0.0315
0.83
0.0338
0.89
0.0217
0.57
0.0499
1.32
0.4
0.0502
0.0417
0.83
0.0448
0.89
0.0288
0.57
0.0659
1.31
0.5
0.0595
0.0481
0.81
0.052
0.87
0.034
0.57
0.0786
1.32
0.6
0.0643
0.0498
0.77
0.0545
0.85
0.0368
0.57
0.0854
1.33
0.7
0.068
0.0511
0.75
0.0565
0.83
0.0393
0.58
0.0901
1.33
0.8
0.0708
0.0521
0.74
0.0578
0.82
0.0413
0.58
0.0929
1.31
0.9
0.0736
0.053
0.72
0.0592
0.80
0.0433
0.59
0.0957
1.30
1
0.0757
0.0539
0.71
0.0602
0.80
0.045
0.59
0.0973
1.29
95% one-sided UCLs
0.1
0.0253
0.024
0.95
0.0241
0.95
0.0186
0.74
0.0298
1.18
0.2
0.0498
0.0473
0.95
0.0476
0.96
0.0369
0.74
0.0585
1.17
0.3
0.0736
0.07
0.95
0.0704
0.96
0.0547
0.74
0.0862
1.17
0.4
0.0967
0.0921
0.95
0.0926
0.96
0.0722
0.75
0.113
1.17
0.5
0.114
0.105
0.92
0.106
0.93
0.0841
0.74
0.134
1.18
0.6
0.121
0.107
0.88
0.11
0.91
0.0886
0.73
0.144
1.19
0.7
0.126
0.109
0.87
0.113
0.90
0.092
0.73
0.151
1.20
0.8
0.13
0.109
0.84
0.114
0.88
0.0943
0.73
0.155
1.19
0.9
0.133
0.11
0.83
0.115
0.86
0.0967
0.73
0.16
1.20
1
0.136
0.111
0.82
0.116
0.85
0.0984
0.72
0.162
1.19
-------
The sensitivity analysis for choice of lag reveals that the MLEs of extra risk for the
selected two-piece linear spline model with the knot at 5,750 ppm x days for different lag
periods (0, 5, 10, 15, and 20 years) ranged from about 40% less than (10-year lag) to about 30%
greater than (20-year lag) the estimates for the selected model (15-year lag). The 95% (one-
sided) upper bounds of extra risk ranged from about 25% less than (10-year lag) to about 20%
greater than (20-year lag) the estimates for the selected model. Of these specific models, the
model with the 20-year lag was the best fitting model, based on log likelihood (see Table D-12).
The models for lags of 0, 5, and 10 years had ^-values > 0.05 for inclusion of the exposure terms
(0.11, 0.057, and 0.080, respectively).
For lags of 0 and 10 years, the optimal knot was in the vicinity of 500 ppm x days, and
for the lag of 5 years, the optimal knot was slightly higher (6,250 ppm x days) than the knot for
the selected model (5,750 ppm x days). Comparisons to extra risk results with these different
optimal knots cannot be made without knowledge of the parameter estimates for the regression
coefficients, which the EPA did not obtain from NIOSH because these additional analyses are
outside of the scope of the intended lag sensitivity analyses, as changing the knot results in an
entirely different (and inferior in terms of likelihood) model. With the optimal knot, the models
for lags of 0 and 5 years had ^-values > 0.05 for the exposure terms (0.077 and 0.057,
respectively), and the model with a lag of 10 years had p = 0.049. The exception is the two-piece
linear spline model with a 20-year lag, which had the optimal knot at the same value as two-
piece linear spline model with a 15-year lag (i.e., 5,750 ppm x days [see Table D-2]). This two-
piece linear spline model with a 20-year lag is the best fitting model of all the two-piece spline
models with optimal knots and all the models with the knot at 5,750 ppm x days but with
different lags. As noted above, for the occupational exposure scenarios of interest, the two-piece
linear spline model with a 20-year lag yielded MLEs of extra risk about 30% greater than and
95% (one-sided) upper bounds about 20% greater than those for the selected model (15-year
lag).
D.2. BREAST CANCER MORTALITY
D.2.1. Exposure Distribution among Women and Breast Cancer Deaths in the Cohort
Mortality Study (n = 9,544)
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. Table D-19
shows the distribution of the 102 breast cancer deaths by exposure quartile with a 20-year lag.
The cutpoints are those used in the published data (Steenland et aL 2004). Regarding the
women in the cohort as a whole, cumulative exposure at the end of follow-up, with no lag, had a
D-28
-------
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.
Table D-19. Distribution of cases in Cox regression analysis of breast cancer
mortality after using a 20-year lag
Cumulative exposure, 20-year lag3
Number of breast cancer deaths
0 (Lagged out)
42
>0-646 ppm-days
17
647-2,779 ppm-days
16
2,780-12,321 ppm-days
15
>12,321 ppm-days
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, and26,676 ppm x days. These values are for
the risk sets but should provide a good approximation to the full cohort values.
D.2.2. Modeling of Breast Cancer Mortality Data Using a Variety of Models
D .2.2.1. Cox Regression (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 (log-)linear model, a two-piece
(log-)linear model, a log-transform model, and a cubic spline model. We also ran a number of
analogous models using linear RR models (see Section D.2.2.2 below).
The categorical log RR model for breast cancer mortality was run using the originally
published cutpoints to form four categories above the lagged-out group, as shown in Table D-19.
To graph the categorical points, each category was assigned the midpoint of the category as its
exposure level, except for the last one which was assigned 50% more than the last cutpoint
12,322 ppm-days.
For the two-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 (see Figure D-9). We
also explored knots beyond 7,000 ppm-days by looking at increments of 1,000 ppm-days from 0
to 25,000 (see Figure D-10 shows the results for knots up to 15,000 ppm-days). None of these
outperformed the knot at 700 ppm-days, although Figure D-10 suggests a local maximum
likelihood near 13,000 ppm-days.
D-29
-------
-2 log likelihood for different knots for breast cancer mortality
•-|—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—p
0 1000 2000 3000 4000 5000 6000 7000
KNOT
Figure D-9. Likelihoods vs. knots for the two-piece log-linear model, breast
cancer mortality.
-2 log likelihood for different knots for breast cancer mortality
KNOT
Figure D-10. Likelihoods vs. knots for the two-piece log-linear model, breast
cancer mortality, up to 15,000 ppm-days.
In Figure D-ll below, we show the categorical and two-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.
D-30
-------
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% cutoflF 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 is responsible for the attenuation seen in
the exposure-response at high exposures.
The two-piece log-linear spline model in Figure D-ll 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).
3.5
2.5
5,000
10,000 15,000 20,000
CUMEXP20
LogRR
• Categorical
— — Log RR, 95% cutoff
Log RR, Spline w/ Knot @ 700
25,000 30,000
Figure D-ll. Dose-response models for breast cancer mortality.
Plot of the dose-response relationship of continuous exposure (lagged 20 years) for breast cancer mortality,
with the two-piece log-linear spline, the categorical, and the log-linear 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). Dots represent the categorical results (quartiles).
For comparison purposes, we also show the logarithmic transformation log RR model in
Figure D-12 (which we have not used for risk assessment because it is supralinear in the low
dose region).
D-31
-------
Breast cancer mortality log transformed
-2 I og I i kel i hood is 917. 770
Ckt egor i cal anal yses over I ayed
FR
4 "
.++t-t
H+44++H+H
++
Figure D-12. Breast cancer mortality—log-linear model with log cumulative
exposure.
Plot of the dose-response relationship of continuous exposure (lagged 20 years) for breast cancer mortality,
using a logarithmic transformation log RR model. Dots represent the categorical results (quartiles).
Outputs from the categorical, two-piece log-linear spline, and log-linear RR models are
given below in Tables D-20 to D-24. The two-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, as judged by
conformity with categorical and cubic spline analysis, and may still be useful for risk assessment.
D-32
-------
Table D-20. Categorical output breast cancer mortality, 20-year lag
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
923.433
915.509
AIC
923.433
923.509
SBC
923.433
934.009
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
7.9244
4
0.0944
Score
8.5160
4
0.0744
Wald
8.3993
4
0.0780
Analysis of maximum likelihood estimates
Variable
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
CUM201
1
0.56653
0.33920
2.7894
0.0949
1.762
CUM202
1
0.57236
0.35505
2.5987
0.1070
1.772
CUM203
1
0.67537
0.37632
3.2207
0.0727
1.965
CUM204
1
1.14110
0.40446
7.9598
0.0048
3.130
Table D-21. Two-piece log-linear spline, breast cancer mortality, 20-year lag,
knot at 700 ppm-days
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
923.433
918.037
AIC
923.433
922.037
SBC
923.433
927.287
Testing global null hypothesis: BETA =0
Test
X2
DF
Pr> ChiSq
Likelihood
5.3967
2
0.0673
Score
6.0153
2
0.0494
Wald
5.8857
2
0.0527
Anlysis of maximum likelihood estimates
Parameter
Parameter
estimate
Standard error
X2
Pr> ChiSq
Hazard ratio
LIN_0
0.0006877
0.0004171
2.7178
0.0992
1.001
LIN_1
-0.0006782
0.0004188
2.6229
0.1053
0.999
3Covariance linO and linl: -1.75 x 10 7
D-33
-------
Table D-22. Log-linear model, breast cancer mortality, 20-year lag
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
923.433
920.647
AIC
923.433
922.647
SBC
923.433
925.272
Testing global null hypothesis: BETA =0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
2.7865
1
0.0951
Score
3.7383
1
0.0532
Wald
3.6046
1
0.0576
Analysis of maximum likelihood estimates
Variable
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
CUMEXP20
0.0000122
6.40812E-6
3.6046
0.0576
1.000
Table D-23. Log-transform log RR model, breast cancer mortality, 20-year lag
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
923.433
917.743
AIC
923.433
919.743
SBC
923.433
922.368
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr> ChiSq
Likelihood ratio
5.6908
1
0.0171
Score
5.7676
1
0.0163
Wald
5.7688
1
0.0163
Analysis of maximum likelihood estimates
Parameter
DF
Parameter
estimate
Standard error
X2
Pr> ChiSq
Hazard ratio
LCUM20
1
0.08376
0.03487
5.7688
0.0163
1.087
D-34
-------
Table D-24. Two-piece log-linear spline model, breast cancer mortality, 20-
year lag, knot at 13,000 ppm-days
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
923.433
918.237
AIC
923.433
922.237
SBC
923.433
927.487
Testing global null hypothesis: BETA =0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
5.1963
2
0.0744
Score
5.9044
2
0.0522
Wald
5.7813
2
0.0555
Analysis of maximum likelihood estimates
Variable
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
LIN_0
0.0000607
0.0000309
3.8539
0.0496
1.000
LIN_1
-0.0000583
0.0000371
2.4761
0.1156
1.000
D.2.2.2. Linear Relative Risk Models for Breast Cancer Mortality
Finally, we also ran linear RR models for these data, as shown in Figure D-13 below
(denoted "ERR" models), which also includes the RRs from the log RR categorical model as
shown in other graphs. Again, the linear curve, highly influenced by the upper 5% tail of
exposure, underestimates the categorical points, while the log transform and two-piece spline
capture better the initial increase in risk followed by an attenuation. Parameter estimates for
these models can be found in Table D-25.
D-35
-------
• Categorical
Spline ERR, Knot=700, CUMEXP20
ERR, CUMEXP20
ERR, Log(CUMEXP20)
5,000
10,000 15,000 20,000 25,000 30,000
CUMEXP20
Figure D-13. Linear RR models for breast cancer mortality.
[Editorial note: "ERR" refers to linear RR models.]
Table D-25. Model results for breast cancer mortality, linear RR
modelsb
linear RR model
Parameter(s)
SE
-2 Log-likelihood
CUMEXP20
B = 2.6779 x 10-5
SE = 2.1537 x 10-5
920.122
Log(CUMEXP20)
B = 0.122090
SE = 0.061659
917.841
Spline, knot = 700,
CUMEXP203
B1 = 8.30 x 10-4,
B2 = -8.07 x 10-4
SE1 =6.14 x 10-4,
SE2 =6.19 x 10-4
918.058
SE = standard error.
aCovariance 2 pieces of spline: -3.80 x 1 Or1.
bEditorial note: 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, the EPA used the profile likelihood approach for the linear RR models (Langholz
and Richardson. 20101 which allows for asymmetric CIs. The unit risk estimate for breast cancer mortality
presented in this assessment does not rely on any of the linear RR models, thus CIs calculated using the
profile likelihood method are not shown here.
D-36
-------
D.3. LYMPHOID CANCER MORTALITY (SUBSET OF ALL HEMATOPOIETIC
CANCERS COMBINED) (« = 17,530).
D.3.1. Exposure Distribution in Cohort and among Lymphoid Cases in the Cohort
Mortality Study
The estimated daily exposure to EtO across different jobs and time periods ranged from
0.05 to 77 ppm. Exposure intensities from this broad range were multiplied by the length of time
in different jobs to get estimates of cumulative exposure. The duration of exposure for the lull
cohort at the end of follow-up had a mean of 8.7 years and a standard deviation of 9.3 years.
Cumulative exposure at the end of follow-up, with no lag, had a mean of 27 ppm-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. For additional details
about the exposure and other characteristics of the full cohort and the lymphoid cancer risk sets,
see Section D.5 of Appendix D.
As noted in Section D.l.l, 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. 1.1 for
more discussion.
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.. 20041 and we
also used the same cutpoints as in the publication. Lymphoid cancer consists of non-Hodgkin
lymphoma, lymphocytic leukemia, and myeloma (ICD-9 200, 202, 203, 204). The distribution
of cases for lymphoid cancer mortality is presented in Table D-26.
Table D-26. Exposure categories and case distribution for lymphoid cancer
mortality
Cumulative exposure,
15-year lag3
Male lymphoid cancer
deaths
Female lymphoid cancer
deaths
Total lymphoid cancer
deaths
0 (lagged out)
6
3
9
>0-1,200 ppm-days
2
8
10
1,201-3,680 ppm-days
4
7
11
3,681-13,500 ppm-days
5
5
10
>13,500 ppm-days
10
3
13
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.
D-37
-------
D.3.2. Lag Selection for the Lymphoid Cancer Mortality Data
After the SAB review of the 2014 draft assessment, the issue of lag selection was
revisited. Table D-27 provides -2 log-likelihood results comparing different models with
different lags. Table D-27 also presents the AIC values for the same models to facilitate
comparison with the two-piece spline models, which include an extra parameter. [The knot is
preselected and is not considered a parameter in these analyses, consistent with the SAB's
concept of parsimony (SAB, 2015)1.14
Of the nonspline models (i.e., linear and log-linear cumulative and log cumulative
exposure models), only the models with log cumulative exposure and a 15-year lag were
statistically significant (p = 0.02 for both the linear and log-linear RR models). For the four
spline model options—log-linear or linear, with the knots at the global maximum likelihood or
the local maximum likelihood—the lowest -2 log likelihoods (and AICs) occur with a lag of 15
years in three of the cases. For the log-linear spline model with the knot at the global maximum
likelihood, the lowest -2 log likelihood (and AIC) occurs with no lag, which is not biologically
likely. The next lowest -2 log likelihood (and AIC) occurs with a lag of 15 years, and the AIC is
within 2 AIC units of the lowest value, suggesting a negligible difference in fit. Thus, for
consistency in comparisons and to optimize the best fitting lag overall, a lag of 15 years was
selected for analyzing the lymphoid cancer mortality data. Selecting the lag time based on the
strongest associations is a common statistical approach (Checkowav et al.. 2004). A lag of 15
years is somewhat long for a lymphohematopoietic cancer, but within the range of plausible
values, especially for mortality, as opposed to incidence. Sensitivity of the results to choice of
lag time is examined in Section D.3.5 below.
14 "in some settings the principle of parsimony may suggest that the most informative analysis will rely uponfixing
some parameters rather than estimating them from the data. The impact of the fhed parameter choices can be
evaluated in sensitivity analyses. In the draft assessment, fixing the knot when estimating linear spline model fits
from relative risk regressions is one such example" [page 12 of SAB f2015TI.
D-38
-------
Table D-27. Minus 2 log-likelihood results and AICs for different models
and different exposure lag times
LAG
Minus twice LL
0.0
5.0
10.0
15.0
20.0
To get AIC
LOG-LINEAR
MODELS
CUMEXP
460.8
460.8
461.6
462.4
463.6
add 2
LCUMEXP
462.0
463.5
463.0
458.4
461.6
add 2
2-PIECE
456.4
460.5
460.9
457.8
461.1
add 4
knot3
100
1,575
1,600
125
125
2-PIECE
459.2
458.6
461.8
add 4
alt knotb
775
1,600
1,600
LINEAR MODELS
CUMEXP
460.8
460.8
460.9
461.2
463.1
add 2
LCUMEXP
461.4
463.2
462.8
458.2
461.2
add 2
2-PIECE
458.9
460.6
460.5
457.3
460.8
add 4
knot3
125
1,575
1,600
125
125
2-PIECE
459.3
458.1
461.4
add 4
alt. knotb
775
1,600
1,600
NULL LOG-LINEAR MODELS0
463.9
NULL LINEAR MODELS0
463.5
LAG
AIC
0.0
5.0
10.0
15.0
20.0
LOG-LINEAR EXPOSURE MODELS
CUMEXP
462.8
462.8
463.6
464.4
465.6
LCUMEXP
464.0
465.5
465.0
460.4
463.6
2-PIECE
460.4
464.5
464.9
461.8
465.1
2-PIECE (alt. knot)
463.2
462.6
465.8
LINEAR EXPOSURE MODELS
CUMEXP
462.8
462.8
462.9
463.2
465.1
LCUMEXP
463.4
465.2
464.8
460.2
463.2
2-PIECE
462.9
464.6
464.5
461.3
464.8
2-PIECE (alt. knot)
463.3
462.1
465.4
"knots were obtained by doing a grid search by increments of 100 ppm x days and then interpolating where
appropriate.
bFor models with very low knots, alternate knots were obtained from local maximum likelihoods because of the
small number of cases informing the slope of the low-exposure spline for low knots (see Figure D-14).
The log-linear and linear models were obtained using different SAS procedures which gave different -2LL results
for the null model.
D-39
-------
AIC for different knots in the unlagged cumulative exposure for lymphoid cancer
mortality
4000 6000
Knot
8000 10000
AIC for different knots in the 5 year lagged cumulative exposure for lymphoid cancer
mortality
464.80 -
464.75 -
/
< 464.70 -
V\ /
464.65 -
\ /
464.60 -
\l
1000 2000 3000 4000 5000
Knot
AIC for different knots in the 10 year lagged cumulative exposure for lymphoid cancer
mortality
AIC for difFerent knots in the 15 year lagged cumulative exposure for lymphoid cancer
mortality
464.9
464
464.8
A
463 -
464.7
o
<
V /
o
<
/
464.6
V /
462 -
464.5
Vv
V
1000 2000 3000 4000 5000
Knot
1000 2000 3000 4000 5000
Knot
AIC for different knots in the 20 year lagged cumulative exposure for lymphoid cancer
mortality
466.00 -
A —-
465.75 -
/ \ /
^ 465.50 -
<
465.25 -
\
465.00 -
\
464.75 -
1000 2000 3000 4000 5000
Knot
Figure D-14. AIC vs. knot for different lag periods for two-piece linear spline
models.
(Graphs for the two-piece log-linear spline models were visually indistinguishable from these graphs for
the linear spline models.)
D-40
-------
D.3.3. Modeling of Lymphoid Cancer Mortality Data Using a Variety of Models
D.3.3.1. Cox Regression (Log RR) Models
While the published results in Steenland et al. (2004) focused on males [Table 7 in
Steenland et al. (2004)1, males and females in fact do not differ greatly in categorical results
using a 15-year lag. A formal chunk test (Kleinbaum 1994) for four interaction terms between
exposure and sex 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-28 below shows the categorical odds ratio
results for men and women separately and combined. In the analyses presented here, males and
females are combined.
Table D-28. Lymphoid cancer mortality results by sex
Cumulative exposure,
15-year lag
Odds ratios
(95% CI)
males
Odds ratios
(95% CI)
females
Odds ratios
(95% CI)
combined
0 (lagged out)
1.00
1.00
1.00
>0-1,200 ppm-days
0.91 (0.16-5.23)
2.25 (0.41-12.45)
1.75 (0.59-5.25)
1,201-3,680 ppm-days
2.89 (0.65-12.86)
3.26 (0.56-18.98)
3.15 (1.04-9.49)
3,681-13,500 ppm-days
2.71 (0.65-11.55)
2.16 (0.34-13.59)
2.44 (0.80-7.50)
>13,500 ppm-days
3.76 (1.03-13.64)
1.83 (0.25-13.40)
3.00 (1.02-8.45)
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, sex, and race.
Using log RR models, we used a categorical model, a (log-)linear model, a two-piece
(log-)linear model, and a log-transform model. We also ran a number of analogous models using
linear RR models (see Section D.3.3.2 below).
The categorical model for lymphoid cancer mortality was run using the originally
published cutpoints to form four categories above the lagged-out group, as shown in Table D-28.
To graph the categorical points, each category was assigned the midpoint of the category as its
exposure level, except for the last one which was assigned 50% more than the last cutpoint.
For the two-piece log-linear model, the single knot was chosen at 100 ppm-days based on
a comparison of likelihoods assessed every lOOppm-day from 100 to 15,000. The best
likelihood was at 100 ppm-days. Figure D-15 below shows the likelihood versus the knots.
Figure D-15 also suggests a local maximum likelihood near 1,600 ppm-days.
D-41
-------
Model results for the categorical and two-piece linear log RR models are shown in Tables
D-29 and D-30. Tables D-31 and D-32 give the results for the log-transform model and linear
log RR models; the latter does not fit the data well. Table D-33 shows the model results for the
two-piece log-linear spine model with the knot at the local maximum likelihood of 1,600 ppm-
days.
Figure D-16 shows the graphical results for the categorical, (log-)linear, two-piece (log-)
linear, and log-transform log RR models. There is a very steep increase in risk at very low
exposures. The knot for the two-piece log-linear curve is alow 100 ppm-days. The steep slope
at low exposures may be unrealistic as a basis for risk assessment, dependent as it is on relatively
sparse data in the low-exposure region. Table D-34 lists the cumulative exposures with a 15-
year lag for all the lymphoid cancer cases (e.g., there are no cases below the knot of
100 ppm-days).
We further explored the sensitivity of the log-linear (standard Cox regression) model to
high exposures, by excluding progressively more of the upper tail of exposure. We excluded 5,
10, 20, 30, 40, and 55% of the upper tail of exposure. The 55% cutoff was at 2,000 ppm-days.
The slope of the log-linear exposure-response model increased by 0.4, 1.7, 7.9, 5.6, 26.7, and
113.7 times, respectively, with the exclusion of progressively more data. It is clear that the curve
changes substantially once the top 20% of the exposure range is truncated.
-2 log likelihood for different knots for lymphoid cancer mortality
461.000-
I—h I I I I I I I I * * I I I I I |
458.000 -
460.000 -
459.000 -
457.000 -
1—i—i—i—i—i—i—r
"n-1"
1000
11 I 1
2000
t—i—i—i—i—|—r
3000
"n-1"
5000
"¦"T
7000
0
4000
6000
KNOT
Figure D-15. Likelihoods vs. knots for two-piece log-linear model, lymphoid
cancer mortality.
D-42
-------
1.5
2,5
10,000
CUMLKPlS
* Cjtcftoocjl
--«Lc*«R. GUMEXP15
Lc**R. LOrICuMDcPIS)
. Oft RH Spline-. Kncfl L0O, CuMEXPLS
Figure D-16. Exposure-response models for lymphoid cancer mortality.
Plot of continuous exposure (with 15-year lag) and lymphoid cancer mortality rate ratios estimated using
the two-piece log-linear spline model with the knot at 100 ppm-days overlaid with other log RR curves and
categorical (quartile) points.
D-43
-------
Table D-29. Categorical results for lymphoid cancer mortality, men and
women combined
Model fit statistics
Criterion
Without covariates
With covariates
-2 LOG L
463.912
458.069
AIC
463.912
458.069
SBC
463.912
473.950
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
5.8435
4
0.2111
Score
5.7397
4
0.2195
Wald
5.6220
4
0.2292
Analysis of maximum likelihood estimates
Variable3
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
CUM151
1
0.56036
0.55981
1.0020
0.3168
1.75
CUM 152
1
1.14581
0.56351
4.1344
0.0420
3.15
CUM 153
1
0.89001
0.57391
2.4049
0.1210
2.44
CUM 154
1
1.09998
0.55112
3.9837
0.0459
3.00
3Categorical exposure groups are quartiles of cumulative exposure with 15-year lag; from Table D-26 the
expo sure ranges are >0-1,200, 1,201-3,680, 3,681-13,500, and >13,500 ppm-days.
Table D-30. Results of two-piece log-linear spline model for lymphoid cancer
mortality, men and women combined, knot at 100 ppm-days
Model fit statistics
Criterion
Without covariates
With 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
X2
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
Parameter estimates
Standard error
X2
Pr> ChiSq
Hazard ratio
LIN_0
0.01010
0.00493
4.1997
0.0404
1.010
LIN_1
-0.01010
0.00493
4.1959
0.0405
0.990
D-44
-------
Table D-31. Results of the log-transform log RR model for lymphoid cancer
mortality, both sexes combined
Model fit statistics
Criterion
Without
covariates
With 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
X2
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
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
LCUM15
1
0.11184
0.04861
5.2936
0.0214
1.118
Table D-32. Results of the log-linear model for lymphoid cancer mortality,
both sexes combined
Model fit statistics
Criterion
Without
covariates
With
covariates
-2 LOG L
463.912
462.413
AIC
463.912
464.413
SBC
463.912
466.383
Testing global null hypothesis: BETA = 0
Teset
X2
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
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
CUMEXP15
1
4.7367E-6
3.35285E-6
1.9959
0.1577
1.000
D-45
-------
Table D-33. Results of two-piece log-linear spline model for lymphoid cancer
mortality, men and women combined, knot at 1,600 ppm-days
Model fit statistics
Criterion
Without
covariates
With covariates
2- LOG L
463.912
458.640
AIC
463.912
462.640
SBC
463.912
466.581
Testing global null hypothesis: BETA = 0
Criterion
Without
covariates
With covariates
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
DF
Parameter
estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
LIN_0
1
0.0004893
0.0002554
3.6713
0.0554
1.000
LIN_1
1
0.0004864
0.0002563
3.6014
0.0577
1.000
D-46
-------
Table D-34. Distribution of cumulative exposures with a 15-year lag for the
lymphoid cancer deaths
CUMEXP15
(mm-days)
Frequency
Percent
Cumulative
frequency
Cumulative
percent
0
9
16.98
9
16.98
100.063
1
1.89
10
18.87
130.644
1
1.89
11
20.75
181.819
1
1.89
12
22.64
272.09525
1
1.89
13
24.53
395.421
1
1.89
14
26.42
485.994
1
1.89
15
28.30
493.608
1
1.89
16
30.19
568.53575
1
1.89
17
32.08
777.045
1
1.89
18
33.96
860.77075
1
1.89
19
35.85
1,506.756
1
1.89
20
37.74
1,566.99
1
1.89
21
39.62
1,597.44
1
1.89
22
41.51
1,603.636
1
1.89
23
43.40
1,646.75225
1
1.89
24
45.28
2,147.01925
1
1.89
25
47.17
2,307.05
1
1.89
26
49.06
2,318.89425
1
1.89
27
50.94
2,567.721
1
1.89
28
52.83
2,592.742
1
1.89
29
54.72
3,478.642
1
1.89
30
56.60
3,776.718
1
1.89
31
58.49
4,556.3585
1
1.89
32
60.38
5,643.896
1
1.89
33
62.26
6,981.06375
1
1.89
34
64.15
7,127.132
1
1.89
35
66.04
7,549.875
1
1.89
36
67.92
10,485.87
1
1.89
37
69.81
11,127.772
1
1.89
38
71.70
12,279.195
1
1.89
39
73.58
13,498.377
1
1.89
40
75.47
15,696.735
1
1.89
41
77.36
17,507.77125
1
1.89
42
79.25
18,294.186
1
1.89
43
81.13
18,702.43425
1
1.89
44
83.025
23,611.25325
1
1.89
45
84.91
D-47
-------
Table D-34. Distribution of cumulative exposures with a 15-year lag for the
lymphoid cancer deaths (continued)
CUMEXP15
(ppm-days)
Frequency
Percent
Cumulative
frequency
Cumulative
percent
35,839.34525
1
1.89
46
86.79
43,955.86
1
1.89
47
88.68
49,101.02825
1
1.89
48
90.57
55,334.747
1
1.89
49
92.45
74,666.586
1
1.89
50
94.34
126,761.401
1
1.89
51
96.23
128,092.08075
1
1.89
52
98.11
146,460.07075
1
1.89
53
100.00
After the SAB review of the 2014 draft assessment, Steenland also provided modeling
results for models with a square-root transformation of cumulative exposure. Results for the
log-linear model with square root of exposure are presented in Table D-35.
Table D-35. Model fit statistics and coefficients for log-linear RR model
with square-root of cumulative exposure, with a 15-year lag, lymphoid
cancer mortality
Log-linear RR model
-2 Log-likelihood
(full model)
AIC
/j-val uc11
Parameter(s)
SE
s qrt (CUMEXP15)
460.8
462.8
0.08
B = 2.83 x 10-3
SE = 1.5 x 10-3
SE = standard error.
aFrom likelihood ratio test.
D.3.3.2. Linear Relative Risk Models
Table D-36 shows the model fit statistics and coefficients for the linear RR models.
Results for linear RR models are seen in Figure D-18 (denoted as "ERR" models). They are
quite similar to the log RR results in Figure D-16. Again there is a very steep rise in the
exposure-response curve at very low exposures. The knot for the two-piece linear curve is again
at 100 ppm-days.
D-48
-------
Table D-36. Model fit statistics and coefficients for linear RR models,
lymphoid cancer mortality
linear RR model
-2 Log-
likelihood
(full model)
AIC
/j-val ue11
Parameter(s)
Profile likelihood 95%
one-sided confidence
bounds
CUMEXP15
461.2
463.2
0.13
B = 1.227 x lO-5
LB = -2.2 x 10-6
UB = 4.71 x 10-5
Lo g (CUMEXP15)
458.2
460.2
0.02
B = 0.2083
LB = 0.0183
UB = 0.768
Sqrt(CUMEXP 15)
459.8
461.8
0.053
B = 6.14 x 10-3
NRb
Spline, knot = 100,
CUMEXP 15c-d
457.4
461.4
0.046
B1 = 0.015198
B2 = -0.015179
LB1 = 1.056 x 10-5
UB1 =0.05901
Spline, knot= 1,600,
CUMEXP 15c-d-e
458.1
462.1
0.07
B1 = 7.58 x 10-4
B2 = -7.48 x 10-4
LB1 = 4.52 x 10-6
UB1 =2.983 x 10-3
aFrom likelihood ratio test.
bNot reported: Confidence intervals for linear RR models, in contrast to those for the log-linear RR models, may
not be symmetrical. The EPA did not apply the profile likelihood approach (Langholz and Richardson. 20101
which allows for asymmetric CIs, to develop CIs for this model because the model was not used further in the
assessment.
Tor estimating risks from occupational exposures (see Section 4.7 of the Carcinogenicity Assessment Document),
both pieces of the two-piece linear spline model are used. For the maximum likelihood estimate, for exposures
below the knot, RR = 1 + (B1 x exp); for exposures above the knot, RR = 1 + (B1 x exp + B2 x [exp - knot]). For
the (one-sided) 95% upper confidence limit, the Waid approach is used as an approximation because it was not
possible to obtain a formula for the profile likelihood upper-bound estimates that could be used in the life-table
analysis. Thus, for exposures below the knot, RRu = 1 +([B1 + 1.645 x SE1] x exp); for exposures above the
knot, RRu = 1 + (B1 x exp + B2 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. As shown
in Figure D-17, the Wald upper-bound estimates are about half-way between the MLE RR estimates and the
profile likelihood upper-bound estimates. In the range of occupational exposures of interest (i.e., up to 12, 775
ppm x days)the Wald-basedRRu estimates are about 67% of the profile-likelihood-based RRu estimates.
dCalculating the profile likelihood bounds is computationally difficult and estimating the bounds forB2 was not
pursued here.
Varl = SE12 = (6.32 x 10"4)2 = 3.99 x 10"7; Var2 = SE22 = (6.31 x 10"4)2 = 3.98 x 10"7;
Covariance = -3.99 x 10~7.
D-49
-------
7
RR
-----Wald ub
PL ub
2000 4000 6000 8000 10000 12000 14000
cumulativeexposure (ppm x days; with 15-year lag)
Figure D-17. Comparison ofWald and profile likelihood (one-sided) 95%
upper-bound estimates for two-piece linear spline model.
D-50
-------
10,000
Figure D-18. Linear RR models for lymphoid cancer.
[Editorial note: "ERR" refers to linear RR models.]
• Categorical
ERR, CUMEXP15
ERR, L0g(CUMEXP15)
ERR Spline, Knot=100, CUMEXP15
D.3.4. Supplemental Results: Results for Log Cumulative Exposure Cox Regression
Model with No Lag
Model fit statistics and parameter coefficients for the log cumulative exposure Cox
regression model with no lag are presented in Table D-37.
Table D-37. Results for log cumulative exposure Cox regression model with no
lag
Model fit statistics
Criterion
Without covariates
With covariates
-2 LOG L
463.912
462.014
AIC
463.912
464.014
SBC
463.912
465.984
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
1.8987
1
0.1682
Score
1.8589
1
0.1728
Wald
1.8530
1
0.1734
Analysis of maximum likelihood estimates
Parameter
DF
Parameter estimate
Standard error
X2
Pr > ChiSq
Hazard ratio
LCUMEXP
1
0.10230
0.07515
1.8530
0.1734
1.108
D-51
-------
D.3.5. Sensitivity of (Incidence) Unit Risk Estimates to Change in Lag Period
Sensitivity of the (incidence) unit risk estimates to choice of exposure lag time for the
two-piece linear spline model is summarized in Table D-38.
Table D-38. Comparison of unit risk estimates for lymphoid cancer incidence
from two-piece linear spline models with different lag periods; cumulative
exposure in ppm x days, knot at 1,600 ppm x days
Lag
(years)
-2
log-likelihood
Parameter estimate
for 1st spline segment
(per ppm x day)
Profile likelihood 95%
one-sided
upper-bound estimate for
1st spline segment (per
ppm x day)
ECoi
(ppm)
LECoi
(ppm)
Unit risk
estimate
(per ppm)11
0
459.4
5.9 x 10-4
6.7 x 10-3
7.47 x 10-3
6.57 x 10-4
15.2
5
460.6
1.59 x 10-4
1.549 x 10-3
0.0300
3.07 x 10-3
3.26
10
460.5
2.11 x 10-4
1.427 x 10-3
0.0245
3.63 x 10-3
2.75
15
458.1
7.58 x 10-4
2.983 x 10-3
7.48 x 10-3
1.90 x 10-3
5.26
20
461.4
4.33 x 10-4
1.745 x 10-3
0.0145
3.59 x 10-3
2.79
Calculated for lymphoid cancer incidence; see Section 4.1.1.3.
The sensitivity analysis for choice of lag reveals that the unit risk estimates for the
selected two-piece linear spline model with the knot at 1,600 ppm x days for different lag
periods (0, 5, 10, 15, and 20 years) ranged from about 48% less than (10-year lag) to about 190%
greater than (i.e., 2.9-times) (no lag) the estimate for the selected model (15-year lag). The
models for lags of 0, 5, 10, and 20 years all had /?-values >0.10 for inclusion of the exposure
terms (0.12, 0.23, 0.21, and 0.35, respectively).
D.3.6. Sensitivity of (Incidence) Unit Risk Estimates to Value of Knot
Sensitivity of the (incidence) unit risk estimates to value of knot for the two-piece linear
spline model is summarized in Table D-39, with knots of 1,600 ± 1,000 ppm x days.
D-52
-------
Table D-39. Comparison of unit risk estimates for lymphoid cancer
incidence from two-piece linear spline models with different knot;
cumulative exposure in ppm x days, with lag of 15 years
Knot
(ppm x
days)
-2
log-likelihood
Parameter
estimate for 1st
spline segment
(per ppm x day)
Profile likelihood 95%
one-sided
upper-bound estimate for
1st spline segment (per
ppm x day)
EC01
(ppm)
LEG,,
(H>m)
Unit risk
estimate
(per
ppm)a
600
458.0
2.26 x 10-3
9.27 x 10-3
2.51 x 10-3
6.11 x 10-4
16.37
1,600
458.1
7.58 x 10-4
2.983 x 10-3
7.48 x 10-3
1.90 x 10-3
5.26
2,600
458.8
4.03 x 10-4
1.61 x 10-3
0.0141
3.52 x 10-3
2.84
Calculated for lymphoid cancer incidence; see Section 4.1.1.3.
Unlike with the sensitivity analysis for knot selection for breast cancer incidence (see
Section D.1.7), where the knots were at higher values of cumulative exposure, the sensitivity
analysis for knot selection in the two-piece linear spline model for lymphoid cancer shows
notable differences in the unit risk estimates for knots 1,000 ppm x days below and above the
selected knot of 1,600 ppm x days. The unit risk estimates for these alternate knot values are
about 3 times greater and 50% lower, respectively, than the unit risk estimate for the selected
model (with the knot at 1,600 ppm x days).
D.3.7. Analysis of Age Interaction for the Exposure Terms in the Two-Piece Linear Spline
Model
Table D-40 shows the ^-values for the inclusion of age interaction terms for the spline
exposure regression coefficients. The interaction terms have ^-values well above 0.05,
indicating that the exposure terms are independent of age (i.e., the proportional hazards
assumption is validated).
Table D-40. Evaluation of age interaction for the exposure terms in the
2-piece linear spline model with knot at 1,600 ppm x days; cumulative
exposure in ppm x days, with lag of 15 years
Parameter
/j-valuc for the inclusion of age interaction term
Betal
0.82
Beta2
0.82
D-53
-------
D.3.8. Sensitivity of (Incidence) Unit Risk Estimates to Upper-Bound Estimation
Approach—Wald vs. Profile Likelihood
Sensitivity of the (incidence) unit risk estimates to the approach used to estimate the
upper bound on the first spline piece from the selected two-piece linear spline model is
summarized in Table D-41. According toLaneholz and Richardson (2010), the distribution of
estimated parameters in nonlog-linear models (hazard functions) is often not symmetrical
(because beta is constrained so that the hazard cannot be less than 0) and profile likelihood
confidence intervals are recommended as being more accurate than Wald-type intervals. The
Wald-based result is 40% lower than the profile-likelihood-based estimate.
Table D-41. Comparison of unit risk estimates for lymphoid cancer
incidence from two-piece linear spline model using Wald-based and
profile-likelihood-based upper-bound estimates on the 1st spline piece
Betal
95% one-sided
estimate
Wald SE1
upper-hound estimate
Unit risk
Estimation
(per ppm x
estimate
for 1st spline segment
EC„i
LEG,,
estimate
Approach
day)
(per ppm x day)
(per ppm x day)
(H>m)
(PPm)
(per ppm)11
Wald
7.58 x icr4
6.32 x icr4
1.80 x 10-3
7.48 x 10-3
3.15 x 10-3
3.17
Profile likelihood
7.58 x lO-4
--
2.98 x 10-3
7.48 x 10-3
1.90 x 10-3
5.26
Calculated for lymphoid cancer incidence; see Section 4.1.1.3.
D.3.9. Sensitivity of Occupational Extra Risk Estimates to Change in Lag Period
In Section 4.7, extra risk estimates for lymphoid cancer mortality and incidence from the
selected model are presented for some occupational exposure scenarios of interest (i.e., 35-year
exposures to 8-hour TWAs ranging from 0.1 to 1 ppm between ages 20 and 55 years) because
the scenarios include cumulative exposures above the level at which the unit risk estimate is
valid. Here, the sensitivity of the selected model (i.e., the two-piece linear spline model with the
knot at 1,600 ppm x days and a lag of 15 years) to changes in lag is explored. Parameter
estimates for the two-piece linear spline model with the knot at 1,600 ppm x days and different
lag periods (0, 5, 10, 15, and 20 years) are presented in Table D-42. The Wald approach was
used as an approximation to derive the upper-bound estimates because it was not possible to
obtain a formula for the profile likelihood upper-bound estimates that could be used in the
life-table analysis. As shown in Figure D-17 above, the Wald upper-bound estimates are about
halfway between the MLE RR estimates and the profile likelihood upper-bound estimates. In the
range of cumulative exposures of interest for the occupational scenarios considered in this
D-54
-------
assessment (i.e., up to 12,775 ppm x days, with a 15-year lag), the Wald-based
upper-bound estimates are about 67% of the profile-likelihood-based upper-bound estimates.
The equations for deriving the MLE and upper-bound estimates across the range of exposures are
presented in footnote c of Table D-36. Sensitivity of the extra risk estimates for lymphoid cancer
incidence for the occupational exposure scenarios to changes in exposure lag time for the
two-piece linear spline model with the knot at 1,600 ppm x days is summarized in Table D-43.
Table D-42. Parameter estimates for the two-piece linear spline model
with the knot at 1,600 ppm x days for different lag periods; cumulative
exposure in ppm x days
Lag
(years)
Betal
(per ppm x day)
Beta2
(per ppm x day)
SE1
(per ppm x day)
SE2
(per ppm x day)
Covariance
(per (ppm x day)2)
0
5.9 x 10-4
-5.8 x 10-4
7.58 x 10-4
7.53 x 10-4
-5.71 x 10-7
5
1.59 x 10-4
-1.51 x 10-4
3.65 x 10-4
3.63 x 10-4
-1.33 x 10-7
10
2.11 x 10-4
-2.01 x 10-4
3.6 x 10-4
3.59 x 10-4
-1.29 x 10-7
15
7.58 x 10-4
-7.48 x 10-4
6.32 x 10-4
6.31 x 10-4
-3.99 x 10-7
20
4.33 x 10-4
-4.32 x 10-4
4.05 x 10-4
4.09 x 10-4
-1.65 x 10-7
D-55
-------
Table D-43. Comparison of extra risk estimates for lymphoid cancer incidence from two-piece linear spline
models with different lag periods; cumulative exposure in ppm x days, knot at 1,600 ppm x days
O
G\
8-hour TWA
15-year lag
0-year lag
Ratio to 15-
year-lagged
estimates
5-year lag
Ratio to 15-
year-lagged
estimates
10-year lag
Ratio to 15-
year-lagged
estimates
20-year lag
Ratio to 15-
year-lagged
estimates
MLEs
0.1
0.0240
0.0213
0.89
0.00565
0.24
0.00720
0.30
0.0126
0.53
0.2
0.0331
0.0276
0.83
0.00757
0.23
0.00981
0.30
0.0182
0.55
0.3
0.0343
0.0282
0.82
0.00793
0.23
0.0103
0.30
0.0189
0.55
0.4
0.0349
0.0286
0.82
0.00824
0.24
0.0107
0.31
0.0193
0.55
0.5
0.0354
0.0290
0.82
0.00854
0.24
0.0111
0.31
0.0194
0.55
0.6
0.0359
0.0294
0.82
0.00884
0.25
0.0114
0.32
0.0196
0.55
0.7
0.0362
0.0298
0.82
0.00912
0.25
0.0118
0.33
0.0196
0.54
0.8
0.0365
0.0301
0.82
0.00941
0.26
0.0121
0.33
0.0197
0.54
0.9
0.0369
0.0305
0.83
0.00969
0.26
0.0125
0.34
0.0198
0.54
1
0.0372
0.0308
0.83
0.00998
0.27
0.0128
0.34
0.0198
0.53
95% one-sided UCLs
0.1
0.0558
0.0646
1.16
0.0266
0.48
0.0270
0.48
0.0316
0.57
0.2
0.0762
0.0826
1.08
0.0348
0.46
0.0362
0.48
0.0453
0.59
0.3
0.0784
0.0838
1.07
0.0353
0.45
0.0373
0.48
0.0471
0.60
0.4
0.0794
0.0845
1.06
0.0355
0.45
0.0380
0.48
0.0480
0.60
0.5
0.0800
0.0852
1.07
0.0355
0.44
0.0387
0.48
0.0486
0.61
0.6
0.0806
0.0858
1.06
0.0354
0.44
0.0394
0.49
0.0492
0.61
0.7
0.0808
0.0862
1.07
0.0350
0.43
0.0401
0.50
0.0497
0.62
0.8
0.0811
0.0867
1.07
0.0346
0.43
0.0409
0.50
0.0502
0.62
0.9
0.0813
0.0871
1.07
0.0339
0.42
0.0417
0.51
0.0509
0.63
1
0.0815
0.0875
1.07
0.0331
0.41
0.0425
0.52
0.0516
0.63
-------
The sensitivity analysis for choice of lag reveals that the MLEs of extra risk for the
selected two-piece linear spline model with the knot at 1,600 ppm x days for different lag
periods (0, 5, 10, 15, and 20 years) ranged from about 25% of (5-year lag) to just over 80% of
(no lag) the estimates for the selected model (15-year lag). The 95% (one-sided) upper bounds
of extra risk ranged from about 45% of (5-year lag) to just over 5% greater than (no lag) the
estimates for the selected model. Of these models, the model with no lag was the best fitting
model after the selected model (15-year lag), based on log likelihood (and AIC) (see Table D-
38), and that is the model that had the most similar MLEs and UCLs to the selected model. The
models for lags of 5, 10, and 20 years each had /?-values > 0.20 for inclusion of the exposure
terms, indicating an inadequate fit to the data.
For lags of 5 and 10 years, the optimal knots for the two-piece linear spline model (1,575
ppm x days and 1,600 ppm x days, respectively) were in the vicinity of the selected knot (1,600
ppm x days), so these sensitivity analyses serve as comparisons for the optimal-knot models as
well as for the selected model with alternative knots; however, as noted above, neither the 5- nor
10-year lagged models had a good statistical fit. For the lag of 20 years, the optimal knot was
125 ppm x days, and a local minimum AIC (maximum likelihood) was observed at 1,600 ppm x
days, similar to the case with the 15-year lag (see Figure D-14). Even with the optimal knot,
however, the 20-year lagged linear spline model had an inadequate statistical fit (p = 0.26). For
the linear spline model with no lag, the optimal knot was also 125 ppm x days, and no clear
alternative local minimum AIC was observed (see Figure D-14). Even with the optimal knot, the
unlagged linear spline model had a poorer fit than the selected model (AIC of 462.9 vs. 462.1).
D.4. HEMATOPOIETIC CANCER MORTALITY (ALL HEMATOPOIETIC CANCERS
COMBINED) (« = 17,530)
D.4.1. Exposure Distribution in Cohort and among All (Lympho)hematopoietic Cases in
the Cohort Mortality Study
In modeling 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 cutpoints as in that
publication. The distribution of cases for hematopoietic cancer mortality is presented in Table
D-44.
D-57
-------
Table D-44 Exposure categories and case distribution for hematopoietic
cancer mortality
Cumulative exposure,
15-year lag3
Male hematopoietic cancel
deaths
Female hematopoietic
cancer deaths
Total hematopoietic
cancer deaths
0 (lagged out)
9
4
13
>0-1,200 ppm-days
4
13
17
1,201-3,680 ppm-days
5
10
15
3,681-13,500 ppm-days
8
7
15
>13,500 ppm-days
11
3
14
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.
D.4.2. Modeling of the Hematopoietic Cancer Mortality Data Using a Variety of Models
D.4.2.1. Cox Regression (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)1, 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
sex is not close to significance (/24,5, 4 DF; p = 0,34), although such tests are not very powerful
in the face of sparse data such as these. Table D-45 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.
Table D-45. All hematopoietic cancer mortality categorical results by sex
Odds ratio
Odds ratio
Cumulative exposure,
(95% CI)
(95% CI)
Odds ratio (95% CI)
15-year lag
males
females
combined
0 (lagged out)
1.00
1.00
1.00
>0-1,200 ppm-days
1.23 (0.32-4.74)
3.76 (1.01-17.23)
2.33 (0.93-5.86)
1,201-3,680 ppm-days
2.53 (0.69-9.27)
4.93 (1.01-23.99)
3.46 (1.33-8.95)
3,681-13,500 ppm-days
3.14 (0.95-10.37)
3.31 (0.64-17.16)
3.02 (1.16-7.89)
>13,500 ppm-days
3.42 (1.09-10.73)
2.11 (0.33-13.74)
2.96 (1.12-7.81)
CI = confidence interval.
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
D-58
-------
cancer mortality, only exposure variables were included in the model. Cases and controls were
matched within risk sets on age, sex, and race.
Using log RR models, we used a categorical model, a (log-)linear model, a two-piece
(log-)linear model, and a log-transform model. We also ran a number of analogous models using
linear RR models (see Section D.4.2.2 below).
The categorical log RR model for hematopoietic cancer mortality was run using the
originally published cutpoints to form four categories above the lagged-out group, as shown in
Table D-45. To graph the categorical points, each category was assigned the midpoint of the
category as its exposure level, except for the last one which was assigned 50% more than the last
cutpoint.
For the two-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 (see Figure D-19). In Figure D-20 below we show the categorical, two-piece
(log-)linear spline and log-transform log RR model results.
Model results for the categorical and two-piece (log-)linear log RR models are shown in
Tables D-46 and D-47, and the results of the log-transform and (log-)linear log RR models in
Table D-48 and Table D-49. Again the log-linear model appears to substantially underestimate
the exposure-response relationship and does not provide a good model fit.
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.
D-59
-------
-2 log likelihood for different knots for all hematopoetic cancer mortality
KNOT
Figure D-19. Likelihood vs. knots for two-piece log-linear model, all
hematopoietic cancer.
•
A M
w
\
1
\
1
1
1
1
1
/
1
0 5,000 10,000 15,000 20,000
CUMEXP15
• Categorical
— — Log RR, CUMEXP15
Log RR, Log(CUMEXP15)
Spline Log RR, Knot=500,
CUMEXP15
Figure D-20. Exposure-response models for hematopoietic cancer mortality.
Plot of continuous exposure (with 15-year lag) and all hematopoietic cancer mortality rate ratios estimated
using the two-piece log-linear spline model with the knot at 500 ppm-days overlaid with other log RR
curves and categorical (quartile) points.
D-60
-------
Table D-46. Categorical results for all hematopoietic cancer mortality, men
and women combined, cumulative exposure with a 15-year lag
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
655.643
647.806
AIC
655.643
655.806
SBC
655.643
665.022
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
7.8371
4
0.0977
Score
7.3994
4
0.1162
Wald
7.2354
4
0.1240
Analysis of maximum likelihood estimates
Variable
DF
Parameter estimate
Standard error
X2
Pr > ChiS
Hazard ratio
CUM151
1
0.84746
0.46956
3.2573
0.0711
2.33
CUM 152
1
1.23989
0.48571
6.5166
0.0107
3.46
CUM 153
1
1.10664
0.48943
5.1126
0.0238
3.02
CUM 154
1
1.08360
0.49603
4.7723
0.0289
2.96
Table D-47. Results of two-piece log-linear spline model for all hematopoietic
cancer mortality, men and women combined, cumulative exposure with a 15-
year lag; knot at 500 ppm-days
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
655.643
647.581
AIC
655.643
651.581
SBC
655.643
656.189
Testing global null hypothesis: BETA = 0
Test
X2
DF
Pr > ChiSq
Likelihood ratio
8.0615
2
0.0178
Score
7.5092
2
0.0234
Wald
7.3467
2
0.0254
Analysis of maximum likelihood estimates
Parameter
DF
Parameter
estimate
Standard error
X2
Pr> ChiSq
Hazard ratio
SP11
1
0.00201
0.000731
6.7457
0.0094
1.002
SP12
1
-0.00201
0.0007738
6.7249
0.0095
0.998
D-61
-------
Table D-48. 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
Without
covariates
With covariates
-2 LOG L
655.643
648.825
AIC
655.643
650.825
SBC
655.643
653.129
Testing global null hypothesis: BETA = 0
Test
x2
DF
Pr > ChiSq
Likelihood ratio
6.8177
1
0.0090
Score
6.6260
1
0.0100
Ward
6.5593
1
0.0104
Analysis of maximum likelihood estimates
Parameter
DF
Parameter
estimate
Standard error
x2
Pr > ChiSq
Hazard ratio
LCUM15
1
0.10706
0.04180
6.5593
0.0104
1.113
Table D-49. Results of log-linear model for all hematopoietic cancer morality,
men and women combined, cumulative exposure with a 15-year lag
Model fit statistics
Criterion
Without
covariates
With covariates
-2 LOG L
655.643
645.922
AIC
655.643
656.922
SBC
655.643
659.226
Testing global null hypothesis: BETA = 0
Test
x2
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 estimate
Parameter
DF
Parameter
estimate
Standard error
x2
Pr > ChiSq
Hazard ratio
CUMEXP15
1
3.26052E-6
3.48788E-6
0.8739
0.3499
1.000
D-62
-------
D.4.2.2. Linear Relative Risk Models for Hematopoietic Cancer Mortality
For completeness, we also present the results of the linear RR models below
(see Table D-50 and Figure D-21; 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 two-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.
Table D-50. Model fit statistics and coefficients for linear RR models,
hematopoietic cancer mortality
linear RR model
-2 Log likelihood
(full model)
AIC
/j-value11
Parameter(s)
SEb
CUMEXP15
654.64
656.64
0.32
B = 6.257 x 10-6
SE = 8.187 x 10-6
Lo g (CUMEXP15)
648.13
650.13
0.006
B = 0.2322
SE = 0.1437
Spline, knot = 500,
CUMEXP 15c-d
646.95
650.95
0.01
Bl = 3.673 x 10-3
B2 = -3.668 x lo-3
SE1 =2.345 x 10-3
SE2 =2.345 x 10-3
SE = standard error.
aFrom likelihood ratio test.
bEditorial note: Confidence intervals for linear RR models, in contrastto thoseforthe log-linear RR
models, may not be symmetrical. The EPA did not apply the profile likelihood approach (Langholz and
Richardson. 20101 which allows for asymmetric CIs, to develop CIs for these models because the models
were not used further in the assessment.
cCovariance of two pieces of linear spline: - 5.70 x 10~6.
dFor Wald estimates, for the maximum likelihood estimate, for exposures below the knot,
RR = 1 + (B1 x exp); for exposures above the knot, RR = 1 + (B1 x exp + B2 x [exp - knot]). For the
95% upper confidence limit, for exposures below the knot, RR = 1 + ([Bl + 1.645 x SE1] x exp); for
exposures above the knot, RR = 1 + (Bl x exp +B2 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.
D-63
-------
4
3.5
•
— — — ~
*
£ 2.5
~
/
# Categorical
f
ERR, CUMEXP15
ERR, Log(CUMEXP15)
2
1.5
1
0
5,000 10,000 15,000
CUMEXP15
20,000
Figure D-21. Linear RR models for hematopoietic cancer mortality.
[Editorial note: "ERR" refers to linear RR models.]
D.5. FURTHER CHARACTERIZATION OF THE NIOSH COHORT
D.5.1. Further Characterization of the Exposure Distributions and Other Characteristics
in the Full Cohort
Tables D-51-D-60 and Figures D-22-D-24 summarize characteristics of the M cohort,
which comprises all persons enrolled in the cohort. Within this context, a case is someone who
was ever diagnosed with a lymphoid cancer and a control is someone who was never diagnosed
with any lymphohematopoietic cancer.
Table D-51. Marginal summaries of workers' exposures, and years of entry
to employment and age at end of follow-up in full cohort
N
Mean
Minimum
25th percentile
Median
75th percentile
Maximum
Year of birth
17,148
1940
1884
1931
1943
1950
1968
Year of entry
17,148
1970.6
1938
1967
1971
1975
1986
Exposures (ppm-yr)
17,148
26.7
0.01
1.46
5.60
23.25
135.2
Age at end of follow-up
17,148
56.3
17.5
47.3
54.6
65.2
100.1
D-64
-------
Table D-52. Cumulative exposure to EtO by year of entry to employment in
full cohort
Analysis variable: exposure (ppm-vr)
Year of entry into employment
N
Mean
Minimum
Median
Maximum
< 1965
3,793
52.0
0.03
9.5
1,352
1965-1969
4,307
26.4
0.04
6.3
767
1969-1972
2,983
20.8
0.02
5.3
738
1972-1975
2,626
18.1
0.04
5.0
396
> 1975
3,415
11.0
0.01
3.9
257
Table D-53. Cumulative exposure to EtO by duration of employment in full
cohort
Analysis variable: exposure (ppm-vr)
Duration of employment
N
Mean
Minimum
Median
Maximum
< 0.9 years
3,441
3.2
0.02
1.7
65
0.9-2.6 years
3,386
7.4
0.02
3.3
173
2.6-7.0 years
3,441
18.8
0.01
8.1
367
7-17 years
3,442
41.9
0.01
18.0
638
> 17 years
3,414
62.4
0.01
21.3
1,352
Table D-54. Cumulative exposure to EtO in each of the risk categories in
full cohort
Analysis variable: exposure (ppm-vr)
Risk category3
N
Mean
Mnimum
Median
Maximum
< 1,200 ppm-days
6,627
1.20
0.01
0.99
3.29
1,200-3,680 ppm-days
3,726
5.89
3.29
5.45
10.07
3,680-13,500 ppm-days
3,713
20.14
10.08
13.69
36.97
> 13,500 ppm-days
3,082
114.7
36.97
51.28
1,352
aRisk category outpoints chosen based on exposure distributions for all lymphohematopoietic cancer; same as in
Steenland et al. (2004\
D-65
-------
Annual exposures experienced Py full cohort during employment
Median (black) exposures are shown along with
Year
Figure D-22. Estimated annual exposures experienced by cases and controls
in the full cohort while working1—medians and interquartile ranges2
'Annualexposure histories taken from NIOSH deidentified exposure records; include 382 Workers
ultimately removed from the analysis due to inconsistencies in the record.
2Prior to 1962, fewer than five cases were working in any given year. This resulted in a number of
years where the 25th, 50th, and 75th percentiles of the exposure distribution were identical or nearly
so.
D-66
-------
Annual exposures experienced by full cohort during employment
Mean (black) exposures are shown along with
the 95th percentile of exposures (dashed line)
i i i i
1950 1960 1970 1980
Year
Figure D-23. Estimated annual exposures experienced by cases and controls
in the full cohort while working1—means and 95th percentiles
'Annualexposure histories taken from NIOSB deidentified exposure records; include 382 workers
ultimately removed from the analysis due to inconsistencies in the record.
Table D-55. Sex distribution overtime—case and control sexes by the
year they entered the workforce
< 1950
1950-1960
1960-1970
1970-1980
1980-1990
Cases
Men
4
5
14
4
0
Women
1
4
15
7
0
Controls
Men
172
501
3,171
3,176
547
Women
144
873
4,095
4,026
365
D-67
-------
o -
1S40 1950 1960 1970 1930
Year
Figure D-24. Sex ratios for currently working populations.
Sex ratios are calculated for each year with a working case, and include all persons of case or control status
currently working at least part of thatyear.
Table D-56. Year of entiy to the EtO workforce
N
Me ail
5th percentile
25th
percentile
Median
75th
percentile
95th
percentile
Case
54
1963
1948
1960
1964
1969
1975
Control
17,070
1970
1956
1966
1970
1974
1981
D-68
-------
Table D-57. Age of entry to the EtO workforce
N
Mean
5th
percentile
25th
percentile
Median
75th
percentile
95th
percentile
Case
54
38.55
21.53
29.77
39.17
45.72
54.08
Control
17,070
29.46
18.30
20.97
26.30
36.29
49.65
Table D-58. Duration of employment in the EtO workforce
N
Mean
5th percentile
25th
percentile
Median
75th
percentile
95th
percentile
Case
54
11.14
0.95
2.53
8.99
18.18
31.88
Control
17,070
8.55
0.34
1.18
4.31
14.38
27.36
Table D-59. Year of departure/retirement from the EtO workforce
N
Mean
5th percentile
25th
percentile
Median
75th
percentile
95th
percentile
Case
54
1975
1961
1967
1974
1981
1986
Control
17,070
1978
1965
1971
1977
1985
1996
Table D-60. Age of departure/retirement from the EtO workforce
N
Mean
5th
percentile
25th
percentile
Median
75th
percentile
95th
percentile
Case
54
49.69
29.68
39.90
49.88
61.47
65.83
Control
17,070
38.01
19.88
24.98
35.12
49.69
63.42
D.5.2. Further Characterization of the Exposure Distributions and Other Characteristics
in the Risk Sets
Figures D-25 and D-26 and Table D-61 summarize characteristics of the risk sets, which
each comprise a lymphoid cancer case and its matched set of-100 controls. Controls were
matched on age, plant, race, and sex and randomly selected from the pool of all those who had
survived without lymphohematopoietic cancer to at least the age of the case in that set.
Exposures were truncated at the case failure age within each set. In this context, statistics for
controls are calculated via the average values for each set.
D-69
-------
Boxplot showing Lymphoid cancer case cumulative unlagged exposure
and risk set control mean cumulative unlagged exposure
(distribution mean values shown in orange)
mo -
50"
Boxplot showing Lymphoid cancer case cumulative 15-year lagged
exposure and risk set control mean cumulative 15-year lagged
exposure (distribution mean values shown in orange)
t
_j
I
Boxplot showing Lymphoid cancer case peak unlagged exposure
and risk set control mean peak unlagged exposure
(distribution mean values shown in orange)
Boxplot showing Lymphoid cancer case peak 15-year lagged
exposure and risk set control mean peak 15-year lagged
exposure (distribution mean values shown in orange)
I
D-70
-------
Boxplot showing Lymphoid cancer case unlagged exposure duration
and risk set control mean unlagged exposure duration
(distribution mean values shown in orange)
Boxplot showing Lymphoid cancer case 15-year lagged
exposure duration and risk set control mean 15-year lagged
exposure duration (distribution mean values shown in orange)
Figure D-25. Box plots1 of both unlagged and 15-year lagged cumulative
total exposures, peak exposures, and exposure durations for risk sets.
'Upper and lower sides of box correspond to 75th and 25th percentiles, spanning the interquartile
range (IQR); the line in the middle ofthe boxrepresents the median; the diamond depicts the
mean; the upper/lower whisker extend from the top/bottomof the box to 15 * IQR from the
top/bottomof the box and the points beyond the whiskers are data outside that range.
D-71
-------
Lymphoid cancer case unlagged cumulative exposures
compared to corresponding risk set control mean
unlagged cumulative exposures
SMean control exposure higher {N=35)
Case exposure higher {W=19)
100-
50 -
o-
Lymphoid cancer case 15-year lagged cumulative exposures
compared to corresponding risk set control mean
15-year lagged cumulative exposures
| — |Mean control exposure higher {N=37)
|—I Case exposure higher (W=17)
E
O100-
Lymphoid cancer case unlagged peak exposures compared to
corresponding risk set control mean unlagged peak exposures
RMean control exposure higher {N=31)
Case exposure higher {N=23)
Lymphoid cancer case 15-year lagged peak exposures compared to
corresponding risk set control mean 15-year lagged peak exposures
Mean control exposure higher {N=32)
jCase exposure higher (N=22)
D-72
-------
Lymphoid cancer case unlagged exposure durations compared to
corresponding risk set control mean unlagged exposure durations
Lymphoid cancer case 15-year lagged exposure durations compared
to corresponding risk set control mean 15-year
lagged exposure durations
Figure D-26. Lymphoid cancer case exposures compared to corresponding
risk set control mean exposures for cumulative total exposures, peak
exposures, and exposure durations both unlagged and with a 15-year lag.
Table D-61. Summary of percentage of total case and control individual
exposures in the risk set worker histories that are excluded when the lag of
15 years is imposed3
Cases
Controls
Mean
Median
Mean
Median
Cumulative exposure
33.37%
1.23%
39.62%
12.48%
Peak exposure
23.36%
0%
30.05%
0%
Exposure duration
32.77%
1.94%
40.90%
22.43%
Calculated using the equation:
Unlagged exposure — Lagged exposure
%Exposure excluded by lagging =
Unlagged exposure
D.6. 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
SMean control duration higher (N=36)
Case duration higher (W=18)
D-73
-------
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 healthy 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) 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 healthy 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).
D.7. POSSIBLE INFLUENCE OF EXPOSURE MISMEASUREMENT
Exposure estimation in the EtO studies considered here is subject to errors in
measurement. The method for exposure estimation used here involved assigned estimated
average exposures in a given job, at a given time period in a given plant, to each worker in that
job. Estimated average exposures were taken from observed measurements in a given job, or
estimated likely average exposures in that job derived from a regression model based on
observed measurements (Hornung et al.. 1994). Errors in measurement in this type of situation
are typically errors of the Berkson type, rather than classical errors (Armstrong. 1998. 1990). In
Berkson errors, the model for errors is
exposuretrue = exposureobserved + error
and the error is independent of the observed exposure. The classical error model is
exposureobserved = exposuretrue + error
and the error is independent of the true exposure. Assuming the errors are unbiased (i.e., their
expected value is 0) in the classical error model, it is well known that measurement error will
bias exposure-response coefficients towards the null in regression analyses. However, in the
Berkson error model, exposure-response coefficients will be unbiased in linear regression
models, although their variance may be increased. In log-linear regression models, Berkson
D-74
-------
error in some instances may result in biased exposure-response estimates (Deddens and
Hornunu. 1994; Prentice. 1982). This may occur when the variance of the errors increases with
the true exposure level, which is often the case in occupational studies, when the disease is
relatively rare (also typical), and when the true exposure is distributed log-normally (again
typical of occupational exposures). In this situation, Steenland etal. (2000) have shown that
exposure-response coefficients using cumulative exposure can be biased either upward or
downward. The direction and degree of bias depends on the degree of increase in the variance of
exposure error as exposure level increases and on the variance of duration of exposure. When
the standard deviation of duration of exposure is less than or equal to its mean, as is the case in
the EtO cohort studied here, simulations have shown that the exposure-response coefficients are
approximately unbiased (Steenland etal., 2000). An added complication not considered in the
simulations conducted by Steenland etal. (2000) is the possible correlation between
measurement error and outcome. If this correlation is strong, which may occur when there is a
strong exposure-response relationship, it is important to take it into account. Estimating the
effect of exposure measurement in the presence of this correlation can be done using Bayesian
models and special software (WINBUGS), but the calculations are complex and require a good
deal of time.
Hornung etal. (1994) provide an estimate of the log-normal distribution of measured
exposure based on personal samples, as well as the likely distribution of error in assigning the
job-specific means to estimate individual exposures. Assignment of such job-specific means was
shown to involve some bias as well as random error. This provides a rich source of information
with which one could simulate the effect of measurement error on exposure-response
coefficients. Based on the exposure estimates used in the study, and some assumptions about the
error of such measurement in terms of bias and random error, as well as the assumption of a
Berkson error model, one could simulate what the true job-specific exposure means were likely
to have been, and then in turn simulate likely true personal exposure distributions. Using the
latter in exposure-response analysis, one could estimate the true exposure-response coefficient.
However, such analyses are rather involved and beyond the scope of the current task.
D-75
-------
APPENDIX E. LIFE-TABLE ANALYSIS
A spreadsheet illustrating the extra risk calculation for the derivation of the LECoi for
lymphoid cancer incidence is presented in Table E-1.
E-l
-------
Table E-l. Extra risk calculation" for lymphoid cancer incidence from environmental exposure to 0.00190 ppm
(the LECoi)b using the two-piece linear spline model with knot at 1,600 ppm x days0
W
to
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Interval
number
(i)
Age
interval
All cause
mortality
(xl05/yr)
lymphoid
cancer
incidence
(xl05/yr)
All
cause
hazard
rate
(h*)
Prob of
surviving
interval
(q)
Prob of
surviving
up to
interval
(S)
lymphoid
cancer
hazard
rate (h)
Cond
prob of
lymphoid
cancer
incidence
in
interval
(Ro)
Exp
duration
mid
interval
(xtime)
Cum
exp mid
interval
(xdose)
Exposed
lymphoid
cancer
haz ard
rate (hx)
Exposed
all cause
hazard
rate
(h*x)
Exposed
prob of
surviving
interval
(qx)
Exposed
prob of
surviving
up to
interval
(Sx)
Exposed
cond prob
of
lymphoid
cancer in
interval
(Rx)
1
<1
632.7
1.9
0.0063
0.9937
1.0000
0.0000
0.00002
0
0.00
0.00002
0.0063
0.9937
1.0000
0.00002
2
1^1
27.2
8.5
0.0011
0.9989
0.9937
0.0003
0.00034
0
0.00
0.00034
0.0011
0.9989
0.9937
0.00034
3
5-9
12.0
4.7
0.0006
0.9994
0.9926
0.0002
0.00023
0
0.00
0.00024
0.0006
0.9994
0.9926
0.00023
4
10-14
14.5
3.5
0.0007
0.9993
0.9920
0.0002
0.00017
0
0.00
0.00018
0.0007
0.9993
0.9920
0.00017
5
15-19
50.7
3.4
0.0025
0.9975
0.9913
0.0002
0.00017
2.5
5.27
0.00017
0.0025
0.9975
0.9913
0.00017
6
20-24
87.7
3.5
0.0044
0.9956
0.9888
0.0002
0.00017
7.5
15.82
0.00018
0.0044
0.9956
0.9888
0.00018
7
25-29
97.6
4.3
0.0049
0.9951
0.9845
0.0002
0.00021
12.5
26.37
0.00023
0.0049
0.9951
0.9845
0.00023
8
30-34
111.8
6.0
0.0056
0.9944
0.9797
0.0003
0.00029
17.5
36.91
0.00033
0.0056
0.9944
0.9796
0.00033
9
35-39
141.4
9.1
0.0071
0.9930
0.9742
0.0005
0.00044
22.5
47.46
0.00052
0.0071
0.9929
0.9741
0.00050
10
40^14
206.9
13.8
0.0103
0.9897
0.9673
0.0007
0.00066
27.5
58.01
0.00081
0.0105
0.9896
0.9672
0.00078
11
45^19
327.5
21.0
0.0164
0.9838
0.9574
0.0011
0.00100
32.5
68.56
0.00126
0.0166
0.9835
0.9572
0.00120
12
50-54
497.4
32.9
0.0249
0.9754
0.9418
0.0016
0.00153
37.5
79.10
0.00203
0.0253
0.9751
0.9414
0.00189
13
55-59
714.3
49.0
0.0357
0.9649
0.9187
0.0025
0.00221
42.5
89.65
0.00311
0.0364
0.9643
0.9179
0.00280
14
60-64
1,022.1
72.4
0.0511
0.9502
0.8865
0.0036
0.00313
47.5
100.20
0.00470
0.0522
0.9492
0.8851
0.00406
15
65-69
1,521.5
106.9
0.0761
0.9267
0.8423
0.0053
0.00434
52.5
110.74
0.00711
0.0778
0.9251
0.8401
0.00575
16
70-74
2,341.0
139.5
0.1171
0.8895
0.7806
0.0070
0.00514
57.5
121.29
0.00950
0.1196
0.8873
0.7772
0.00696
17
75-59
3,746.0
176.0
0.1873
0.8292
0.6944
0.0088
0.00557
62.5
131.84
0.01226
0.1908
0.8263
0.6896
0.00770
18
80-84
6,164.8
198.6
0.3082
0.7347
0.5758
0.0099
0.00492
67.5
142.38
0.01415
0.3125
0.7316
0.5699
0.00692
Ro =
0.03055
Rx =
0.04022
extra risk= (Rx-Ro)/(l-Ro) =0.00998
-------
Table E-l. Extra risk calculation" for lymphoid cancer incidence from environmental exposure to 0.00190 ppm (the
LECoi)b using the two-piece linear spline model with knot at 1,600 ppm x days0 (continued)
Column A:
Column B:
Column C:
Column D:
Column E:
Column F:
Column G:
Column H:
Column I:
Column J:
Column K:
Column L:
Column M:
Column N:
Column O:
Column P:
Interval index number (i).
5-yrage interval (except <1 and 1-4) up to age 85.
All-cause mortality rate for interval i (x 105/yr) (2008-2012 data from CDC).
Lymphoid cancer incidence rate for interval i(x 105/yr) (2008-2012 SEER data).d
All-cause hazard rate for interval i (h*i) (= all-cause mortality rate x number of years in age interval).6
Probability of surviving interval i (without being diagnosed with lymphoid cancer) (qi) [= exp(-h*i)]. 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 andtoexclude 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.
Probability of surviving up to interval i (without having been diagnosed with lymphoid cancer) (Si) (Si = 1; Si = Si-i x q;-i, fori> 1).
Lymphoid cancer incidence hazard rate for interval i (hi) (= lymphoid cancer incidence rate x number of years in interval).
Conditional probability of being diagnosed with lymphoid cancer in interval i [= (h;/h*i) x Si x (1—qi)], 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).
Exposure duration at midinterval (taking into account 15-yrlag) (xtime).
Cumulative exposure midinterval (xdose) (= exposure level [i.e., 0.00190 ppm] x 365/240 x 20/10 x xtime x 365) [365/240 x 20/10 converts continuous environmental
exposures to corresponding occupational exposures; xtime x 365 converts exposure duration in years to exposure duration in days].
Lymphoid cancer incidence hazard rate in exposed people for interval i (hx) (= hi x (1 + p x xdose), where [3 = 0.002983 perppm x day is theprofile likelihood 95%
(one-sided) upper-bound estimate for the regression coefficient for the first spline of the two-piece linear spline model (see Section 4.1.1.2); note that the cumulative
exposures are below the knot of 1,600 ppm x days for each interval, so only the first spline is used.
All-cause hazard rate in exposed people for interval i (h*x) [= h*i + (hx - hi)].
Probability of exposed people surviving interval i (without being diagnosed with lymphoid cancer) (qxi) [= exp(-h*xi)].
Probability of exposed people surviving up to interval i (without having been diagnosed with lymphoid cancer) (Sx) (Sxi = 1; Sx = Sx-i x qx-i, for i> 1).
Conditional probability of exposed people being diagnosed with lymphoid cancer in interval i [= (hx/h*x) x Sx x (1-qx)] (Rx, the lifetime probability ofbeing diagnosed
with lymphoid cancer for exposed people = the sum of the conditional probabilities across the intervals).
aUsing the methodology of BEIR (1988).
bThe estimated 95% lower bound on the continuous exposure level that gives a 1% extra lifetime risk of lymphoid cancer incidence.
cBased on the results of Steenland et al. (2004). reanalyzed by Steenland for both se?es combined (see Appendix D), with a 15-year lag, as described in Section 4.1.1.
dBackground cancer incidence rates are used to estimate extra risks for cancer incidence under the assumption that the exposure-response relationship for cancer
incidence is the same as that for cancer mortality (see Section 4.1.1.3).
Tor 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 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
-------
Table E-l. Extra risk calculation" for lymphoid cancer incidence from environmental exposure to 0.00190 ppm (the
LECoi)b using the two-piece linear spline model with knot at 1,600 ppm x days0 (continued)
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).
-------
APPENDIX F. EQUATIONS USED FOR WEIGHTED LINEAR REGRESSION
OF CATEGORICAL RESULTS
[Source: Rothman (1986), p. 343-344]
Linear model: RR = 1 + bX
where RR = rate ratio, X= exposure, and b = slope.
Slope (b) can be estimated from the following equation:
X" v X" v
6 = ^2 ill
s
j=2
WJXJ
(F-l)
where / specifies the exposure category level and the reference category (j = 1) is ignored.
The standard error of the slope can be estimated as follows:
SE(b)
1
J=2
(F-2)
wjxj
The weights, 117, are estimated from the confidence intervals (as the inverse of the variance):
Var(RRj) « RRJ2Vcir[\n(RR (.)] « RR2
J/J J
ln(RRj)-ln(RRj)
2x1.96
(F-3)
where RRj is the 95% upper bound on the RR, estimate (for the jth exposure category), and RRi is
the 95% lower bound on the RRj estimate.
F-l
-------
APPENDIX G. MODEL PARAMETERS IN THE ANALYSIS OF ANIMAL
TUMOR INCIDENCE
Table G-l. Analysis of grouped data, NTP (1987) mouse study"; multistage
model parameters
Tumor
Multistageb
polynomial
degree
qo
I
1*"
1
qz
(mg/m3)-2
(mg/m3)-2
p value
(y2 goodness
of fit)
Males
Lung adenomas
plus carcinomas
1
2.52 x 10-!
1.52 x lO"2
0.92
Females
Lung adenomas
plus carcinomas
2
3.87 x lO-2
0.0
4.80 x lO-4
0.39
Malignant
lymphoma
3
1.74 x lO"1
0.0
0.0
1.13 x lO-5
0.18
Uterine carcinoma
2
0.0
0.0
9.80 x lO"5
0.90
Mammary
carcinoma
ld
2.27 x lO"2
1.09 x lO"2
-
aThe exposure concentrations were 0, 50, and 100 ppm These were adjusted to continuous exposure.
bP(d) = 1 - exp[-(qo + qid + q2d2 + ... + qkdk)], where d is inhaled ethylene oxide exposure concentration.
cEven though qi is zero in some cases, the upper bound of qi is nonzero.
dThe 100-ppm dose was deleted; the fit was perfect with only two points to fit.
Table G-2. Analysis of grouped data from the Lynch etal. (1984a,c) study of
male F344 rats"; multistage model parameters
Tumor
Multistageb
polynomial degree
qo
I
1
/j-val uc
(y2 goodness of fit)
Splenic mononuclear cell leukemia
lc
3.12 x 101
1.48 x lO"2
-
Testicular peritoneal
mesothelioma
1
3.54 x lO"2
6.30 x lO"3
0.34
Brain mixed-cell glioma
1
0
1.72 x lO-4
0.96
aThe exposure concentrations were 0, 50, and 100 ppm These were adjusted to continuous exposure.
bP(d) = 1 -exp[-(qo + 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.
G-l
-------
Table G-3. Analysis of grouped data from the Garman et al. (1985) and
Snellings et al. (1984) reports on F344 rats"; multistage model parameters
Tumor
Multistageb
polynomial degree
qo
qi
(mg/m3)"1
/j-val uc
(y2 goodness of fit)
Males
Splenic mononuclear cell leukemia
1
1.63 x 10"1
8.56 x lO"3
0.34
Testicular peritoneal mesothelioma
1
2.38 x lO-2
4.74 x lO"3
0.68
Primary brain tumors
1
5.88 x lO"3
2.92 x lO"3
0.46
Females
Splenic mononuclear cell leukemia
1
1.08 x 10"1
2.37 x lO"2
0.75
Primary brain tumors
1
5.94 x lO"3
1.65 x lO-3
0.80
aThe exposure concentrations were 0, 10, 33, and 100 ppm These were adjusted to continuous exposure.
bP(d) = 1 -exp[-(qo + qid+ q2d2+ ... + qkdk)], where d is inhaled ethylene oxide exposure concentration.
Table G-4. Time-to-tumor analysis of individual animal data from the NTT
(1987) mouse study"; multistage-Weibull modelb parameters
Tumor
Multistage
polynomial degree
qo
I
1
z
Males
Lung adenomas plus
carcinomas
1
3.44 x 101
2.03 x lO"2
5.39
Females
Lung adenomas plus
carcinomas
1
5.35 x lO"2
1.76 x lO"2
7.27
Malignant lymphoma
1
1.91 x lO-1
8.80 x lO-3
1.00
Uterine carcinoma
1
0.0
3.81 x lO-3
3.93
Mammary carcinoma
1
3.78 x lO-2
5.10 x lO-3
1.00
aThe exposure concentrations were 0, 50, and 100 ppm These were adjusted to continuous exposure.
bP(d, t) = 1 - exp[-(qo + qi d + q2d2 +.... + qkdk) x (t - to)2], where d is inhaled ethylene oxide exposure
concentration. The length of the study was 104 weeks. The times t and to as expressed in the above formula are
scaled so that the length of the study is 1.0. Then, qois dimensionless, and the coefficients qk are expressed in units
of (mg/m3)~k.
G-2
-------
APPENDIX H. SUMMARY OF 2007 EXTERNAL PEER REVIEW AND
PUBLIC COMMENTS AND DISPOSITION
A draft of this assessment document entitled Evaluation of the Carcinogenicity of
Ethylene Oxide (dated August 2006) (U.S. EPA, 2006a) was available for public comment and
underwent a formal external peer review in accordance with EPA guidance on peer review (U.S.
EPA, 2006b). At the request of the EPA's Office of Research and Development, the EPA
Science Advisory Board (SAB) convened a panel of 15 experts external to the Agency to review
the ethylene oxide (EtO) assessment document. An external peer-review meeting was held in
January 2007, and a final peer-review report was released in December 2007 (SAB, 2007).
The primary purpose of this draft assessment was to review and characterize the available
data on the carcinogenicity of EtO and to estimate the lifetime unit cancer risk from inhalation
exposure. The SAB panel was asked to comment primarily on three main issues including
carcinogenic hazard, cancer risk estimation, and uncertainty associated with the hazard
characterization and quantitative risk estimation. The SAB panel was charged with answering a
number of specific questions that addressed key scientific issues relevant to the assessment. The
comments made by the panel in the Executive Summary of the SAB report (SAB, 2007) in
response to the charge questions are presented verbatim below followed by the EPA's responses;
the comments and responses are arranged by charge question.
In addition, a number of comments from the public were received during the public
comment period. An extract of the significant scientific public comments and the EPA's
responses are also included in a separate section of this appendix (see Section H.2).
Following the 2007 SAB review, a revised draft was developed and released for public
comment in July 2013. The comments on the 2013 draft are summarized in Appendix K along
with the EPA's responses. The 2013 draft was further revised in response to the public
comments and submitted for additional SAB review in August 2014. Comments on the 2014
SAB review draft and the EPA's responses are presented in Appendix I.
H.l. SAB PANEL COMMENTS
The statement of the issues as contained in the Agency's charge to the SAB panel are
listed below in italics followed by (1) the panel's comments quoted directly from the Executive
Summary of the panel's report (SAB, 2007) and (2) the Agency's response to the comments.
Issue 1: Carcinogenic Hazard (see Section 3 and Appendix A of the EPA Draft
Assessment)
H-l
-------
Do the available data and discussion in the draft document support the hazard conclusion that
EtO is carcinogenic to humans based on the weight-of-evidence descriptors in EPA '.s
2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a)? In your response,
please include consideration of thefollowing:
1. a. EPA concluded that the epidemiological evidence on EtO carcinogenicity was strong, but
less than completely conclusive. Does the draft document provide sufficient description of the
studies, balanced treatment of positive and negative results, and a rigorous and transparent
analysis of the data used to assess the carcinogenic hazard of ethylene oxide (EtO) to
humans? Please comment on the EPA's characterization of the body of epidemiological data
reviewed. Considerations include: a) the consistency of the findings, including the
significance of differences in results using different exposure metrics, b) the utility of the
internal (based on exposure category) versus external (e.g., SMR and SIR) comparisons of
cancer rates, c) the magnitude of the risks, and d) the strength of the epidemiological evidence.
SAB COMMENT: A majority of the Panel agreed with the conclusion in the draft document
that the available evidence supports a descriptor of "Carcinogenic to Humans" although some
Panel members concluded that the descriptor "Likely to be Carcinogenic to Humans" was more
appropriate. There was consensus that the epidemiological data regarding ethylene oxide
carcinogenicity were not in and of themselves sufficient to provide convincing evidence of a
causal association between human exposure and cancer. Differing views as to the appropriate
descriptor for ethylene oxide were based on differences of opinion as to whether criteria
necessary for designation as "Carcinogenic to Humans" in the absence of conclusive evidence
from epidemiologic studies were met. The majority of Panel members thought that the
combined weight of the epidemiological, experimental animal, and mutagenicity evidence was
sufficient to conclude that EtO is carcinogenic to humans.
The Panel concluded that the assessment would be improved by: 1) a better introduction
to the hazard characterization section, including a brief description of the information that will be
presented; 2) a clear articulation of the criteria by which epidemiologic studies were judged as to
strengths and weaknesses; 3) addition of a more inclusive summary figure and/or table at the
beginning of section 3.0; and 4) inclusion of material now provided in Appendix A of the draft
assessment to within the main body of that assessment.
The Panel agreed with the EPA in their reliance on "internal" estimates of cancer rates
rather than "external" comparisons (SMR, SIR) due to well recognized limitations to the latter
method of analysis.
H-2
-------
The Draft Assessment characterizes the magnitude of the unit risk estimate associated
with EtO as "weak". This finding is substantiated by the epidemiologic evidence where a
relatively small number of excess cancers are found above background even among highly
exposed individuals. However, the magnitude of risk suggested by the unit risk estimate is
somewhat at odds with this concept. Subsequent recommendations in our report try to address
this apparent inconsistency.
EPA RESPONSE: As supported by the majority of the panel, the EPA is retaining the
conclusion that the combined weight of the epidemiological, laboratory animal, and mutagenicity
evidence is sufficient to characterize EtO as carcinogenic to humans. Some panel members were
of the opinion that the descriptor "Likely to be Carcinogenic to Humans" was more appropriate,
as they found the epidemiological evidence to be weak and the data insufficient to conclude that
key precursor events were observed in humans fSAB (2007), p. 10], The EPA and the majority of
the SAB panel disagree that the epidemiological evidence is weak. The EPA has strengthened
the summary review of these data in the human evidence section (see Section 3.1) and in the
hazard characterization section (see Section 3.5.1). In addition, the revised assessment
specifically addresses the precursor data for rodents and humans. While the databases for
humans and rodents contain different types of studies, the EPA did not find any inconsistency
and concluded that the data support a finding of a mutagenic mode of action (relevant to
humans), a finding with which the SAB concurred. The EPA has expanded the discussion of
these data, specifically in Sections 3.3.3.2, 3.3.3.3, and 3.4.1.
In response to the panel recommendations, the EPA has added an introduction at the
beginning of Chapter 3 that provides a brief description of the information presented in the
chapter and has provided a clearer explanation of the criteria used to evaluate the strengths and
weaknesses of epidemiological studies (at the beginning of Section 3.1). With respect to the
recommendation to put material from Appendix A into the main body of the document, the EPA
determined that the in-depth level of detail in Appendix A was not appropriate for the main body
of the document. Instead, the EPA has added two shorter summary tables of the
lymphohematopoietic cancer (see Table 3-1) and breast cancer (see Table 3-2) findings in the
various epidemiology studies to Section 3.1.1. The EPA has also added a cross-reference to
summary Table A-5 in Appendix A at the beginning of Section 3.1. The main body of the
document provides a summary of the findings of all the epidemiological studies, referencing
Appendix A for further details.
The EPA notes that the panel agreed with the EPA's use of "internal" estimates rather
than "external" comparisons.
H-3
-------
The 2006 draft assessment did not refer to or characterize the magnitude of the unit risk
associated with EtO exposure as "weak." Rather, it was with respect to the Hill considerations
for causality (Hill. 1965) in the weight-of-evidence analysis for hazard characterization
(see Section 3.5.1) that the draft assessment noted that there was little strength in the association,
as reflected by the modest magnitude of the (relative) risk estimates from the epidemiology
studies. The exposure-response models used to develop the unit risk estimates are derived from
the NIOSHdata and are thus consistent with the results of the NIOSH epidemiology study, as
can be seen in the figures depicting RR versus exposure for the various exposure-response
models. The unit risk estimates are derived from these exposure-response models and are
similarly consistent with the results of the NIOSH study, as long as they are used in the low-
exposure range, as intended. Because the exposure-response relationships for the cancers of
interest in the NIOSH study are generally supralinear, the unit risk estimates will overpredict the
NIOSH results if applied to exposure levels that correspond to the region of the exposure-
response relationships where the responses plateau.
1. b. Are there additional key published studies or publicly available scientific reports that are
missing from the draft document and that might be useful for the discussion of the
carcinogenic hazard of EtO?
SAB COMMENT: The Panel agreed that the discussion of endogenous metabolic production
of ethylene oxide and the formation of background adducts should be expanded.
The Panel believed that the description of studies of DNA adduct formation resulting
from EtO exposure appears incomplete and superficial. This discussion should be expanded—
both in terms of the number of studies cited and the depth of the discussion.
Since ethylene is metabolized to EtO, some members recommended the inclusion of the
ethylene body of literature for consideration. Most members were hesitant about adding them to
the document, but if added, they cautioned that a discussion of the caveats associated with their
interpretation relative to ethylene oxide should be included.
EPA RESPONSE: The discussion of endogenous metabolic production of EtO and its
significance and contribution to the formation of background adducts in rodents and humans has
been expanded (see Sections 3.3.2 and 3.3.3.1 and Section C.7 of Appendix C). A discussion of
the endogenous production of ethylene during normal physiological processes and its
metabolism to EtO under certain conditions has been added (see Section C.7 of Appendix C). It
should be noted that the endogenous production of EtO due to the metabolism of endogenous
H-4
-------
ethylene will be present in all test animals or subjects (including controls); hence, this factor is
considered inherently in the analysis of effects of EtO exposure.
The discussion of DNA adduct formation resulting from EtO exposure has also been
expanded to add breadth and depth (see Section 3.3.3.1 and Section C. 1.1 of Appendix C).
Section C.l.l of Appendix C includes discussion of general DNA adduct formation, sensitivity
of the methods used to detect DNA adducts, and DNA adduct studies, both in vitro and in vivo,
that have been conducted in animals and humans.
The EPA agrees with the majority of the panel that data on (exogenous) ethylene should
not be included in the assessment. One caveat provided on page 12 of the SAB report is that the
ethylene bioassays administered ethylene concentrations with such low EtO equivalents that
they would appear "to be below the limit of detection for a tumor response over the spontaneous
background in the F344 rat." Thus, the ethylene data would not be very informative for the EtO
assessment, for which there are already adequate EtO bioassays.
The EPA considered all 34 references listed by the SAB panel in its report (p. 13-15),
and the revised draft cites all but 10 of them. The 10 references that were not cited were
considered to be not particularly relevant or necessary to the assessment: one was on propylene
oxide, one was on N-nitrosocompounds, two were on ethylene, two were related to OSHA's
review of EtO, two were mutagenicity papers from the 1970s published in a Russian journal,
one was a 1979 mutagenicity paper published in a French journal, and the last was a paper on
the emission of EtO from the frying of foods.
1. c. Do the available data and discussion in the draft document support the mode-of-action
conclusions?
SAB COMMENT: The Panel agreed with the Draft Assessment conclusion of a mutagenic
mode of action. However, an expanded discussion of the formation of DNA adducts and
mutagenicity is warranted.
EPA RESPONSE: The EPA has expanded the discussion of DNA adduct formation (see
Section 3.3.3.1 and Section C.l.l of Appendix C), mutagenicity (see Section 3.3.3 and Sections
C.2-C.5 of Appendix C), and possible mechanisms (see Section 3.4) in the revised assessment
document.
l.d. Does the hazard characterization discussion for EtO provide a scientifically balanced and
sound description that synthesizes the human, laboratory animal, and supporting (e.g., in
vitro) evidencefor human carcinogenic hazard?
H-5
-------
SAB COMMENT: While some members of the Panel found the hazard characterization section
of the Draft Assessment to be satisfactory, a majority expressed concerns that this section did not
achieve the necessary level of rigor and balance. An issue in this characterization, particularly in
the face of epidemiological data that are not strongly conclusive, is whether the presumed
precursor events leading to cancer in animals, such as mutations and/or chromosomal
aberrations, are observed in humans. This issue needs to be addressed in greater detail.
EPA RESPONSE: A more rigorous and balanced hazard characterization was incorporated into
the revised assessment (see Section 3.5.1). To address the issue of precursor events, the
genotoxicity (see Section 3.3.3 and Appendix C) and mode-of-action (see Section 3.4.1) sections
have been revised to provide a more complete and balanced discussion of EtO-induced precursor
events in laboratory animals and humans. As addressed in the EPA response under Charge
Question 1.a above, while the databases for humans and rodents contain different types of
genotoxicity studies, the EPA did not find an inconsistency in EtO-induced precursor events and
concluded that the data support a finding of a mutagenic mode of action (relevant to humans) and
that the key precursor events are anticipated to occur in humans (see Sections 3.3.3.2, 3.3.3.3,
3.4.1, and 3.5.1).
Issue 2: Risk Estimation (Section 4 and Appendices C and D of the EPA Draft Assessment)
Do the available data and discussion in the draft document support the approaches taken by
EPA in its derivation of cancer risk estimates for EtO? In your response, please include
consideration of the following:
2. a. EPA concluded that the epidemiological evidence alone was strong but less than
completely conclusive (although EPA characterized the total evidence—from human,
laboratory animal, and in vitro studies—as supporting a conclusion that EtO is "carcinogenic
to humans "). Is the use of epidemiological data, in particular the Steenland et al. (Steenland
et al., 2004; Steenland et al., 2003) data set, the most appropriate for estimating the magnitude
of the carcinogenic risk to humans from environmental EtO exposures? Are the scientific
justifications for using this data set transparently described? Is the basis for selecting the
Steenland et al. data over other available data (e.g., the Union Carbide data) for quantifying
risk adequately described?
SAB COMMENT: The Panel concurred that the NIOSH cohort is the best single
epidemiological data set with which to study the relationship of cancer mortality to the full range
H-6
-------
of occupational exposures to EtO. That said, the Panel encouraged the EPA to broadly consider
all of the epidemiological data in developing its final Assessment. In particular, the Panel
encourages the EPA to explore uses for the Greenberg etal. (1990) data including leukemia and
pancreatic cancer mortality and EtO exposures for 2,174 Union Carbide workers from its two
Kanawha Valley, West Virginia facilities. [Also described in (Teta etal.. 1999; Teta etal..
1993)1.
The Panel encouraged the EPA to investigate potential instability that may result from
interaction between the chosen time metric for the dose response model and the treatment of time
in the estimated exposure (i.e., log cumulative exposure with 15 year lag) that is the independent
variable in that dose-response model.
EPA RESPONSE: The EPA re-evaluated all of the epidemiological studies with quantitative
exposure-response data and has revised the assessment to include an expanded discussion of
study selection, including a summary table of important considerations, in Section 4.1, as well as
expanded discussions of the exposure assessments for the Union Carbide (see Appendix A,
Section A.2.20) and NIOSH(see Appendix A, Section A.2.8) studies.
In regard to the possible use of other epidemiologic data for exposure-response
modeling, the assessment document includes a detailed discussion of the studies of workers at
the Union Carbide facilities in West Virginia. Since the 2007 SAB review, analyses of the data
from an extended follow-up (through 2003) of the Union Carbide cohort, focused on the 1,896
EtO production workers who did not work in the chlorohydrin unit, have been published by
Swaen et al. (2009) and Valdez-Flores et al. (2010). This cohort is about one-tenth the size of
the NIOSH cohort. At the end of the 2003 follow-up, only 27 lymphohematopoietic cancer
deaths (including 12 leukemias and 11 NHLs) were observed in the cohort. Thus, even after
extended follow-up, the number of cases is small compared to the NIOSH study, which had 74
lymphohematopoietic cancer deaths, 53 from lymphoid cancers.
Furthermore, the Union Carbide study has a less extensive exposure assessment than the
NIOSH study. In part, the deficiency is inherent in a chemical production setting, where it is
difficult to find workers with relatively uniform work histories that involve relatively constant
exposure to EtO. The exposure assessment used by Swaen et al. (2009) for the Union Carbide
study was relatively crude, based on just a small number of department-specific (high-, medium-,
and low-exposure intensity) and time period-specific (1925-1939, 1940-1956, 1957-1973, and
1974-1988) categories, and with exposure estimates for only a few of the categories derived
from actual measurements (see Section A.2.20 of Appendix A for the details). This is in contrast
to the sterilization plants studied by NIOSH, where workers can be grouped into relatively
common jobs/work zones, facilitating assignment of exposure. Furthermore, extensive sampling
H-7
-------
data (2,350 measurements from 1975 to 1986, reduced to 205 annual job-specific means,
representing 80% of the data; another 20% were not included but used as a validation sample)
were used in the NIOSH study to estimate exposure in different jobs and years. Such sampling
data were not used in estimating exposures in the Union Carbide cohort. Finally, the NIOSH
regression model for estimating EtO exposure included data not only on job/work zone, but also
on variables such as size of sterilizer, type of product, freshness of product, and exhaust systems
for sterilizers. This regression model explained 85% of the variance in the EtO validation data
set. As a result, the exposure estimates in the NIOSH study are expected to be more accurate.
In addition to its larger size, greater number of cases, and more thorough exposure
assessment, the NIOSH study had other advantages over the Union Carbide cohort, such as the
inclusion of female workers and the absence of occupational coexposures, as documented in
Section 4.1. Furthermore, because of the lack of comparability in the exposure estimates across
the two studies, it is not possible to group together the NIOSH cohort and the Union Carbide
cohort for a rigorous combined quantitative exposure-response analysis. Thus, the EPA used the
NIOSH study alone as the basis for quantitative risk estimates, consistent with the concurrence
of the SAB panel that the NIOSH study is the best single study for that purpose.
The EPA has investigated the issue about potential instability that may result from
interaction between the chosen time metric for the dose-response model and the treatment of
time in the estimated exposure (e.g., log cumulative exposure with 15-year lag) in the NIOSH
cohort and does not consider it to be a substantial problem. The concern is apparently that the
15-year lag in the exposure metric, which discounts the most recent exposures, may cause an
over-reliance in the exposure-response analysis on exposures which were estimated prior to
1978, which may be less accurate because the NIOSH exposure model assumed that the effect of
calendar year was constant before 1978. As discussed by Hornung et al. (1994), including the
engineering controls in the NIOSH exposure model could not completely explain the decreases
in EtO levels observed since the late 1970s. Thus, Hornung et al. (1994) also included calendar
year in the model as a surrogate for improvements in work practices, above and beyond the
engineering controls, resulting from increased awareness in the late 1970s of the potential
carcinogenicity of EtO. Fitting the measurement data from 1976 to 1985 showed that the effect
of calendar year on exposure estimates was maximal between 1976 and about 1978-1979 and
reduced exposure estimates after that. Thus, the calendar year effect in the exposure model was
fixed at 1978 for years prior to 1978. Assuming the effect of calendar year to be constant before
1978 was both consistent with the available data for exposure levels prior to 1978 and reasonable
given that the increasing awareness of EtO carcinogenicity in the late 1970s could explain the
calendar year effect decreasing exposures only after that time. Exposure estimates prior to 1978
H-8
-------
were then determined entirely by the other variables in the model, for which data were available
for the years before 1978.
There is inevitably more uncertainty about the estimation of exposures prior to 1976,
when there were no sampling data. However, the use of a 15-year lag is unlikely to appreciably
increase the uncertainty that exists in the cumulative exposure estimates due to potential
measurement error in the pre-1976 exposure-level estimates, given that in the follow-up through
1998, exposures in the lagged out period for most workers would be substantially lower than
exposure before the lag came into effect. See Section D.5.2 of Appendix D for more information
on the impacts of the 15-year lag on exposure estimates.
2.b. Assuming that Steenland et al. (Steenland et al.. 2004; Steenland et al.. 2003) is the most
appropriate data set, is the use of a linear regression model fit to Steenland et al. 's categorical
results for all lymphohematopoietic cancer in males in only the lower exposure groups
scientifically and statistically appropriate for estimating potential human risk at the lower end
of the observable range? Is the use of the grouping of all lymphohematopoietic cancer for the
purpose of estimating risk appropriate? Are there other appropriate analytical approaches
that should be considered for estimating potential risk in the lower end of the observable
range? Is EPA's choice of a preferred model adequately supported andjustified? In
particular, has EPA adequately explained its reasons for not using a quadratic model
approach such as that of Kir man et al. (2004) ? What recommendations wouldyou make
regarding low-dose extrapolation below the observed range?
SAB COMMENT: The Panel identified several important shortcomings in the linear regression
modeling approach used to establish the point of departure for low dose extrapolation of cancer
risk due to EtO [note added by the EPA: more detailed comments provided by the SAB panel
about the linear regression approach and the EPA's responses are presented beginning on page
H-25], The Panel was unanimous in its recommendation that the EPA develop its risk models
based on direct analysis of the individual exposure and cancer outcome data for the NIOSH
cohort rather than the approach based on published grouped data that is presently used. The
suggested analysis will require EPA to acquire or otherwise access individual data and develop
appropriate methods of analysis. The Panel recommends that the Agency allocate the
appropriate resources to conduct this analysis.
The Panel was divided on whether low dose extrapolation of risk due to environmental
EtO exposure levels should be linear (following Cancer Guideline defaults for carcinogenic
agents operating via a mutagenic mode of action) or whether plausible biological mechanisms
argued for a nonlinear form for the low dose response relationship. With appropriate discussion
H-9
-------
of the statistical and biological uncertainties, several Panel members strongly advocated that both
linear and nonlinear calculations be considered in the final EtO Risk Assessment.
In conjunction with its recommendation to use the individual NIOSH cohort data to
model the relationship of cancer risk to exposures in the occupational range, the Panel
recommended that the Agency explore the use of the lull NIOSH data set to estimate the cancer
slope coefficients that will in turn be used to extrapolate risk below the established point of
departure. The use of different data to estimate different dose response curves should be
avoided unless there is both strong biologic and statistical justification for doing so. The Panel
believed this justification was not made in the Agency's draft assessment.
Although the analysis based on total lymphohematopoietic (LH) cancers might have
value as part of a complete risk assessment, the rationale for this aggregate grouping needs to be
better justified. The Panel recommends that data be analyzed by subtype of LH cancers (e.g.
lymphoid, myeloid) and strong consideration be given to these more biologically justified
groupings as primary disease endpoints.
The Panel was divided in its views concerning the appropriateness of estimating the
population unit risk for LH cancer based only on the NIOSH data for males. Several Panel
members pointed out that a standard approach in cancer epidemiology and risk analysis begins
by conducting separate dose-response analyses on males and females and combining the data
only if the results are similar. Conducting separate analyses for males and females is also the
standard practice when analyzing data from animal carcinogenicity bioassays. A second
approach to dealing with the possibility of gender differences in response is to include gender as
a fixed effect in the statistical modeling of the data and determine whether gender or its
interaction with other predictors (e.g., gender x exposure) are significant explanatory variables.
If so, the combined model with the estimated gender effects could be used directly or separate,
gender-specific dose response analysis would be performed. If not, the gender effects could be
dropped and the model re-estimated for the combined male and female data. In addition, the
Agency should test whether the male/female differences are mitigated by use of alternate
disease endpoints discussed in the previous paragraph.
EPA RESPONSE: The above comment from the panel addresses a variety of issues and the
EPA's responses to some of these issues are comparatively detailed; thus, the EPA has
subdivided the response into separately titled subsections to make it easier to read.
EPA response on the modeling of the individual-level data: In response to the SAB
comments, the EPA conducted extensive analyses using the individual-level (continuous)
exposure and cancer outcome data for the NIOSH cohort. These analyses are described in
H-10
-------
Section 4.1.1.2 for lymphoid cancer modeling and Section 4.1.2.3 for breast cancer incidence
modeling (no further analyses were done with the all lymphohematopoietic cancer data because
lymphoid cancer estimates are preferred or with the breast cancer mortality data because the
incidence data set is preferred). These sections also include summary tables of the key models
examined and the factors considered in model selection (see Tables 4-4 and 4-12 for lymphoid
cancer and breast cancer incidence, respectively). More details on the various models and the
model results are provided in Appendix D.
The underlying problem that makes the EtO data sets from the NIOSH cohort difficult to
model (for the purposes of environmental risk assessment) is that the exposure-response
relationships, particularly for lymphoid cancer and breast cancer mortality, are supralinear (i.e.,
the responses rise relatively steeply at low exposures and then attenuate or "plateau").
Supralinear exposure-response relationships are inherently difficult to model for the purposes of
environmental risk assessment (i.e., to estimate risk at low exposures) because the standard
single-parameter exposure-response models tend to exaggerate the low-exposure slope in order
to simultaneously fit the plateauing at higher exposures. One approach attempted by the EPA, in
consultation with Steenland, to address this difficulty was to use two-piece spline models, which
provide more flexibility and allow for the lower-exposure and higher-exposure data to befit with
different spline segments.
For the breast cancer incidence data, the EPA was able to develop several continuous
models that provided reasonable fits to the individual-level exposure data across the entire range
of the data, consistent with the SAB recommendations. The best-fitting of these models, the
two-piece linear spline model, now forms the basis for the EPA's unit risk estimate for breast
cancer incidence (see Section 4.1.2.3).
For lymphoid cancer, however, despite the extensive modeling efforts, the various
alternative continuous models investigated, including the two-piece spline models, proved
problematic, as explained in detail in the text (see Section 4.1.1.2). In particular, the statistically
significant models predicted extremely steep slopes in the low-dose region. Thus, the EPA has
retained the approach used in the 2006 external review draft assessment and has based the
preferred unit risk estimates for lymphoid cancer on a linear regression using the categorical
data, excluding the highest exposure group. In consideration of the SAB recommendation,
however, unit risk estimates from the most suitable alternative model based on the continuous
exposure data were developed and added to the assessment for comparison purposes.
While the EPA understood and appreciated the SAB's recommendation and did much work
to model the individual-level data for lymphoid cancer, it should be noted that modeling of grouped
data is an important and well-recognized statistical methodology and its use is consistent with EPA
guidance, policy, and past practice. For example, the EPA's 2005 Guidelines for Carcinogen Risk
H-ll
-------
Assessment (U.S. EPA. 2005a) specifically recognize the use of linear modeling of grouped
epidemiological data ("For epidemiologic studies, including those with grouped data, analysis by
linear models in the range of observation is generally appropriate unless the fit is poor," p. 3-11). In
addition, the EPA's approach of using a weighted linear regression through the categorical
relative risk estimates follows established statistical procedures (van Wijngaarden and Hertz-
Picciotto, 2004; Rothman, 1986).
The breast cancer mortality data displayed similar extreme supralinearity, and the optimal
two-piece spline model yielded an unrealistically steep low-dose slope estimate; thus, the EPA
again used a linear regression of the categorical data, excluding the highest exposure group (see
Section 4.1.2.2). In consideration of the SAB recommendation, however, a unit risk estimate for
breast cancer mortality from the most suitable alternative model based on the continuous
exposure data was developed and added to the assessment for comparison purposes. The breast
cancer mortality data, however, are not critical to the assessment because the breast cancer
incidence data set is preferred (see Section 4.1.2.3).
EPA response on the use of a nonlinear approach to low-exposure extrapolation:
The EPA has given careful consideration to the range of perspectives provided in the SAB report
on the issue of low-dose extrapolation, including the viewpoint expressed by several panel
members who advocated that both linear and nonlinear calculations be considered in the EtO
assessment. It is the EPA's judgment, as detailed below, that the inclusion of a nonlinear
approach is not warranted.
As discussed in Chapter 3 of the assessment, EtO is a DNA-reactive, mutagenic, multisite
carcinogen in humans and laboratory animal species; as such, it has the hallmarks of a compound
for which low-dose linear extrapolation is strongly supported. The EPA's Guidelines for
Carcinogen Risk Assessment (U.S. EPA, 2005a) specifically note the use of low-dose linear
extrapolation for "agents that are DNA-reactive and have direct mutagenic activity." Appendix
A of the SAB report also provides support for low-dose linearity for genetically acting agents,
noting, for example, that additivity to background carcinogenic processes at low doses is
expected to result in incremental risk that approaches linearity, as discussed by Crump et al.
(1976). By comparison, the Guidelines recommend that, "A nonlinear approach should be
selected when there are sufficient data to ascertain the mode of action and conclude that it is not
linear at low doses and the agent does not demonstrate mutagenic or other activity consistent
with linearity at low doses." The EPA's analysis indicates that EtO does not meet any of those
conditions. For EtO, there is sufficient weight of evidence to support a mutagenic/genotoxic
MO A, without compelling evidence of additional or alternative MO As being operative
(see Section 3.4.1).
H-12
-------
The EPA specifically considered a two-hit MO A proposed by Kirman etal. (2004) to
support a (nonlinear) quadratic model for leukemia. The Kirman et al. (2004) proposal was
based on several assumptions, and the EPA concluded that the evidence was inadequate to
substantiate the assumptions supporting use of the quadratic model, as discussed in detail in
Section 3.4 of the assessment.
With regard to the particular comments of the SAB members advocating presentation of a
nonlinear approach, the SAB report (p. 23) suggests that linear extrapolation "is a conservative
assumption, given EtO's reactivity (which will diminish the amount reaching the nucleus),
mutagenic mode of action, and that it is generated endogenously" and that "[s]ome repair seems
likely and some threshold probably exists." The evidence is ample, however, that EtO from both
endogenous and exogenous sources reaches the nucleus and forms adducts (see Section 3.3.3.1
and Section C.l.l of Appendix C), and more recent data from Marsden et al. (2009) specifically
demonstrate (nonsignificant) increases of DNA adducts for very low exposures to exogenous
EtO (see Section 3.3.3.1). Any diminution of the amount of EtO reaching the nucleus is
expected to affect the slope of the low-dose linear relationship but not linearity per se. Similarly,
the fact that endogenous EtO is present and that some repair takes place is not considered
evidence against low-dose linearity because the low doses of exogenous EtO are expected to
contribute to background carcinogenic processes for the common cancers, such as lymphoid
cancer and breast cancer, associated with EtO exposure. The SAB report itself, in that same
paragraph presenting the argument for nonlinearity (p. 23), acknowledges that a "linear model
per se at low doses is acceptable."
Additional reasons for using a nonlinear approach expanded upon in Appendix C of the
SAB report were largely general suppositions that (1) DNA adducts may show a nonlinear
response when identical adducts are formed endogenously, and (2) mutations do not have linear
relationships with exposure but exhibit an "inflection point." However, more recent data from
Marsden et al. (2009) support a linear exposure-response relationship for EtO exposure and DNA
adducts. Although they caution that their study was not designed to test for linearity and that
some of the adduct levels induced at low EtO concentrations are below the limit of accurate
quantitation, Marsden et al. (2009) reported statistically significant linear dose-response
relationships (p < 0.05) for exogenous adducts in all three tissues examined (spleen, liver, and
stomach) and measured increases of DNA adducts from exogenous EtO exposure above those
from endogenous EtO for very low exposures to exogenous EtO, as discussed in detail in the
assessment (see Section 3.3.3.1 and 4.5), providing evidence against the first reason proposed in
support of a nonlinear approach in Appendix C of the SAB report. In support of the second
reason, Appendix C of the SAB report presents two EtO-specific mutation data sets; however,
the EPA's analysis of these data sets, summarized below, finds that they are in fact consistent
H-13
-------
with low-dose linearity. In summary, the EPA's review of studies addressing dose-response
patterns for adduct formation and mutagenesis by EtO finds these data to be supportive of the
inferences made in the EtO assessment [and more broadly in the EPA's Guidelines for
Carcinogen Risk Assessment (U.S. EPA 2005a)1 regarding the plausibility of linear,
nonthreshold, low-dose dose-response relationships for the biological effects of EtO, which is
mutagenic and directly damages DNA.
The EPA further notes that the supralinear exposure-response relationships from the
NIOSH data at low occupational exposures argue against the existence of a "threshold," practical
or otherwise, at exposure levels anywhere near the POD. Also, the rodent bioassays do not
suggest an absence of increased cancer risk at their lowest exposure levels.
Analysis of the EtO mutagenicity data sets presented in Appendix C of the SAB Report:
In Appendix C in the SAB report, one reviewer provided slides (numbers 25 and 26)
showing dose-response data for hprt mutations in mice exposed to either EtO or to ethylene. For
ethylene, a model estimate of an EtO-equivalent concentration was used to represent metabolism
of ethylene to EtO. In both cases, mutations at the hprt locus ofT-cells isolated from spleens of
Big Blue mice were quantified. The EtO study results are from Walker et al. (1997), and it
appears that the ethylene results are derived from experiments presented in Walker et al. (2000).
In the latter case, there are some differences in the estimated EtO equivalents and the hprt
mutation frequencies between the values given in the slide and those reported by Walker et al.
(2000). The EPA performed statistical analyses using the data presented in slide 26 of Appendix
C.
To examine these data, the EPA first analyzed the EtO data set (Walker et al., 1997)
using maximum likelihood estimation (MLE). The EPA then looked at the consistency of the
ethylene data set (Walker et al., 2000) with the EtO data set. The EtO data were fit with a linear
model using a log-normal distribution of the individual animal response measurements due to the
low mutant frequency that causes skewness of the data. As shown in Figure H-l, this model
provided an adequate fit to the EtO data (open circles represent individual animal data for the
EtO exposures; model goodness-of-fit p = 0.09; variance fit assuming homogeneous variance in
log scale, p = 0.64). The MLE of the model is plotted (geometric mean [solid line] as an
estimation of the median response along with the lower and upper 2.5 percentiles of the model
[dashed lines]). The second, ethylene-derived, data set is plotted on the same graph (closed
circles). The predicted EtO-equivalents from the ethylene data set fall well below the lowest
dose level used in the EtO experiment, a range in which the EtO-based model would predict only
a small response (i.e., no more than a 25% increase in mutation rate above background, a level
H-14
-------
that cannot be expected to be detectable given the variability in the EtO experimental data; see
Figure H-l). The fact that the ethylene exposures did not show measureable increases in hprt
mutations is consistent with the modeled EtO results.
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).
Note, however, that all medians of the ethylene-derived data are at or below the
EtO-based model and one of the points is below the lower 2.5 percentile of the model, indicating
that this point is unlikely to be consistent with the same model. To further investigate the
compatibility of the data from the two experiments, the EPA analyzed the combined data set by
including a term that represents the source of the data (the EtO vs. ethylene experiments) into the
modeling (as above). This experimental variable was significant (p < 0.05), indicating that there
is a systematic difference in response between the EtO and ethylene-derived data. As a further
check, the EPA refit the data using an exponential model that provided an MLE fit with a degree
of upward curvature (but still having low-dose linear behavior). Using a categorical
experimental variable within this experiment also indicated a systematic dependence of results
on data source (EtO vs. ethylene), indicating that this finding was not dependent on the choice of
a straight-line dose-response model. As an additional sensitivity analysis, the EPA reran the
H-l 5
-------
modeling using the 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, the EPA concluded that combining
the ethylene data with EtO data in evaluating dose-response relationships for the hprt mutations
might not be appropriate.
Slide 27 of the SAB report presents data from Nivard etal. (2003) on the frequency of
recessive lethal mutations in Drosophila exposed to EtO [frill data set presented in Vogel and
Nivard (1998)1. Plotting of mutation rate versus EtO concentration for wild-type Drosophila on
nonlog-transformed axes shows a downward curving (supralinear) relationship indicating greater
potency of EtO (per unit exposure) at low exposures as compared with high exposures (see
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, the EPA's 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 or an "inflection point" at low doses of EtO; that is, they do not indicate a low-dose
nonlinear or threshold-type dose-response pattern. Thus, the EPA's review finds these data to be
supportive of the inferences made in the assessment [and more broadly in the EPA's 2005
Guidelines for Carcinogen Risk Assessment (U.S. EPA, 2005a)l regarding the plausibility of
linear, nonthreshold, low-dose dose-response relationships for the carcinogenic effects of EtO,
which is mutagenic and directly damages DNA.
H-16
-------
c
0
y
0
Q.
CT
C
0
15
cr
4—
c
o
¦-I—»
CO
-t—'
15
o
c\i
o
o
o
200
400
600
800
1000
Ethylene oxide concentration, ppm
Data: Nivard (2003)/Vogel and Nivard
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, the panel
comment pertains only to the occupational exposure scenarios. This is addressed in the EPA's
response to the SAB comment on Charge Question 2.d below.
EPA response on lymphohematopoietic cancer groupings: As recommended by the
panel, the primary risk estimates in the revised assessment are based on the analysis of the
lymphohematopoietic cancer subtype of lymphoid cancers (see Section 4.1.1.2), which was the
H-17
-------
subtype with the strongest evidence of an EtO association in the NIOSH data set (Steenland et
al.. 2004). Analysis based on total lymphohematopoietic cancers is also included for
completeness and comparison purposes.
EPA response on the use of only the male data for lymphohematopoietic cancers:
Subsequent analyses by Steenland determined that there was not a statistically significant
difference between the lymphohematopoietic cancer results for males and females (see Sections
D.3.3.1 and D.4.2.1 of Appendix D). Thus, in the revised assessment, data on males and females
were combined as appropriate, and unit risk estimates are now based on lymphoid cancers for
males and females combined and breast cancer in females.
The following additional comments on page 31 of the SAB panel report under "2.b.
Methods of Analysis: 7. Statistical issues," are quoted verbatim below followed by the
EPA's responses:
SAB COMMENT:
7. Statistical issues
Pages 29-49 of the draft Evaluation outline the EPA's proposed approach to estimation
of the Inhalation Unit Risk for EtO. In addition to the general issues of estimation and model-
based extrapolation described above, there are a number of statistical assumptions and methods
used in this approach that deserve mention.
Conditional on the cancer slope factor results from the weighted least squares regression
analysis, the life table (BEIR IV) approach to the determination of the LEC01 is programmed
correctly.
The life table methodology that is the basis for the BEIR IV algorithm is designed to
estimate excess mortality and is not readily adapted to modeling excess risk for events
(incidence) that do not censor observation on the individual in population under study. The
methodology for substituting the mortality slope to an excess risk computation for HL cancer
incidence requires the assumption of a proportional rate of incidence/mortality across the cancer
types that are included in the grouped analysis. This is generally not a viable assumption. The
Panel therefore discourages the use of the BEIR IV algorithm for extrapolation of the cancer
mortality algorithm to estimation of excess cancer incidence.
Several Panel members commented on the use of the upper confidence limit for the
estimated slope coefficient as the basis for estimating anLECoi. The Panel encourages the EPA
to present unit risk estimates based on the range of ECo lvalues corresponding to the lower 95%
H-18
-------
confidence limit, the point estimate, and the upper 95% confidence limit for the estimated cancer
slope coefficients from the final dose-response models.
EPA RESPONSE: The above comment from the panel addresses a variety of issues and the
EPA has subdivided the response into separately titled subsections to make reading it easier.
EPA response on using the Committee on the Biological Effects of Ionizing
Radiation [BEIR] approach to estimate incidence risks: In this assessment, the EPA's
preferred unit risk estimates are those for cancer incidence, not mortality, as the cancers
associated with EtO exposure (lymphohematopoietic, in particular lymphoid, and breast cancers)
have substantial survival rates. Regarding the breast cancer incidence estimates, the assumption
that a cancer mortality exposure-response model applies to cancer incidence was not needed
because the model used for the breast cancer incidence estimates was based on incidence data.
In addition, although the BEIR approach was designed for mortality estimates, the EPA believes
it has made a suitable adjustment to the approach by redefining the population at risk as those
alive and without a diagnosis of breast cancer at the beginning of the age interval (rather than
those alive at the beginning of the interval). This adjustment was not made in the life-tables for
the lymphoid cancer estimates because, unlike for breast cancer incidence rates, lymphoid cancer
incidence rates (actually, the differential rates obtained by subtracting the mortality rates from
the incidence rates) are negligible in comparison to the all-cause mortality rates.
Regarding the lymphoid cancers, the SAB provided the relevant comment that
mathematically the BEIR formula would apply to the case where there is a proportional rate of
incidence/mortality across the cancer types that are included in the grouped analysis. The EPA
considered this in its application of the BEIR formula. The fact that the ratios of incidence to
mortality are not strictly proportional contributes some uncertainty to the incidence estimates for
the grouping of lymphoid cancers, but not a large amount. Uncertainties in using the life-table
analysis approach to seek to develop reasonable estimates for incidence risk, including those
noted by the SAB, are acknowledged in the assessment, and the impact of nonproportionality
among cancer types is one of the uncertainties discussed (see Section 4.1.1.3). As illustrated in
the assessment, these uncertainties do not have a major impact on the final risk estimates. The
incidence unit risk estimate is about 120% higher than (i.e., 2.2 times) the mortality-based
estimate, which is consistent with the relatively high survival rates for lymphoid cancers.
Potential concern that the incidence unit risk values might be overestimated would come
primarily from the inclusion of multiple myeloma because that subtype has the lowest
incidence:mortality ratios (and thus, if that subtype were driving the increased mortality
observed for the lymphoid cancer grouping, then including the incidence rates for the other
H-19
-------
subtypes, which have higher incidence mortality ratios, in the cause-specific background rates in
the life-table might inflate the incidence estimates). Multiple myelomas, however, constitute
only 25% of the lymphoid cancer cases, and there is no evidence that multiple myeloma is
driving the EtO-induced excess in lymphoid cancer mortality (25% is below the proportion of
multiple myeloma deaths one would expect in the cohort based on age-adjusted background
mortality rates of multiple myeloma, NHL, and chronic lymphocytic leukemia, and these 3
subtypes have the same pattern of mortality rates increasing as a function of age mostly above
age 50, so the comparison with lifetime background rates is reasonable). Thus, using the total
lymphoid cancer incidence rates is not expected to result in an overestimation of the incidence
risk estimates; if anything, the incidence risks would likely be diluted with the inclusion of the
multiple myeloma rates.
The panel's suggestion to not use the BEIR approach for development of cancer
incidence estimates for lymphoid cancer would not allow for the development of the desired
cancer incidence risk estimates. Deriving incidence estimates from mortality data is consistent
with EPA guidance, which suggests making adjustments to reflect the relationship between
incidence and mortality fU.S. EPA (2005a), p. 3-121. A possible alternative approach involving
a crude survival adjustment to the mortality-based estimates would yield results with greater
uncertainty than those from the life-table approach used. No alternative approaches were
identified by the SAB. In the absence of an appropriate alternative approach to estimate risks of
cancer incidence, the EPA has retained the application of the BEIR (life-table) approach, which
it judges to provide a reasonable estimate of incidence risks. The EPA recognizes the
uncertainties and assumptions outlined by the panel and has expanded the discussion of these in
the carcinogenicity assessment (see Section 4.1.1.3). However, the EPA notes that deriving
mortality estimates as the sole cancer risk estimates for lymphohematopoietic cancer would
substantially underestimate cancer risk. In addition, the EPA presents the mortality-based
estimates for comparison, and as discussed above, the lymphoid cancer incidence unit risk
estimate is about 120% higher than (i.e., 2.2 times) the mortality-based estimate, which is
considered reasonable, given the high survival rates for lymphoid cancers.
EPA response on the use of upper and lower confidence limits: In both the 2006 and
revised drafts of the EtO assessment, the EPA presents 95% (one-sided) lower bounds and
central estimates of the ECoisas well as standard errors for the regression coefficients used in the
modeling, which provide information about the variability in the modeled slope estimate. The
EPA's Guidelines for Carcinogen Risk Assessment (U.S. EPA, 2005a) also recommend the
calculation of a 95% upper bound on the central estimate (in this case the ECoi) related to the
POD "to the extent practicable" ITJ.S. EPA (2005a), p. 1-14], and this value has been added to
H-20
-------
the revised assessment for the selected breast cancer incidence model (see Section 4.1.2.3, Table
4-13, footnote j, based on the profile likelihood confidence limits for the regression coefficient).
However, for the linear regression model used as the basis for the lymphoid cancer unit risk
estimate, it was not practicable to calculate such a value, as it was undefined. Although there
were models for lymphoid cancer from which upper bounds could have been calculated, the
linear regression model was selected as the basis for the POD for the express purpose of
obtaining a realistic slope estimate for the low-exposure region (see Section 4.1.1.2) and not for
providing a realistic upper-bound estimate fortheECoi.
The EPA considered the SAB panel comment encouraging the EPA "to present unit risk
estimates based on the range ofECoi values corresponding to the lower 95% confidence limit,
the point estimate, and the upper 95% confidence limit." However, as a consequence of the
two-step approach used by the EPA to generate cancer potency estimates from a POD rather than
directly from the statistical model used to estimate the POD, potency estimates below the
response level corresponding to the POD are no longer associated with the statistical model.
Linear extrapolation from a POD that is the 95% (one-sided) lower bound on the central estimate
of the exposure concentration associated with the selected (benchmark) response level (e.g., the
LECoi) might be generally expected to yield a reasonable upper bound on cancer risk for that
data set (although not strictly a statistical "95%" upper bound). In contrast, estimates involving a
linear extrapolation from the upper bound on that central estimate are not generally meaningful
and could be misleading if they are mistaken for lower bounds on potency, as the actual
exposure-response relationship may exhibit some sublinearity below the response level
corresponding to the POD. Thus, it has not been EPA practice to develop potency estimates
based on the upper 95% confidence limit on the ECoi, and the EPA did not undertake to develop
any for this assessment. (The EPA does present the standard upper-bound unit risk estimates
based on the LECois [e.g., Table 4-22] as well as "0.01/ECoi" estimates [Table 4-23].)
2. c. Is the incorporation of age-dependent adjustment factors in the lifetime cancer unit risk
estimate, in accordance with EPA '.s Supplemental Guidance (U.S. EPA, 2005b), appropriate
and transparently described?
SAB COMMENT: In accordance with EPA guidance, the Draft Assessment applied an Age
Dependent Adjustment Factor (ADAF) to adjust the unit risk for early life exposure. While the
majority of the Panel felt that the application of a default value by the Agency was appropriate
due to lack of data, the description in the Draft Assessment was not adequate, particularly for
those not familiar with the EPA's Supplemental Guidance.
H-21
-------
EPA RESPONSE: The EPA has added a new subsection (see Section 4.4) detailing the
application of the ADAFs.
2.d. Is the use of different models for estimation of potential carcinogenic risk to humans
from the higher exposure levels more typical of occupational exposures (versus the lower
exposure levels typical of environmental exposures) appropriate and transparently described
in Section 4.5?
SAB COMMENT: While the method was transparently described, most of the Panel did not
agree with the estimation based on two different models for two different parts of the dose
response curve (see response to 2b). The use of different data to estimate different dose response
models curves should be avoided unless there is both strong biological and statistical justification
for doing so. The Panel believed this justification was not made in the Agency's draft report.
EPA RESPONSE: For the breast cancer incidence risk estimates, a single model, the two-piece
linear spline model, is now recommended for the occupational exposure scenarios. The two-
piece linear spline model is a unitary model comprised of two linear pieces or segments with
different slopes that are joined at a point referred to as a "knot." The two-piece linear model has
the flexibility to represent situations, such as with EtO, where the relationship between exposure
level and response changes over the range of exposure. For lymphoid cancer risk estimates, the
preferred model for the occupational exposure scenarios of interest to the EPA, the
log-cumulative exposure Cox regression model, is applicable over the entire range of
occupational exposure scenarios of interest. A second model, the linear regression of the
categorical results, is provided should exposure scenarios involving lower exposures be of
interest at some future time or to other parties. Thus, two models are presented for the lower-
exposure exposure scenarios, but just one of the models is recommended for the higher-exposure
exposure scenarios; users have the option of using a single model across the range of exposure
scenarios or of transitioning across models, depending on the exposure scenarios of interest, and
some further guidance on choice of approach has been added in Section 4.7 of the revised
assessment. As discussed in the assessment, the log-cumulative exposure model, which provides
a good fit to the data in the plateau and is suitable for exposure scenarios with cumulative
exposures in that region, is not appropriate for the low-exposure region (i.e., below the range of
the occupational scenarios presented in this assessment) because such a steep increase in slope is
considered to be biologically implausible and the good statistical global fit of the model should
not be over-interpreted to infer that the model provides a meaningful fit to the low-exposure
region. Likewise, the linear regression used to model the lower-dose exposure groups is not
Ft-22
-------
intended to reflect the exposure-response relationship in the higher-exposure region. Hence, for
lymphoid cancer, the use of both models may be required to cover a broader range of
occupational exposure scenarios. Table 4-19 of the assessment shows how results from the two
models compare over a range of exposure scenarios for which either model might be used.
2.e. Are the methodologies used to estimate the carcinogenic risk based on rodent data
appropriate and transparently described? Is the use of "ppm equivalence" adequatefor
interspecies scaling of EtO exposures from the rodent data to humans?
SAB COMMENT: The ppm equivalence method is a reasonable approach for interspecies
scaling of EtO exposures from rodent data to humans. If the use of animal data becomes more
important (i.e., the principal basis for the ethylene oxide unit risk value), more sophisticated
approaches such as PBPK modeling should be considered.
EPA RESPONSE: The EPA notes the panel's support for the use of the ppm equivalence
method. As the unit risk value is based on human data, the consideration of more sophisticated
models was not warranted.
Issue 3: Uncertainty (Sections 3 and 4 of the EPA Draft Assessment)
1. EPA '.s Risk Characterization Handbook requires that assessments address in a transparent
manner a number of important factors. Please comment on how well this assessment clearly
describes, characterizes and communicates the following:
a. The assessment approach employed;
b. The use of assumptions and their impact on the assessment;
c. The use of extrapolations and their impact on the assessment;
d Plausible alternatives and the choices made among those alternatives;
e. The impact of one choice versus another on the assessment;
f Significant data gaps and their implications for the assessment;
g. The scientific conclusions identified separately from default assumptions and policy calls;
h. The major risk conclusions and the assessor's confidence and uncertainties in them; and
i. The relative strength of each risk assessment component and its impact on the overall
assessment.
SAB COMMENT: The Panel has responded to Charge Questions 1 and 2 and has tried to
incorporate their comments regarding Charge Question 3 within those responses. A separate
H-23
-------
response for Charge Question 3 was not deemed necessary since issues of uncertainty were
addressed in the responses to charge questions 1 and 2 [p. 9],
The following are detailed comments on the regression modeling used in the draft ethylene
oxide assessment quoted from the SAB ethylene oxide panel report (related to Charge
Question 2.b; p. 24—26) and the EPA response:
SAB COMMENT:
2. Linear regression model for categorical data
The Panel identified several important shortcomings in the linear regression modeling
approach used to establish the point of departure for low dose extrapolation of cancer risk due to
EtO. Based on its review of the methods and results presented at the January 17,18, 2007
meeting, the Panel was unanimous in its recommendation that the EPA develop its risk models
based on direct analysis of the individual exposure and cancer outcome data for the NIOSH
cohort. The Panel understands that these data are available to EPA analysts upon request to the
CDC/NIOSH. The Panel recognizes the burden that areanalysis of the individual data places on
the EPA ORD staff but given the important implications of the risk assessment, this burden is
well justified to achieve the best scientific and statistical treatment of all the available
epidemiological data.
The following paragraphs present the statistical basis for the Panel's assessment of the
linear regression model approach and the use of categorized exposure and outcome data.
The approach described in the Draft Assessment uses a model based on categories
defined by cumulative exposure ranges for male subjects in the NIOSH cohort. Steenland et al.
identified several models that provide a significant (p < 0.05) fit to the exposure data; however,
the EPA has elected to use model-based relative rate parameter estimates for categories of 15
year lagged, cumulative exposure. In Steenland etal. (2004) this model was not one that
provided a significant fit to the NIOSH data (p = 0.15 for the likelihood ratio test of P = [01, P2,
P3, P4] = 0). The use of the weighted least squares regression fit of a linear regression line
through the three data points defined by the estimated rate ratios and mean cumulative exposures
for the first three exposure categories of the Steenland etal. 15 year lag, cumulative exposure
category model is not a robust application of this technique. The Panel identified four
weaknesses in the approach.
a) Model-based dependent variable: The dependent variables are model-based estimates
of rate ratios for exposure categories. The rate ratio values used in the weighted least squares
regression are derived from a cumulative exposure model (15 year lag) in which the estimated
regression parameters in the proportional hazards regression model are not significantly different
H-24
-------
from 0 at a = 0.05 (p = 0.15). In Steenland et al. (2004). the only individually based
(proportional hazards) model that fits the data for males in the NIOSH cohort is a model for log
of individual exposure through t-15 years.
b) Grouped data regression: The weighted least squares fit applies estimates of variance
for the individual rate ratios under that assumption that these inverse weighting corrections
correctly adjust for heteroscedasticity of residuals in the underlying regression model.
Historically, models for grouped proportions applied adjustments of this type but it is by no
means a preferred technique when the underlying individual data are available. The "ecological
regression" model per Rothman (Rothman and Greenland. 1998) is subject to bias due to within
group heterogeneity of predictors and unmeasured confounders. The heterogeneity in the
grouped model involves the range of exposures within the collapsed categories. The unmeasured
confounders include variables (other than gender) that affect the potency of exposure or may
have produced gross misclassification based on the original exposure model estimation for the
individual (Hornung et al., 1994).
c) The model fitting does not conform exactly to the Rothman (1986) procedure: The
1998 (Second edition) of Rothman (Rothman and Greenland, 1998) describes the technique for
estimating this risk from grouped data in Chapter 23. In that updated version of the original
monograph the model that is fitted is:
Expected (Rate I Exposure) = B0+Bl * Mean(Exposure)
The objective is to estimate the rate ratio (for exposure 0=no, l=yes, or equivalently for a
one unit increase in the exposure metric). That estimator is then:
rr = 1 + / B0
The model estimated by the EPA method is:
Expected (rr / Exposure) = B* * Mean(Exposure)
In the former, the variance in the estimation of the rate ratio is a function of the variance
of the estimated slope and the variance in the estimated baseline hazard, represented by the
estimated intercept. This variance is present in the estimation of the baseline hazard in the
Steenland et al. (2004) estimation of the rate ratios but is not present in the EPA adaptation to the
linear rate ratio model. The EPA approach permits no intercept (>0) for the background
H-25
-------
exposure or any allowance for an effect of true non-zero exposures in the internal control group
(exposures less than 15 years).
In general, the use of categorical exposure ranges is not the optimal strategy for using
epidemiologic data. When continuous data are categorized and then used in dose response
modeling, it amounts to starting with a lull range of exposures, collapsing that range into
somewhat arbitrary boundaries and then deriving a continuous dose response model for an even
larger range of exposures.
Categorizing continuous variables results in a host of issues:
Assumption that the risk within the category boundaries is constant.
It is not known whether a given categorization is representative of the data since
there are many ways of categorizing.
Loss of power and precision by spending degrees of freedom on each category.
Misclassification at category boundaries (this can be minimized by choosing
cutpoints where relatively few observations are present).
Categorizations can be manipulated to show the desired results.
The Panel acknowledged that techniques such as the linear regression method described
by Rothman and Greenland (1998) or Poisson regression may be the most appropriate techniques
when only grouped or categorized data are available for estimating the dose/response model.
However, the original NIOSH cohort data are available at the individual level and this permits
the use of models such as the Cox regression models employed by Steenland etal. (2004) that
utilize the lull information in the individual observations. If categories of exposure (as opposed
to individual exposure estimates) must be used, the crude rates should be computed for a large
number of equally spaced exposure ranges and the Rothman and Greenland (1998) model fitted
to these multiple points.
EPA RESPONSE: The EPA agrees that it may be generally preferable to develop risk models
on the basis of direct analysis of individual exposure and cancer outcome data. The 2006 draft
assessment included the presentation of models based on fitting Cox regression models to
individual exposure-outcome data for EtO. The Cox regression models with log cumulative
exposure provided reasonable fits to the data, as described by Steenland et al. (2004) and in the
2006 draft assessment. However, the EPA concluded that these models represented
exposure-response relationships that were excessively sensitive to changes in exposure level in
the low-dose region, and thus, were not biologically realistic. That is, in the low-dose region,
H-26
-------
these models would yield extremely large changes in response for small changes in dose level.
Accordingly, the judgment was that these models would not be suitable as the basis for low-dose
unit risk values. This is what led the EPA to use the regression methodology with the published
grouped data. The grouped data regression methodology is considered to be a valid procedure
for analysis of such data, and, as mentioned above with respect to Charge Question 2.b, the
EP A's Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a) specifically recognize the use
of linear modeling of grouped epidemiological data (U.S. EPA. 2005a); therefore, the EPA has
retained its use for some endpoints (lymphoid cancer and breast cancer mortality) in the revised
assessment and implemented it as described by Rothman (1986) [also described in van
Wijngaarden and Hertz-Picciotto (2004)1.
The EPA followed the panel's recommendation and performed additional analyses of the
individual data in collaboration with Steenland. The work performed by Steenland is described
in Appendix D of the revised assessment. Working with Steenland, alternative models based on
direct analysis of all individual data using (1) linear relative risk models (Laneholz and
Richardson. 2010) and (2) two-piece linear and log-linear spline models [e.g., Rothman etal.
(2008)1 were developed and evaluated. In the revised assessment, linear low-dose risk
estimates based on the two-piece linear spline model (using the Langholz-Richard son linear
relative risk approach) were used for breast cancer incidence risk estimates. Additional
responses to specific comments follow:
a) Model-based dependent variable: The rate ratios for the exposure categories were not
all statistically significant, likely due to loss of power from categorizing the data (in the draft
that the SAB reviewed, which was based on the results in males only, it is true that none of the
RR estimates for the lower three quartiles was statistically significantly elevated; in the revised
draft, based on both sexes, the RR estimate for the 2nd quartile is statistically significant). The
fact that the log cumulative Cox regression model is statistically significant for the continuous
exposure data, however, establishes that there is an exposure-response trend for these data.
Despite the lack of statistical significance for some of the categorical RR estimates, the EPA
used the categorical results because they provide the best available estimates of the RRs for the
limited exposure ranges reflected in each category, and these estimates were felt to be adequate,
particularly for the three lowest quartiles (the highest exposure quartiles, which represent large,
open-ended exposure ranges, were excluded from the linear regression models), for use in the
linear regression model.
b) Grouped data regression: The panel comments identify assumptions inherent in the
method. The EPA does not believe, however, that these assumptions preclude the use of the
Rothman model in the context of the EtO cancer risk estimation. The EPA disagrees with the
H-27
-------
suggestion that unmeasured confounders may have produced gross misclassification and
somehow impaired the exposure model estimation for individuals. The estimation performed
by NIOSH to estimate individual worker exposure (Hornung et al.. 1994) was extensive and
detailed. The resulting model used to estimate worker exposure accounted for 85% of the
variation in average EtO exposure (see Section 4.1 and Section A.2.8 of Appendix A). Thus,
unmeasured confounding, while possible, is unlikely to be substantial. The EPA agrees with the
panel that the exposure analysis of Hornung etal. (1994) is an example of an "exemplary
quantitative analysis of likely errors in exposure estimates." In response to the panel's
suggestion that the Hornung analysis represents an "invaluable opportunity" for further analysis
of the impact of possible errors in exposure estimation, the EPA investigated the possible use of
the "errors in variables" approach (page 27 of the panel report). Steenland visited the NIOSH
offices in Cincinnati in order to review the data and assess whether it would support an errors-
in-variables analysis. Unfortunately, the electronic data files used in the exposure analysis were
no longer available, so that analysis based on the errors-in-variables approach was not possible.
c) The EPA reviewed the statistical procedure for modeling categorical data using the
methodology in Rothman (1986). This review confirmed that the Rothman procedure was
followed closely. The equations used, which are the same as those in Rothman (1986) (pp.
341-344), are described in Appendix F. The equations are also provided in van Wijngaarden
and Hertz-Picciotto (2004). The Rothman (1986) procedure, which is appropriate for
case-control data such as the NIOSH data, is based on estimating the effect at each response
level relative to the reference or baseline level. Thus, the effect estimates are relative rates
(odds ratios), not absolute rates as used in the approach of Rothman and Greenland (1998) cited
by the SAB. The rate ratio in the referent group (i.e., those with estimated cumulative
exposure = 0) is 1.0, by definition and without an associated estimate of variability; hence, there
is no intercept term in the model. As described by Rothman (1986) (p. 345), variability in the
reference category is necessarily entrained in estimates of the slope. As Rothman (1986) points
out, this can result in loss of estimation efficiency but nevertheless yields a valid estimate of
trend. Thus, while it is true, as the comment states, that this procedure may not be optimal in a
theoretical sense, it can provide a useful mechanism for estimating linear trend. The panel
acknowledges that a linear regression may be the most appropriate approach when only grouped
data are available. The EPA agrees but would add that when the objective is low-dose risk
estimation, the approach may yield the most useful results from a pragmatic perspective. The
availability of individual data does not preclude the use of the Rothman (1986) grouped data
regression methodology. [See also the summary and review of the paper by Valdez-Flores and
H-28
-------
Sielken (2013) in Section J. 3.1 of Appendix J for a discussion of limitations in estimating the
intercept when conducting a linear regression of the categorical results for the EtO data sets.]
In the case of the EtO data, it was possible to derive theoretically correct models via
direct analysis of the individual data. In the case of the breast cancer incidence data, this
approach yielded a model that provided a suitable basis for unit risk estimation. For the other
data sets (breast cancer mortality, lymphoid cancer mortality), however, most of the models
derived using all the individual data were not useful for unit risk estimation because of
excessive sensitivity in the low-dose range. The large sensitivity of the models to small changes
in low-dose values results in unstable low-dose risk estimates lacking in biological plausibility;
thus, the Rothman procedure was used. In consideration of the SAB recommendation, however,
unit risk estimates from the most suitable alternative models for lymphoid cancer and breast
cancer mortality based on the continuous exposure data were developed and added to the
revised assessment for comparison with the results of the linear regression of the categorical
results, which was still the preferred model for reasons detailed in the revised assessment (see
Sections 4.1.1.2 and 4.1.2.2).
Responses to SAB panel 'bullet' comments (contained within the SAB comment on page H-26
above):
• Assumption that the risk within the category boundaries is constant.
EPA RESPONSE: The EPA is not assuming that within-category risk is constant. Instead, the
assumption is that observed risk within a category may be averaged over a category even though
there may be a trend within the category. This is a conventional approach in epidemiological
analyses in which categorical analysis is used.
• It is not known whether a given categorization is representative of the data since there are many
ways of categorizing.
EPA RESPONSE: The data groupings used in the EPA analyses were based on sound
statistical principles and standard epidemiological practice and were subject to peer review
through the publications of Steenland et al. (2003) and Steenland et al. (2004). The categories
were generally quartiles based on the distribution of cumulative exposures for the cases of the
cancer of interest, resulting in essentially the same number of cancer cases per quartile, a typical
approach in epidemiological studies.
H-29
-------
• Loss of power and precision by spending degrees of freedom on each category.
EPA RESPONSE: There is some loss of power and precision in categorization. This can result
in a failure to find a statistically significant effect when in fact there is a meaningful effect in the
data.
• Misclassification at category boundaries (this can be minimized by choosing cut points where
relatively few observations are present).
EPA RESPONSE: Misclassification can occur at category boundaries; however, this is
expected to have a small impact on overall results. Moreover, the likely consequences of
misclassification across boundaries are that if an RR is overestimated in one category, the RR in
an adjacent category will be underestimated. Using a linear regression model across the
categories may serve to smooth out some of this misclassification, if there is any.
• Categorizations can be manipulated to show the desired results.
EPA RESPONSE: This may be possible, but no manipulation of the EtO data was performed
by the EPA to show "desired results." The data categories used in the EPA analyses were
established a priori in the Steenland (2004; 2003) publications. The panel's recommendation to
use "a large number of equally spaced exposure ranges" was not practical for lymphoid cancer
because of the relatively small number of deaths.
H.2. PUBLIC COMMENTS
A number of public comments were received that addressed a range of technical issues
related to the inhalation carcinogenicity of EtO. A number of comments were also received that
are generally directed at what are referred to as "risk management" issues and, as such, are not
addressed here. In the following, summaries of comments on technical risk assessment issues
submitted by the public are provided followed by the EPA's responses (note that some duplicate
comments were omitted).
PUBLIC COMMENT 1.0: The Draft Cancer Assessment Fails to Meet the Rigorous Standard
of Quality Required Under the Information Quality Act and Cancer Guidelines. The Draft
Cancer Assessment is "influential information" as set forth under the Information Quality Act
(IQA) and therefore is subject to a rigorous standard of quality. EPA guidance and the
Guidelines for Carcinogen Risk Assessment (Cancer Guidelines) (U.S. EPA. 2005a) require a
H-30
-------
rigorous standard of quality, which necessitates ensuring that the Draft Cancer Assessment uses
scientifically defensible analytical and statistical methods and has a higher degree of
transparency than information considered noninfluential, particularly regarding the application of
uncertainty factors in EPA's dose-response assessment and risk characterization. The Draft
Cancer Assessment demonstrably fails to meet either the standard set forth under the IQA or the
Cancer Guidelines. EPA must, therefore, substantially revise the assessment before the final EtO
Integrated Risk Information System (IRIS) Risk Assessment (IRIS Assessment) is publicly
disseminated or relied upon for any regulatory purposes.
EPA RESPONSE: Comments received from the SAB and from the public have been addressed
and the EtO carcinogenicity assessment has been revised. It is the EPA's position that as a result
of the extensive development, review, reanalysis, and revision, the revised assessment follows
the EPA's 2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA. 2005a), uses
scientifically defensible analytical and statistical methods, and meets a high standard of
transparency. As such, the revised assessment is consistent with the EPA's Information Quality
Guidelines.15
PUBLIC COMMENT 2.0: EPA failed to use all available epidemiologic data, including the
Union Carbide Corporation (UCC) data and all the National Institute of Occupational Safety and
Health (NIOSH) data that were available at the time EPA conducted its assessment.
EPA RESPONSE: The assessment describes and considers all relevant epidemiological data
available at the time the assessment was conducted, including all the NIOSH and UCC data. The
Union Carbide data and the publications that this public commentator referred to were evaluated
and included in the assessment. The EPA also reviewed articles describing additional follow-up
and analysis of the Union Carbide data that have been published after the panel's report was
finalized. Ultimately, the EPA came to the conclusion that the shortcomings inherent in the
Union Carbide data, particularly the crude assignment of exposure levels to subjects in the UCC
cohort, are fundamental, and as a consequence, the data are not suitable for credible quantitative
analysis of the carcinogenic risk due to exposure to EtO. In the NIOSH data, exposure estimates
were based on a very large number of exposure measurements and a sophisticated modeling
approach (Hornung etal., 1994), which took into account job category and other factors such as
15U.S. EPA (2002) Guidelines for Ensuring and Maximizing the Quality, Objectivity, Utility, and Integrity of
Information Disseminated by EPA. https://www.epa.gov/sites/production/files/2015-08/documents/epa-info-
qualitv-guidelines.pdf
H-31
-------
product type, exhaust controls, age of product, cubic feet of sterilizer, and degree of aeration.
Hence, prediction and assignment of exposure levels for different workers in the NIOSH study
would be expected to be much better than the more simplistic assignment methods used in the
Union Carbide study. Although the recent follow-up of the UCC cohort has now been reported,
there still remains a rather small number of cancers (27 lymphohematopoietic cancers vs. 79 in
the NIOSH cohort and 12 vs. 31 NHLs). Consequently, for example, there was a 50% excess of
NHL in the 9+ years of employment category in the Union Carbide study (Swaen et al.. 20091
but it was based on only five cases and was thus not statistically significant. Also, the UCC
cohort is restricted to men, making an analysis of breast cancer, which was seen to have a
significant increase among female workers with high EtO exposures in the NIOSH cohort,
impossible. In sum, the Union Carbide and NIOSH cohorts are not comparable on a number of
levels, and the NIOSH cohort remains superior as a basis for exposure-response analyses. In the
NIOSH cohort, exposure-response analyses are likely to involve much less misclassification of
exposure and are based on greater numbers, and thus, would be expected to be more reliable.
Analyses of the important breast cancer endpoint are only possible with the NIOSH cohort. See
also the EPA's response to comments on Charge Question 2.a above.
PUBLIC COMMENT 3.0: EPA inappropriately based its evaluation on summaries of statistics
available in various publications, rather than the primary source data, review of which and
reliance upon are essential to conduct valid dose-response modeling. EPA should have based its
calculations on readily available NIOSH data for individual subjects from the cohort mortality
study.
EPA RESPONSE: The statistics used in the draft assessment were obtained from published
journal articles describing the analysis of the NIOSH data. They are summary and categorical
statistics that are commonly used in epidemiological research. The methodology for using such
categorical data to perform dose-response analysis is well established in the epidemiological
literature and is described in Rothman (1986), pp. 343-344, and van Wijngaarden and Hertz-
Picciotto (2004). The categorical and summary statistics used by the EPA are constructed from
the individual data in the NIOSH study. It is possible to perform analyses and construct models
via direct analysis of the individual data and in some cases this is a preferable approach. In fact,
the draft EPA assessment presented the results of such analyses in the form of the Cox regression
models that were based on direct analysis of the individual data with exposure as a continuous
variable. These models provided reasonable fits to the data. However, it was the judgment of
the EPA that these models generated estimates of risk in the low-dose region that were
excessively sensitive to changes in exposure level, and therefore, would not be suitable as the
H-32
-------
basis for low-dose unit risk values. This is what led the EPA to use the regression methodology
with the published grouped data. The EPA, in consultation with Steenland, performed analyses
to fit additional models to the continuous exposure NIOSHdata. The work by Steenland is
described in Appendix D of the revised assessment. Working with Steenland, the EPA
developed and evaluated sets of models using the individual data, including (1) linear relative
risk models (Langholz and Richardsoa 2010) and (2) two-piece linear and log-linear spline
models [e.g., Rothman et al. (2008)1. In the revised assessment, linear low-dose estimates based
on the two-piece spline model and using the Langholz-Richardson linear approach were used for
breast cancer incidence risk estimates. See also the EPA's response to comments on Charge
Question 2.b above.
PUBLIC COMMENT 4.0: EPA Statistical Analysis of the Data Is Flawed and Other Incorrect
Procedures Grossly Overestimate Risk. Key flaws include:
PUBLIC COMMENT 4.1: EPA's risk assessments are invalid, based on linear regressions on
odds ratios (ORs), rather than on individual subject data;
EPA RESPONSE: The odds ratios referenced are summary statistics. Regression on
categorical or summary statistics such as odds ratios is a valid statistical approach. See the
response to Comment 1.2 and response to the SAB panel comment on this issue (Charge
Question 2.b above).
PUBLIC COMMENT 4.2: EPA fails to include all available epidemiologic data;
EPA RESPONSE: This comment refers to the Union Carbide data. See response to Comment
2.0 and response to the SAB panel comment on this issue (Charge Question 2.b above).
PUBLIC COMMENT 4.3: EPA's rationale and methodology for exclusion of the highest
exposure group is inappropriate;
EPA RESPON SE: The EPA did not use the data from the highest exposure group in estimating
the unit risk because it was evident that the relationship between exposure and response changed
over the range of exposure. The general pattern in the data indicated a steep increase in response
in the low exposure range with a leveling or plateau in the high exposure range. Inclusion of the
data from the highest exposure levels in either a Cox regression model or a linear regression
yielded overall estimated relationships that were not suitable for risk assessment. Analyses
conducted by Steenland excluding various percentages of the highest exposures confirmed that
H-33
-------
the highest exposures are attenuating the slopes in such models (see Section D.3.3.1 of Appendix
D). Although the Cox regression models with log cumulative exposure provided adequate fits to
the different data sets, estimates of risk in the low-dose region were overly sensitive to changes
in dose level, and thus, not biologically realistic. In order to obtain a suitable result for risk
estimation at low exposures, the EPA used a linear regression model and excluded the highest
exposure group in the draft assessment. An additional justification for not including the highest
exposure category is that it represents a large, open-ended exposure range, which is less easily
represented by a single exposure value, such as the mean exposure used for the narrower lower
quartiles of exposure, for the purposes of the linear regression. The EPA's Benchmark Dose
Technical Guidance (U.S. EPA. 2012) recognizes analyses omitting high-dose data points, when
these data are not compatible with the development of suitable descriptive statistical analyses, as
a viable analytical approach.
For the revised assessment, the EPA investigated the use of two-piece spline models that
modeled the data as a combination of two splines or segments, one that increased steeply in the
lower dose region joined with a second that increased at a lower rate in the higher dose region.
This approach has the advantage of including all the (individual) data and incorporating into the
overall model the change in the relationship over the observed range of exposure.
PUBLIC COMMENT 4.4: EPA's use of the heterogeneous broad category of distinct diseases
of lymphohematopoietic (LH) cancers as the response increases sample size at the expense of
validity and, thereby, reduces the ability to identify a valid positive dose-response relationship.
EPA RESPONSE: The EPA uses the narrower, less heterogeneous category of lymphoid
cancer data for the primary risk estimates in the revised assessment.
PUBLIC COMMENT 5.0: Certain Policy Decisions EPA Implements in the Draft Cancer
Assessment Are Scientifically Unsupported, Overly Conservative, Inappropriate and Have Not
Been Reviewed by a Science Advisory Board. EPA made several policy decisions that
compounded greatly the inherent conservatism in the risk estimates. These include, among
others: (1) EPA's reliance on the lower bound of the point of departure, rather than the best
estimate when using human data; (2) use of background incidence rates with mortality-based
relative rates, thereby relying on unsupported assumptions that bias results; (3) EPA's
assumption of an 85-year lifetime of continuous exposure and cumulative risk, rather than the
more traditional 70-year lifetime; and (4) the application of adjustment factors for early-life
exposures.
H-34
-------
EPA RESPONSE: The EtO assessment has been reviewed by the SAB and the EPA has
responded to their comments and revised the assessment. With regard to (1), use of the lower
bound on the point of departure is consistent with the EPA's 2005 Guidelines for Carcinogen
Risk Assessment (U.S. EPA 2005a); (2), background incidence rates were used with mortality-
based relative rates because the EPA's objective is to estimate incidence risk not mortality risk
and making adjustments to the analysis when one has only mortality data is consistent with the
EPA's 2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA 2005a) (see also the EPA's
response to this issue under the further statistical issues subsection at the end of Charge Question
2.b above); (3), the EPA did not assume an 85-year lifetime, rather exposures were considered up
to age 85 (i.e., actual age-specific mortality and disease rates to age 85 were used in a life-table
analysis; because most individuals die before age 85 years, the overall average lifespan from the
analysis is about 75 years); (4), the EPA's application of adjustment factors for early life
exposures in the EtO assessment was in accordance with the recommendations in the EPA's
supplemental guidelines and the scientific data supporting the supplemental guidelines (U.S.
EPA, 2005b). The application of these adjustment factors in this assessment was endorsed by
the SAB. Moreover, the EPA's 2005 Guidelines for Carcinogen Risk Assessment (U.S. EPA
2005a) and Supplemental Guidelines (U.S. EPA, 2005b) were both reviewed by the SAB.
PUBLIC COMMENT 6.0: EPA Improperly Relies Entirely on Males in Its Assessment of
Lymphohematopoietic (LH) Cancer Mortality. To be scientifically defensible, EPA's LH cancer
risk characterization must include both males and females, consistent with a "weight-of-
evidence" approach that relies on all relevant information. In the NIOSH retrospective study,
increased risks of LH cancer were observed in males but not females, even though the NIOSH
cohort was large and diverse, and consisted of more women than men. EPA's exclusive reliance
on male data is scientifically unsound without a mechanistic justification for treating males and
females differently with respect to LH, which the analysis lacks.
EPA RESPONSE: In the revised assessment, the lymphohematopoietic cancer unit risk
estimates are based on data for both sexes.
PUBLIC COMMENT 7.0: EPA's Draft Risk Estimates for Occupational Exposure Levels
Rely on Invalid and/or Inappropriate Models. The models used to estimate risks from
occupational exposure are flawed because they generate supralinear results, regardless of the
observed data. These estimates also suffer from the same invalid methodology used in the
environmental risk estimates. EPA must employ a dose-response model that would generate
results consistent with the observed data.
H-35
-------
EPA RESPONSE: It is the underlying data that indicate a supralinear exposure-response
relationship, particularly for lymphoid cancer, all lymphohematopoietic cancer, and breast cancer
mortality, as suggested by the categorical results as well as by the poorer fits of the Cox
regression models with untransformed cumulative exposure data.
PUBLIC COMMENT 8.0: EtO is Considered by Many to be a Weak Mutagen and EPA
Should Consider This in Proposing a Unit Risk Factor. A chemical's mutagenic potency is
necessarily related to its carcinogenic potency. If genotoxicity is considered the means by which
a chemical induces cancer, it follows that it will not induce cancer under conditions where it does
not induce mutations, at either the chromosome or gene level, thus providing a mechanistic basis
for estimating carcinogenicity. EtO has been shown only to be a weak mutagen; therefore, it
should not be automatically considered a human carcinogen and certainly not a potent
carcinogen. In addition, no treatment-related tumors were observed in rats exposed to EtO, even
at the 100 ppm concentration level, at the 18 month sacrifice, and the most sensitive tumor type
(i.e., splenic mononuclear cell leukemia) did not significantly increase in the exposed rats until
23 months, almost the end of their lifetime of exposures (Snellings et al.. 1984). EPA's analysis
should have reconciled these findings with its estimation of EtO's carcinogenic potency, but the
analysis does not do so.
EPA RESPONSE: The EPA does not consider the mutagenicity and carcinogenicity findings to
be in conflict with the potency estimates. EtO is a relatively weak mutagen when compared to
strong mutagens such as cancer chemotherapeutic agents and diepoxides but not necessarily
when compared to other environmental mutagens. Also, EtO is clearly carcinogenic in mice and
rats. The inhalation unit risk estimate based on human data is notably larger than that based on
rodent data (about 23 times larger), and the reasons for this discrepancy are unknown; however,
such species differences are not unusual.
It would not be surprising if there was no statistically significant increase in tumors at 18
months in the Snellings etal. (1984) study. Because of the latency for cancer development,
tumors generally occur later in life. Furthermore, only 20 animals per sex per dose group were
killed at 18 months (and tissues from the animals in the low- and mid-dose group only got
microscopically examined in the presence of a gross lesion), so there is low power to detect an
effect. Nonetheless, Snellings etal. (1984) did report that incidences of brain tumors, which are
a rather uncommon tumor type in F344 rats, were increased in the mid- and high-dose groups at
the 18-month kill. In addition, for testicular peritoneal mesotheliomas, Snellings etal. (1984)
H-36
-------
reported that when the rats with unscheduled deaths were included in the evaluation, EtO
exposure appeared to be related to an earlier occurrence of mesotheliomas.
PUBLIC COMMENT 9.0: EPA's Risk Estimates Do Not Pass Simple Reality Checks.
PUBLIC COMMENT 9.1: The results of the Draft Cancer Assessment (resulting in negligible
risk only at levels less than a part per trillion), are not reasonable when compared with the results
generated for other substances that are considered potent mutagens and/or potent carcinogens,
and do not comport with the results of other assessments EPA has undertaken.
EPA RESPON SE: The procedures used in this assessment comport with those used in other
assessments the EPA has undertaken. Differences in relative potency across chemicals based on
exposure levels may reflect differences in absorption, distribution, metabolism, excretion, or the
pharmacodynamics of the chemicals.
PUBLIC COMMENT 9.2: The Draft Cancer Assessment grossly over predicts the observed
number of cancer mortalities in the study upon which it is based by more than 60-fold.
EPA RESPONSE: The unit risk estimates are derived from, and are consistent with, the results
of the NIOSH epidemiology study, as long as they are used in the low-exposure range, as
intended. Because the exposure-response relationships for the cancers of interest in the NIOSH
study are generally supralinear, the unit risk estimates will overpredict the NIOSH results if
applied to the region of the exposure-response relationships where the responses plateau. The
potency estimates derived in the assessment are constructed for use with low dose levels
consistent with environmental exposure and are not appropriate for use with exposures in
occupational settings, as stated explicitly in the document. Occupational exposure scenarios are
addressed in Section 4.7 of the assessment document. Extra risks associated with occupational
exposures are in the "plateau" region of the exposure-response relationships, and thus, increase
proportionately less than risks in the low-dose region.
PUBLIC COMMENT 9.3: EPA's de minimis value from the Draft Cancer Assessment is 2 to
3 orders of magnitude below the endogenous level of EtO that is produced naturally in humans.
EPA RESPONSE: The EPA's risk estimates are for risk above background. The issue of
endogenous levels is addressed in the revised assessment. See Section 4.5 for a discussion of the
specific issue raised in this comment.
H-37
-------
PUBLIC COMMENT 9.4: EPA's draft unit risk values for EtO are unreasonably large, given
the evidence of carcinogenicity in a large body of epidemiology studies that is not conclusive,
the weak mutagenicity data, and the lack of cancer response in rodents until very late in life.
EPA must make the best use of all of the epidemiology, toxicology and genotoxicity data for EtO
that provide valid information on the relationship between exposure and cancer response to
improve the reasonableness of the unit risk values for EtO.
EPA RESPONSE: The EPA believes that it has made the best use of the available information
in revising the assessment. The EPA's evaluation of the weight of evidence concludes that the
epidemiological evidence is strong (see Section 3.5.1). In addition, the unequivocal evidence of
rodent carcinogenicity and the supporting mechanistic evidence add sufficient weight for the
characterization of "carcinogenic to humans" (see Section 3.5.1), which is beyond what is
needed to support the derivation of quantitative risk estimates. This is thoroughly presented in
the assessment and was supported by the SAB review. The unit risk estimates are derived from,
and are consistent with, the results of the large, high-quality NIOSH epidemiology study. See
also the response to Comment 8.0 above.
PUBLIC COMMENT 10.0: The Draft Cancer Assessment Does Not Use the Best Available
Science as Required under the Information Quality Act and Cancer Guidelines.
PUBLIC COMMENT 10.1: EPA based its evaluation on summaries of statistics available in
various publications. These data, however, are not sufficient to conduct valid dose-response
modeling. EPA should have based its calculations on readily available National Institute of
Occupational Safety and Health (NIOSH) data for individual subjects from the cohort mortality
study.
EPA RESPONSE: See response to Comment 3.0.
PUBLIC COMMENT 10.2: EPA did not use all available epidemiologic data, including the
Union Carbide Corporation (UCC) data and all NIOSH data that were available at the time EPA
conducted its assessment. In particular, the Greenberg etal. (1990) UCC study reported the
consistency of the death certificate diagnosis with a pathology review of medical records for
leukemia cases, a validation not conducted for cases in the NIOSH study.
EPA RESPONSE: The EPA considered all the available epidemiological data, including
NIOSH and UCC data, and the publications that the American Chemistry Council referred to in
H-38
-------
its comments. See response to Comment 2.0 for more details on why the UCC data were not
used for the derivation of quantitative risk estimates.
PUBLIC COMMENT 11.0: EPA Should Recognize That EtO Is Both a Weak Mutagen and
Weak Animal Carcinogen.
EPA RESPONSE: The lull text of this comment was essentially the same as Comment 8.0 and
is addressed in the EPA's response to that comment above.
PUBLIC COMMENT 11.1: Among 26 alkylating agents studies by Vogel and Nivard (1998),
EtO showed the second lowest carcinogenic potency.
EPA RESPONSE: The Vogel and Nivard (1998) study is not relevant to the EPA's assessment
of the carcinogenicity of EtO. Most of the substances considered by Vogel and Nivard (1998)
are chemotherapeutic chemicals that are, by design, intended to be strong alkylating agents.
PUBLIC COMMENT 11.2: Previous assessments of EtO inhalation time to tumor in rats
showed that the increased risks observed at higher experimental doses did not extend to the
lowest experimental dose. To comply with the Cancer Guidelines, EPA should include these and
other relevant animal data in a weight-of-evidence characterization of EtO.
EPA RESPONSE: The carcinogenicity data reviewed in Section 3.2 reveal that, of 13
exposure-response relationships for the tumor types associated with EtO exposure from the three
rodent bioassays, all but one show an increased incidence at the lowest exposure level, although
not all the increases are statistically significant at that level.
PUBLIC COMMENT 12.0: EPA's Risk Estimates Do Not Pass Simple Reality Checks.
PUBLIC COMMENT 12.1: [This was the same as Comment 9.1 above.]
PUBLIC COMMENT 12.2: The results of the Draft Cancer Assessment are at odds with
EPA's conclusion that EtO is a potent (de minimis level < 1 ppt) human carcinogen and EtO's
potency seen in animal studies.
EPA RESPONSE: The risk estimates based on the rodent data are over an order of magnitude
lower than (-1/23) the estimate based on the human data, for unknown reasons, but species
differences are not unusual and human data are generally preferred over rodent data for
H-39
-------
quantitative risk estimates because the uncertainties due to interspecies extrapolation are
avoided.
PUBLIC COMMENT 12.3: EPA's draft unit risk values for EtO are not applicable to the
general public. The Draft Cancer Assessment grossly over predicts the observed number ofLH
cancer mortalities in the study upon which it is based by more than 60-fold. Further, EPA's de
minimis value is about 50 times lower than the lowest ambient concentration found at remote
coastal locations. Based upon PBPK simulations, endogenous concentrations of EtO in humans
are approximately 400-1700 times greater than EPA's proposed de minimis value of 0.00036
parts per billion.
EPA RESPONSE: The unit risk estimates are derived from, and are consistent with, the results
of the NIOSH epidemiology study, as long as they are used in the low-exposure range, as
intended; see response to Comment 9.2 above. Endogenous and ambient concentrations of EtO
could be contributing to background rates of lymphohematopoietic cancer and breast cancer
incidences, which are appreciable. The EPA values are not implausible upper-bound estimates.
H-40
-------
APPENDIX I. EPA RESPONSES TO SAB COMMENTS ON 2014 EXTERNAL
REVIEW DRAFT
This Appendix provides responses to the comments received from the Science Advisory
Board (SAB) in their August 7, 2015 report (SAB, 2015) following their review of the EPA's
2014 SAB review draft (U.S. EPA, 2014a, b). A similar draft was reviewed by the public in
2013 (U.S. EPA, 2013a, b), and responses to the public comments are presented in Appendix K.
Responses to SAB (SAB, 2007) and public comments on the EPA's 2006 external review draft
(U.S. EPA, 2006a) are compiled in Appendix H. In response to charge questions, Appendices K
(then L) and H were specifically reviewed by the SAB during their review of the 2014 draft.
Public comments to the SAB on the 2014 draft are not addressed directly in this appendix;
however, the SAB had all of the public comments for consideration and some of the public
comments were explicitly reflected by the SAB in their comments to the EPA.
I. SAB RECOMMENDATIONS IN SAB LETTER TO THE ADMINISTRATOR WITH
EPA RESPONSES
1. COMMENT: Overall the SAB finds the agency has been highly responsive to the 2007
SAB recommendations. The SAB finds that the National Institute of Occupational Safety
and Health (NIOSH) dataset is still the most appropriate dataset to use and concurs with
the agency's decision to not use the Union Carbide Corporation cohort data. The
statistical and epidemiological issues in this assessment are complex and the agency is to
be commended for conducting the additional exposure-response modeling in response to
the 2007 SAB recommendations. The SAB believes that the advice and
recommendations in this report can be addressed relatively quickly and that the draft
assessment should move forward to be finalized.
EPA RESPONSE: Consistent with the SAB's concurrence, the EPA has continued
to use the NIOSH data set as the basis for the quantitative risk estimates and has not
derived any estimates from the Union Carbide Corporation cohort data.
2. COMMENT: The draft assessment employed lagged exposure estimates in the
derivation of cancer risk estimates. Although there is a scientific rationale for a period of
latency between biologically important exposures and subsequent cancer incidence or
mortality, the SAB did not find a strong biological or statistical argument supporting the
particular selected latency periods applied for breast and lymphoid cancers. The EPA is
1-1
-------
encouraged to perform a sensitivity analysis of various latency periods to determine what
effect this selection had on risk estimates.
EPA RESPONSE: The lag period defines an interval before death, incidence, or end
of follow-up during which any exposure is excluded from the calculation of the
exposure metric. The EPA re-examined lag selection for both the lymphoid cancer
mortality (see Section D.3.2 of Appendix D) and breast cancer incidence
(see Section D.1.2 of Appendix D) data sets with a larger group of models than was
considered in the 2014 draft and has again selected 15 years as the lag for each data
set (endpoint). Sensitivity analyses were conducted with lags of 0, 5, 10, and 20
years to determine the effect of lag selection on the unit risk estimates
(see Sections D.1.6 and D.3.5 of Appendix D) and on the extra risk estimates for the
occupational exposure scenarios in Section 4.7 (see Sections D.l.ll and D.3.9 of
Appendix D). For breast cancer, unit risk estimates from the selected model with the
alternate lag periods varied by at most 35% from the primary estimate derived with
the selected lag period of 15 years. For the occupational exposure scenarios, the
upper-bound extra risk estimates varied by at most about 25% from the estimates
derived with the selected lag. For lymphoid cancer, the unit risk estimates from the
selected model with the alternate lag periods ranged from about 48% less than to
about 190%) greater than the estimate derived with the selected lag period of 15 years.
For the occupational exposure scenarios, the upper-bound extra risk estimates varied
by at most about 55% from the estimates derived with the selected lag.
3. COMMENT: A number of different statistical models were examined for estimating
breast cancer incidence risk from low exposure to EtO. The draft assessment presents a
number of considerations used in the selection of the preferred model. The SAB
generally concurs with the selection of the two-piece spline model for estimating breast
cancer incidence. However, the SAB has recommendations on improving the
considerations used for model selection, including less reliance on the Akaike
information criterion (AIC). However, if AIC is used for model selection, it should be
used appropriately. There should be a priori considerations regarding the nature of the
functional form being applied. Specifically, the SAB recommends prioritizing functional
forms of the exposure that allow regression models with more local fits in the low
exposure range (e.g., spline models). The draft assessment also presents risk estimates
from other "reasonable models." Although much of this approach is scientifically
appropriate, the SAB finds that a clear definition of "reasonable models" is lacking and
1-2
-------
encourages some modifications and more transparency in the presentation. The SAB also
provides recommendations on prioritizing statistical considerations in the selection of
models. Any model that is to be considered reasonable for risk assessment must have a
dose-response form that is both biologically plausible and consistent with the observed
data.
EPA RESPONSE: The EPA has followed the SAB's recommendations for model
selection. Model selection for both the breast cancer incidence (see Section 4.1.2.3)
and lymphoid cancer (see Section 4.1.1.2) data prioritizes functional forms that allow
more local fits in the low-exposure range (e.g., spline models), relies less on AIC, and
includes consideration of biological plausibility. In addition, the EPA has confirmed
that the AIC is being used appropriately—the different models being compared were
fit using the same measures, the models had the same outcome variable, and the
models were estimated with software routines that calculate AIC in the same way.
[The EPA has determined that SAS proc NLP, which was used for the linear RR
models for lymphoid cancer, consistently yielded -2 log likelihoods and AICs almost
0.4 units lower than those from proc PHREG, which was used for the log-linear
models, for the same models (when the log-linear models were also run in proc NLP),
including the null model. This small discrepancy is assumed to be related to
computational processing differences. For breast cancer incidence, procNLMIXED
was used for the linear RR models, and this proc generated the same -2 log
likelihoods and AICs as did proc PHREG.] The EPA has improved the clarity and
transparency of the discussion of model selection, and the EPA no longer
distinguishes a subset of "reasonable models." The EPA continues to use the
two-piece spline model for the breast cancer incidence data, consistent with SAB
concurrence (see Section 4.1.2.3).
4. COMMENT: For lymphoid cancer, the draft assessment presents a linear regression of
categorical results using dose categories as the preferred model for the derivation of the
unit risk estimate for low exposure to EtO. The SAB prefers the use of continuous
individual-level exposure data over the use of categorical results. The linear regression
of categorical results should not be selected unless the individual exposure model results
are biologically implausible. The SAB recommends presentation of multiple estimates of
the unit risk in sensitivity analyses and an updated justification of model selection.
1-3
-------
EPA RESPON SE: In response to SAB comments, the EPA has changed its model
selection for lymphoid cancer from the linear regression of categorical results to a
model based on individual-level exposure data. The EPA presents unit risk estimates
from multiple models for comparison (see Table 4-7) and has updated the
justification for model selection (see Section 4.1.1.2). Consistent with SAB
recommendations, the model selection now emphasizes use of the individual-level
data, prioritization of functional forms that allow more local fits in the low-exposure
range (e.g., spline models), the principle of parsimony, less reliance on AIC, a
weighing of biological and statistical considerations, and prioritization of models that
can be used for both environmental exposures and the occupational exposure
scenarios. As a result of these model selection emphases, the EPA has selected the
two-piece linear spline model with the knot at 1,600 ppm x days for the lymphoid
cancer data (see Section 4.1.1.2).
5. COMMENT: The SAB suggests that the agency consider using the same model for both
environmental and occupational exposures. The use of different models for
environmental and occupational exposures should only be done with sufficient
justification.
EPA RESPONSE: In response to the SAB comments, the EPA now uses two-piece
spline models, which can be applied to both environmental (see Section 4.1) and
occupational (see Section 4.7) exposures, for both lymphoid cancer and breast cancer
incidence.
6. COMMENT: The uncertainty discussions are generally clear, objective, and
scientifically appropriate, but they can be improved and extended. Considerations about
uncertainty directly pertaining to the analyses reported can be separated into uncertainty
due to the data themselves (particularly from reliance on a single data set), and
uncertainty of the results given the data. The SAB recommends adding descriptive
summaries of the characteristics of the NIOSH cohort, better quantification of the results
from the various models (such as reporting unit risk estimates and comparisons in
sensitivity analyses), and down-weighting epidemiologic results based on external
standards that may be subject to bias due to the healthy worker effect.
EPA RESPON SE: In response to the SAB comments, the uncertainty discussion has
been restructured to address uncertainty due to the data themselves
1-4
-------
(see Section 4.1.4.1) separately from uncertainty of the results given the data
(see Section 4.1.4.2). In addition, more descriptive summaries of the characteristics
of the NIOSH cohort have been incorporated into the assessment, including sex
distribution over time, age and year of entry to the EtO workforce, duration of
employment in the EtO cohort, age and year of departure/retirement from the EtO
cohort, cumulative total and peak exposures for individual cases and controls,
percentage of total case and control individual exposures in the worker histories that
are excluded when various lags are imposed, and mean, median, and 25th, 75th, and
95th percentile values for annual exposures among cases and controls (see Section D.5
of Appendix D). Additional sensitivity analyses have also been included in the
assessment, and all the results are reported as unit risk estimates. For both breast
cancer incidence and lymphoid cancer, a variety of models are compared, and the
sensitivity of the selected models to different lag periods and knots is examined.
Also, mortality and incidence estimates as well as upper-bound estimation approaches
are compared. For breast cancer incidence, further analyses include comparisons of
the subcohort with interviews and the lull cohort, of total breast cancer and invasive
breast cancer only, and of the lull (selected) model and the model with the exclusion
of the nonexposure covariates (breast cancer risk factors). For lymphoid cancer,
results are compared to the results for all lymphohematopoietic cancers. All of the
EPA's quantitative estimates are based on internal comparisons.
7. COMMENT: The draft assessment presents an accurate, objective, and transparent
summary of published studies on EtO genotoxicity. The SAB agrees that the weight of
the scientific evidence from epidemiological studies, laboratory animal studies and in
vitro studies supports the general conclusion that the carcinogenicity of EtO in laboratory
animals and humans is mediated through a mutagenic mode of action. The SAB finds
that several areas of the draft assessment can be improved to enhance the clarity of
presentation and to provide a more detailed interpretation of findings within the context
of more recent advances in the understanding of the biology of cancer and has specific
recommendations and suggestions for revision detailed in the report.
EPA RESPONSE: The EPA has retained the conclusion that there is sufficient
weight of evidence that a mutagenic mode of action is operative in EtO
carcinogenicity, but in response to SAB comments, the EPA has strengthened the
presentation of the evidence. Section 3.3 (genotoxicity) has been revised to
synthesize the information used to support a mutagenic MO A in a more systematic
1-5
-------
and complete manner, including the addition of a substantially expanded summary
that integrates study information in terms of dose-response and temporal relationships
(see Section 3.3.3.4). Section 3.4 has been revised and reorganized to more clearly
discuss the mechanisms by which the genotoxic effects might be instrumental in EtO
carcinogenesis (see Section 3.4.1.1), particularly in the target organs (see Sections
3.4.1.2 and 3.4.1.3). In addition, the EPA has revised and expanded Table 3-6 (now
3-8) and provided additional summary tables (see Tables 3-6, 3-7, 3-9, and 3-10).
(For more details on specific revisions, see responses to detailed comments in Part
II. 5, below.)
8. COMMENT: Appendix H of the draft assessment provides a summary of the 2007 SAB
comments and the EPA's response to the comments. The responses are transparent,
objective, and for the most part, accurate (exceptions are noted in the current report). In
particular, the SAB supports the expanded discussion of endogenous EtO provided in the
draft assessment and has suggestions for further improvement; agrees with the decision
not to include a toxicity value for EtO based upon nonlinear extrapolation and recognizes
and agrees with revisions to strengthen support for a classification of EtO as
"carcinogenic to humans."
EPA RESPONSE: Appendix H is largely unchanged. The EPA's response to any
exceptions noted in the 2015 SAB report are addressed where they arise
(see Section II.6.a and n.6.b below; pages 1-35 to 1-44).
9. COMMENT: In general, the literature review of new studies presented in Appendix J
appears complete. The logic and progression of the review is clearly supported. The
clarity can be improved by distinguishing between statements made by study authors and
statements made by the EPA. The SAB concurs that inclusion of the new studies would
not substantially alter the findings of the assessment, with the exception of the Mikoczy
et al. (2011) study of Swedish sterilization workers, which can strengthen support for the
hazard characterization of EtO and provide support for the modeling of the NIOSHdata.
EPA RESPONSE: The EPA has clarified what are Agency interpretations and what
are study author statements. In addition, the EPA has incorporated discussion of the
Mikoczy etal. (2011) study into the main body of the report, supporting the hazard
characterization of the epidemiological evidence on breast cancer
(see Sections 3.1 and 3.5) and the supralinear exposure-response relationship
1-6
-------
observed with the NIOSH data (see Section 4.1.4). Also, a comparison was done of
the Mikoczy et al. (2011) RR estimates with predicted RR values from the selected
model derived from the NIOSH data (see Section 4.1.4 and Section J.2.2 of
Appendix J), (see also response to Comment II.7.a.i below.)
10. COMMENT: Appendix L [now K] presents public comments on the July 2013 draft of
the assessment and EPA responses to the public comments. The SAB finds that overall,
the EPA has been very responsive to the public comments. The responses are thorough,
clear, and appropriate.
EPA RESPONSE: In response to specific comments detailed in Section II.8.a below
(pages 1-47 to 1-54), the EPA has strengthened a few of the responses.
II. COMMENTS FROM THE SAB REPORT
1. More detailed comments regarding lag (p. 7-8 of SAB report)
a. COMMENT: [T]he SAB recommends the methods used to determine minimum
latency estimates in the CDC 9/11 Working Group Guidelines (Howard, 2013) as a
good framework for assessing latency in cancer onset. However, the disease-specific
latency selections in the guidelines are specific to the World Trade Center Health
Program and 9/11 agents, and are not relevant to the EtO draft assessment.
EPA RESPON SE: The EPA is interested in an optimal lag and not the minimum
lag and has used standard epidemiological methods to determine an optimal lag
(see Sections D.1.2 and D.3.2 of Appendix D). Nonetheless, the EPA has
reviewed the CDC guidelines (Howard, 2013) and found that the method that the
EPA used is also one of the methods discussed in the CDC guidelines—"Method
4A: Statistical Modeling—Estimates of cancer latency obtained by statistical
modeling in epidemiologic studies of the association between exposure to an
agent and a type of cancer."
b. COMMENT: The SAB encourages the EPA to refine the discussion of this
uncertainty with a paragraph in the body of the assessment and a summary of an
analysis (detailed in an appendix) that examines the sensitivity of estimates of unit
risks over the plausible range of latency periods (i.e., 0-20 years). [...] The SAB
1-7
-------
encourages the EPA to formalize the presentation and discussion of the quantitative
results for the sensitivity analysis of exposure lags that is currently included in
Appendix D, focusing on the sensitivity of the EPA's recommended models and a
strongest competitor(s) to the length of the assumed latency period. The body of the
draft assessment should include a short summary of the quantitative results of the
sensitivity analysis described in detail in the appendix, accompanied by a qualitative
discussion of how the results of the sensitivity analysis should factor into an overall
assessment of the biological and statistical uncertainty of the unit risk estimates
derived under the alternative models of exposure risk.
EPA RESPON SE: The EPA has conducted the sensitivity analyses recommended
by the SAB. These are summarized in Sections 4.1.1.3 (lymphoid cancer) and 4.1.2.3
(breast cancer), discussed qualitatively in the context of overall uncertainty in
Sections 4.1.4 (sources of uncertainty) and 4.5 (conclusions regarding the unit risk
estimates), and detailed in Sections D.1.6 and D.3.5 of Appendix D. For breast
cancer, unit risk estimates from the selected model with the alternate lag periods
varied by at most 35% from the primary estimate derived with the selected lag period
of 15 years, and a comparison is made with the strongest competitor, a 20-year lag.
For lymphoid cancer, the unit risk estimates from the selected model with the
alternate lag periods ranged from about 48% less than to about 190% greater than the
estimate derived with the selected lag period of 15 years, and there is no good
competitor. See also response to Comment 1.2 above.
c. COMMENT: In summary, the SAB agrees that it is scientifically plausible, and
even likely, for there to be a period of latency between biologically important
exposures and subsequent cancer incidence or mortality.
EPA RESPON SE: The EPA agrees and has selected lag periods of 15 years of
lymphoid cancer mortality (see Section D.3.2 of Appendix D) and 15 years for
breast cancer incidence (see Section D. 1.2 of Appendix D). See also response to
Comment 1.2 above.
2. More detailed comments regarding breast cancer incidence model selection
a. Bulleted summary recommendations regarding model selection (p. 11 of SAB
report)
1-8
-------
COMMENT: The SAB requests that the EPA provide better documentation of
the NIOSH study data, particularly with respect to exposure.
EPA RESPON SE: The EPA has provided the additional details requested by
the SAB, as detailed in SAB comments on Charge Question 4 (see Comment
II.4.b.iii below); these are summarized in tables and figures in Section D.5 of
Appendix D. Some of the new cohort details summarized in Section D.5
include mean, median, minimum, maximum, and 25th and 75th percentiles of
cumulative exposure in the lull cohort; cumulative exposures by year of entry
and by duration of employment; sex distribution over time; distributions of
year of entry, age of entry, duration of employment, and age and year of
departure/retirement; distributions of cumulative and peak exposures for
individual cases and controls; percentages of total case and control individual
exposures in the worker histories that are excluded when the 15-year lag is
imposed; and mean, median, and 25th, 75th, and 95th percentile values for
annual exposures among cases and noncases in the cohort.
COMMENT: In selecting models for use in risk assessment, the SAB
recommends less reliance on the AIC for model selection. If AIC is used for
model selection, it should be used appropriately.
EPA RESPONSE: The EPA has followed the SAB's recommendations for
model selection, including less reliance on AIC (see Section 4.1.2.3). In
addition, the EPA has confirmed that the AIC is being used appropriately.
See also response to Comment 1.3 above and Comment n.2.d.iii and its
response below.
COMMENT: The SAB recommends prioritizing functional forms of the
exposure that allow regression models with more local fits in the low-exposure
range (e.g., spline models).
EPA RESPONSE: The EPA has followed the SAB's recommendations for
model selection, including prioritizing functional forms that allow more local
fits in the low-exposure range, and the EPA's selected model is a two-piece
spline model (see Section 4.1.2.3).
1-9
-------
COMMENT: Any model that is to be considered reasonable for risk assessment
must have a dose-response form that is both biologically plausible and consistent
with the observed data.
EPA RESPONSE: Consistent with SAB concurrence (SAB letter to the
Administrator), the EPA has selected a two-piece linear spline model. This
general model shape is biologically plausible and the selected model is
consistent with the data (see Section 4.1.2.3).
COMMENT: Sensitivity analyses should be reported for a range of results and
should include the target quantity of interest (unit risk, excess risk). Although not
all models are equally reasonable from a risk assessment perspective, lull and
transparent reporting of the target parameters of interest provides valuable
context.
EPA RESPONSE: As detailed in Comment II.2.b.iv and its response below,
the EPA reports ranges of results for various sensitivity analyses and, in
response to SAB recommendations, these are reported as unit risk estimates
(or, for the occupational exposure scenarios, as estimates of extra risk [see
Section 4.7]).
More detailed comments from the text regarding model selection (p. 8-11 of
SAB report)
COMMENT: There is not enough detail provided for the NIOSH exposure data
for the SAB to determine the appropriateness of the data. Therefore the SAB
response is conditional on the assumption that the NIOSH exposure data are
appropriate.
EPA RESPONSE: The EPA has provided the additional information
requested by the SAB in the detailed SAB comments on Charge Question 4
(see Comment II.4.b.iii below); these are summarized in tables and figures in
Section D.5 of Appendix D. In particular, the SAB had concerns about some
exposure data presented by public commenters, also detailed in SAB
comments on Charge Question 4, and these concerns are addressed in the
I-10
-------
response to those comments (see Comment II.4.b.iv below). In brief, contrary
to public comments made at the SAB meeting, theNIOSHEtO exposure
patterns were not anomalous, but rather reflected the underlying changes in
variables predicting exposure over time. One of the principal drivers of the
NIOSH exposure levels was the cubic feet of the sterilizers used, and sterilizer
volumes increased over time in some plants.
COMMENT: Although generally the EPA's model selection for breast cancer
incidence is scientifically appropriate, it could be described more clearly and
transparently. The EPA is encouraged to revise the discussion of the Cox model,
or more generally, relative risk models, to use terminology that can be directly
linked with the published literature. Prentice (1985) provides examples of this
terminology and a discussion of relative risk models. Terminology describing the
behavior of the models at the low-exposure range should be clearly defined,
particularly terms that are used to make judgments, such as "unstable."
EPA RESPON SE: The EPA has improved the clarity and transparency of
the discussion of model selection (see Section 4.1.2.3). In addition, the EPA
has summarized the terminology it applies to the relative risk models at the
beginning of Chapter 4, using terminology in the published literature
(Laneholz and Richardson, 2010). The EPA has clearly defined any terms
used to describe low-exposure model behavior.
COMMENT: The SAB supports the prioritization of incidence data and the
choice of data to use for the breast cancer incidence analyses. The SAB also
concurs with the reliance on analyses based on the individual estimates of
cumulative exposure for risk assessment (in contrast to categorized exposure or
other exposure metrics such as duration). Exposure duration is not as informative
for risk assessment because the magnitude of exposure is not part of duration.
Using an exposure lag is more biologically plausible than using no lag. The SAB
commends the EPA for considering and documenting the results for a variety of
different model specifications in terms relevant for the ultimate risk assessment.
In particular, a good choice is the linear spline structure used to parameterize the
exposure covariate in the relative risk function under an exponential (exp(f(x))) or
linear (1 +flx)) relative risk model. A spline parameterization off(x) has the
advantage of allowing the shape of the relative risk function to vary over the
range of exposure while ensuring that the behavior of the function in the low-
1-11
-------
exposure range is not unduly influenced by the highest exposures. The linear
spline parameterization has the disadvantage that it has a "corner" and a smooth
dose-response function would be preferred. The draft assessment uses a cubic
spline model to address this, but ultimately the simpler linear spline model was
selected as the preferred model. The ECoi from the cubic spline model is similar
to the one from the linear spline model and the SAB concurs with the EPA's
preference for the much simpler linear spline model parameterization, recognizing
the virtue of simplicity and transparency of reporting. Alternatives to using
cumulative exposure in the model as a single untransformed term are
log-transformation and square root transformation. These alternatives are less
desirable because they produce more global fits to the entire exposure range,
which would give the higher exposures more influence (compared to the more
local spline models) on the fitted dose-response in the low-exposure range of the
data. [.. SJpline models have the advantage of being sensitive to local behavior in
the data. They can also be chosen to be parsimonious (an example is a 2-piece
linear spline). Models fit to exposure categories are similarly sensitive to local
behavior in the data, but they require more parameters to be estimated and are
thus less parsimonious than the spline models considered in the assessment.
EPA RESPONSE: The EPA agrees and has retained the prioritization of the
incidence data and use of the subcohort with interviews from the NIOSH
incidence study (Steenland et al.. 2003). The EPA has also retained use of the
cumulative exposure metric and the lag period. In addition, the EPA has
retained its preference for the two-piece linear spline model.
iv. COMMENT: The SAB has some concern about the number of models that were
fit to the data because over-reliance on the best-fitting results can lead to
statistical artifacts [such as "random high bias" which has been defined in the
context of hypothesis testing; e.g., see Fleming (2010)1. At this stage of the EtO
risk assessment, the SAB's concern with the large number of models that have
been explored can best be addressed by striving for comprehensive reporting of
model results; i.e., sensitivity analyses should be reported for a range of results.
These should include sensitivity to the functional form of the model (both the
choice of relative risk function and the functional form of exposure within).
Other aspects of the analysis should also be considered such as inclusion of
confounding variables, choice of lag, and cohort (full cohort vs. those with
1-12
-------
interviews). The SAB recommends inclusion of tables documenting the various
estimates of the target parameter of interest (which is predominantly the unit risk
estimate) from the many models that were considered for the risk assessment.
Although not all models are equally reasonable from a risk assessment
perspective, lull and transparent reporting of the target parameters of interest
provides valuable context. Appropriate use of appendices and thoughtfully
designed tables in the main report can minimize the potential for confusion that
may result from reporting so many estimates.
EPA RESPONSE: Additional sensitivity analyses have been included in the
assessment, and all the results are reported as unit risk estimates. A variety of
models are compared, both with different relative risk functions and different
functional forms of exposure within, for the full incidence cohort and the
subcohort with interviews (see Table 4-15). In addition, the sensitivity of the
selected model to different lag periods, different knots, and the exclusion of
covariates (breast cancer risk factors) is examined (see Tables D-12, D-13,
and D-14 of Appendix D). All the models that were investigated are
presented in the assessment.
v. COMMENT: [T]he draft assessment states that low-dose extrapolation was
performed for risk assessment, but the document does not state whether or not the
doses considered for the unit risk estimates were outside the range of the NIOSH
exposure data. For instance, as given by the conversion shown in footnote "e" of
Table 4-13, 5,800 ppm-days corresponds to 0.075 ppm (with the correction to the
formula that one divides by 365). The tenth percentile of the breast cancer
incidence data corresponds to 157 ppm-days of exposure and 17 incident cases
have nonzero exposure at or below this level (using a 15-year lag; see Table D-l).
Using the same formula, this corresponds to 0.00202 ppm. The LEC01 from the
preferred model is 0.00576 ppm, more than twice 0.00202 ppm, suggesting there
is no low-dose extrapolation in these data. Because there is no low-dose
extrapolation in these data, there is less uncertainty of the unit risk estimate than
would be otherwise present.
EPA RESPONSE: Even though lifetime cumulative exposures from
environmental exposure may overlap the low end of the range of lagged
cumulative exposures from occupational exposure, the exposure-response
1-13
-------
model is based on the M range of the occupational exposure data and not just
the data in the range of environmental exposures, which are too sparse to
model on their own. Even with a two-piece spline model which gives a more
local fit to the low-exposure data, there is uncertainty about the exposure-
response relationship specifically in the range of environmental exposures.
The point of departure (LECoi) is intended to be at the low end of the
"observable" range (i.e., the range of exposures for which the study might be
able to detect a significant increase in risk), but it is still substantially above
typical environmental exposure levels (according to the EPA's 2005 National
Air Toxics Assessment data, the average exposure concentration of EtO from
all sources [including background] in the United States is 0.0062 |io/m3 [3.4 x
10"6ppm]; the average background concentration is 0.0044 [j,g/m3 [2.4 x 10"6
ppm]), and thus, there is uncertainty about the low-exposure extrapolation
from the point of departure (6.75 x 10~3ppm, or 12 (J,g/m3, for breast cancer
incidence). For lymphoid cancer, only 2 of the 13 cases below the knot
(-15%) are also below the point of departure. For breast cancer, about 25% of
the cases below the knot are also below the point of departure, roughly
corresponding to the lowest 1.5 deciles (see Table D-l of Appendix D), but,
for example, as one can see from Table D-3 of Appendix D, the variability of
the low-exposure data is such that the two lowest deciles both have RR
estimates <1, and thus, are not by themselves consistent with the unit risk
estimate, illustrating the uncertainty that still exists in the low-exposure
extrapolation. The EPA has added a footnote to the uncertainty discussion
(see Section 4.1.4) to clarify that there is still considered to be low-exposure
extrapolation from the point of departure.
vi. COMMENT: In conclusion, the SAB concurs with the EPA's selected model for
the breast cancer incidence data. However, it could be described more clearly and
transparently and the SAB prefers a somewhat different set of criteria for
selecting the most appropriate model. There are clear advantages to relying on
parsimonious regression models directly fit to the individual-level cumulative
exposure data using spline models to parameterize exposure. In addition, biologic
plausibility and other external information (such as corroborating information
from other studies) should help inform the model selection. For example, the
incidence rate ratio (IRR) results reported for the Swedish sterilization workers
study by Mikoczy etal. (2011) could be used to support the selected model. The
1-14
-------
task of selecting a final model is more challenging when a set of plausible models
gives widely disparate unit risk estimates. [Comments in Part n.2.d below]
provide[] further advice on how to prioritize potentially plausible models.
Ultimately though, the SAB expects that this preferred approach will result in
selecting the same or a very similar model to the one selected by the EPA.
EPA RESPONSE: As noted in response to Comment 1.3 above, the EPA has
improved the clarity and transparency of the discussion of model selection and
has followed the SAB's recommendations for model selection
(see Section 4.1.2.3). The EPA has also cited the Mikoczy et al. (2011) study
as supporting the selected model (see Section 4.1.4). Ultimately, the EPA
selected a model virtually the same as that in the 2014 draft assessment—a
two-piece linear spline model with the knot at 5,750 ppm x days.
Bulleted summary recommendations regarding discussion of "reasonable
models" (p. 13 of SAB report)
COMMENT: Revise the discussion to provide more clarity and transparency as
well as making the disposition easier to follow.
EPA RESPONSE: The EPA no longer distinguishes a group of "reasonable
models" and so this particular discussion has been eliminated. The EPA has
improved the clarity and transparency of the discussion comparing models.
COMMENT: Discarding a model because the fitted curve is "too steep" is only
acceptable when there is scientific justification.
EPA RESPONSE: The EPA has reworded the discussion and no longer
discounts models as "too steep."
COMMENT: Clearly articulate the criteria for determining that models are
reasonable as well as providing transparent definitions for frequently used terms
such as "too steep," "unstable," "problematic," and "credible."
1-15
-------
EPA RESPON SE: The EPA has generally omitted these terms from the
revised discussion. If used, they are clearly defined or not used in discussions
of model selection.
COMMENT: Assign weight to various models based on a modified combination
of biological plausibility and statistical considerations; use somewhat different
considerations for comparing AICs than those currently employed in the draft
assessment.
EPA RESPONSE: See detailed Comment II.2.d.i and response below.
COMMENT: Use a different set of emphases in the priorities for the most
reasonable models; detailed suggestions are provided by the SAB in this response.
EPA RESPONSE: The EPA no longer distinguishes a group of "reasonable
models"; however, the EPA has adopted the SAB's recommended emphases
in the overall model selection. See detailed Comment n.2.d.i and response
below.
More detailed comments from the text regarding discussion of "reasonable
models" (p. 11-13 of SAB report)
COMMENT: The SAB recommends assigning weight to certain types of models
based on a modified combination of biologic plausibility and statistical
considerations, and using somewhat different considerations for comparing AICs
than those currently employed in the draft assessment. Regarding statistical
considerations about various models, the SAB recommends a different set of
priorities for establishing the most reasonable models and gives guidance on the
preference for their ordering. First, prioritization should be given to regression
models that directly use individual-level exposure data. [...] Second, among
models fit to individual-level exposure data, models that are more tuned to local
behavior in the data should be relied on more heavily. Thus, spline models should
be given higher priority over transformations of the exposure. Third, the principle
of parsimony should be considered. [,..I]n some settings the principle of
parsimony may suggest that the most informative analysis will rely upon fixing
some parameters rather than estimating them from the data. The impact of the
1-16
-------
fixed parameter choices can be evaluated in sensitivity analyses. In the draft
assessment, fixing the knot when estimating linear spline model fits from relative
risk regressions is one such example. Use of AIC can assist with adhering to this
principle of parsimony, but its application cannot be used naively and without
also including scientific considerations.
EPA RESPONSE: Consistent with SAB recommendations, the model
selection now emphasizes use of the individual-level data, prioritization of
functional forms that allow more local fits in the low-exposure range (e.g.,
spline models), the principle of parsimony (e.g., fixing the knot), less reliance
on AIC, a weighing of biological and statistical considerations, and
prioritization of models that can be used for both environmental exposures
and the occupational exposure scenarios. In addition to the statistical
considerations specified by the SAB, the EPA considered the biological
plausibility of the exposure-response shapes for the cancer endpoints and
overall consistency with the observed data in selecting the final models.
ii. COMMENT: [0]ne advantage of fitting and examining a wide range of models
is to get a better understanding of the behavior of the data in the exposure regions
of interest. For instance, the models shown in Table 4-13 and Figures 4-5 and
4-6 can be compared, ideally with one or more of these presentations augmented
with a few more model fits, including the square root transformation of
cumulative exposure, linear regression of categorical results given more
categories, and several additional 2-piece linear spline models with different
knots. From the comparisons, it is clear that these data suggest a general pattern
of the risk rising very rapidly for low-dose exposures and then continuing to rise
much more slowly for higher exposures. It is reassuring to observe that many of
the fitted models reflect this pattern even though they have different sensitivity to
local data.
EPA RESPONSE: The EPA has added square-root transformation models to
Tables 4-13 and 4-14 and Figure 4-6. Additional linear regressions of
categorical results were not conducted because the EPA has prioritized the
individual-level data, consistent with SAB advice. Additional two-piece
linear spline models with different knots are considered in the sensitivity
analyses in Section D.1.7 of Appendix D. The EPA agrees that it is reassuring
1-17
-------
that many of the fitted models reflect a general pattern of the risk rising
rapidly for low-dose exposures and then continuing to rise more slowly for
higher exposures. Ultimately, the two-piece spline model was selected,
consistent with SAB concurrence.
iii. COMMENT: The application of AIC for selecting models is acceptable within
some constraints as outlined in the following discussion. [...] (The following
discussion is intended to be fairly comprehensive and thus covers points that the
SAB did not identify as problematic in the draft assessment.) AIC is an
appropriate tool to use for model selection for both nested and non-nested models,
provided these models use the same likelihood formulation and the same data.
AIC is not the preferred way to characterize model fit. For model selection, (1)
AIC is not an appropriate tool for comparing across different models that are fit
using different measures, such as comparing a Poisson vs. least squares fit to
count data; (2) one should not use AICs to compare models using different
transformations of the outcome variable; and (3) comparing AICs from models
estimated using different software tools, including different implementations
within the same statistical package can be challenging because many calculations
of AIC remove constants in the likelihood from the estimated AIC. These AIC
features require that users interested in comparing AICs across different software
routines (even those within one statistical package) understand exactly what
likelihood is being maximized and how the AIC is calculated. AIC can be used to
compare the same regression model with the same outcome variable and different
predictors whether or not these models are nested. This gives a consistent
estimate of the mean-squared prediction error (MSPE), which is one criterion for
choosing a model. Finally, the theory behind this MSPE criterion can break down
with a large number of models. Thus, naive applications of AIC for model
selection can be problematic (but are not necessarily so in any particular
application). In particular, differences in AICs could be an artifact of how the
calculation was done. This is a possible difference between the linear and
exponential relative risk models applied to the breast cancer incidence data.
EPA RESPONSE: The EPA notes that the SAB comment identifies general
situations in which comparing AICs might be inappropriate, but it does not
state that any of those situations arose in the EtO analyses. Nevertheless, the
EPA has confirmed that the AIC is being used appropriately—the different
1-18
-------
models being compared were fit using the same measures, the models had the
same outcome variable, and the models (both linear and exponential relative
risk models) were estimated with software routines that calculate AIC in the
same way. See also response to Comment 1.3 above.
e. More detailed comments from the text regarding knot selection (p. 13 of SAB
report)
i. COMMENT: The method used to identify the knots involves a sequential search
over a range of plausible knots to identify the value at which the likelihood is
maximized. This is scientifically appropriate and a practical solution that is
transparently described.
EPA RESPONSE: The EPA applies the same approach to knot selection in
the revised assessment.
3. More detailed comments regarding lymphoid cancer model selection
a. Bulleted summary recommendations regarding rationale for selecting linear
regression of categorical results (p. 15 of SAB report)
i. COMMENT: The SAB recommends that the linear regression of categorical
estimates of lymphoid cancer mortality risk not be selected as the preferred model
unless the individual exposure model results are biologically implausible.
EPA RESPON SE: The EPA no longer relies on the linear regression of
categorical results as the preferred model but, rather, has selected a two-piece
linear spline model based on the individual-level data. See also response to
Comment 1.4 above.
ii. COMMENT: In deriving unit risk estimates under a linear regression model for
risk by exposure category the use of category median exposure rather than the
mean exposure is recommended.
EPA RESPONSE: The EPA considers the mean exposure to be most
suitable in this context (i.e., for RR as a linear function of cumulative
1-19
-------
exposure, with bounded categories); however, because the EPA no longer
relies on the linear regression of categorical results as the preferred model, it
does not impact the conclusions of the assessment.
COMMENT: The SAB recommends presentation of multiple estimates of the
unit risk derived under the alternative models for individual and categorized
exposures.
EPA RESPONSE: The EPA has expanded Table 4-7 to include more
alternative models (e.g., linear two-piece spline models and other linear
models of individual-level data as well as square-root-transformation models),
and to present estimates of the unit risk for each model. Alternative linear
regressions of categorical results were not conducted because the EPA has
prioritized the individual-level data, consistent with the SAB's advice.
More detailed comments from the text regarding model selection (p. 14-15 of
SAB report) mostly contain recommendations regarding the linear regression of
categorical results in the event that the EPA retained that approach for the preferred
model, but as the EPA has selected a two-piece linear spline model based on the
individual-level data for the revised assessment, these comments are no longer
relevant.
Bulleted summary recommendations regarding model selection for estimating
low-exposure cancer risks and cancer risks from occupational exposure
scenarios (p. 15 of SAB report)
COMMENT: As noted in the response to Charge Question 3a, the SAB
recommends that the linear regression of categorical estimates of lymphoid cancer
mortality risk not be selected as the preferred model unless the individual
exposure model results are biologically implausible.
EPA RESPON SE: The EPA no longer relies on the linear regression of
categorical results as the preferred model but, rather, has selected a two-piece
linear spline model based on the individual-level data.
1-20
-------
COMMENT: The SAB finds the rationale for the selection of the preferred
exposure-response model for lymphoid cancer to be lacking and not transparently
communicated. The SAB refers to the response to Charge Questions 2a and 2b
for general recommendations to strengthen the model selection rationale and
transparency in the discussion of model inputs and model fitting for the lymphoid
cancer data.
EPA RESPONSE: Consistent with SAB recommendations, the model
selection now emphasizes use of the individual-level data, prioritization of
functional forms that allow more local fits in the low-exposure range (e.g.,
spline models), the principle of parsimony, less reliance on AIC, a weighing
of biological and statistical considerations, and prioritization of models that
can be used for both environmental exposures and the occupational exposure
scenarios.
More detailed comments regarding model selection for estimating low-exposure
cancer risks and cancer risks from occupational exposure scenarios (p. 15 of
SAB report)
COMMENT: The SAB suggests that the EPA consider using the same model
for both environmental and occupational exposures. The use of different models
needs sufficient justification.
EPA RESPONSE: Consistent with SAB recommendations, the EPA now
uses the same model for both environmental and occupational exposures. See
also response to Comment 1.5 above.
More detailed comments from the text regarding the derivation of lymphoid
cancer incidence estimates from mortality data (p. 15-16 of SAB report)
COMMENT: The approach used for deriving risk estimates for lymphoid cancer
incidence and the rationale for using this approach are explained transparently and
are scientifically appropriate. However, if the draft assessment were also
intended for a broad audience, the approach could be more transparently
described. The SAB suggests the EPA go through some more crudely estimated
approaches so general readers can understand clearly all the different aspects of
1-21
-------
obtaining the unit risk and excess risk estimates without having to rely on the
more complex life table analyses. If the EPA judges it to be informative, the SAB
suggests that extra lifetime risk be presented in terms of the number of lymphoid
cancers that are due to the exposure to EtO in the cohort.
EPA RESPON SE: The EPA has added a more crude approach to illustrate
the derivation of the estimates (see Sections 4.1.1.2 and 4.1.2.3). In addition,
the estimated numbers of lymphoid and breast cancers in the cohort that are
due to EtO exposure, assuming the selected exposure-response models, are
shown in Table 1-1.
Table 1-1. Number of cancer cases in the cohort attributable to
EtO exposure, assuming the selected models
Cancer type
Mean
exposure
(ppm x
days, with
15-year
lag)3
Selected
model
RR
estimate
from
selected
modelb
Attributable
fraction0
Total
cases
in
cohort
Number of
cases in
cohort
attributable
to EtO
exposure11
Lymphoid
cancer
mortality
8,704
Two-piece
linear spline
model with
knot at 1,600
ppm x dayse
2.28
0.56
53
30
Breast cancer
incidence
(subcohort
with
interviews)
9,230
Two-piece
linear spline
model with
knot at 5,750
ppm x daysf
1.56
0.36
233
83
aFrom the risk sets.
bCalculated from selected model at mean exposure.
cAttributable fraction = (RR-1)/RR.
dNumber of attributable cases = attributable fraction x total cases.
ePi = 7.58 x icr4; p2 = -7.48 x lCr4
f|3i = 8.978 x lO-5; p2= -7.786 x lCr5
ii. COMMENT: [T]he risk estimates (Table 4-5, for example) would benefit by
expressing these in scientific notation, rather than a list of leading zeros.
1-22
-------
EPA RESPON SE: In cases in which there is more than one leading zero,
most results are now expressed in scientific notation, consistent with SAB
recommendations.
4. More detailed comments regarding the qualitative discussions of uncertainty in the
cancer risk estimates
a. Bulleted summary recommendations regarding the qualitative discussions of
uncertainty in the cancer risk estimates (p. 19 of SAB report)
i. COMMENT: The SAB recommends that the EPA consolidate the current
discussion of exposure uncertainty that appears in various sections of Appendices
D and H and also to include in the body of the draft assessment a qualitative
discussion of the statistical uncertainty that is associated with the model-based
predictions of annual exposures.
EPA RESPONSE: Instead of consolidating the discussions in Appendices D
and H, the EPA has expanded the discussion of exposure uncertainty in
Section 4.1.4.2.1, which is the main section of the document in which
exposure uncertainty is addressed. Information on the statistical uncertainty
that is associated with the model-based predictions of annual exposures,
however, is lacking. The EPA had considered an errors-in-variables analysis,
as discussed in Section D.7 of Appendix D (see also "grouped data
regression" response on page H-28 of Appendix H); however, it was
determined that such an analysis would be very time consuming and involve a
lot of assumptions, and the analysis was deemed to be beyond the scope of
this assessment.
ii. COMMENT: To better characterize the NIOSH worker samples and their
exposure profiles, the SAB recommends that key demographic, work history and
exposure characteristics of the NIOSH cases and controls be summarized in
descriptive tables or figures in the body of the EtO risk assessment report.
EPA RESPONSE: The EPA has included the recommended results in
Section D.5 of Appendix D.
1-23
-------
COMMENT: The EPA should ensure that they obtain a copy of the NIOSH
individual data including all relevant data released from NIOSH to members of
the public.
EPA RESPONSE: In response to SAB recommendations, the EPA has
obtained the relevant publicly available data, which include the cohort
mortality data and some exposure data pertaining to modeled exposure levels
for the 13 plants.
COMMENT: The SAB repeats its recommendation from previous charge
questions that there be improvements in the quantification of the results from the
models that were fit as a way of improving the qualitative discussion of
uncertainty. Specifically, unit risks should be reported and compared in
sensitivity analyses for a rich set of models.
EPA RESPONSE: Consistent with SAB recommendations, the EPA now
reports unit risks for all the comparisons. See also response to Comment 1.6
above.
COMMENT: The SAB recommends down-weighting all epidemiological results
that are based on external standards (e.g., standardized mortality ratio,
standardized incidence ratio).
EPA RESPONSE: All of the EPA's quantitative estimates are based on
internal comparisons.
More detailed comments from the text regarding the qualitative discussions of
uncertainty in the cancer risk estimates (p. 16-19 of SAB report)
COMMENT: The uncertainty discussions are generally clear, objective, and
scientifically appropriate but they can be improved and extended. Considerations
about uncertainty directly pertaining to the analyses reported can be separated into
1) uncertainty due to the data (particularly from reliance on a single dataset), and
2) uncertainty of the results.
1-24
-------
EPA RESPONSE: The uncertainty discussion has been restructured to
address uncertainty due to the data themselves (see Section 4.1.4.1) separately
from uncertainty of the results given the data (see Section 4.1.4.2).
ii. COMMENT: The SAB supports the use of the NIOSHEtO worker cohort
described in Steenland etal. (2004) and Steenland et al. (2003) as the primary
data source for the modeling of cancer risk from EtO exposures. This is
consistent with the support for the data source in the previous SAB (2007) review.
The support of the NIOSH data is founded on study documentation of the original
exposure measurements, procedures for exposure estimation (Hornung et al.,
1994) and historical modeling (prediction) of exposures that occurred before the
time period in which actual exposure measurements were systematically
collected. All such model-based reconstructions of exposure data are subject to
variable and potentially systematic sources of error (i.e., bias). [...] Appendices D
and H of the current draft assessment provide a comprehensive response to most
of the key questions of data or model uncertainty that were raised in the SAB
(2007) review (see the response to Charge Question 5b [Section n.6 below]).
[,..T]he SAB recommends that the EPA consolidate the current discussion of
exposure uncertainty that appears in various sections of Appendices D and H and
also to include in the body of the draft assessment a qualitative discussion of the
statistical uncertainty that is associated with the model-based predictions of
annual exposures.
EPA RESPONSE: Instead of consolidating the discussions in Appendices D
and H, the EPA has expanded the discussion of exposure uncertainty in
Section 4.1.4.2.1, which is the main section of the document in which
exposure uncertainty is addressed. Information on the statistical uncertainty
that is associated with the model-based predictions of annual exposures,
however, is lacking. The EPA had considered an errors-in-variables analysis,
as discussed in Section D.7 of Appendix D (see also "grouped data
regression" response on page H-28 of Appendix H); however, it was
determined that such an analysis would be very time-consuming and involve a
lot of assumptions, and the analysis was deemed to be beyond the scope of
this assessment.
1-25
-------
COMMENT: On page 17 of the SAB report, the SAB recommends a list of
characteristics of the NIOSH cases and controls to be summarized in tables or
figures - gender distribution over time, year of entry and age of entry to the EtO
workforce, duration of employment in the EtO cohort, and age and year of
departure/retirement from the EtO cohort - as well as a list of exposure
characteristics to summarize - box plot of cumulative total and peak exposures
for individual cases and controls, time plot of the distribution of computed mean,
median, and 25th, 75th, and 95th percentile values for annual exposures among
cases and controls, and summary of percent of total case and control individual
exposures in the worker histories that are excluded when various lags are imposed
(e.g., 5, 10,15 and 20 years).
EPA RESPONSE: Each of these characteristics has been summarized in
tables and figures in Section D.5 of Appendix D.
COMMENT: The SAB is also concerned that public commenters had exposure
data from the NIOSH cohort that the EPA did not have. For instance, a few
selected graphs were presented in public comments to the Augmented CAAC that
indicated exposure predictions for four jobs in two of the fourteen plants showed
lower exposures in some or all years prior to 1975. The SAB was provided only a
few carefully selected examples, and thus was unable to assess the extent of these
surprising data. This is an uncertainty that can easily be ruled out. Upon
reviewing the model equation in Hornung et al. (1994), the SAB finds the
surprising historical behavior to be unlikely and could be explained by changes in
processes in specific plants, rather than some failure of the model to capture
historically larger exposures.
EPA RESPONSE: The EPA has obtained some exposure data from NIOSH
and has ascertained that, contrary to public comments made at the SAB
meeting, the NIOSH EtO exposure patterns are not anomalous, but rather
reflect the underlying changes in variables predicting exposure over time.
One of the principal drivers of the NIOSH exposure levels was the cubic feet
of the sterilizers used [see Table III, Hornung etal. (1994)1. It was not
uncommon in these plants for sterilizer volume to have increased over time as
the demand for EtO-sterilized products increased. Increased sterilizer volume
generally resulted in higher predicted average exposures until the late 1970s,
1-26
-------
when increased controls were used after it became known that EtO might be
dangerous. Table 1-2 shows the sterilizer volume, as well as the
model-predicted EtO levels, from the first example for Plant 1 presented at the
SAB meeting (Dept OI/Oper MP). The sterilizer volume in this plant
increased until the mid-1970s and then decreased, and predicted exposure
levels followed the same pattern. The other example presented at the SAB
meeting for this plant, using Dept OQ/Oper AF, exhibits the same
concordance.
Table 1-2. Plant 1, sterilizer volume and predicted EtO
exposure levels by year
Plant 1, Dept OI, Oper MP
Year
Sterilizer volume (cubic ft)
Predicted EtO level (ppm)
1966
650
2
1967-1968
1,300
4.3
1969-1975
2,250
9.1
1976-1977
1,600
5.9
1978-1979
650
2
Plant 5 follows a similar pattern. Table 1-3 shows the sterilizer volume, as well as
the model-predicted EtO levels, from the first example for Plant 5 presented at the
SAB meeting (Dept 1, Oper ZZ). The predicted exposure levels across time again
follow closely the sterilizer volume. The same concordance is seen for the 2nd
example in Plant 5 (Dept Ol/Oper 82).
Table 1-3. Plant 5, sterilizer volume and predicted ETO
exposure levels by year
Plant 5, Dept 1, Oper ZZ
Year
Sterilizer volume (cubic ft)
Predicted ETO level (ppm)
1943-50
887
6
1951-61
1,679
15
1962-70
1,304
10
1971-72
1,964
18
1973-76
2,624
26
1977-78
3,284
32
1-27
-------
COMMENT: Although the SAB concurs with the EPA's decision to rely solely
on the NIOSHdataset for the risk assessment, the use of only one dataset is a
source of uncertainty. This uncertainty can be reduced by highlighting how the
Swedish sterilization workers data (Mikoczy et al.. 2011) help support the
conclusions reached from the NIOSHdata.
EPA RESPONSE: In response to SAB recommendations, the EPA has cited
the Mikoczy et al. (2011) study as supporting the conclusions reached from
the NIOSHdata; this is discussed in the context of reducing the uncertainty
associated with using a single study in Section 4.1.4.1 (see also response to
Comment II.7.a.i below).
COMMENT: The SAB recommends better quantification of the results from the
models that were fit as a way of improving the qualitative discussion of
uncertainty. In particular, as has been noted in responses to previous charge
questions, the unit risks should be reported and compared in sensitivity analyses
for a rich set of models. This could include analyses that e.g., differ according to
the various outcomes, subcohorts, link functions, functional forms of the exposure
(i.e., exposure parameterizations), exposure metrics, exposure lags (see response
to Charge Question 1), confounder adjustments, and standard error estimation
approaches (Wald vs. profile likelihood). Such information would provide
context for the unit risk behavior across the range of plausible models. The SAB
also encourages consideration of focusing the reporting of sensitivity analyses on
the target parameters of interest (unit risk, excess risk).
EPA RESPONSE: Additional sensitivity analyses have been added to the
assessment, and all the results are reported as unit risk estimates. For both
breast cancer incidence and lymphoid cancer, various models are compared,
including different relative risk functions and functional forms of the
exposure, and the sensitivity of the selected models to different lag periods,
knots, and upper-bound estimation approaches is examined. For breast cancer
incidence, additional analyses include comparisons of the subcohort and the
full cohort, of total breast cancer and invasive breast cancer only, and of the
full model and the selected model with the nonexposure covariates (breast
cancer risk factors) excluded. For lymphoid cancer, results are compared to
the results for all lymphohematopoietic cancers. Sensitivity analyses were not
1-28
-------
conducted for alternative exposure metrics because it is unclear how to derive
unit risk estimates for metrics such as duration and peak exposure, and as
noted in Comment II.2.b.iii, the SAB concurred with the reliance on analyses
based on the individual estimates of cumulative exposure for risk assessment.
vii. COMMENT: If feasible, consideration of additional analyses using alternative
exposure metrics is suggested. The December 4, 2014 EPA memo (U.S. EPA,
2014) notes that four exposure metrics were already considered by the agency. If
additional metrics are available, it would be valuable to consider these as well.
EPA RESPONSE: No additional exposure metrics are available, and as
noted in the response to Comment II.4.b.vi above, it was not considered
feasible to derive unit risk estimates for alternative exposure metrics.
viii. COMMENT: The SAB encourages consideration of the following points in the
document, either directly in the uncertainty discussion, or in other places, as
appropriate. The first two points are observations, the third is a recommendation.
a) The dose-response model indicated by the NIOSH cohort that suggests risk
increases sharply for low exposures and then increases further but less steeply
for higher exposures. The biologic plausibility of this functional form is
uncertain, and evidence that there are mechanistic explanations that support
this form will inform the risk assessment.
EPA RESPONSE: The EPA is not aware of a mechanistic explanation
for the shape of the exposure-response relationship in the NIOSH cohort
data but notes that the SAB found it "reassuring to observe that many of
the fitted models reflect this pattern" for breast cancer incidence data (p.
12 of the SAB report), and the same is true for the lymphoid cancer data.
Similarly, the SAB noted that the results of the Mikoczy et al. (2011)
study could be used to support the selected model for breast cancer
incidence (p. 10 of SAB report). The EPA now cites the Mikoczy et al.
(2011) study as supporting the selected model (see Section 4.1.4).
b) The analysis of NIOSH data relies on cumulative exposure as the dose metric.
Given the status of the exposure data, it is unlikely that other more refined
1-29
-------
exposure information can be used to better understand the mechanisms ofEtO
exposure in cancer initiation. Furthermore, it is often difficult to determine
mechanisms from epidemiological data, particularly when these data are
limited.
EPA RESPON SE: The EPA has considered the issue and agrees that it is
unlikely that other more refined exposure information can be used to
better understand the mechanisms of EtO exposure in cancer initiation and
that it is often difficult to determine mechanisms from epidemiological
data, particularly when these data are limited, as is the caes with EtO.
c) The SAB recommends down-weighting all epidemiological results that are
based on external standards (e.g., standardized mortality ratio, standardized
incidence ratio). The presence of the healthy worker effect cannot be denied
in these occupational data and the use of an external standard for comparison
does not avoid healthy worker types of biases.
EPA RESPON SE: The EPA agrees that internal comparisons are
superior to external comparisons, and all of the EPA's quantitative
estimates are based on internal comparisons.
5. More detailed comments regarding genotoxicity discussions
a. Bulleted summary recommendations regarding genotoxicity discussions (p.
19-21 of SAB report)
i. COMMENT: Table 3.6 should be revised to specify the sites involved and the
relative importance (weight) assigned to each of the individual studies presented.
In addition, a new table should be added to show the dose-response relationships
for the formation of DNA adducts and the in vivo genotoxic effects in humans and
comparative model systems.
EPA RESPONSE: Table 3.6 has been revised and is now Table 3-8. A
similar table (see Table 3 -7) was created showing the cytogenetic effects in
laboratory animals. It has been made clearer that most of the studies are of
peripheral blood lymphocytes. The relative importance of the studies is
1-30
-------
considered primarily as a function of the genotoxic endpoint investigated and
the estimated level of exposure. A discussion regarding the relative
importance (qualitative weight) of various genotoxicity endpoints has been
included (see Section 3.3.3.3), and the studies in Table 3-8 have been
arranged roughly in order of increasing estimated exposure concentration.
The studies of chromosomal aberrations and sister chromatid exchanges are
generally positive at the higher exposure levels, while the data on micronuclei
at the higher exposure levels are more limited. A "Comments" column has
also been added to the table providing more study details. In addition, a
similar table presenting the results from studies reporting DNA adducts and/or
mutations following in vivo exposures in humans or laboratory animals has
been added (see Table 3-6), along with two new summary tables showing the
temporal and dose-response relationships for the in vivo formation of DNA
adducts and mutations (see Table 3-9) or cytogenetic effects (see Table 3-10)
in humans and laboratory animals.
COMMENT: The rationale for decisions made regarding model selection for
calculations of unit risk should be presented in this section, and elsewhere, within
the context of MO A considerations and the initial key biological events involved
in mutagenesis and carcinogenesis.
EPA RESPONSE: The models used for the epidemiologic data are
essentially empirical curve-fitting models, and it is unclear how the available
biological data can be used to guide general model selection. In one specific
case, considerations of the biological data did inform the decision not to use a
two-hit quadratic model for lymphohematopoietic cancers
(see Section 3.4.1.2). In addition, the conclusion of a mutagenic mode of
action resulting from direct EtO-DNA interactions occurring in the absence of
any evidence for concurrent cytotoxicity or alternative modes of action
(see Sections 3.4.1.1, 3.4.1.4, 3.4.2, and 3.4.3) provides support for the use of
linear, low-exposure extrapolation for the derivation of the unit risk estimate
(e.g., Section 4.1.1.2).
COMMENT: Although the description of the database was found to be
adequate, the synthesis of the information used to support a mutagenic MOA
should be presented in a more systematic and complete manner. Section 3.4
1-31
-------
should be reorganized around a broader evidence base for a mutagenic MOA to
more clearly establish the framework for defining mutagenic MOA. Key
elements of this framework, as informed by a recent review by Eastmond (2012)
should include [details of the sub-bullets are presented in the SAB report]:
o Characterization of Molecular Alterations
o Characterization of mutagenic or clastogenic effects
EPA RESPONSE: Section 3.3.3 (genotoxicity) has been revised to
synthesize the information used to support a mutagenic MOA in a more
systematic and complete manner, including more detailed characterization of
the molecular alterations (e.g., Section 3.3.3.1) and mutagenic
(see Section 3.3.3.2) and other genotoxic (see Section 3.3.3.3) effects. In
addition, a table (see Table 3-7) summarizing the cytogenetic effects in
laboratory animals (comparable to the previous Table 3-6 [now Table 3-8]
for humans) and a table (see Table 3-6) summarizing the dose-response
information on DNA adducts and mutations in humans and laboratory animals
have been added. Furthermore, a substantially expanded summary
(see Section 3.3.3.4) of the genotoxicity section summarizes and integrates the
mutagenicity and genotoxicity information and includes two new tables
summarizing the temporal and dose-response findings for DNA adducts and
mutations (see Table 3-9) and for cytogenetic effects (see Table 3-10) in
humans and laboratory animals. Section 3.4 has been revised and reorganized
to more clearly discuss the mechanisms by which the genotoxic effects might
be instrumental in a mutagenic mode of action (see Section 3.4.1.1),
particularly in the target organs (see Sections 3.4.1.2 and 3.4.1.3).
iv. COMMENT: In the absence of further mechanistic information, evidence for
DNA interactions coupled with consistency in the occurrence of
mutagenic/clastogenic effects provides a sound basis for applying a mutagenic
MOA to risk assessment. Additional data that may be informative in revising the
draft to support a mutagenic MOA includes [details of the sub-bullets are
presented in the SAB report]:
o Genotoxic Effects in Cancer Target Organs
o Non-linearities
o Temporal Relationships
o Alternative Mechanisms
1-32
-------
o Summary of Cancer MO A
EPA RESPONSE: Building on the information in Section 3.3, Section 3.4
has been revised and reorganized to more clearly discuss the mechanisms by
which the genotoxic and mutagenic effects might be instrumental in EtO-
induced carcinogenesis (see Section 3.4.1.1), particularly in the target organs
(see Sections 3.4.1.2 and 3.4.1.3), i.e., how a mutagenic mode of action might
be operating. The support for low-exposure linearity from the DNA adduct
data at very low EtO doses (Marsden et al.. 2009) is discussed in more detail
in the derivation of the unit risk estimate (see Section 4.5); however, cross-
referencing to that discussion has been added to Section 3.3.3.1. Temporal
relationships are addressed in the expanded summary in Section 3.3.3.4 and in
the new Tables 3-9 and 3-10. A short section (see Section 3.4.2) has been
added on alternative mechanisms. A revised summary of the evidence for a
mutagenic mode of action is provided in Section 3.4.1.4.
Bulleted suggestions regarding genotoxicity discussions (p. 21 of SAB report)
COMMENT: Inclusion of additional experimental details about the separation
of endogenous from exogenous adducts as reported by Marsden et al. (2009)
would help provide biological perspective for issues related to risk assessment
considerations, especially linearity versus non-linearity of dose-response
relationships.
EPA RESPON SE: The discussions of the Marsden et al. (2009) study in
Section 3.3.3.1 and in Section C.7 of Appendix C were expanded. This study
reported linear dose-response relationships for N7-HEG adducts in the three
tissues evaluated from exogenous EtO dosing down to very low doses.
COMMENT: The genotoxicity section would be improved by consideration of
the role that differences in DNA repair capacity between different target cells in
different tissues plays in relative vulnerability to mutagenesis. For example,
genes known to regulate vulnerability of breast cancer in women, such as
BRAC1, BRAC2 and XRCC1, are known to regulate DNA repair pathways in
breast tissue (Shi et al., 2004; Hu et al., 2002). This line of thinking can help to
1-33
-------
inform the biological bases to better understand the shape of the dose response in
the low-dose region of the NIOSH dataset.
EPA RESPONSE: This material is covered in Section C.6of Appendix C.
Mention has also been added to Sections 3.4.1.1, 3.4.1.2, and 3.4.1.3. There
was insufficient information to elucidate a basis for the supralinear
exposure-response relationships observed for lymphoid and breast cancers in
the NIOSH study.
COMMENT: In light of the above discussion, the organization of the text can
also be revised to include information about known differences in mutagenic and
carcinogenic pathways for EtO at different tumor sites, as well as the degree to
which biochemical differences at the cellular or tissue level differentially impact
MO A. Furthermore, references made in page 3-29 to the levels of different
adducts are presented without making a clear and necessary distinction between
the putative or assigned biological impact for N-7 versus 0-6 DNA adducts.
EPA RESPONSE: Not much is known about different pathways operating at
different tumor sites, but the text has been more clearly organized to discuss
possible mechanisms specific to lymphohematopoietic cancers
(see Section 3.4.1.2) and breast cancer (see Section 3.4.1.3). Also, to the
extent that there is information, the sensitivities of different tissues to
EtO-induced mutagenicity and genotoxicity are discussed in these sections. In
addition, the discussion of the different adducts and their biological
implications in Section 3.3.3.1 has been expanded.
More detailed comments from the text regarding the genotoxicity discussions (p.
19 of SAB report)
COMMENT: Section 3.3.3 and Appendix C of the draft assessment present an
accurate, objective and transparent summary of the results of research studies
published up to July 2013 on EtO genotoxicity. The SAB agrees that the weight
of the scientific evidence from epidemiological studies, laboratory animal studies
and in vitro studies supports the general conclusion that the carcinogenicity of
EtO in laboratory animals and humans is mediated through a mutagenic mode of
action (MOA). Indeed, the genotoxicity database has firmly established that EtO
1-34
-------
is a direct-acting agent, as evidenced by the formation of DNAadducts and highly
reproducible, positive effects in a variety of in vitro and in vivo mutation and
clastogenesis assays. The genotoxic studies examined showed adducts,
mutagenesis, or clastogenesis at the bioassay doses and in some cases lower
(Donner et al.. 2010; Marsden et al.. 2009; Recio et al.. 2004; Walker et al..
1997). providing evidence of dose-response concordance for a mutagenic mode of
action.
EPA RESPONSE: The EPA has retained the conclusion that there is
sufficient weight of evidence that a mutagenic mode of action is operative in
EtO carcinogenicity, but in response to SAB comments, the EPA has
strengthened the presentation of the evidence (e.g., with the expanded
discussion of temporal and dose-response relationships in Section 3.3.3.4).
6. More detailed comments regarding Section H.1 of Appendix H—responses to SAB
comments regarding the 2006 draft
a. Bulleted summary recommendations regarding Section H.1 (p. 28 of SAB
report)
i. COMMENT: Consider adding a brief introductory summary of purpose and
highlights to each chapter 2, 3 and 4 to improve the readability of the assessment
document.
EPA RESPONSE: Text boxes containing a brief summary of the purpose
and the major conclusions of the chapter have been added to the beginning of
Chapters 2, 3, and 4.
ii. COMMENT: Expand the description of endogenous sources of EtO to include
formation from external exposure to ethylene.
EPA RESPONSE: Discussion of the conversion of exogenous ethylene to
EtO has been added to Section 3.3.3.1.
1-35
-------
COMMENT: Summarize the key highlights of Dr. Steenland's further analysis
as they reflect on the reliability of the cumulative exposure with 15-year lag
metric used in the dose-response assessment.
EPA RESPON SE: In response to SAB comments, the EPA has summarized
the key highlights of the response that Steenland provided to the 2007 SAB
comments on Charge Question 2. a (see revised pages H-8 to H-9 of Appendix
H).
More detailed comments from the text regarding Section H.1 (p. 21-28 of SAB
report)
COMMENT: Appendix H provides a summary of the SAB (2007) peer review
comments, followed by the agency's response. Overall, the EPA was highly
responsive to the comments and recommendations presented in the SAB (2007)
report. Responses are transparent, objective, and for the most part, accurate
(exceptions are noted in the current review). The agency should be commended
for this effort because this was a particularly challenging undertaking given the
lack of consensus in the SAB (2007) report on several issues critical to key
outcomes of the draft assessments. The EPA not only addressed all major
consensus recommendations but also responded specifically to both the majority
and minority opinions whenever divergent views were expressed.
EPA RESPONSE: The EPA thanks the committee.
COMMENT: There are some recommendations or suggestions of the SAB
(2007) peer review that are not implemented in the current draft assessment [...].
Feedback regarding these agency decisions is provided in the detailed response to
this charge question and in responses to other charge questions. This feedback
can be summarized as follows:
1. The SAB finds that EtO likely acts by a mutagenic MO A and therefore its
potency should be modeled according to a linear low-dose model. EPA's
Guidelines for Carcinogen Risk Assessment (U.S. EPA 2005a) note the
following: "A nonlinear extrapolation method can be used for cases with
sufficient data to ascertain the mode of action and to conclude that it is not
linear at low doses " (p. 3-23). The SAB finds that the empirical data for
EtO and its MOA are consistent with a linear low-dose extrapolation and the
1-36
-------
database does not provide the type of evidence that the Cancer Guidelines
would find sufficient to support a nonlinear MO A, which precludes the need
for the presentation of nonlinear modeling approaches.
2. The SAB concurs with the decision not to use the Union Carbide Cohort data
for unit risk derivation, but suggests that the agency discuss the weight of the
evidence of the UCC, NIOSH, and Swedish sterilization workers studies. More
suggestions regarding the Swedish sterilization workers study can be found in
the response to charge question 6.
3. The SAB suggests that the EPA consider using the same model for both
environmental and occupational exposures.
4. The SAB agrees with the decision to not move the contents of Appendix A to
the main body of the draft assessment.
EPA RESPONSE: Consistent with SAB concurrence, the EPA has retained
the use of linear low-exposure extrapolation, as well as the decisions not to
use the UCC data for unit risk derivation and not to move the contents of
Appendix A into the body of the assessment. In addition, in response to SAB
comments, the EPA has considered the implications of the UCC and Mikoczy
et al. exposure-response relationships in the uncertainty discussions of the unit
risk estimates derived from the NIOSH data (see Section 4.1.4.1), and the
EPA now uses the same exposure-response model for both environmental and
occupational exposures for both cancer endpoints (see Sections 4.1.1, 4.1.2,
and 4.7).
iii. COMMENT: This charge question asks specifically about responses to
comments on endogenous EtO (p. H-4), a nonlinear approach (P. H-13 toH-17),
and the cancer hazard characterization. Each of these topics is addressed in the
detailed response to the charge question, but can be summarized as follows: (1)
The SAB supports the expanded discussion of endogenous EtO provided in the
draft assessment and has suggestions for further improvement; (2) as noted above,
the SAB agrees with the decision not to include a toxicity value for EtO based
upon nonlinear extrapolation, but recommends a more balanced and objective
discussion of the subject; and (3) the SAB recognizes and agrees with revisions to
strengthen support for a classification of EtO as "carcinogenic to humans."
EPA RESPONSE: In response to specific comments detailed further below
(and above), the EPA has revised the assessment. For example, discussion of
the conversion of exogenous ethylene to EtO has been added to Section
3.3.3.1 (see response to Comment II.6.a.ii above) and cross-referencing has
1-37
-------
been added to the discussion of endogenous EtO exposure in Section 3.3.3.1
to link to the discussion of the relevance of endogenous EtO to the unit risk
estimate in Section 4.5 (see response to Comment II.6.b.vi below). In
addition, the discussion of not using a nonlinear approach has been made
more balanced. For example, cautions about the Marsden et al. (2009) study
have been added (e.g., page H-13 of Appendix H). Furtheremore, the related
discussions of genotoxicity (see Section 3.3.3) and mode of action (see
Section 3.4) have been made more comprehensive and balanced. Also, the
EPA has included discussion of two more recent studies (Zhang et al„ 2015b;
Zhang et aL 2015 a), which provide further support for a mutagenic mode of
action and for oxidative stress not being an additional mode of action of
concern (see Section J.4.1 of Appendix J).
COMMENT: The SAB agrees with the decision not to transfer in tola materials
from Appendix A - Critical Review of the Epidemiological Evidence to the main
body of the assessment. The addition of the two brief summary tables on the
hematopoietic and breast cancer studies is a good alternative for strengthening the
chapter. This choice is consistent with the NRC (2011) recommendations that the
main body of the assessment focus on the critical data, rationales, and analyses
used to support the unit risk derivation and that, as much as possible, detailed
description of key and other studies or analyses be summarized in appendices
with appropriate cross-referencing in the main body of the assessment. If
anything, the current document could benefit from transferring more materials to
appendices, although it is acknowledged that the current draft assessment is not
intended to conform completely to the new format for IRIS assessments.
EPA RESPONSE: In the interest of minimizing further delays in the
finalization of this assessment, the EPA has not further condensed the main
assessment text.
COMMENT: The EPA also clarified its designation of the unit risk estimate as
"weak" in the prior draft assessment, and section 3.5.1 of the current draft
assessment provides a good evaluation of the strength of the weight of the
evidence in term of Hill's criteria for causality.
EPA RESPONSE: No response needed.
1-38
-------
vi. COMMENT: Based on the discussion presented in the assessment and
considering the weight of the evidence from human and animal studies, the SAB
finds EPA's conclusion on endogenous exposure to EtO to be supported.
Nonetheless (and also in light of the analyses presented on pages H-15 to H-17
and further insights derived from the SAB recommendations in the response to
Charge Question 5a- Section 3.5 of this report), it appears that recognizing this
source of metabolic EtO and briefly expanding on its relevance to the assessment
would complete the description of sources of endogenous EtO and their relative
importance for adduct formation. This could be readily done in detail in
Appendix C with the expanded, but succinct description added to Chapter 3.0 and
cross-referenced to the appendix.
EPA RESPON SE: The relevance of endogenous EtO exposure to the
assessment is discussed in Section 4.5 in the context of the use of low-dose
linear extrapolation in deriving the unit risk estimate. Cross-referencing to
this section has been added to the discussion of endogenous EtO exposure in
Section 3.3.3.1.
vii. COMMENT: The EPA added 24 of the 34 additional references recommended
by the panel. There was no explanation for the reasons for not including 10 of the
references suggested for inclusion.
EPA RESPONSE: All 34 of the references were considered; however, some
of them were not particularly relevant to the assessment (e.g., one was on
N-nitrosocompounds). The text in Appendix H has been expanded to provide
reasons for the exclusions.
viii. COMMENT: The SAB finds that the EPA has been responsive in providing an
expanded and more balanced discussion of human and animal studies of precursor
events that support a mutagenic MO A. However, the supportive evidence for a
mutagenic MOA needs further enhancement and discussion as indicated in the
SAB response to Charge Question 5a (Section 3.5 of this report).
EPA RESPONSE: In response to SAB comments, the EPA has strengthened
the presentation of the evidence supporting a mutagenic MOA. Section 3.3
1-39
-------
(genotoxicity) has been revised to synthesize the information in a more
systematic and complete manner, including the addition of a substantially
expanded summary that integrates study information in terms of
dose-response and temporal relationships (see Section 3.3.3.4). Section 3.4
has been revised and reorganized to more clearly discuss the mechanisms by
which the genotoxic effects might be instrumental in EtO carcinogenesis
(see Section 3.4.1.1), particularly in the target organs
(see Sections 3.4.1.2 and 3.4.1.3). In addition, the EPA has revised and
expanded Table 3-6 (now 3-8) and provided additional summary tables
(see Tables 3-6, 3-7, 3-9, and 3-10). (For more details on specific revisions,
see responses to detailed comments in Part n.5 above.)
ix. COMMENT: The selection of the NIOSH cohort and the decision not to
combine these data with the Union Carbide cohort is better and more
transparently justified in the revised draft assessment. [...] The SAB concurs with
this assessment of the UCC data and concurs with the decision not to include the
UCC data. However, the SAB considers that a more detailed description of the
NIOSH cohort is needed as it relates to the derivation of exposure metrics, as
indicated in the SAB response to Charge Question 2 (Section 3.2 of this report)
for the current draft assessment.
EPA RESPON SE: The EPA has provided the additional details requested by
the SAB, as indicated in the SAB comments on Charge Question 2 and
detailed in SAB comments on Charge Question 4 (see Comment II.4.b.iii
above); these are summarized in tables and figures in Section D.5 of
Appendix D. Some of the new cohort details summarized in Section D.5
include mean, median, minimum, maximum, and 25th and 75th percentiles of
cumulative exposure in the full cohort; cumulative exposures by year of entry
and by duration of employment; sex distribution over time; distributions of
year of entry, age of entry, duration of employment, and age and year of
departure/retirement; distributions of cumulative and peak exposures for
individual cases and controls; percentages of total case and control individual
exposures in the worker histories that are excluded when the 15-year lag is
imposed; and mean, median, and 25th, 75th, and 95th percentile values for
annual exposures among cases and noncases in the cohort.
1-40
-------
x. COMMENT: It is not known if Dr. Steenland received only the comment as
presented in the Executive Summary of the SAB (2007) report, or the more
detailed discussion in pages 20-22 of the SAB (2007) report. The SAB considers
that, although consultation with Dr. Steenland on the technical aspects of this
recommendation is appropriate because of his intimate knowledge of the exposure
model developed for the NIOSHEtO cohort studies, the EPA should have
provided its own response to the SAB (2007) recommendation. Dr. Steenland
indicates that he was not completely sure about the meaning of the
recommendation and proceeded to present a set of reasonable arguments as to
why the bias introduced by using this metric would not alter the analysis
appreciably. It is also important to note that the exposure estimates likely to be of
lower reliability (because there were no exposure measurement data that could be
included in the exposure model prior to 1979) are also likely to be higher than the
more recent exposures and, therefore, would play a less important role in the
current derivation of the point of departure (POD). The response, however, has
not completely clarified the issue of potential estimate instabilities introduced by
interactions between time-varying predictor variables and the log cumulative
exposure with a 15-year lag exposure estimate. This issue is addressed in the
SAB response to Charge Question 2 (Section 3.2 of this report) for the current
draft assessment.
EPA RESPON SE: The EPA has addressed the SAB comments raised
regarding the NIOSH exposure estimates in response to other charge questions
(e.g., see response to Comment n.4.b.iii above). For example, Section D.5 of
Appendix D now presents time plots of the distribution of computed mean,
median, and 25th, 75th, and 95th percentile values for annual exposures among
cases and controls (see Figures D-22 and D-23), as well as a summary of the
distribution of cumulative exposures as a function of year of entry into
employment (see Table D-52). The EPA has added sensitivity analyses in
which unit risk estimates (see Sections D.1.6 and D.3.5) and extra risk
estimates for occupational exposure scenarios (see Sections D.l.ll andD.3.9)
are derived from the selected models using cumulative exposure with different
lag periods. Furthermore, in response to SAB comments, the EPA deleted
Steenland's response on the issue of potential instability resulting from the
interaction of the lag and the treatment of time in the exposure model and now
provides its own response (pages H-8 to H-9 of Appendix H).
1-41
-------
xi. COMMENT: The EPA was highly responsive in addressing concerns about the
use of categorical data for POD derivation and contracted with Dr. Steenland, the
principal investigator of the NIOSH studies, to perform multiple analyses of the
NIOSH cohort data (including use of individual and categorical exposure
estimates) using alternative modeling approaches. In addition, there was also an
attempt to expand on the error analysis of the NIOSH cohort exposure estimation,
although this could not be accomplished because the data files used in that
assessment were no longer available. Results from the extensive additional
analysis are detailed and well described in the current draft assessment, both in
Chapter 4 and in Appendix D, together with the rationale for supporting the
decisions by EPA in model selection. Problems with the implementation of the
recommendations are described clearly. Outcomes from alternative models are
summarized both in tables and graphical form, with justification for the preferred
models. It is important to emphasize that Dr. Steenland's involvement in the
additional analyses is a strength of the revised draft not only because of his
intimate familiarity with the NIOSH cohort studies but his expertise in exposure
modeling and occupational epidemiology. The revised assessment for breast
cancer risk incidence is based on continuous exposure data. The analysis for LH
cancer subtype is based on the NIOSH cohort lymphoid cancer results (results for
all LH cancers are also presented) for both genders (no statistically significant
gender differences were found). Results for individual and categorical data
models are presented; EPA preferred the non-categorical model.
EPA RESPONSE: No response required.
xii. COMMENT: Although there are still significant concerns with the final
selection of modeling approaches for derivation of unit risk in the current draft
assessment (see the responses to Charge Questions 1-4, Sections 3.1-3.4 of this
report), the EPA should be commended for the effort and the commitment of
resources to address the comments and recommendations in the SAB (2007)
report. Likewise, the EPA considered the SAB's extensive comments on the
rationale for non-linear low-dose extrapolation including additional analysis of
experimental animal data on mutations by EtO (pages H-15 to H-19 of Appendix
H), concluding that the evidence did not indicate low-dose, non-linear
extrapolation or threshold dose-response patterns. Thus, the rationale (including
1-42
-------
more expansion on EPA guidance) for using low-dose, linear extrapolation is
improved and stronger in the current assessment, but some concerns remain (see
responses to Charge Questions 1-3 and 6, Sections 3.1-3.3 and Section 3.6 of this
report).
EPA RESPONSE: No response required here; the EPA has addressed the
SAB concerns related to the exposure-response modeling in the context of the
other SAB comments (e.g., the EPA has revised its model selection for the
lymphoid cancer data and now uses a continuous exposure model; see
response to Comment 1.4 above).
xiii. COMMENT: Concerns about the suitability of life table methodology for
determination ofLECOl have been addressed. The EPA provides a convincing
rationale (especially since alternative approaches are not available) for using the
BEIR IV algorithm. The response to the request to present the range unit risk
estimates for the upper and lower 95% confidence limits of the EC01 is also
reasonable.
EPA RESPONSE: No response required.
xiv. COMMENT: The EPA also responded in detail to the comments provided in
Appendix A of the SAB (2007) report. Many of the comments referred to the use
of categorical exposure metrics and regression on group data that are also the
subject of the current SAB review and are reflected in the responses to Charge
Questions 1-3 (Sections 3.1-3.3 of this report).
EPA RESPONSE: No response required.
xv. COMMENT: The SAB finds this [response regarding expanded discussion of
application of ADAFs (Section4.4)] to be responsive to the SAB (2007) comment.
EPA RESPONSE: No response required.
xvi. COMMENT: The SAB suggests that the EPA consider using the same model
for both environmental and occupational exposures. (Please refer to the response
to Charge Question 3 - Section 3.3 of this report).
1-43
-------
EPA RESPONSE: The EPA is now using the same model for both
environmental and occupational exposures for both cancer endpoints.
xvii. COMMENT: The SAB agrees with EPA's response [on the use of ppm
equivalency for interspecies scaling of EtO exposure],
EPA RESPONSE: No response required.
xviii. COMMENT: SAB comments on uncertainty in the current draft assessment are
reflected in the response to Charge Question 4 (Section 3.4 of this report).
EPA RESPONSE: No response required.
7. More detailed comments regarding Appendix J—new studies
a. Bulleted summary recommendations regarding Appendix J (p. 29 of SAB
report)
i. COMMENT: Specific suggestions for expanded inclusion of the Swedish
sterilization workers study results (Mikoczy etal., 2011) for breast cancer
include:
• Discussion of the study should be moved to a more central position in the draft
assessment.
• The Swedish sterilization worker study should be incorporated into an overall
weight of evidence assessment of EtO effects at low doses.
• Consideration of using the word "strong" in its Bradford-Hill strength of
association analysis.
• Consideration of characterizing the exposure assessment as high quality in light
of the results of the exposure matrix for the early period of the study being
validated by hemoglobin adduct levels (Haemar et al., 1991).
• Consideration of a quantitative risk assessment based on the breast cancer data
in the study.
•Alternately, consideration of applying NIOSH estimates to the Swedish
sterilization workers study to assess the consistency of findings with:
o Low dose exposure
1-44
-------
o Attenuation of risk with higher exposures
o The observation of increased breast cancer risk with 16 more years of
follow-up (latency)
EPA RESPONSE: The EPA has incorporated discussion of the Mikoczy et
al. (2011) study into the main body of the report. For example, the study is
now considered in the weight-of-evidence analysis and supports the
characterization of the epidemiological evidence on breast cancer as "strong"
(see Sections 3.1 and 3.5). In addition, the study is cited as supporting the
supralinear exposure-response relationship observed with the NIOSH breast
cancer incidence data (see Section 4.1.4). Also, a comparison was done of the
Mikoczy etal. (2011) RR estimates with predicted RR values from the
selected model derived from the NIOSH data; the selected model
underestimated the Mikoczy etal. (2011) results (see Section 4.1.4 and
Section J.2.2 of Appendix J).
COMMENT: Consideration of separating agency interpretation of study
findings from those of the studies' authors.
EPA RESPONSE: The EPA has clarified what are Agency interpretations
and what are study author findings.
COMMENT: Consideration of an expanded review of recent studies, including
summary reviews, with specific focus on issues related to mode of action.
EPA RESPON SE: The EPA has added to Appendix J some more recent
studies with significant new information pertaining to mode of action (Zhang
et al.. 2015b; Zhang et al.. 2015a); however, at this stage of development of
the assessment, the Agency did not further consider new studies unless they
provided important new information.
COMMENT: Consideration of emphasizing the importance of internal
comparisons in occupational studies.
EPA RESPON SE: The preference for internal comparisons is listed among
the considerations in evaluating epidemiological studies at the beginning of
1-45
-------
Section 3.1. The statement that "[i]nternal comparisons are considered
superior to external comparisons in occupational epidemiology studies
because internal comparisons help control for the healthy worker effect and
other factors that might be more comparable within a study's worker
population than between the workers and the general population" occurs later
in Section 3.1, but was also brought forward as a footnote to the
considerations at the beginning of the section.
b. More detailed comments from the text regarding Appendix J (p. 28-29 of SAB
report)
i. COMMENT: In general, the logic and progression of the literature review are
clearly supported. However, in the descriptions and assessments of the new
studies, it is not entirely clear which statements are made by the study authors and
which are made by the EPA. The discussion of the Kiran et al. (2010)
case-control study is thorough. The conclusion that the Kiran etal. (2010) study
overall supports the draft assessment is reasonable. The conclusion that small
numbers of participants in the Morgan etal. (1981) and Ambroise etal. (2005)
studies preclude more detailed analysis, but warrant inclusion in the review is
reasonable. The summary of the Valdez-Flores and Sielken (2013) study
discussion in Appendix J-3 is thorough, but parts of the discussion should be
moved to the main body of the draft assessment. The SAB generally agrees that
the new studies in Appendix J do not substantially alter the findings of the
assessment with the exception of the Swedish sterilization workers study
(Mikoczy et al.. 2011; Hagmar et al.. 1991). This study of EtO sterilization
workers, with detailed exposure data at low doses with documented substantial
effects on breast cancer has stronger implications than suggested in the draft
assessment. The strong breast cancer results at low-dose exposures in this study
greatly add to the overall findings. The observation of a 2.5-3.5-fold significantly
elevated risk of breast cancer associated with low cumulative exposure in this
study demonstrates strong evidence of carcinogenicity.
EPA RESPONSE: The EPA has clarified what are Agency interpretations
and what are study author findings. The Morgan et al. (1981) and Ambroise
et al. (2005) studies were ultimately omitted because the Morgan et al. (1981)
study was included in the EPA's 1985 EtO health assessment (U.S. EPA
1-46
-------
1985) and the Ambroise et al. (2005) study, in addition to being too small to
be informative, does not report EtO-specific results. The EPA has
incorporated discussion of the Mikoczy et al. (2011) study into the main body
of the report (see response to Comment II.7.a.i above for details). The
discussion of the Valdez-Flores and Sielken (2013) study has been
substantially shortened and has not been incorporated into the main report
because the issue of modeling the categorical data is no longer of paramount
importance, as the assessment now relies on models of the continuous
exposure data for both cancer endpoints in the NIOSH study.
8. More detailed comments regarding Appendix K (then L) and Section H.2 of
Appendix H—responses to public comments on the 2013 and 2006 drafts,
respectively
a. Detailed comments from the text regarding Appendix K (p. 29-32 of SAB
report) (no bulleted summary recommendations were provided in response to
this charge question)
i. COMMENT: Appendix K presents a summary of the EPA responses to public
comments on the July 2013 draft assessment. The section begins with a brief and
clear summary of the comments received.
EPA RESPONSE: No response required.
ii. COMMENT: Before assessing the responses of the EPA to each of the specific
comments, a general assessment of the nature of the comments received by the
EPA, which primarily came from industry or industry organizations, is presented.
In addressing this charge question, the primary focus is to evaluate the quality and
thoroughness with which the EPA responded to the public comments rather than
to evaluate the issues raised as these are covered in the responses to the other
charge questions in the current report.
EPA RESPONSE: No response required.
iii. COMMENT: Comment 1: This comment claims that the EPA failed to follow
NRC (2011) guidelines and failed to apply a systematic and weight-of-evidence
1-47
-------
approach. The EPA response is clear but could be even stronger. There are
several places in the draft assessment where the weight-of-evidence approach is
discussed and justified. To strengthen the response to this question, some more
detail listing places in the draft assessment where NRC (2011) and EPA
guidelines as well as the systematic and weight-of-evidence approach are
explained and justified would be helpful. There was additional comment on the
use of NIOSH breast cancer incidence data that were not publically available.
The EPA response clearly described their adherence to the EPA Information
Quality Act Guidelines, which do not require all raw epidemiology data be
publically available. Constraints due to confidentiality were also noted.
EPA RESPON SE: In response to SAB comments, the EPA has strengthened
the response to the public comment by including more details of where the
considerations used in the evaluation of the epidemiological studies
(see Section 3.1), weight-of-evidence analysis (see Section 3.5),
characterization of the cancer hazard (see Section 3.5), selection of the
epidemiology study(ies) for quantitative risk estimation (see Section 4.1), and
selection of exposure-response models (see Section 4.1) can be found.
COMMENT: Comment 2: The comment states that the EPA did not properly
explain the criteria used to evaluate studies and deem them to be of high quality
for inclusion in their analysis. A summary of the characteristics used by EPA in
the EtO assessment was revised in order to more clearly respond to this public
comment. Criteria used to evaluate data quality are now discussed in much more
detail than in the previous document.
EPA RESPONSE: No response required.
COMMENT: Comment 3: The comment states that lymphohematopoietic and
lymphoid cancers should not be grouped because they are derived from different
cells of origin. The response clearly states the rationale for grouping these
together and notes that the SAB (2007) report agreed with the logic of that
grouping for comparison purposes. This response is clear and appropriate.
EPA RESPONSE: No response required.
1-48
-------
vi. COMMENT: Comment 4: The comment states that the evidence for breast
cancer is too weak. The response notes that the document acknowledges that the
breast cancer database is more limited than that for other cancers. Further, the
response notes that the SAB (2007) report accepted the derivation of a unit risk
factor based on that database. This response is clear and appropriate.
Additionally, the EPA could also discuss the animal model data (Parsons et al..
2013; NTP. 1987) and Swedish sterilization workers study data (Mikoczy et al..
2011) to provide further support for breast cancer as a potential hazard from EtO
exposure.
EPA RESPON SE: As suggested by the SAB, the EPA has expanded the
response to include additional supporting information.
vii. COMMENT: Comment 5: The comment notes that EtO is a weak mutagen.
Both the response and the draft assessment never claim that EtO is a strong
mutagen. The "weakness" of EtO as a mutagen as compared to many anticancer
compounds and other reactive epoxides is clearly stated. In their response, the
EPA provides further justification by noting that there is seldom a good
correlation between mutagenic and carcinogenic potencies. This response is clear
and appropriate.
EPA RESPONSE: No response required.
viii. COMMENT: Comment 6: The comment states that a mutagenic MOA is not
supported by the most recent scientific evidence; other MO As, specifically
oxidative stress and cell proliferation, should be considered. There are two major
issues here with regard to the MOA. First, the database concerning the MOA is
rather complex, which the draft assessment and the EPA response acknowledge.
Second, and most significantly, the Parsons etal. (2013) study cited in the
comment is considered to be flawed and does not adequately argue that other
MO As besides direct mutagenesis are involved. The response clearly states that
there is no support for the conclusions in Parsons etal. (2013). In the response,
the EPA cites another recent study (Nagy et al„ 2013) that does not support
oxidative stress. The response also provides a detailed discussion of the problems
of inferring too much from K-ras mutation data. Even fewer data exist to support
1-49
-------
a proliferative MO A. The EPA response methodically presents the reasoning
behind this conclusion.
EPA RESPONSE: No response required. However, the EPA notes that
more recent studies (Zhang etaL 2015b; Zhang et al.. 2015a) similarly do not
support the oxidative stress hypothesis; these studies have been added to
Appendix J (see Section J.4.1).
COMMENT: Comment 7: The comment criticizes the EPA for failing to
incorporate the Union Carbide Corporation (UCC) data into the dose-response
assessment. It goes on to state that the NIOSH exposure assessment also suffered
from limitations. The EPA response is concise and clear. This issue is discussed
in detail in the draft assessment and was supported by the SAB (2007) report.
The NIOSH study meets the criteria of being a high-quality study much more
strongly than the UCC data. This response is well-supported and appropriate.
The SAB concurs with the EPA decision to not combine UCC EtO exposure data
with those from the NIOSH study.
EPA RESPONSE: No response required.
COMMENT: Comment 8: This comment criticizes the EPA for using summary
data rather than the individual data in the modeling of breast cancer mortality and
lymphoid cancer despite the SAB (2007) recommendations. Two key points are
made in the response. First, the rationale for the modeling procedures used and
their consistency with the previous recommendations in the SAB (2007) report
are noted. Second, the response notes that the current document adds additional
models based on continuous exposure data and has added them to the assessment
for comparison purposes. This response is appropriate. However, the SAB
suggests that the model should only apply to low-dose exposures and that a range
of doses should be specified over which the model applies.
EPA RESPONSE: The assessment has been revised so that models based on
the continuous exposure data are used for both cancer endpoints. Both
selected models are linear two-piece spline models, and the assessment notes
that the linear extrapolations from these models are valid for exposures below
the knots.
1-50
-------
COMMENT: Comment 9: A comment from two sources criticized the EPA for
using a non-peer-reviewed supralinear spline model. The response notes that the
model was published in 2011. Further, the response notes that use of the model
will receive additional review by the SAB. This response is clear and appropriate.
EPA RESPONSE: No response required.
COMMENT: Comment 10: A comment was made regarding other concerns
about the modeling procedures used and how they lead to over-prediction of
cancer deaths in the NIOSH study. In response to concerns raised by the two
publications cited in the comment, the EPA provided additional discussion in
Appendix J to specifically address concerns raised with respect to the Valdez-
Flores and Sielken (2013) study. The response further suggested that the
referenced citations did not provide convincing evidence of flaws in the modeling.
Further, the EPA notes that the potential degree of over-prediction is far less than
that claimed in the comment and the two papers. This response is appropriate.
EPA RESPONSE: No response required.
COMMENT: Comment 11: A comment was made from three sources that the
EPA should present both linear and nonlinear extrapolation approaches. This
subject is discussed at great length in the draft assessment and in Appendix H.
The response further notes that the SAB (2007) report agreed that there was
presently insufficient evidence to support use of a nonlinear extrapolation
approach. This response is appropriate.
EPA RESPONSE: No response required.
COMMENT: Comment 12: A comment was made from two sources that
combining breast cancer and lymphoid cancer unit risk estimates is not justified,
and that the EPA did not discuss competing risks, different background
populations, incidence vs. mortality, and the use of different exposure-response
models. In their response, the EPA first notes that breast cancer and lymphoid
cancers were first modeled separately and then combined. The rationale for
combining these unit risk estimates is explained in detail in the draft assessment.
1-51
-------
Further, the subject of competing and background risks is also discussed in detail
in the draft assessment. Finally, the response concludes by noting the distinction
between cancer incidence and cancer status. Standard practice in IRIS
assessments is to estimate total cancer risk and not just the risk from individual
cancer types; this practice is consistent with EPA guidelines and NRC
recommendations. This response is appropriate.
EPA RESPONSE: No response required.
xv. COMMENT: Comment 13: A comment was made from three sources that the
EPA should reexamine its risk determination given background and endogenous
levels of EtO and that the EPA's risk estimates are unrealistically high. The EPA
response explains how background rates for the cancers of interest have been
taken into account in the risk determination. They also note that in one of the
comments an unrealistic exposure concentration was used in arguing their point.
This response is appropriate.
EPA RESPONSE: No response required.
xvi. COMMENT: Comment 14: Two sources commented that the EPA should not
be deriving occupational exposure limits for EtO. The EPA response makes two
clarifications. First, the EPA's Office of Pesticide Programs (OPP) is indeed
responsible for deriving occupational exposure limits. Second, and more
importantly, the response notes that such a derivation was not conducted in the
present risk determination. Rather, the response notes that with the models used
for the EtO cancer data, the unit risk estimate is not appropriate for the full range
of occupational exposure scenarios of interest to OPP. For the purposes of OPP,
the assessment provides sample risk estimates for exposure scenarios of interest to
OPP for its own risk assessment of sterilization uses of EtO.
EPA RESPON SE: The EPA has clarified in its response that the Agency
does not set "occupational exposure limits" for EtO but has the authority to
consider occupational risks in labeling and regulation decisions.
1-52
-------
xvii.
COMMENT: Overall Analysis of EPA Response to Public Comments in
Appendix K (thenL): The responses provided by the EPA are focused, generally
complete, and appear to be delivered in good faith.
EPA RESPON SE: The EPA confirms that the responses were delivered in
good faith.
xviii. COMMENT: In addition to evaluating the EPA response (Charge Question 7) to
public comments received on the July 2013 draft assessment, the EPA also
presented their responses to public comments received on the 2006 draft
assessment (U.S. EPA 2006a) in Appendix H. Some of the comments were
addressed by changes made in the current assessment. For example, one criticism
was that the 2006 draft assessment (U.S. EPA, 2006a) had an improper reliance
on data from only one sex. The current draft assessment uses data from both
sexes. Another example was the EPA response to Comment 7 regarding the
modeling procedures. Although the EPA response to the comment on the 2006
draft assessment (U.S. EPA 2006a) was very brief and lacked sufficient detail,
these issues are extensively addressed in the current draft assessment and the
accompanying appendices. Several other comments were redundant with public
comments made on the 2013 draft assessment. Examples include comments on
EtO mutagenicity, lack of use of the UCC database, and the use of summary data
versus individual data. In summary, the previous EPA responses in Appendix H
as well as the changes that were instituted in the current draft assessment
adequately and appropriately respond to the public comments on the 2006 draft
assessment (U.S. EPA, 2006a).
EPA RESPONSE: No response required.
1-53
-------
APPENDIX J. SUMMARY OF MAJOR NEW STUDIES SINCE THE
LITERATURE CUTOFF DATE
The cutoff date for literature inclusion into the main body of this assessment was June 30,
2010. At that time, the analyses and text were largely completed, with the exception of a few
focused issues which remained for discussion and review. An updated literature search was done
in 2013, involving a systematic literature search for the time frame from January 2006 to May
2013 to ensure that no major studies were missed from the time of the first external review draft
in 2006 until the cutoff date and to determine if any significant new studies had been published
since the cutoff date that might alter the findings of the assessment. No studies were identified
that would impact the assessment's major conclusions. Nonetheless, two new studies of high
pertinence to the assessment had been published since the cutoff date, and these studies are
reviewed briefly in this Appendix for transparency and completeness. Two additional highly
pertinent studies published after the May 2013 literature search were identified from public
comments received in October 2013 on the July 2013 public review draft of the EtO
carcinogenicity assessment. These additional new studies similarly would not affect the
assessment's major conclusions but are reviewed briefly here for transparency and completeness
and to be responsive to the public comments. A final updated literature search, using the same
approach as for the 2013 search, was conducted for the time period from May 2013 through
August 2016. Once more, no studies were identified that would impact the assessment's major
conclusions; however, two new studies of high pertinence to the assessment were published in
that time frame, and these studies are also reviewed briefly in this appendix.
The Appendix first provides a description of the systematic literature search that was
conducted to identify relevant new studies (see Section J.l) and then provides the reviews of the
two major new studies identified in the May 2013 literature search (see Section J.2), the two
additional major studies identified from the 2013 public comment period (see Section J.3), and
the two major new studies identified in the 2016 literature search (see Section J.4). Sections J.2
and J.3 were part of the external review draft (U.S. EPA, 2014a, b) that was reviewed by the
SAB in late 2014 (SAB, 2015); Section J.4 discusses studies published after completion of that
review draft.
J.l. SYSTEMATIC LITERATURE SEARCH
Systematic literature searches were conducted in May 2013, covering the time frame
from January 2006 to May 2013, and September 2016, for the time period of May 2013 through
August 2016. The searches were conducted using the LitSearch tool in the EPA's HERO
database, and the following three literature databases were searched: PubMed, Web of Science,
J-l
-------
and ToxNet. The search terms involved Ethylene Oxide AND (carcinogenicity OR cancer OR
mutagenicity OR mutation OR genotoxicity).
The May 2013 search identified 372 references, of which 56 were determined to be
potentially relevant.16 The disposition of the 56 potentially relevant references is summarized in
Table J-1. In brief, for the purposes of this carcinogenicity assessment, 26 references that were
primarily discussions of methods studies or exposure studies17 or were reviews or other
secondary source material were not generally considered further. The remaining 30 references
were given further consideration to see if they represented major new studies. No new studies
were identified that 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 in
2013
Category
References
Disposition
Exposure studies
Davis et al. (2006)
Lin et al. (2007)
Tateo andBononi (2006)
Not considered further.
Methods studies
Ahn and Shin (2006)
Tretvakova et al. (2012)
Wu et al. (2011)
Not considered further.
Reviews or other
secondary source
material
Brown and Rushton (2012)
Butterworth and Chapman (2007)
Chan et al. (2006)
Farmer and Singh (2008)
Grosse et al. (2007)
Hoenerhoff et al. (2009)
Jarabek et al. (2009)
Keshavaetal. (2006a)
Keshavaetal. (2006b)
Manserviei et al. (2010)
McCarthy et al. (2009)
Mosavi-Jarrahi et al. (2009)
Okada et al. (2012)
Smith-Bindman (2012)]
Not considered further.
16In 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. (20071 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 notrelevant to the assessment and was excluded from further
consideration.
17This refers to general exposure studies; exposure studies related to any of the epidemiological studies of EtO
would be considered further.
J-2
-------
Table J-l. Disposition of 56 new references identified as potentially relevant
(continued)
Category
References
Disposition
Reviews or other
secondary source
material
(continued)
Snedeker (2006)
Steinhausenet al. (2012)
Weiderpass et al. (2011)
Won (2010)
WHO. 2008 [same asIARC (2008)]
Not considered further.
IARC (2008)
Already cited in the assessment.
Cancer studies
Kiran et al. (2010)
Mikoczv et al. (2011)
Reviewed in Section J.2.
Swaen et al. (2009)
Already cited in the assessment.
van Balen et al. (2006)
Not considered further. Primarily a study of risks to
farmers. EtO left outof analysis because too few study
subjects were exposed to it. Subjects were part of the
EPILYMPH studv analyzed bv Kiran et al. (2010) (see
Section J.2.1).
Fondelli et al. (2007)
Not considered further. No EtO-specific results.
Kim et al. (2011)
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 processes ofcases.
Genotoxicity/
mutagenicity studies
Donneretal. (2010)
Godderis et al. (2006)
Hone et al. (2007)
Houle et al. (2006)
Marsdenet al. (2007)
Marsdenet al. (2009)
Tompkins et al. (2008)
Yone et al. (2007)
Already cited in the assessment.
Mazon et al. (2009)
Tomoa et al. (2006)
Tompkins et al. (2009)
Citations added to the assessment.
Huane et al. (2011)
Not considered a major new study. Largely an
exposure study; examined us e of urinary N7-HEG as a
biomarker of EtO exposure in EtO-exposed workers
and smokers in Taiwan.
Lindbera et al. (2010)
Not considered further. This study examined use of a
micronucleus assay for detecting genotoxic damage in
mouse alveolar Type II and Clara cells—HO 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).
Mazon et al. (2010)
Not considered further. Focused on a specific repair
gene product in E. coli.
Parsons et al. (2012)
Tompkins et al. (2006)
Not considered further. Published abstracts, not full
papers.
J-3
-------
Table J-l. Disposition of 56 new references identified as potentially relevant
(continued)
Category
References
Disposition
Other
Sielken and Valdez-Flores (2009a)
Sielken and Valdez-Flores (2009b)
Already cited in the assessment.
Swenbere et al. (2008)
Valdez-Flores et al. (2010)
Haufroid et al. (2007)
Citation added.
Kensler et al. (2012)
Not relevant; focused on chemoprevention.
Steenland et al. (2011)
Not considered further. Peer-reviewed publication of
analyses already in the assessment.
Valdez-Flores etal. (2011)
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 Darterbv the same authors (Valdez-Flores et
al.. 20101
Swenbere et al. (2011)
Not considered further. Largely a review; focused on
implications of endogenous adducts forrisk
assessment—this issue is already addressed in the
assessment (e.g., at the end of Section 4.5 and in the
responses to SAB comments in Appendix H).
EPILYMPH = population-based case-control study of lymphoma in six European countries.
The September 2016 search identified 180 references, of which 17 were determined to be
potentially relevant. The disposition of the 17 potentially relevant references is summarized in
Table J-2. Eight references that were primarily discussions of methods studies or exposure
studies or were reviews or other secondary source material were not considered further. The
remaining 9 references were given further consideration to see if they represented major new
studies. No new studies were identified that would impact the assessment's major conclusions.
Two references were considered highly pertinent studies, and these are reviewed briefly in
Section J.4 of this Appendix.
J-4
-------
Table J-2. Disposition of 17 new references identified as potentially relevant
in 2016
Category
References
Disposition
Exposure studies
Gabriel et al. (2013)
Jacob et al. (2013)
Kloth et al. (20141
St Helen et al. (20141
Not considered further.
Methods studies
Brehenv etal. (20141
Not considered further.
Reviews or other
secondary source
material
Bukowska (20151
Eastmond et al. (20141
Konduracka et al. (20141
Not considered further.
Cancer studies
Yuan et al. (20141
Not considered further. Case-control study of lung
cancer and urinary metabolites of a variety of pollutants
- the urinary metabolite of EtO was not associated with
increased risk of lung cancer.
Genotoxicity/
mutagenicity studies
Naev et al. (20131
Parsons et al. (20131
Already cited in the assessment (see Section J.3).
Zhaneetal. (2015b 1
Reviewed in Section J.4.
Philippin et al. (20141
Not considered further. Focused on the capacity for
Af7-alkylguanine adducts to induce mutagenicity in
E. coli.
Zhaneetal. (20161
Not considered further. Study of in silico modeling
using Pearson's hard and soft acids and bases theory to
estimate the activation energies and other chemical
characteristics of 36 epoxides and correlate these
calculated activation energies against previously
published mutagenicity results in S. typhimurium strain
TA100.
Other
Valdez-Flores andSielken (20131
Already cited in the assessment (see Section J.3).
Zhaneetal. (2015al
Reviewed in Section J.4.
Filser et al. (20131
Not considered further. Study of EtO levels in blood
from ethylene exposure.
J.2. REVIEWS OF MAJOR NEW STUDIES IDENTIFIED IN THE 2013 LITERATURE
SEARCH
As discussed in Section J.l, a systematic literature search was conducted in 2013 to
determine whether any significant new or missed studies had been published since January 2006.
No new studies were identified that would impact the assessment's major conclusions.
Nonetheless, two studies of high pertinence to the assessment had been published since the June
2010 cutoff date for literature inclusion. The two studies are epidemiology studies of
occupational exposure toEtO. These studies are reviewed briefly here for transparency and
completeness, and key features of the studies are summarized in Table J-3.
J-5
-------
Table J-3. New 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
Population-based
case-control study
involving 22 centers
in 6 European
countries (Czech
Republic, France,
Germany, Italy,
Ireland, Spain)
[EPILYMPH study]
Kiran et al. (201 CD
2,347 cases
(1,314 male,
1,033 female);
2,463 controls
(1,321 male,
1,142 female),
matched on
sex, age group,
and residence
area
1.2% of study population
defined as ever-exposed
(31 cases, 27 controls)
All lymphoma
(no. cases/no. controls) OR (95% CI)
Unexposed (2,316/2,436) 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)
/?-trend = 0.246
Would vary by
individual participant
because it is not an
industry-based study;
however, inclusion of
farm work and
occupational exposure
to solvents in the
regression model did
not affect the risk
estimates
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
EPILYMPH = population-based case-control study of lymphoma in six European countries.
-------
Table J-3. New 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
Two plants that
produced disposable
medical equipment,
Sweden
Mikoczv et al.
2,171
(862 men,
1,309 women)
Exposure levels were up
to 75 ppm in 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.
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) [casesl SIR (95% CI)
Fluorochlorocarbons,
methyl formate
(1:1 mixture with EtO)
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 cancer 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.
(2011)
Same cohort as
(Haemar et al..
1995; Haemar et al..
19911 followed an
additional 16 years
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
DDm x vears [casesl IIR (95% CI)
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)
-------
Table J-3. New 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
(continuedfrom 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
DDm x vr [casesl IIR (95% CD
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)
-------
J.2.1. Kiran etal. (2010)
Kiran et al. (2010) investigated occupational exposure to EtO in a population-based
case-control study of lymphoma in six European countries (the "EPILYMPH study"). Cases
(n = 2,347) were consecutive adult patients with a first diagnosis of lymphoma, classified under
the 2001 World Health Organization lymphoma classification system, in 1998-2004 at 22
centers in the six countries. Controls from Germany and Italy were randomly selected from the
general population, matched on sex, 5-year age group, and residence area. Controls from the
Czech Republic, France, Ireland, and Spain were matched hospital controls with diagnoses other
than cancer, infectious diseases, and immunodeficient diseases (total controls = 2,463).
Participation rates were 88% in cases, 81% in hospital controls, and 52% in population controls.
All study subjects were interviewed in person using the same structured questionnaire, which
included questions on sociodemographic factors, lifestyle, health history, and complete work
history (including all lull-time jobs held for >1 year). For each job, information was collected on
type of industry, tasks performed, machines used, and exposure to 35 specific agents (or groups
of agents) of interest, including EtO. Supplemental questionnaire modules for specific
occupations were used to get additional details on job sand exposures of interest.
Exposure was evaluated in each center by specially trained industrial hygienists who
reviewed all the questionnaires and assessed frequency and intensity of exposure to each agent
on a 4-point scale (unexposed and low, medium, and high exposures) as well as confidence in the
assessment (low, medium, or high). Most of the exposed cases and controls were classified with
medium or high confidence, although a greater proportion of cases than controls were thus
classified (23/31 vs. 15/27). Because of the low prevalence (1.2%) of EtO exposure in the study,
the medium and high categories of exposure frequency and intensity were combined in the
statistical analyses. A cumulative exposure score for EtO was also developed for each study
subject, integrating duration, frequency, and intensity of exposure; these scores were then
categorized as above or below the median score among exposed subjects.
Risk was assessed for all lymphoma, B-cell lymphoma (which represented 80% of all the
lymphoma cases), and the most common subtypes of B-cell lymphoma. The OR was calculated
using unconditional logistic regression, adjusting for age, sex, and center. Including education,
farm work, and exposure to solvents in the model, reportedly did not change the risk estimates
(results not shown). Linear trends for the exposure metrics were calculated using the Wald test
for trend.
Because of the low prevalence of EtO exposure in the study (1.2%), the number of
exposed cases and controls was limited (31 and 27, respectively), especially for analyses of
lymphoma subtypes. Results for all lymphoma for ever exposed and for the highest exposure
J-9
-------
category for each of the different exposure metrics are presented in Table J-3. Increased risks
were observed for ever exposed and for the highest exposure category for each of the exposure
metrics, and the OR for medium or high frequency of exposure was statistically significant (4.3;
95% CI 1.4, 13.0). However, none of the trend tests was statistically significant. The overall
association appeared to be stronger using hospital controls; however, when considering only
subjects whose EtO exposures were assessed with medium or high confidence, the increased
ORs were similar using either hospital or population controls. Results were similar when only
B-cell lymphoma, which represented the majority of all lymphomas, was evaluated. The
strongest associations were generally observed for chronic lymphocytic leukemia, and ^-values
for trend were <0.051 for all the exposure metrics for that lymphoma subtype. The investigators
note that while random variation related to the low prevalence might account for some positive
results, their combined probability test (Fisher method) indicated that the chance probability of
an upward trend in chronic lymphocytic leukemia across the four metrics assumed to be
independent (confidence, frequency, intensity, and duration) was 0.003.
In conclusion, this study adds further support to the weight-of-evidence finding obtained
in Chapter 3 of strong, but less than conclusive, evidence of a causal association between EtO
exposure and lymphohematopoietic cancers in humans. Because only categorical exposures
were assessed, no quantitative risk estimates can be derived from this study.
J.2.2. Mikoczy etal. (2011)
This study is an update of the Hagmar et al. (1991) and Hagmar etal. (1995) studies
discussed in Section 3.1 of the assessment and in Section A.2.11 of Appendix A. The first
update (Hagmar etal.. 1995) had a median follow-up time of only 11.8 years; this update extends
the follow-up period through 2006, providing an additional 16 years of follow-up. The cohort
consists of 2,171 (1,309 females and 862 males18), employed for at least 1 year prior to 1986, at
two Swedish facilities that sterilized medical equipment using EtO (Plant A sterilization
operations ran from 1970 to 1994; Plant B sterilization operations ran from 1964 to 2002). Vital
status and emigration data at the end of follow-up were obtained from the Swedish population
registry, cause of death for 1972-2006 was obtained from Statistics Sweden, and malignant
tumor data for 1972-2006 were obtained from the Swedish Cancer Registry. At the end of
follow-up, the mean age of the cohort was 56 years and the cohort had contributed
58,305 person-years of risk; 171 cohort members had died (7.9%) and 126 (5.8%) had emigrated
and were of unknown vital status. Mean duration of employment in the cohort was 6.3 years.
18Without 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: Hagmar et al.. 1991Y
J-10
-------
In the original study (Hagmar et al.. 1991). individual cumulative exposure estimates
were derived based on job-exposure matrices for each plant and exposure level estimates
determined up to 1986. While exposure levels were high in the early years of the operations
(e.g., peak levels of 75 ppm in 1964 in Plant B and exposure levels up to 40 ppm in 1970 in
Plant A), 8-hour TWA levels had decreased to below 1 ppm by 1985 [see Hagmar etal. (1991)
and Section A.2.11 of Appendix A for more details on the original exposure assessment]. For
this update, worker histories for the 1,303 workers who were still employed at the two plants at
the end of the original study (1986) were extended up until the cessation of sterilization
operations in the plants, and exposure estimates for the follow-up period were determined from
yearly statutory industrial hygiene measurements of EtO from 1986 on. Because of the low
exposure levels after 1985, the impact of updating the cumulative exposure estimates was low
(the largest impact was reportedly on the 90th percentile, which changed from 1.17 to
1.29 ppm x years). The mean and median cumulative exposures for the 2,020 cohort members
for whom job titles were available were 2.92 ppm x years and 0.13 ppm x years, respectively.
Standardized mortality and incidence ratios (SMRs and SIRs) were obtained by
comparing the number of deaths or incident cases observed to the number expected based on
5-year age group-, cause-, calendar year-, and sex-specific rates in the county (external
referents). For cancer incidence (but not mortality), internal analyses were also conducted using
Poisson regression analyses, adjusted for age group, sex, and calendar period, with no induction
(latency) period. In the internal analyses, incidence rate ratios (IRRs) were calculated by
comparing the incidence rates for the two highest cumulative exposure quartiles with that for the
50% of workers with cumulative exposures below the median of 0.13 ppm x years (internal
referents). [Internal analyses are generally preferred by the EPA over external analyses because
the referents are from the same cohort as the exposed subjects, potentially reducing confounding
as well as the healthy worker effect, which can mask an increase in risk; however, in this study,
some of the advantages of internal analyses may be mitigated by the absence of an unexposed
referent group, which could itself dampen relative risk estimates.]
Results for cancer mortality and incidence for the cancer types of interest (i.e.,
lymphohematopoietic cancers and female breast cancer) are summarized in Table J-3. For
lymphohematopoietic cancers, nonsignificant increases in SMRs and SIRs were reported. For
the incidence data, the internal analysis shows no exposure-related association for
lymphohematopoietic cancers, although the EPA found this analysis to be relatively
uninformative for these cancers, given the small number of cases (five cases in each of the two
highest exposure quartiles and seven cases in the referent group of workers with cumulative
exposures below the median), the generally low estimated cumulative exposures, and the absence
of an unexposed referent group. The EPA also noted that data were not reported or analyzed for
J-ll
-------
the subgrouping of "lymphoid" cancers. In a crude comparison, ignoring the exposures in the
referent group, the lack of a lag period, and the fact that the results are for all
lymphohematopoietic cancers rather than lymphoid cancers, the EPA compared the Mikoczy et
al. (2011) results with RR estimates obtained from the selected model for lymphoid cancer based
on the NIOSHdata (two-piece linear spline model with knot at 1,600 ppm x days; with a 15-year
lag; see Section 4.1.1.2 and Figure 4-7) using the mean cumulative exposures for the highest two
quartiles in the Mikoczy et al. (2011) study. The results obtained from the selected model for the
NIOSH data were within the 95% confidence interval for the RR estimates reported by Mikoczy
et al. (2011) (see Table J-4).
Table J-4. Comparison of Mikoczy etal. (2011) RR estimates with those
obtained using the selected models based on the NIOSH study
Exposure
group
(ppm-years)
Mean
cumulative
exposure
(ppm-vr f
Mean
cumulative
exposure
(ppm-days)
Reported RR estimate
(n; 95% CI)
RR estimate from model
based on NIOSH datab
Lympho he mato -
poietic cancer
Breast cancer
Lymphoi d
cancer
Breast
cancer
0-0.13
0.0745
1.00 (7)
1.00 (10)
0.14 -0.21
0.1737
63.40
1.17
(5; 0.36-3.78)
2.76
(14; 1.20-6.33)
1.05
1.01
>0.22
14.1846
5,177
0.92
(5; 0.28-3.05)
3.55
(17; 1.58-7.93)
2.25
1.46
Personal communication from Zoli Mikoczy to Jennifer Jinot, U. S. EPA, 15 September 2015.
bIgnoring the 15-year lag in model and the nonzero exposure in the referent group.
For breast cancer mortality (results not shown), a "slight but nonsignificant decrease" in
the SMRwas reported. With a 15-year induction period included, the SMRfor breast cancer
was reportedly "somewhat increased." For workers with cumulative exposures above the
median, with a 15-year induction period, a "higher than expected" SMR, which was not
statistically significant, was reported.
For breast cancer incidence (41 incident cases), SIRs were nonsignificantly decreased,
both with and without a 15-year induction period. Internal analyses resulted in statistically
significant increases in the IRRs for the two highest cumulative exposure quartiles as compared
to the 50% of workers with cumulative exposures below the median (see Table J-3), despite
having a low-exposed rather than an unexposed referent group.
The EPA noted that the cumulative exposure estimates for this study were very low
compared to those in other studies. For example, in the Swaen et al. (2009) study of the UCC
J-12
-------
cohort of male EtO production workers, the estimated average cumulative exposure was
67.16 ppm x years. In the more comparable NIOSH cohort of sterilization workers, cumulative
exposure estimates at the end of follow-up for the full cohort, which included workers with <1
year of employment, had a mean of 27 ppm x years and median of 6 ppm x years (see
Appendix D, Section D.l), and in particular, the mean cumulative exposure at the end of
follow-up in the breast cancer incidence study cohort, which only included workers with >1 year
of employment, was 37.0 ppm x years. Yet, the breast cancer incidence RRs for the categorical
exposure groups reported in Steenland et al. (2003) for the NIOSH breast cancer incidence study
were lower than those observed in the Mikoczy etal. (2011) study.
Thus, if unit risk estimates for breast cancer incidence were derived based on the
Mikoczy etal. (2011) study, they would be higher than the estimates calculated from the NIOSH
study. The EPA did not derive such estimates, however, because the reported grouped results
from the Mikoczy etal. (2011) study are not well suited for derivation of a unit risk estimate, the
EPA does not have the individual data to model, and the NIOSH study is preferred as the basis
for the unit risk estimate in any event (see Section 4.1). Instead, as a crude comparison, ignoring
the exposures in the referent group and the lack of a lag period, the EPA compared the Mikoczy
et al. (2011) results with RR estimates obtained from the selected model for breast cancer
incidence based on the NIOSH data (two-piece linear spline model with knot at
5,750 ppm x days; with a 15-year lag; see Section 4.1.1.2 and Figure 4-9) using the mean
cumulative exposures for the highest two quartiles in the Mikoczy etal. (2011) study. The
results obtained from the selected model for the NIOSH data were below the lower bound of the
95% confidence interval for the RR estimates reported by Mikoczy etal. (2011) (see Table J-4);
i.e., the selected model used to derive the unit risk estimate for breast cancer incidence in this
assessment underestimates the IRRs observed in the Mikoczy etal. (2011) study.
The EPA could not determine the reasons for the discrepancy between the observed IRRs
and the predictions from the model based on the NIOSH data. As noted above, the cumulative
exposure estimates for the Mikoczy etal. (2011) study are lower than those for the NIOSH study.
At the high end, two of the NIOSH plants had jobs with historical exposure levels as high as
those estimated for the Mikoczy etal. (2011) study (Haemar etal., 1991), but most of the
NIOSH plants had lower estimated exposure levels (see Table J-5). However, exposure
durations are shorter in the Mikoczy et al. (2011) study and more person-years would have
accrued in more recent time periods, when exposure levels in the Swedish plants were lower. A
less rigorous approach was used to estimate historical exposure levels for the plants in the
Mikoczy etal. (2011) study than the regression model that was developed for the NIOSH study.
Measurement data were available from 1973 for one plant ("A") and 1975 for the other ("B");
for earlier exposures, estimates were constructed taking into account information on changes in
J-13
-------
production methods and environmental controls, subjective memories, and time trends (Hagmar
et al.. 1991). However, Plant A started operations in 1970 and Plant B in 1964, so the historical
reconstructions did not have to go very far back in time and are thus probably subject to less
uncertainty than most such retrospective reconstructions. Another major difference between the
two studies is that there were many fewer breast cancer cases in the Mikoczy etal. (2011) study
(41 incident cases [33 with >15 years since time of first exposure] vs. 233 cases [at least 170
with >15 years since time of first exposure] in NIOSH's subcohort with interviews).
Additionally, there was no information on potential breast cancer risk factors in the Mikoczy et
al. (2011) study, as was available for the NIOSH subcohort, although accounting for these
factors made little difference in the unit risk estimate derivation from the NIOSH data (see
Section D.1.8 of Appendix D).
Table J-5. Comparison of highest exposure levels estimated for the
NIOSH cohort plants with those in the Mikoczy etal. (2011) study
plants"
Plant
Highest exposure level (ppm)
~ Years
Mkoczv et al. <20111 cohort plant
A
40
1970-1972
B
75
1964-1966
NIOSH cohort plantb
1
14
1969-1975
2
19
1976-1977
4
4
1971-1978
5
77
1977-1978
6
77
1977-1978
7
17
1969-1978
8
25
1967-1978
9
3
1969-1979
10
24
1974-1978
11
20
1970-1978
12
17
1972-1978
13
25
1970-1977
14
5
1976-1979
a8-hour TWAs for jobs/operations with the highest exposure levels per plant from NIOSH exposure data
and Hagmar et al. (19911 compared for the earliest time periods of the two Mikoczv etal. (2011) study
plants.
bPlant 3 did not have exposure data.
J-14
-------
In conclusion, the EPA finds that the nonsignificant increases in SMRs and SIRs for
lymphohematopoietic cancers reported in this study are consistent with an increase in
lymphohematopoietic cancer risk but, overall, the study is underpowered for the analysis of
lymphohematopoietic cancers and contributes little to the weight of evidence for these cancers.
For breast cancer incidence, however, the statistically significant exposure-related increases in
internal analyses add support to the weight-of-evidence finding obtained in Chapter 3 of strong,
but less than conclusive, evidence of a causal association between EtO exposure and female
breast cancer in humans. Although the Mikoczy et al. (2011) results are consistent with a higher
unit risk estimate for breast cancer incidence than that obtained from the NIOSH study results,
the Mikoczy et al. (2011) results support the general supralinear exposure-response relationship
(i.e., steeper rise at lower exposure levels and then aplateauing of response at higher exposure
levels) observed in the NIOSH study.
J-15
-------
J.3. REVIEWS OF MAJOR STUDIES IDENTIFIED BETWEEN THE 2013
LITERATURE SEARCH AND THE 2014 SAB REVIEW DRAFT
Two additional major studies were identified from public comments on the July 2013
public review draft of the EtO carcinogenicity assessment, and a third study related to one of
those studies was also discovered after the May 2013 literature search. These three studies are
reviewed briefly here. These new studies would not affect the assessment's major conclusions
but are reviewed here for transparency and completeness and to be responsive to the public
comments.
J.3.1. Valdez-Flores and Sielken (2013)
Valdez-Flores and Sielken (2013) criticized the approach employed by the EPA in earlier
drafts of the EtO carcinogenicity assessment of using a weighted linear regression of the RR
estimates based on categorical exposure groups to derive exposure-response relationships for
lymphoid cancer mortality and breast cancer mortality, stating that exposure-response modeling
is best based on individual data. While the EPA does not agree with aspects of the Valdez-Flores
and Sielken (2013) paper [see Section J.3.1 of Appendix J in (U.S. EPA 2014a)l, the EPA is no
longer using the weighted linear regression of the categorical results as a selected model, and
thus, the issues raised by Valdez-Flores and Sielken (2013) are not relevant to the current
assessment.
J.3.2. Parsons et al. (2013) [and Nagy etal. (2013)1
As part of a larger study to examine potential key events in EtO-induced mouse lung
carcinogenesis, Parsons et al. (2013) exposed Big Blue B6C3Fi mice to various concentrations
of EtO by inhalation for 4, 8, or 12 weeks (0, 10, 50, 100, or 200 ppm for 4 weeks or 0, 100, or
200 ppm for 8 or 12 weeks) and analyzed the levels of three specific K-ra.s codon 12 mutations
(GGT^GAT, GGT^GTT, and GGT^TGT) in lung DNA samples using ACB-PCR
(allele-specific competitive blocker PCR). Parsons et al. (2013) presented the first results to be
published from this larger study. K-ra.s mutations were investigated because K-ra.s mutations,
and more specifically codon 12 mutations, were identified in all of the lung tumors evaluated
from EtO-exposed mice in the NTP cancer bioassay (Hong etal.. 2007). Of the codon 12
mutations in the 23 mouse lung cancers evaluated, 21 were GGT—>GTT mutations. Parsons et
al. (2013) suggest that because 8-oxo-dGadducts19 preferentially cause G:C—>TA mutations, an
early increase of the GGT—>GTT (and/or GGT—>TGT) mutation relative to the GGT—>GAT
19Same as 8-hydroxy-2'-deo?yguanosine (8-OHdG) adducts.
J-16
-------
mutation would support the hypothesis that EtO causes oxidative stress in the mouse lung,
resulting in the formation of 8-oxo-dG adducts.
Because many of the K-ras mutant fraction (MF) measurements were below the limit of
accurate ACB-PCR quantification (10 5), differences among treatment groups were assessed by
analyzing the numbers ofMFs greater than and less than 10 5 using a Fisher's exact test. Parsons
et al. (2013) reported that for the GTT mutation at 4 weeks of EtO exposure, a significant
increase in MF compared to concurrent controls occurred only in the 100-ppm group.
"Surprisingly," as Parsons etal. (2013) noted, the MFs at 8 weeks in both the 100- and 200-ppm
groups were statistically significantly decreased relative to concurrent controls, and at 12 weeks,
the 200-ppm group had statistically significant decreases. A similar pattern was observed for the
GAT mutation, with statistically significant increases in the 50-, 100-, and 200-ppm groups at 4
weeks and statistically significant decreases in the 100- and 200-ppm groups at 8 weeks
compared to concurrent controls. The EPA noted that MFs were decreased in the 100- and
200-ppm groups at 12 weeks as well, but the results were almost all greater than 10 5, and thus,
the trend would not be apparent using the Fisher's exact test. For the TGT mutation, all of the
measurements were less than 10 5, and the investigators performed no further analyses. Parsons
et al. (2013) also reported a "surprising amount of variability" in the GTT and GAT MF results
among the 4-, 8-, and 12-week control groups, with the 8-week control GTT results and the 8-
and 12-week control GAT results being statistically significantly increased compared to their
respective 4-week control results.
Instead of observing an early preferential increase in GTT mutations, as anticipated,
Parsons etal. (2013) reported an early induction of both GAT and GTT mutations, with a greater
induction of the GAT mutation, which they note is the main K-ra.s codon 12 mutation observed
by Hong et al. (2007) in "spontaneous" mouse lung tumors (11 of 17 K-ra.s codon 12 mutations
in 108 lung tumors from control B6C3Fi mice in the NTP 2-year cancer studies were GAT
mutations). To explain these findings and the irregular pattern of results in which GTT and GAT
MFs "did not accumulate straightforwardly with cumulative [EtO] dose or duration of treatment"
and because "no induction of cytotoxicity or apoptosis was detectable" in another part of the
larger study [results not presented by Parsons etal. (2013)1, Parsons etal. (2013) proposed the
following biphasic response. Parsons et al. (2013) hypothesized that "[EtO] may have caused a
low level of oxidative stress and produced negatively charged molecules that modify Ras and
Ras signaling...leading to an early expansion of K-ra.s mutant clones" but at higher EtO
concentrations or longer exposure durations, or both, the amplification of existing K-ra.s
mutations switches to the selective senescence or death of K-ra.s mutant cells. No explanation is
proposed for the erratic control results.
J-17
-------
The EPA notes several limitations of the Parsons et al. (2013) study and its reported
findings. First, the study is looking at only three specific base-substitution mutations in one
specific codon of one specific gene. Given that carcinogenesis is a multifaceted process,
involving numerous genes, and that EtO can induce a variety of different types of mutation and
other genotoxic effects, one should not infer too much about the mode of action for EtO-induced
mouse lung carcinogenesis from this one study. In addition, the high degree of variability in
most of the dose group MF results and the instability of the control results across different
exposure durations suggest that the assay results might be unreliable. Nonetheless, Parsons et al.
(2013) proposed some elaborate pathways to explain the "surprising" time- and dose-response
patterns they observed. A more straightforward explanation for the highly variable dose group
results, the erratic control group results, and the irregular time- and dose-response patterns is
measurement error associated with the assay.
However, even if EtO caused a low level of oxidative stress and modified Ras signaling,
resulting first in amplification and then in the death of K-ra.s mutant cells, as Parsons etal.
(2013) proposed, which might explain some of their irregular time- and dose-response patterns
(the erratic control results are still unexplained), their hypothesized explanation does not
constitute a complete mode of action for the EtO-induced lung carcinogenicity observed in the
NTP mouse cancer bioassay. Moreover, the Parsons et al. (2013) study, which found decreased
levels of GAT and GTT mutations at 8 and 12 weeks compared to concurrent controls, does not
elucidate the observations by Hong etal. (2007) of later-occurring K-ra.s codon 12 GTT or GAT
mutations in all of the lung tumors evaluated from EtO-exposed mice in the NTP 2-year cancer
bioassay.
Furthermore, these hypotheses have no independent support to date. In fact, this proposal
disagrees with another 2013 study (Nagy etal., 2013)indicating that lung epithelial cells are
relatively sensitive to the DNA alkylating effects of EtO and relatively resistant to oxidative
DNA damage and that EtO does not induce oxidative damage. To investigate the relative
susceptibility of different cell types to different types of DNA damage, Nagy et al. (2013)
exposed human lung epithelial cells, peripheral blood lymphocytes, and keratinocytes for 1 hour
in vitro to different concentrations (previously determined to be subcytotoxic) of EtO (TWA
concentrations of 0, 16.4, 32.1, 55.5, or 237.5 [j,M)to assess alkylating damage or hydrogen
peroxide (0, 1, 2, 5, or 10 [j,M)to assess oxidative damage. DNA damage was determined using
the comet assay, and oxidative damage was detected by incorporating a step involving incubation
with formamidopyrimidin DNA-glycosylase (Fpg)—a lesion-specific restriction endonuclease
that can recognize oxidized purines and pyrimidines—into the assay. Nagy et al. (2013) reported
that linear regression analyses showed a statistically significant positive correlation between EtO
exposure and DNA damage as measured by both tail length and tail DNA for all three cell types.
J-18
-------
The shallowest slope was for keratinocytes for both DNA damage parameters. The slope for the
tail length parameter was higher for lung epithelial cells than for lymphocytes across the applied
concentration range, and the slope for the tail DNA parameter was higher for lung epithelial cells
than for lymphocytes across the concentration range for all but the highest concentration. A
statistically significant positive correlation also was found between hydrogen peroxide exposure
and oxidative DNA damage measured by both tail length and tail DNA for all three cell types.
For oxidative DNA damage, however, the shallowest slope was for lung epithelial cells for both
DNA damage parameters. Nagy et al. (2013) also reported that the oxidative potential of EtO
was similarly evaluated, and no evidence of Fpg-dependent oxidative DNA damage was found in
the examined cells at the applied concentrations (data not presented). Likewise, the more recent
mouse lung studies by Zhang et al. (Zhang et al., 2015b; Zhang et al., 2015a) provided little
support for the oxidative stress hypothesis (see Section J.4.1 below).
In addition, the EPA notes that none of the results presented by Parsons et al. (2013)
preclude direct genotoxic effects of EtO. For example, Parsons etal. (2013) also reported that
increased ell MFs were observed in lung tissues from the same EtO-exposed mice and that MFs
increased significantly with EtO concentration at 8 and 12 weeks (results to be published
separately), indicating that direct genotoxicity from EtO can occur elsewhere in the DNA.
Furthermore, even the K-ra.s codon 12 mutations that Parsons et al. (2013) investigated can result
directly from EtO—Parsons etal. (2013)themselves noted that the GAT mutation can result
from EtO-induced 06-HEG adducts, and even if 8-oxo-dG adducts from oxidative stress
preferentially cause G:C—>TA mutations as indicated by Parsons et al. (2013), a variety of
mutagens are known to cause G:C—>TA mutations as well (DeMarini, 2000).
J.4. REVIEW OF MAJOR STUDIES IDENTIFIED IN THE 2016 LITERATURE
SEARCH
Two additional major studies were identified after the 2014 SAB review draft (U.S. EPA,
2014a, b), in the September 2016 literature search. These new studies would not affect the
assessment's major conclusions but are reviewed briefly here for completeness.
J.4.1. Zhang et al. (2015a) and Zhang et al. (2015b)
In two studies published separately, Zhang etal. (2015a) and Zhang etal. (2015b)
exposed male B6C3Fi mice to < 200 ppm EtO via whole-body inhalation for either 4 or 12
weeks and measured the resulting impact on lung levels of glutathione conjugates (Zhang et al.,
2015a) or purine nucleotides and adducts (Zhang etal., 2015b), using liquid chromatography
coupled with mass spectrometry to improve the simultaneous detection of various endpoints.
Specifically, positive ions generated by electrospray ionization following separation by reverse
phase chromatography were quantified using selective reaction monitoring in a Q-trap and
J-19
-------
tandem mass spectrometry. To evaluate the effects of EtO exposure on lung levels of reduced
and oxidized glutathione (GSH and GSSG, respectively), as well as 2-hydroxyethylated
glutathione (HESG) resulting from EtO-alkylation of reduced glutathione, Zhang etal. (2015a)
exposed male mice to 0, 10, 50, 100, or 200 ppm EtO for 6 hours/day, 5 days/week, for 4 weeks.
The intra- and inter-day relative variation and accuracy were acceptable (< 13% and 87-113%,
respectively), and the lower limit of quantification (LLOQ) was reported to be 0.002 [j,g/mL (~ 2
ppb) for all three analytes, which seems to the EPA to be sufficiently sensitive considering that
GSH is present at [j,mol/g levels in rodent tissues (Pilon et al., 1988). To evaluate the effects of
EtO exposure on purine nucleotide adduction, Zhang et al. (2015b) exposed male B6C3Fimice
to 0, 100, or 200 ppm EtO on a similar schedule for a longer duration of 12 weeks. As with the
glutathione conjugates, the intra- and inter-day relative variation and accuracy (< 19% and 87-
120%), respectively) for a variety of guanine and adenine nucleotide adducts were acceptable,
and the LLOQs ranged from low ppt for DNA adducts to ppm for the unmodified purine
nucleotides.20 The authors did not evaluate N3-HEA, which was previously reported in the
spleens of F344 rats after 4 weeks of exposure to 300 ppm EtO (Walker et al., 1992), but did
evaluate two other products of adenine n-alkylation (Nl-HedA and N6-HedA). While the
authors did not evaluate the formation of the predominant EtO-guanine alkylation product, N7-
HEG, they did measure the levels of guanine adducts likely to result from reactive oxygen
species activity directly (8-OHdG), or indirectly following lipid peroxidation (CrotondG and
N2,3-EthenodG) (see Footnote 20 for abbreviations).
In both studies, the lungs were not perfused, but were excised and snap-frozen
immediately following the final exposure period. While Zhang etal. (2015a) reported exposing
groups of 20 mice to each concentration evaluated and then combining 50 mg of lung tissue from
subgroups of 4 mice to create five analytical samples for each concentration, this process was not
clearly described in Zhang etal. (2015b), although they also reported five analytical samples per
exposure group, and may have pooled tissue from multiple mice in a similar manner. The study
authors presented tables of the biological sample measurements but did not provide any
extensive analysis or qualitative discussion of the results, or perform any statistical analysis, in
either report (Zhang etal., 2015b; Zhang et al., 2015a); thus, the EPA conducted its own
statistical analyses. In mice exposed to < 200 ppm for 4 weeks, lung levels of both GSH and
GSSG decreased with increasing exposure concentrations, exhibiting a dose-response
2°0.025, 0.00125, 0.025, 0.00125, 0.025, 0.01, 2,342, and 2,500 ng/mL for 8-hydro?y-2'-deo?yguanosine (8-OHdG),
a-methyl-y-hydro?y-l,N2-propano-2'-deoxyguanosine (CrotondG), N2,3-etheno-2'-deo?yguanosine
(N2,3-EthenodG), 06-(2-hydro?yethyl)-2'-deo?5'guanosine (06-HEdG), l-(2-hydroxyethyl)-2'-deoxyadenosine (Nl-
HEdA), N6-(2-hydro?yethyl)-2'-deo?y adenosine (N6-HEdA), 2'-deo?yguanosine (dG), and 2'-deoxyadenosine (dA),
respectively.
J-20
-------
relationship consistent with a linear trend (see Table J-6), although by pairwise comparisons only
concentrations > 100 ppm induced statistically significant decrements in both endpoints [Table
J-6; from Table 4 in Zhang etal. (2015 aVI. The EPA notes that this observation is consistent
with previous measurements of blood EtO-hemoglobin adducts in mice and rats indicating the
potential for significant tissue glutathione depletion following exposures >100 ppm
(see Section 3.3.2). While both GSH and GSSG decreased in the lungs of mice with increasing
EtO exposure, the ratio of GSH:GSSG (a redox couple commonly evaluated as a measure of
cellular oxidative stress) did not change. Furthermore, while below the limit of quantification in
control samples, levels of the EtO-GSH alkylation product HESG increased in what also
appeared to be a linear relationship with increasing exposure concentrations > 10 ppm.
The EPA found it interesting that Zhang etal. (2015a) could not quantify HESG levels
from the lungs of control mice, considering that N7-HEG DNA adducts resulting from
endogenous EtO alkylation have been reported in various tissues including lungs from
unexposed rats and mice [e.g., Wu et al. (1999a); Walker etal. (1992); see Section 3.3.3.4],
While presumably endogenous EtO would also form HESG adducts at some level in the mouse
lung, these background levels must have been at least 10 times lower than the 14.3 [j,g/g average
levels resulting from 10 ppm EtO exposure, given the stated LLOQ of 0.002 [j,g/mL (Zhang et
al., 2015a). Levels of both nonoxidized glutathione (e.g., GSH + HESG) and total glutathione
(GSH + GSSG + HESG) remained similar or may have increased marginally with treatment.
The decrease in lung GSH and GSSG levels concomitant with increased HESG levels, together
with the constant ratio of GSHGSSG, following exposure to EtO concentrations from 10 to
200 ppm indicated to the authors that the lung GSH depletion resulted from EtO alkylation to
HESG and not oxidation to GSSG. While the ratio of GSH:GSSG was unchanged, the ratio of
free reduced total other glutathione (i.e., GSH:[GSSG + HESG]) decreased dramatically from
13.1 in controls to 0.8 in the lungs of mice exposed to 200 ppm (see Table J-6), suggesting to the
EPA that the capacity of the lungs to withstand oxidative stress induced by some other
exogenous source may be severely compromised at higher EtO concentrations, consistent with
previous reports evaluating EtO-hemoglobin adduction (see Section 3.2.2).
J-21
-------
Table J-6. Evaluation of reported measurements of GSH, GSSG, and
HESG; averages (SD) of pooled samples [jig/g tissue; Zhang et al. (2015a)]a
Pooled
exposure
group
(ppm)
GSH
(from
Table 4)
GSSG
(from
Table 4)
HESG
(from
Table 4)
GSH +
HESG
(from
Table 4)
GSH/GSSG
(from Table
4)
GSSG +
HESG
(calc)
GSH/[GSSG +
HESG] (calc)
GSH +
GSSG +
HESG
(calc)
0
702
(59.0)b
53.4 (4.78)b
-------
EtO-exposed rats at any lower concentrations (see Section 3.3.3.4). From the Zhang et al.
(2015b) mouse lung adduct data, the EPA's statistical analyses indicated that levels of 06-HEdG
adducts increased in an apparently linear manner, with statistically significant increases observed
in the 200-ppm samples by pair-wise comparisons (see Table J-7). Levels of both adenine
adducts (i.e., Nl-HEdA and N6-HEdA) were unquantifiable in the control samples but were
present in samples from mouse lungs in both exposure groups, with significant increases
observed in samples from 200 versus 100 ppm. The EPA notes that along with the potentially
mutagenic Oe-HEdG adduct, both Nl-HEdA and N6-HEdA could also be promutagenic, as
adducts in these positions may interfere with base-pairing interactions. 8-OHdG can be formed
by direct reaction of reactive oxygen species such as superoxide or hydroxyl radicals, but levels
of this adduct were unaffected by EtO exposures <200 ppm (see Table J-7). While Zhang et al.
(2015b) did not remark on changes in CrotondG adducts, which can form following lipid
peroxidation, CrotondG levels appeared to have increased with exposure, and the levels in lung
samples following 200-ppm exposures were significantly increased compared with control
samples, although to a lesser extent than 06-HEdG or either adenine adduct.
Table J-7. Evaluation of reported measurements of various DNA adducts;
averages (SD) [n = 5 analytical samples; Zhang etal. (2015b)]a
Group (ppm)
06-HEdG
(A/dG)b
8-OHdG
(A/dG)b
CrotondG
(A/dG)b
Nl-HEdA
(A/dG)b
N6-HEdA
(A/dG)b
0
0.229 (0.167)c-d
42.0 (7.78)
0.120 (0.0302)d
< LOQ
< LOQ
100
0.467 (0.365)
39.3 (8.47)
0.147 (0.0324)
1.96 (0.152)
1.39 (0.0897)
200
0.743 (0.118)*
46.0 (3.97)
0.190 (0.0165)**
6.97 (1.79)**
4.27 (0.540)***
3 Authors presented no statistical analysis of results. Average (standard deviation: SD) datareported in Table 4 of
Zhang etal. f2015b') were evaluated within each column by the EPA using one-way ANOVA for C^-HEdG,
8-OHdG, and CrotondG, and student's t-test with Welch's correction for Nl-HEdA and N6-HEdA. Significant
changes compared with the 0-ppm group by Dunnett's multiple comparison posttest,orbetween the 100- and
200-ppm groups for Nl-HEdA andN6-HEdA by unpaired student's t-test with Welch's correction for unequal
variance, are indicated by *(p < 0.05), ** (p < 0.01), and *** (p < 0.001).
bAdducts x 106/dG levels = A/dG.
cLevels were < LOQ in 2/5 samples; for purpose of average (SD) calculations, samples < LOQ were considered to be
equal to the LOQ for the 06-HEdG (0.00125 ng/mL), and A/dG were calculated using this LOQ value x 106/dG
concentration presented in Table 4 for each sample.
dMeans significantly differed amongst the treatment groups for each endpoint by 1-way ANOVA (p < 0.05), anda
linear trend was present between means and row number, p < 0.01, using test for linear contrast (posttestfortrend) in
Graphpad fhttp://www.graphpad.com/guides/prism/6/statistics/indexhtm?stat posttesttrend.htm).
J-23
-------
While neither reactive oxygen species nor oxidized lipids (e.g., malondialdehyde or
thiobarbituric acid reactive substances [tBARs] levels), were measured directly in either study,
the lack of decrease in the ratio of GSH:GSSG (Zhang et al.. 2015a) or increase in 8-OHdG
levels (Zhang et al.. 2015b). both routinely evaluated as markers of cellular oxidative stress,
coupled with the limited increase in CrotondG adducts and the inability to detect N2,3-ethenodG,
both formed following lipid peroxidation, suggest to the EPA that oxidative stress is not induced
in the lungs of mice following 4-12 weeks of exposure to < 200 ppm EtO (Zhang et al.. 2015b;
Zhang etal., 2015a). In addition, the EPA finds that the Zhang etal. (2015b) study supports the
identification of 06-HEdG as a direct product of EtO reactivity, consistent with previous in vitro
and in vivo reports (see Section 3.3.3.1), and adds coherence to the available database by
observing an exposure-related increase in lung 06-HEdG levels at lower concentrations than
previously evaluated (i.e., 100-200 ppm vs. 300 ppm), quantification in another rodent species
(i.e., mice vs. rats), and even detection in the majority of unexposed lung samples (3/5),
suggesting that endogenous EtO may be responsible for a low background level of this
potentially mutagenic DNA adduct. Furthermore, the significant increases in other potentially
mutagenic purine adducts (i.e., CrotondG, Nl-HEdA, and N6-HEdA) is consistent with the novel
mutational spectra preferentially affecting purine nucleotides reported in lung and other tumors
from EtO-exposed male and female B6C3Fi mice [e.g., NTP (1987); Hong etal. (2007); Houle
et al. (2006); see Section 3.3.3.4] and the conclusion that EtO-induced rodent tumors likely arise
via a mutagenic mode of action following the direct formation of mutagenic EtO-DNA adducts.
J-24
-------
APPENDIX k. SUMMARY OF PUBLIC COMMENTS RECEIVED ON THE
JULY 2013 PUBLIC COMMENT DRAFT AND EPA RESPONSES
The EPA's Science Advisory Board (SAB) reviewed an external review draft of the
ethylene oxide (EtO) carcinogenicity assessment in 2007 (see Appendix H). Following that
review, a revised draft was developed and released on July 23, 2013 for a 45-day public
comment period. In response to requests from the American Chemistry Council's (ACC's)
Ethylene Oxide Panel, the Ethylene Oxide Sterilization Association (EOSA), and Balchem
Corporation, the public comment period was extended from September 5 to October 11, 2013.
During the public comment period, 16 sets of comments were received, not including the
three requests to extend the public comment period. The major substantive science comments
came from four groups. The first of these, the Breast Cancer Fund (docket #0043), expressed
agreement with the EPA's hazard and mode of action (MO A) conclusions. The comments from
the remaining three groups [American Chemistry Council's Center for Advancing Risk
Assessment Science and Policy (ARASP) (#0055), EOSA (#0056), and ACC (#0057)] largely
overlapped. A summary of the substantive science comments from these latter three groups and
the EPA's responses is provided below. The comments have been synthesized and paraphrased
and are organized roughly to follow the order of the carcinogenicity assessment. The complete
set of public comments is available in Docket ID No. EPA-HQ-ORD-2006-0756-0035 at
http://www. regulations. gov.
The July 2013 draft was further revised in response to the public comments and
submitted for additional SAB review in August 2014. Comments on the 2014 SAB review draft
and the EPA's responses are presented in Appendix I.
1. COMMENT: EPA failed to comply with multiple guidelines, including Information Quality
Act guidelines and 2011 National Academy of Sciences [NRC] recommendations.
Specifically, EPA failed to apply a transparent and systematic weight-of-evidence approach
in both qualitatively and quantitatively assessing the cancer risks, did not base the assessment
on the best available science, and used National Institute for Occupational Safety and Health
(NIOSH) breast cancer incidence data that are not available to the public. (ACC, EOSA)
EPA RESPONSE: The EPA has complied with applicable guidelines. The EtO assessment
was largely developed before the IRIS program started implementing the 2011 NRC
recommendations and formalizing approaches to conducting and documenting systematic
review. Although not presented in the formalized manner IRIS has been developing, the
EtO assessment provides a valid and transparent weight-of-evidence analysis based on
K-l
-------
the best available science. Considerations used in assessing the epidemiological studies
are summarized at the beginning of Section 3.1, and the considerations used in the
weight-of-evidence analysis for carcinogenic hazard are detailed in Section 3.5.1,
culminating in a synopsis of how the evidence fits the lines of evidence for the
characterization of "carcinogenic to humans" laid out in the EPA's 2005 Guidelines for
Carcinogen Risk Assessment. Considerations used in selecting the epidemiology
study(ies) for quantitative risk estimation are summarized in Section 4.1, along with
considerations used in selection of exposure-response models. A systematic literature
search was conducted from January 2006. Major new studies identified in the literature
search as well as even more recent studies noted by the ACC in its public comments have
been added to Appendix J. The charge to the SAB includes questions addressing
adequacy, transparency, and clarity of the assessment and completeness of the appendix
on new studies. With respect to the breast cancer incidence data, the EPA's Information
Quality Act guidelines do not require that all underlying raw epidemiology data be
publicly available; they allow for confidentiality constraints.
2. COMMENT: Data quality evaluation should clearly describe the criteria used to deem a
study as high quality. (ARASP)
EPA RESPON SE: The EtO assessment discusses general characteristics used to evaluate
epidemiology studies and notes numerous characteristics that supported the determination
that the NIOSH study was a "high-quality" study, for example, high-quality exposure
estimates (as discussed in Section A.2.8 of Appendix A), large size, adequate follow-up,
inclusion of males and females, absence of other occupational exposures, and use of
internal comparisons. The assessment has been revised to summarize these
characteristics clearly in one location (see Footnote 13 in Section 3.5.1).
3. COMMENT: Lymphohematopoietic and lymphoid cancers should not be grouped because
they are derived from different cells of origin. (ARASP, ACC, EOSA)
EPA RESPONSE: The EPA did appropriately combine lymphoid cancers, as the
"lymphoid" cancer category is a grouping of cancers with a common
lymphohematopoietic cell lineage (multiple myeloma and most lymphocytic leukemias
and non-Hodgkin lymphomas develop from B-lymphocytes). The 2007 SAB panel
supported the use of this grouping. The larger lymphohematopoietic cancer grouping is
provided solely for comparison because many of the epidemiologic studies do not present
data for a lymphoid cancer grouping.
4. COMMENT: The evidence for breast cancer is too weak. (ACC, EOSA)
K-2
-------
EPA RESPONSE: Although the epidemiological database for breast cancer is more limited
(i.e., few studies with sufficient numbers of female breast cancer cases) than that for
lymphohematopoietic cancers, the EPA determined that the available evidence is
sufficient to consider breast cancer a potential hazard from EtO exposure. In addition,
the epidemiological database is strengthened by the follow-up study (Mikoczy etal.,
2011) of the Swedish cohort of sterilizer workers first reported on by Hagmar et al.
(Hagmar etal.. 1995; Hagmar et al.. 1991) (see Section J.2.2 of Appendix J), and the
epidemiological evidence is supported by the finding of mammary gland carcinomas in
female mice exposed to EtO by inhalation (NTP, 1987) and by mechanistic data
(see Section 3.4.1.3). The 2007 SAB panel did not object to the derivation of unit risk
estimates based on the available breast cancer evidence.
5. COMMENT: EtO is a weak mutagen. (ACC, EOSA)
EPA RESPONSE: The EPA agrees that EtO is a relatively weak mutagen compared to the
anticancer agents and the other reactive epoxides investigated in the Vogel and Nivard
(1998) paper. Vogel and Nivard (1998) compared 37 anticancer agents, which are
generally highly mutagenic by design, and four epoxides, including EtO, one of which
was a cross-linking diepoxide.
The EPA notes, however, that there is generally no strong correlation between
potency in short-term mutagenicity and genotoxicity tests and carcinogenic potency. For
example, for the Ames assay, Fetterman etal. (1997) found a "very weak" relationship
between quantitative mutagenic and carcinogenic potencies. In addition, EtO is highly
volatile and concentrations can become much reduced over the course of an in vitro
assay, making potency from such assays difficult to determine.
6. COMMENT: A mutagenic MoA is not supported by the most recent scientific evidence;
other MoAs, specifically oxidative stress and cell proliferation, should be considered. (ACC)
EPA RESPONSE: The 2007 SAB panel concurred with the EPA's conclusion at that time
that a mutagenic MOA was operative in the carcinogenicity of EtO. In its 2013 public
review draft, the EPA presented more recent information and found this information to be
supportive of the earlier conclusion of a mutagenic MOA. New information presented by
the ACC is not sufficient to alter that conclusion. Other MO As proposed by the ACC are
speculative.
As evidence against a direct mutagenic MOA, the ACC cites a paper by Parsons
et al. (2013). This study and its limitations are discussed in detail in Section J.3.2 of
Appendix J. In brief Parsons etal. (2013) investigated only one type of mutation (base
substitution mutations) in one codon (12) of one gene (K-ra.s) in one tissue (mouse lung)
for exposure durations up to 12 weeks. Given that carcinogenesis is a multifaceted
K-3
-------
process, involving numerous genes, and that EtO can induce a variety of different types
of mutation and other genotoxic effects, one cannot infer too much about the MOA for
EtO-induced mouse lung carcinogenesis from this one study. In addition, the high degree
of variability in the mutant fraction results for most of the dose groups and the instability
of the control results across different exposure durations suggest that the assay results
might be unreliable. To attempt to explain the irregular time- and dose-response patterns
that they observed (e.g., statistically significant increases in specific K-ra.s mutant cells at
4 weeks but statistically significant decreases at 8 and 12 weeks compared to controls),
Parsons etal. (2013) hypothesized that EtO causes a low level of oxidative stress that
modifies Ras signaling, resulting first in the amplification and then in the death of K-ra.s
mutant cells (the erratic control results are still unexplained). That hypothesized
explanation for the irregular results, however, does not constitute a complete MOA for
the EtO-induced lung carcinogenicity observed in the NTP mouse cancer bioassay and
does not explain the observations by Hong et al. (2007) of later-occurring K-ras codon 12
mutations in all lung tumors evaluated from EtO-exposed mice in the NTP 2-year cancer
bioassay. A more straightforward explanation for the highly variable dose group results,
the erratic control group results, and the irregular time- and dose-response patterns is
measurement error associated with the assay.
Thus far, there is no independent support for the hypotheses of Parsons et al.
(2013). In fact, the proposed hypotheses are at odds with a Nagy et al. (2013) study of
human cells in vitro. Using the sensitive comet assay, Nagy et al. (2013) found that lung
epithelial cells are relatively susceptible to the DNA alkylating effects of EtO and
relatively resistant to oxidative DNA damage (induced by hydrogen peroxide) compared
to peripheral blood lymphocytes and keratinocytes. In addition, Nagy etal. (2013) found
no evidence that EtO induced oxidative DNA damage in the examined cells at the applied
concentrations.
Furthermore, as Parsons et al. (2013) and the ACC acknowledged, the results
Parsons et al. (2013) presented do not preclude direct genotoxic effects of EtO. Direct
effects of EtO could include K-ra.s mutations as well as genotoxic effects elsewhere in
the DNA.
Moreover, any inferences about K-ra.s mutations that one can draw from the
Parsons etal. (2013) study are not necessarily generalizable to other cancer types. Codon
12 of the K-ra.s gene was selected for investigation because Hong et al. (2007) had
observed mutations in this K-ra.s codon in all 23 lung tumors they evaluated from
EtO-exposed mice in the NTP 2-year cancer bioassay. However, Hong et al. (2007)
observed other patterns of K-ra.s mutations, involving other codons, in other tumors
(Harderian gland and uterine tumors) from the NTP mouse bioassay.
In support of an oxidative stress MOA, the ACC also cites work by Marsden et al.
(2009). Marsden et al. (2009) used sensitive detection techniques and an approach
designed to quantify endogenous N7-HEG adducts and exogenous N7-HEG adducts
K-4
-------
separately to measure the amounts of endogenous and exogenous N7-HEG adducts
occurring in rat liver, spleen, and stomach following EtO treatment (see also
Section 3.3.3.1 and Appendix C). In addition to direct DNA adduct formation via
alkylation observed in the liver, spleen, and stomach, Marsden et al. (2009) observed an
indirect effect of EtO exposure on endogenous N7-HEG adduct formation in the liver and
spleen and hypothesized that EtO also could cause adduct formation indirectly by
inducing oxidative stress, which might in turn induce the endogenous formation of
ethylene, which can be metabolized to EtO.
As discussed in the EtO assessment (see Section 3.3.3.1 and Appendix C),
although not statistically significant, increases in exogenous adducts were observed at the
lowest dose in liver and spleen, and Marsden et al. (2009) noted that the exogenous
adduct data are consistent with a linear dose-response relationship (p < 0.05) in all three
tissues examined. In addition, more substantial relative increases in exogenous adducts
appear to be occurring at lower doses than for endogenous adducts [see Table 1 of
Marsden et al. (2009)1. Thus, even if the speculative oxidative stress MO A is also
operative in liver and spleen at higher doses, it does not rule out direct genotoxic effects
of EtO. Moreover, liver and spleen (the parenchymal tissue) are not known target organs
for EtO-induced carcinogenicity and the results do not seem to be generalizable to other
tissues, as there was no evidence of increased endogenous adducts in the stomach, where
there were clear, statistically significant increases in exogenous adducts for all but the
lowest dose.
Regarding cell proliferation, the ACC offers no solid evidence that such an effect
is induced by EtO exposure. The ACC acknowledges that no generalized mitogenesis
occurred in the lung in the Parsons et al. (2013) study. Nor was cytotoxicity or apoptosis
detectable (Parsons et al., 2013). Similarly, in the Nagy et al. (2013) study mentioned
above, all observed genotoxic effects occurred at subcytotoxic doses. Cytotoxicity also
has not been an issue in other toxicity and genotoxicity studies of EtO; thus, regenerative
proliferation resulting from EtO-induced cytotoxicity is not credible as a key component
of a MOA for EtO-induced carcinogenesis.
The ACC suggests that the observation of early increases in the GAT K-ra.s codon
12 mutation in the Parsons etal. (2013) study supports a mitogenesis MOA because the
GAT mutation is the most common K-ra.s mutation observed in spontaneous mouse lung
tumors. G:C—>A:T mutations do not just occur spontaneously, however; they can be
induced by a variety of agents, including EtO (see Section J.3.2). Furthermore, as
discussed above and in Section J.3.2, there is considerable uncertainty pertaining to the
Parsons etal. (2013) results, and to explain some of the irregular time- and dose-response
patterns observed, Parsons et al. (2013) proposed first amplification and then death of
K-ra.s mutant cells, so how the Parsons et al. (2013) findings support mitogenesis as a
MoA for EtO-induced carcinogenesis is unclear. The ACC also proffers the claim by
Parsons etal. (2013) that no single type of DNA adduct correlates with the K-ra.s codon
K-5
-------
12 mutations observed as evidence of a mitogenesis MO A; however, EtO induces
multiple types of DNA adducts and Parsons et al. (2013) themselves acknowledged that it
could "be postulated that a combination of different types of DNA damage could lead to
the profile of induced K-ras mutation..."
7. COMMENT: EPA failed to incorporate the Union Carbide Corporation (UCC) data into the
dose-response assessment. The NIOSH exposure assessment also suffered from limitations.
(ACC, EOSA)
EPA RESPON SE: As recommended by the 2007 SAB panel, the EPA considered using the
UCC data and determined that they were not of sufficient quality to add useful
information to the NIOSH study's data for the derivation of unit risk estimates (seethe
reasons discussed in detail in the assessment [e.g., Section A.2.20 of Appendix A] and in
the responses to the SAB comments [p. H-6 to H-8]). Thus, the EPA decided to use the
NIOSH data as the basis for the exposure-response modeling (see also Section 4.1).
Although no exposure assessment is without limitations, the NIOSH regression
model includes a number of relevant variables and had a high validity when tested against
independent data (see Section A.2.8 for details). The approach used to derive the UCC
exposure estimates was much less rigorous and there is considerable uncertainty in the
resulting estimates. The 2007 SAB panel supported the use of the NIOSH study as a
basis for risk estimates.
8. COMMENT: Despite SAB recommendations, EPA used summary data rather than the
individual data in the modeling of breast cancer mortality and lymphoid cancer. (ACC,
EOSA)
EPA RESPON SE: As documented in the assessment and in the responses to SAB
comments (p. H-12 and H-13), the EPA investigated multiple models based on the
individual continuous exposure data, including a log-linear model. For the breast cancer
incidence data, the EPA was able to develop several continuous models that provided
reasonable fits to the individual-level exposure data across the entire range of the data
(see Section 4.1.2.3), consistent with the SAB recommendations.
For lymphoid cancer, however, despite the extensive modeling efforts, the various
alternative continuous models investigated—including the two-piece spline models—
proved problematic, as explained in detail in the text (see Section 4.1.1.2). In particular,
the statistically significant models predicted extremely steep slopes in the low-dose
region. Thus, the EPA has retained the approach of using a linear regression of the
categorical data, excluding the highest exposure group, as the basis for the preferred unit
risk estimates for lymphoid cancer. The EPA notes that modeling of grouped data is also
an important and well-recognized statistical methodology and its use is consistent with EPA
guidance, policy, and past practice. The breast cancer mortality data were similarly
K-6
-------
difficult to model due to extreme supralinearity, and the optimal two-piece spline model
yielded an unrealistically steep low-dose slope estimate; thus, the EPA again used a linear
regression of the categorical data, excluding the highest exposure group, as the basis for
the preferred estimate (see Section 4.1.2.2). (The breast cancer mortality data are not
critical to the assessment because the breast cancer incidence data set is preferred.)
Since the July 2013 public comment draft, however, unit risk estimates for
lymphoid cancer and breast cancer mortality from the most suitable alternative models
based on the continuous-exposure data were developed and added to the assessment for
comparison purposes.
9. COMMENT: EPA used a non-peer-reviewed supralinear spline model. (ACC, EOSA)
EPA RESPON SE: The spline model the EPA used for the breast cancer incidence data was
the best fitting of the continuous models considered, and others have used this model
with similar data sets to estimate risk. The breast cancer modeling work was published in
a peer-reviewed journal (Steenland et al.. 20111 and the EtO spline model will receive
further SAB review. Moreover, the two-piece spline model used is not inherently
supralinear; it is a flexible model that can accommodate sub linear or supralinear (or
linear) exposure-response relationships. The EtO two-piece spline models become
supralinear models because the underlying exposure-response relationships of the data to
which they are being fitted are supralinear.
10. COMMENT: There are a number of modeling issues in addition to those mentioned in
other comments, specifically flaws discussed in Valdez-Flores and Sielken (2013) and
Valdez-Flores etal. (2010) and over-predictions of the cancer deaths in the NIOSH study.
(ACC, EOSA)
EPA RESPON SE: The EPA did not find that Valdez-Flores and Sielken (2013) or Valdez-
Flores et al. (2010) provided convincing evidence of flaws in the modeling. The EPA
addressed the issues presented by Valdez-Flores et al. (2010) in the July 2013 assessment
(see Section A.2.20 of Appendix A). Discussion of the new Valdez-Flores and Sielken
(2013) study has been added to Appendix J (see Section J. 3.1). In light of issues raised
by Valdez-Flores and Sielken (2013). text was added to the assessment clarifying the
model comparisons in some of the figures of Chapter 4.
How the predicted numbers of deaths for the cohort study are being calculated is
unclear from the submitted comments; thus, the specific claims could not be evaluated.
The EPA notes, however, that the ACC is no longer claiming that the observed number of
cancer mortalities is overpredicted "by more than 60-fold." In Appendix I of the ACC
comments, the claim is made that the lymphoid cancer mortality is overpredicted by
"1.87- to 3.26-fold" and breast cancer mortality is overpredicted by "1.24- to 1.84-fold."
K-7
-------
These estimates are based on the upper confidence limits on the models, however; a more
suitable basis for comparison with the observed deaths is the maximum likelihood
estimates (MLEs) of the models. According to Figure E.l in the ACC's Appendix I, the
best estimate from the MLE of the model for lymphoid cancer mortality is only about a
1.6-fold difference, and Figure A.1 suggests less than a 1.3-fold difference for breast
cancer mortality.
11. COMMENT: EPA should present both linear and nonlinear extrapolation approaches.
(ARASP, ACC, EOS A)
EPA RESPONSE: The EPA notes that some members of the 2007 SAB panel
recommended that the EPA include a nonlinear approach; this view was not a consensus
position—some panel members thought that such an approach should be included, but
others thought a nonlinear approach was not warranted. The EPA considered available
information and opinions presented by SAB members and concluded that there was
insufficient evidence for a nonlinear approach. This conclusion and its basis are
discussed in detail in the responses to SAB comments in Appendix H of the draft
assessment (p. H-13 to Ft-18). Part of the charge for the second SAB review will be to
consider the EPA's responses to the comments of the first SAB panel, including the
EPA's judgment not to include a nonlinear approach. New information presented by the
ACC is not sufficient to alter the determination not to include a nonlinear approach (see
the EPA's response to Comment 6 above).
12. COMMENT: Combining breast cancer and lymphoid cancer unit risk estimates is not
justified, and EPA did not discuss competing risks, different background populations,
incidence vs. mortality, and the use of different exposure-response models. (ACC, EOSA)
EPA RESPONSE: When combining cancer types in a dose-response model, it is desirable
that the cancer types have a common origin. In contrast, when combining unit risk
estimates (for cancer types that have been modeled separately) to derive a total cancer
unit risk estimate, it is desirable that the cancer types be independent. Thus, in the EtO
assessment, breast cancer and lymphoid cancers were modeled separately, and then the
unit risk estimates were combined to develop a total cancer unit risk estimate. It is
standard practice in IRIS assessments to estimate total cancer risk and not just the risk
from individual cancer types, and this practice is consistent with EPA guidelines (U.S.
EPA, 2005a) and National Research Council recommendations (NRC, 1994).
In terms of extra risks (above background) from environmental exposure levels of
EtO, the likelihood of co-occurrence of EtO-induced breast and lymphoid cancers is
negligible. In addition, considering the risks from both cancer types occurring in a single
K-8
-------
individual is not "double-counting" if the cancer types are independent with respect to
EtO exposure.
The total cancer unit risk estimate is intended to apply to the general population,
of which females comprise a substantial portion. For a risk estimate for males only, the
unit risk estimate for lymphoid cancer alone is presented in the assessment also. The
issue of different background populations (male and female) is now addressed in the
assessment.
The unit risk estimates that are being combined are for cancer incidence, so no
inconsistency exists with respect to cancer status. Similarly, the unit risk estimates that
are being combined are linear slopes, so no inconsistency exists with respect to the model
form being combined, either (the exposure-response models used to derive the unit risk
estimates are irrelevant to the combing of the unit risk estimates).
13. COMMENT: EPA should reexamine its risk determination given background and
endogenous levels of EtO; EPA's risk estimates are unrealistically high. (ARASP, ACC,
EOS A)
EPA RESPONSE: The unit risk estimates the EPA developed are for extra risk (i.e., above
background); background and endogenous levels of EtO, which would be relevant to (the
true) background risk, are not integral to the development of the estimates of extra risk.
As discussed in the assessment (see Section 4.5), given the high background rates of
lymphoid and breast cancers (lymphoid cancers have a background lifetime incidence
risk on the order of 3%, while the background lifetime incidence risk for breast cancer is
on the order of 15%), EPA does not consider the risk estimates for exogenous exposure to
be inconsistent with the data on background and endogenous levels.
According toEPA's 2005 National Air Toxics Assessment data, the average
exposure concentration of EtO from all sources (including background) in the United
States is 0.0062 (J,g^m3; the average background concentration is 0.0044 [j,g/m3. Using the
EPA's draft unit risk estimates, adjusted for assumed increased early-life susceptibility,
upper-bound estimates of the cancer risk resulting from a lifetime exposure to the average
concentration from all sources are roughly 1 lymphoid cancer case for every 220,000
people and 1 breast cancer case for every 120,000 women; the upper-bound estimates
resulting from a lifetime exposure to the average concentration above background (i.e.,
from known sources) (0.0018 (J,g/m3) are roughly 1 lymphoid cancer case for every
770,000 people and 1 breast cancer case for every 410,000 women. The calculations the
ACC provided were for an unrealistic exposure concentration of 1 ppb (1.8 (J,g/m3).
14. COMMENT: EPA should not derive occupational exposure limits for EtO. (ACC, EOSA)
K-9
-------
EPA RESPONSE: The EPA does not set "occupational exposure limits" for EtO; however,
the EPA's Office of Pesticide Programs (OPP)has a regulatory interest in occupational
exposures resulting from sterilization uses of EtO, as the EPA has the legal authority to
consider occupational risks in pesticide labeling and registration decisions under FIFRA
(Federal Insecticide, Fungicide, and Rodenticide Act). Typically, OPP uses the IRIS unit
risk estimates for its risk assessments of occupational exposures, which is valid when the
exposure-response model is reasonably linear over the relevant range of exposures. With
the models used for the EtO cancer data, however, the unit risk estimate is not
appropriate in the lull range of the occupational exposure scenarios of interest to OPP.
Thus, the assessment provides sample risk estimates for exposure scenarios of interest to
OPP for its risk assessment of sterilization uses of EtO. These estimates are not
"occupational exposure limits," and OPP will conduct its own risk assessment based on
current exposure estimates. OSHAand NIOSHhad the opportunity to review an earlier
draft EtO assessment during the interagency review phase of the IRIS assessment
process.
K-10
-------
REFERENCES
Abeles. FB; Heggestad, HE. (1973). Ethylene: An urban air pollutant. J Air Waste Manag Assoc
23: 517-521. httpy/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. Mutat Res 585: 120-126.
http//dx.doi. org/10.1016/i.mrgentox.2005.04.009
Agarwai A; Unfer, R; Mallapragada, SK. (2007). Investigation of in vitro biocompatibility of
novel pentablock copolymers for gene delivery. JBiomed Mater Res A 81: 24-39.
http://dx.doi.Org/10.1002/ibm.a.30920
Agurelj 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, H -S; Shin, H, -S. (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. org/10.1016/i.jchromb.2006.06.003
Ambroise, D; Moulin, JJ; Squinazi, F; Protois, JC; Fontana, JM; Wild, P. (2005). Cancer
mortality among municipal pest-control workers. Int Arch Occup Environ Health 78:
387-393. httpy/dx.doi.org/10.1007/s00420-004-0599-x
Appelgren, L, -E; Eneroth, G; Grant, C; Lanstrom, -L; 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; Malloy, EJ; Eisen, EA. (2007). Reducing healthy worker survivor bias by
restricting date of hire in a cohort study of Vermont granite workers. Occup Environ Med
64: 681-687. httpy/dx.doi.org/10.1136/oem.2006.031369
Armstrong, BG. (1990). The effects of measurement errors on relative risk regressions. Am J
Epidemiol 132:1176-1184.
Armstrong, BG. (1998). Effect of measurement error on epidemiological studies of
environmental and occupational exposures. Occup Environ Med 55: 651-656.
http y/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. http//dx.doi.org/10.1097/00001648-
199403000-00009
Bastlova, T; Andersson, B; Lambert, B; Kolmaq A. (1993). Molecular analysis of ethylene
oxide-induced mutations at the HPRT locus inhuman 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. In Health Risks of Radon and Other
Internally Deposited Alpha-Emitters. Washington, DC: National Academy Press.
http//dx.doi. org/10.17226/1026
Benson, LO; Teta, MJ. (1993). Mortality due to pancreatic and lymphopoietic cancers in
chlorohydrin production workers. Br JIndMed 50: 710-716.
http//dx.doi.org/10.1136/oem.50.8.710
R-l
-------
Beranek. DT. (1990). Distribution of methyl and ethyl adducts following alkylation with
monofunctional alkylating agents [Review], MutatRes 231: 11-30.
Bisanti, L; Maggjni, M; Raschetti, R; Alegjani, SS; Ippolito, FM; Caffari, B; Segnaa N; Ponti,
A. (1993). Cancer mortality in ethylene oxide workers. Br JInd Med 50: 317-324.
Boffetta, P; van der Hei O; Norppa, H; Fabianova, E; Fucic, A; Gundy, S; Lazutka, J; Cebulska-
Wasilewska, A; Puskailerova, D; Znaor, A; Kelecsenyi, Z; Kurtinaitis, J; Rachtaa J;
Fornj 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/aie/kwi367
Bolt HM. (2000). Carcinogenicity and genotoxicity of ethylene oxide: new aspects and recent
advances [Review], Crit Rev Toxicol 30: 595-608.
httpy/dx.doi.org/10.1080/10408440008951121
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. httpy/dx.doi.org/10.1007/s002040050451
Bolt HM; Peter, H; Fost, U. (1988). Analysis of macromolecular ethylene oxide adducts
[Review], Int Arch Occup Environ Health 60: 141-144.
httpy/dx.doi.org/10.1007/BF00378688
Bonassi S; Znaor, A; Ceppi, M; Lando, C; Chang. WP; Holland, N; Kirsch-Volders, M; Zeiger,
E; Baa S; Barale, R; Bigatti, MP; Bolognesi, C; Cebulska-Wasilewska, A; Fabianova, E;
Fucic, A; Hagmar, L; Joksic, G; Martelli A; Migliore, L; Mirkova, E; Scarfr MR; Zijno,
A; Norppa, H; Fenech, M. (2007). An increased micronucleus frequency in peripheral
blood lymphocytes predicts the risk of cancer in humans. Carcinogenesis 28: 625-631.
httpy/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.
httpy/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], Mutat Res 678: 76-94.
http y/dx. doi. or g/10.1016/j. mrgentox.2009.05.006
Breheny, D; Cunningham, F;Kilford, J; Payne, R; Dilloa D; Meredith. C. (2014). Application of
a modified gaseous exposure system to the in vitro toxicological assessment of tobacco
smoke toxicants. Environ Mol Mutagen 55: 662-672. httpy/dx.doi.org/10.1002/em.21876
Brittoa DW; Tornqvist, M; van Sittert, NJ; Watsoa 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.
Browa CD; Asghariaa B; Turner, MJ; Fennell TR. (1998). Ethylene oxide dosimetry in the
mouse. Toxicol Appl Pharmacol 148: 215-222. http://dx.doi.org/10.10Q6/taap.1997.8349
Browa T; Rushtoa L. (2012). Occupational cancer in Britain haematopoietic malignancies:
Leukaemia, multiple myeloma, non-Hodgkins lymphoma. Br J Cancer 107: S41-S48.
httpy/dx.doi.org/10.1038/bic.2012.117
Bukowska, B. (2015). [Hemoglobin adducts as biomarkers of human exposure to selected
xenobiotics] [Review], Postepy Higieny i Medycyny Doswiadczalnej 69: 668-680.
httpy/dx.doi.org/10.5604/17322693.1156936
R-2
-------
Butterworth. BE; Chapmaa 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.Org/10.1016/i.yrtph.2007.07.004
Chaa C, -C: Shie. R. -H; Chang. T. -Y; Tsai D. -H. (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. http//dx.doi.org/10.1016/0006-3002(63)91283-6
Checkowav, H; Pearce, NE; Kriebel, D. (2004). Research methods in occupational
epidemiology. In Research Methods in Occupational Epidemiology (2 ed.). New York,
NY: Oxford University Press.
http//dx.doi. org/10.1093/acprof:oso/9780195092424.001.0001
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.
http//dx.doi.org/10.1016/0165-1218(85)90013-8
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.
http//dx.doi.org/10.1136/oem.2003.008268
Crump, KS; Hoel, DG; Langlev, CH; Peto, R. (1976). Fundamental carcinogenic processes and
their implications for low dose risk assessment. Cancer Res 36: 2973-2979.
Csanady, GA; Denk, B; Ptitz, 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
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; HBartsch (Eds.),
Relevance to Human Cancer of N-Nitroso Compounds, Tobacco Smoke and Mycotoxins
(pp. 107-112). Lyon, France: International Agency for Research on Cancer.
Davis, FG; Erdai S; Williams, L; Bigner, D. (2006). Work exposures to animal
neurocarcinogens. Int J Occup Environ Health 12: 16-23.
http//dx. doi.org/10.1179/oeh.2006.12.1.16
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. Mutat Res 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 and Occupational Health: Current
Applications, Limitations, and Future Prospects (pp. 77-85). Westport, CT: Auburn
House.
Dellarco, VL; Generoso, WM; Sega, GA; Fowle, J. R; Jacobson-Kram, D. (1990). Review of
the mutagenicity of ethylene oxide [Review], Environ Mol Mutagen 16:85-103.
http//dx. doi. org/10.1002/em.2850160207
R-3
-------
DeMarini DM. (2000). Influence of DNA repair on mutation spectra in Salmonella [Review],
Mutat Res 450: 5-17. http://dx.doi.org/10.1016/S0Q27-5107(00)00013-0
Divine, BJ; Amanollahi, KS.Q986). Ethylene oxide and cancer [Letter], JAMA 256: 1726-1727.
Donner. EM: Wong. BA; James. RA; Prestoa 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
Eastmond, DA. (2012). Factors influencing mutagenic mode of action determinations of
regulatory and advisory agencies. Mutat Res 751: 49-63.
http//dx.doi. org/10.1016/i.mrrev.2012.04.001
Eastmond, DA; Keshava, N; Sonawane, B. (2014). Lymphohematopoietic cancers induced by
chemicals and other agents and their implications for risk evaluation: An overview
[Review], Mutat Res Rev Mutat Res 761: 40-64.
http//dx.doi. org/10.1016/i.mrrev.2014.04.001
Ehrenberg, L; Gustafssoa 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: Lund: forf
Ehrenberg„ L; Hallstrom, T. (1967). Haematologic studies on persons occupationally exposed to
ethylene oxide. In Radiosterilization of Medical Products (Budapest, 5-9 June 1967):
Proceedings Series - International Atomic Energy Agency 89 (pp. 327-334). (SM 92/26;
STI/PUB/157). Vienna, Austria: 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. http://dx.doi.org/10.1016/Q165-l 110(81)90034-8
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.
http J/dx. doi. or g/10.1016/0027- 5107(77)90017- 3
Eide, I; Zhao, C; Kumar, R; Hemminki, K; Ky, W; Swenberg„ JA. (1999). Comparison of (32)P-
postlabeling and high-resolution GC/MS in quantifying N7-(2-Hydroxyethyl)guanine
adducts. Chem Res Toxicol 12: 979-984. http://dx.doi.org/10.1021/tx9900391
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; Sepai, O;
Tang„ Y, -S. (1993). Identification of endogenous electrophiles by means of mass
spectrometric determination of protein and DNA adducts [Review], Environ Health
Perspect 99: 19-24. http//dx.doi.org/10.2307/3431452
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], Mutat Res 659:
68-76. http J/dx. doi. org/10.1016/i.mrrev.2008.03.006
R-4
-------
Farooqi. Z; Tornqvist, M; Ehrenberg, L; Nataraiaa AT. (1993). Genotoxic effects of ethylene
oxide and propylene oxide in mouse bone marrow cells. Mutat Res 288: 223-228.
http://dx. doi.org/10.1016/0027-5107(93)90088-W
Fennell TR; Macneela. JP; Morris. RW; Watsoa M; Thompsoa 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.
Fettermaa BA; Kim BS; Margolia BH; Schildcrout JS; Smith, MG; Wagner, SM; Zeiger, E.
(1997). Predicting rodent carcinogenicity from mutagenic potency measured in the Ames
Salmonella assay [Review], Environ Mol Mutagen 29: 312-322.
http://dx.doi.org/10.1002/(SICr>l 098-2280(1997^)29:3 <312: :AID-EM12>3.0.CQ;2-H
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. Arch Toxicol 66:157-163.
Filser, JG; Kessler, W; Artati, A; Erbach, E; Fallen T; Kreuzer, PE; Li Q; Lichtmannegger, J;
Numtip, W; Kleia D; Putz, C; Semder, B; Csanady, GA. (2013). Ethylene oxide in blood
of ethylene-exposed B6C3F1 mice, Fischer 344 rats, and humans. Toxicol Sci 136: 344-
358. http://dx.doi.org/10.1093/toxsci/kft218
Fleming, TR. (2010). Clinical trials: Discerning hype from substance. Ann Intern Med 153: 400-
406. http://dx.doi.org/10.7326/0003-4819-153-6-201009210-00008
FondeE MC; Costantini, AS: Ercolanelli, M; Pizzo, AM: Maltoni, SA; Quina 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.
Fost, U; Marczynski, B; Kasemana R; Peter, H. (1989). Determination of7-(2-
hydroxyethyl)guanine with gas chromatography/mass spectrometry as a parameter for
genotoxicity of ethylene oxide. Arch Toxicol Suppl 13: 250-253.
httpy/dx.doi.org/10.1007/978-3-642-74117-3 43
Fuchs, J; Wullenweber, U; Hengstler, JG; Bienfait, HG; Hilti 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.
httpy/dx.doi.org/10.1007/s002040050080
Gabriel, S; Steinhausea M; Van Gelder, R. (2013). Identification of work-related exposure to
carcinogenic substances in Germany. WIT Trans Ecol Environ 174:85-102.
httpy/dx. doi.org/10.2495/AIRl 30081
Galloway, SM; Berry, PK; Nichols, WW; Wolmaa SR; Soper, KA; Stolley, 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.
httpy/dx.doi.org/10.1016/0165-1218(86)90082-0
Gardner, MJ; Coggoa D; Pannett, B; Harris, EC. (1989). Workers exposed to ethylene oxide: A
follow up study. Occup Environ Med 46: 860-865.
httpy/dx.doi.org/10.1136/oem.46.12.860
Garry, VF; Hozier, J; Jacobs, D; Wade, RL; Gray, DG. (1979). Ethylene oxide: Evidence of
human chromosomal effects. Environ Mutagen 1:375-382.
httpy/dx.doi.org/10.1002/em.2860010410
R-5
-------
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. http://dx.doi.org/10.1002/em.2850160209
Generoso. WM: Cain. KT: Hughes. LA: Sega. GA; Bradea PW; Gosslee. DG: Shelby. MP.
(1986). Ethylene oxide dose and dose-rate effects in the mouse dominant-lethal test.
Environ Mol Mutagen 8: 1-7. http://dx.doi.org/10.1002/em.2860080102
Generoso, WM: Rutledge, JC: Cain. KT: Huuhes, LA; Downing. DJ. (1988). Mutagen-induced
fetal anomalies and death following treatment of females within hours after mating. DNA
Repair 199: 175-181. http://dx.doi.org/10.1016/Q165-l 161(88)90245-2
Godderis, L; Aka, P; Matecuca, R; Kirsch-Volders, M; Lison. 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.org/10.1016/i.tox.2005.11.021
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.org/10.1136/oem.47.4.221
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; Baan, R; Straif, K; Secretary 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/S1470-
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.
http //dx. doi. or g/10.1016/S0027- 5107(99)00026-3
Hagmar, L; Mikoczy, Z; Welinder, H. (1995). Cancer incidence in Swedish sterilant workers
exposed to ethylene oxide. Occup Environ Med 52: 154-156.
http//dx. doi. org/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. http://dx.doi.org/10.1007/BF0Q386377
Hallier, E; Langhof, T; Dannappei D; Leutbecher, M; Schroder, K; Goergens, HW; Mtiller, 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.
http //dx. doi. or g/10.1007/BF01973 3 04
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-
R-6
-------
transferase and epoxide hydrolase polymorphisms. Cancer Epidemiol Biomarkers Prev
16: 796-802. httpy/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. Mutat Res 304: 229-234.
http J/dx. doi. or g/10.1016/0027- 5107(94)90215-1
Hill, AB. (1965). The environment and disease: Association or causation? Proc R Soc Med 58:
295-300.
Hoenerhoff MJ; Hong„ HH; Ton, T, -V; 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 J/dx. doi. or g/10.1177/0192623 3 093 51726
Hogstedt, B; Gullberg„ B; Hedner, K; Kolnig„ A, -M; Mitelman, F; Skerfving„ S; Widegreq B.
(1983). Chromosome aberrations and micronuclei in bone marrow cells and peripheral
blood lymphocytes in humans exposed to ethylene oxide. Hereditas 98: 105-113.
httpy/dx.doi.org/10.1111/i. 1601-5223.1983.tb00585.x
Hogstedt, C; Aringer, L; Gustavsson, 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 och Halsa 49: 1-
32.
Hogstedt, C; Aringer, L; Gustavsson, A. (1986). Epidemiologic support for ethylene oxide as a
cancer-causing agent. JAMA 255: 1575-1578.
httpy/dx.doi.org/10.1001/iama. 1986.03370120053022
Hogstedt, C;Malmqvist, N; Wadman, B. (1979a). Leukemia in workers exposed to ethylene
oxide. JAMA 241: 1132-1133. httpy/dx.doi.org/10.1001/iama. 1979.03290370036024
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. httpy/dx.doi.org/10.1136/oem.36.4.276
Hogstedt, LC. (1988). Epidemiological studies on ethylene oxide and cancer: An updating. Lyon,
France.
Hong„ H, -HL; Houle, CD; Ton, T, -VT; 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.
httpy/dx.doi.org/10.1080/01926230601063839
Hornung, RW: Greife, AL; Stavner, 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 JIndMed25: 825-836.
httpy/dx.doi.org/10.1002/aiim.4700250607
Houle, CD: Tog T, - VT: Clavtoa N: Huff, J: Hong, H, -HL: Sills, RC. (2006). Frequent p53 and
H-ras mutations in benzene- and ethylene oxide-induced mammary gland carcinomas
from B6C3F1 mice. Toxicol Pathol 34: 752-762.
httpy/dx.doi.org/10.1080/01926230600935912
Howard, J. (2013). Minimum latency & types or categories of cancer (pp. 1-9). Atlanta, GA:
Centers for Disease Control and Prevention: WTC Health Program.
httpsy/www. cdc.gov/wtc/pdfs/wtchp minlatcancer2013-05-01.pdf
R-7
-------
Hu. JJ; Smith. TR; Miller. MS; Lohmaa K; Case. LP. (2002). Genetic regulation of ionizing
radiation sensitivity and breast cancer risk. Environ Mol Mutagen 39:208-215.
http://dx.doi.org/10.1002/em. 10058
Huang. C. -C: Shih. W. -C: Wu. C. -F: Chea M. -F: Chea Y. -L: Lin. Y. -H: Wu. K. -Y. (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 Commun Mass Spectrom 22: 706-710.
http//dx.doi. org/10.1002/rcm.3414
Huang. C. -C: Wu. C. -F: Shih. W. -C: Chea M. -F: Chea C. -Y; Chiea Y. -C: Liou. S. -H:
Chiang„ S, -Y; Wu, K, -Y. (2011). Comparative analysis of urinary N7-(2-
hydroxyethyl)guanine for ethylene oxide- and non-exposed workers. Toxicol Lett 202:
237-243. httpy/dx.doi.org/10.1016/i.toxlet.2011.02.009
IARC (International Agency for Research on Cancer). (1994a). IARC monographs on the
evaluation of carcinogenic risks to humans: Ethylene (pp. 45-71). Lyon, France:
International Agency for Research on Cancer.
http//mono graphs. iarc. fr/ENG/Monographs/vol60/
IARC (International Agency for Research on Cancer). (1994b). IARC monographs on the
evaluation of carcinogenic risks to humans: Ethylene oxide (pp. 73-159). Lyon, France:
International Agency for Research on Cancer.
httpy/mo no graphs, iarc. fr/ENG/Monographs/vol60/
IARC (International Agency for Research on Cancer). (2008). IARC monographs on the
evaluation of carcinogenic risks to humans: 1,3-butadiene, ethylene oxide and vinyl
halides (vinyl fluoride, vinyl chloride and vinyl bromide) [IARC Monograph], Lyon,
France: International Agency for Research on Cancer.
httpy/mo no graphs, iarc. fr/ENG/Monographs/vol97/
Jacob, P. HI; Abu Raddaha, AH; Dempsey, D; Havel C; Peng„ M; Yu, L; Benowitz, NL. (2013).
Comparison of nicotine and carcinogen exposure with water pipe and cigarette smoking.
Cancer Epidemiol Biomarkers Prev 22: 765-772. httpy/dx.doi.org/10.1158/1055-
9965.EPI-12-1422
Jarabek, AM; Pottenger, LH; Andrews, LS; Casciano, D; Embry, MR; Kim, JH; Prestoa RJ;
Reddv, MY; 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.
httpy/dx.doi.org/10.1080/10408440903164155
Jenssea D; Ramei C. (1980). The micronucleus test as part of a short-term mutagenicity test
program for the predicition of carcinogenicity evaluated by 143 agents tested. Mutat Res
75: 191-202. httpy/dx.doi.org/10.1016/0165-1110(80)90014-7
Johansoa G; Filser, JG. (1992). Experimental data from closed chamber gas uptake studies in
rodents suggest lower uptake rate of chemical than calculated from literature values on
alveolar ventilation. Arch Toxicol 66: 291-295. http://dx.doi.org/10.1007/BF02307176
Joyner, RE. (1964). Chronic toxicity of ethylene oxide. Arch Environ Occup Health 8: 700-710.
httpy/dx.doi.org/10.1080/00039896.1964.10663741
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. httpy/dx.doi.org/10.1002/em. 10167
R-8
-------
Kelsev. KT; Wiencke. JK; Eisea 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.
Kensler. TW; Ng, D; Carmella. SG: Chea M: Jacobsoa LP: Munoz, A: Egner, PA: Chea JG:
Qiaa GS: Chea TY: Fahey, JW: Talalay, P; Groopmaa JD: Yuaa J, -M; 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 y/dx. doi.org/10.1093/carcin/bgr229
Keshava, N; Jinot, J: Sonawane, B. (2006a). An evaluation of mutagenic mode of action for
carcinogenicity: Ethylene oxide [Abstract], Environ Mol Mutagen 47: 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, H.-J: Korallus. U. (1990). A multicentre mortality study of
workers exposed to ethylene oxide. Br J Ind Med 47: 182-188.
Kim E, -A; Lee, H -E; Ryu, H -W; Park, S, -H; Kang„ -S. (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
Kiraa 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/EDE0b013e3181f4cc0f
Kirmaa CR; Sweeney, LM; Teta, MJ; Sielkea 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. httpy/dx.doi.org/10.1111/i.0272-4332.2004.00517.x
Kleinbaum DG. (1994). Logistic regression: A self-learning text. In Logistic Regression: A Self-
Learning Text. New York, NY: Springer-Verlag New York.
http y/dx. doi. or g/10.1007/978-1-4757-4108-7
Kligermaa AD; Erexsoa 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
Kloth, S; Baur, X; Goea T; Budnik, LT. (2014). Accidental exposure to gas emissions from
transit goods treated for pest control. Environmental Health 13: 110.
httpy/dx.doi.org/10.1186/1476-069X-13-110
Koepke, SR; Kroeger-Koepke, MB; Bosaa W; Thomas, BJ; Alvord, WG: Michejda, CJ. (1988).
Alkylation ofDNAin 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.
httpy/dx.doi.org/10.1016/0167-8817(85)90053-7
Kolmaa A; Chovanec, M. (2000). Combined effects of gamma-radiation and ethylene oxide in
human diploid fibroblasts. Mutagenesis 15: 99-104.
http y/dx. doi.org/10.1093/mutage/15.2.99
R-9
-------
Kolmaa A; Chovanec. M; Osterman-Golkar. S. (2002). Genotoxic effects of ethylene oxide,
propylene oxide and epichlorohydrin in humans: Update review (1990-2001) [Review],
Mutat Res 512: 173-194. http://dx.doi.org/10.1016/S1383-5742(02)00067-4
Kolmaa A: N aslund. M. (1987). Mutagenicity testing of ethylene oxide in Escherichia coli
strains with different repair capacities. Environ Mol Mutagen 10:311-315.
http://dx.doi.org/10.1002/em.2850100310
Kolmaa A; N aslund. M; Calleman, CJ. (1986). Genotoxic effects of ethylene oxide and their
relevance to human cancer [Review], Carcinogenesis 7: 1245-1250.
http://dx.doi.Org/10.1093/carcin/7.8.1245
Konduracka, E; Krzemieniecki, K; Gajos. G. (2014). Relationship between everyday use
cosmetics and female breast cancer [Review], Pol Arch Med Wewn 124: 264-269.
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.
http://dx.doi.Org/10.1093/carcin/16.3.483
Lambert, B; Anderssoa B; Bastlova, T; Hou. S, -M; Hellgrea 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. http://dx.doi.org/10.2307/3431943
Langholz, B; Richardsoa 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
Laurent, CH; 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. http://dx.doi.org/10.1007/BF0Q378726
Leclercq, L; Laurent, C; De Pauw, E. (1997). High-performance liquid
chromatography/electrospray mass spectrometry for the analysis of modified bases in
DNA: 7-(2-hydroxyethyl)guanine, the major ethylene oxide-DNA adduct. Anal Chem 69:
1952-1955. http://dx.doi.org/10.1021/ac9607673
Lerda, D; Rizzi, R. (1992). Cytogenetic study of persons occupationally exposed to ethylene
oxide. Mutat Res 281: 31-37. httpy/dx.doi.org/10.1016/0165-7992(92)90033-E
Leutbecher, M; Langhof T; Peter, H; Fost, U. (1992). Ethylene oxide: Metabolism in human
blood and its implication to biological monitoring. Arch Toxicol 15: 289.
httpy/dx.doi.org/10.1007/978-3-642-77260-3 38
Lewis, SE; Barnett, LB; Feltoa C; Johnsoa FM; Skow, LC; Cacheiro, N; Shelby, MP. (1986).
Dominant visible and electrophoretically expressed mutations induced in male mice
exposed to ethylene oxide by inhalation. Environ Mol Mutagen 8: 867-872.
httpy/dx.doi.org/10.1002/em.2860080609
Li, F; Segal A; Solomoa JJ. (1992). In vitro reaction of ethylene oxide with DNA and
characterization of DNA adducts. Chem Biol Interact 83:35-54.
http y/dx. doi. or g/10.1016/0009-2797(92)90090- 8
Lia J, -S; Chuang„ KT; Huang„ MS; Wei, K, -M. (2007). Emission of ethylene oxide during
fiying of foods in soybean oil. Food Chem Toxicol 45: 568-574.
httpy/dx.doi.org/10.1016/i.fct.2006.10.002
R-10
-------
Lindberg, HK; Falck. GC; Catalaa J: Santonea 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: Chea YH; Yang. T; Hsieh. LL; Chea CJ: Wu. TN. (1999). Increased
chromosome-type chromosome aberration frequencies asbiomarkers of cancer risk in a
blackfoot endemic area. Cancer Res 59: 1481-1484.
Lorenti Garcia, C; Darroudi, F; Tates, AD; Natarajaa 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/S 1383-5718(01)00149-8
Lucas. LJ: Teta. MJ. (1996). Breast cancer and ethylene oxide exposure [Letter], Int J Epidemiol
25: 685-686. http//dx.doi.org/10.1093/iie/25.3.685
Lynch. DW; Lewis, TR; Moormaa WJ; Burg„ JR; Gulati. DK; Kaur, P; Sabharwal, PS. (1984).
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. http//dx.doi.org/10.1016/0041-008X(84)90031-0
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.CQ;2-E
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.lQ16/S1383-
5718(99)00129-1
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; Valgjmigli, 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.
Marsdea DA; Jones, DJ; Brittoa RG; Ognibene, T; Ubick, E; Johnsoa GE; Farmer, PB; Browa
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.l 158/0008-5472.CAN-08-
4233
Marsdea DA; Jones, DJ; Lamb, JH; Tompkins, EM; Farmer, PB; Browa 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/10.102 l/tx600264t
Mayer, J; Warburtoa 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
R-ll
-------
ethylene oxide exposed workers and controls. Mutat Res 248: 163-176.
httpy/dx.doi.org/10.1016/0027-5107(91)90098-9
Mazon, G; Philippin, G; Cadet J: Gasparutto, D; Fuchs, RP. (2009). The alkyltransferase-like
ybaZ gene product enhances nucleotide excision repair of 0(6)-alkylguanine adducts in
E. coli. DNARepair 8: 697-703. http://dx.doi.ore/10.1016/j.dnarep.2009.01.022
Mazon, G; Philippic G; Cadet J: Gasparutto, D; Modesti, M; Fuchs, RP. (2010).
Alkyltransferase-like protein (eATL) prevents mismatch repair-mediated toxicity induced
by 06-alkylguanine adducts in Escherichia coli. ProcNatl Acad Sci USA 107: 18050-
18055. httpy/dx.doi.org/10.1073/pnas. 1008635107
McCarthy, MC; O'Brien, TE; Charrier, JG; Hafiier, 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.ore/10.1289/ehp. 11861
Mikoczy, Z; Tinnerberu, 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.ore/10.3390/iierph8062009
Morean, RW; Claxton, KW; Divine, BJ; Kaplan, SD; Harris, VB. (1981). Mortality among
ethylene oxide workers. J Occup Environ Med 23: 767-770.
http://dx.doi. ore/10.1097/00043764-198111000-00011
Mosavi-Jarrahi, A; Mohaeheehi, MA; Kalaehchi, B; Mousavi-Jarrahi, Y; Kolahj 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. Korhaz- es orvostechnika 33: 17-22.
Naev, K; Adany, R; Szucs, S; Adam, B. (2013). Susceptibility of lung epithelial cells to
alkylating genotoxic insult. Environ Mol Mutagen 54: 682-689.
http://dx.doi. ore/10.1002/em. 21800
Natarajan, AT; Preston, RJ; Dellarco, V; Ehrenbere„ L; Generoso, W; Lewis, S; Tates, AD.
(1995). Ethylene oxide: Evaluation of genotoxicity data and an exploratory assessment of
genetic risk [Review], Mutat Res 330: 55-70. http://dx.doi.ony i 0.1016/0027-
5107(95)00036-1
NCI (National Cancer Institute). (1978). Bioassay of 1,2-dichloroethane for possible
carcinogenicity (CAS No. 107-06-2). In National Cancer Institute Carcinogenesis
Technical Report Series, no 55. (NCI-CG-TR-55). Bethesda, MD: National Institute of
Health, National Cancer Institute, Division of Cancer Cause and Prevention,
Carcinogenesis Testing Program.
Nivard, MJ; Czene, K; Seeerback, D; Voeei EW. (2003). Mutagenic activity of ethylene oxide
and propylene oxide under XPG proficient and deficient conditions in relation toN-7-(2-
hydroxyalkyl)guanine levels in Drosophila. Mutat Res 529: 95-107.
http J/dx. doi. or e/10.1016/S0027- 5107(03)00111 - 8
Normaa 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.
http://dx.doi.ore/10.1093/iie/24.2.276
NRC (National Research Council). (1994). Science and judgment in risk assessment. In Science
and Judgment in Risk Assessment (pp. 672). Washington, DC: National Academy Press.
http J/dx. doi. or e/10.17226/2125
R-12
-------
NRC (National Research Council). (2011). Review of the Environmental Protection Agency's
draft IRIS assessment of formaldehyde. Washington, DC: The National Academies Press.
http J/dx. doi. or g/10.17226/13142
NTP (National Toxicology Program). (1987). Toxicology and carcinogenesis studies of ethylene
oxide (CAS no 75-21-8) in B6C3F1 mice (inhalation studies). Natl Toxicol Program
Tech Rep Ser326: 1-114.
Nygrea J; Cedervalj B; Eriksson, S; Dusinska, M; Kolmaa A. (1994). Induction of DNA strand
breaks by ethylene oxide in human diploid fibroblasts. Environ Mol Mutagen 24: 161-
167. http://dx.doi.org/10.1002/em.2850240304
Oesch, F; Heagstler, JG; Araad, 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.
httpy/dx.doi.org/10.1007/sl 1356-011-0550-0
Olsea GW; Lacy, SE; Bodner, KM; Chau, M; Arceneaux, TG; Cartmilj 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. httpy/dx.doi.org/10.1136/oem.54.8.592
Ong. T; Bi. Ft, -K: Xing. S: Stewart. J: Moormaa 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.
http y/dx. doi. org/10.1002/em.2850220306
Otteneder, M; Lutz, WK. (1999). Correlation ofDNAadduct levels with tumor incidence:
Carcinogenic potency of DNA adducts [Review], Mutat Res 424: 237-247.
http y/dx. doi. or g/10.1016/S0027- 5107(99)00022-6
Parsons, BL; Manjanatha, MB; Myers, MB; McKim, KL; Waag„ 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 [Abstract], Environ Mol Mutagen 53:
S62.
Parsons, BL; Manjanatha, MG; Myers, MB; McKim, KL; Shelton, SD; Wang„ Y; Gollapudi BB;
Moore, NP; Haber, LT; Moore, MM. (2013). Temporal changes in K-ras mutant fraction
in lung tissue of big blue B6C3F mice exposed to ethylene oxide. Toxicol Sci 136: 26-38.
httpy/dx.doi.org/10.1093/toxsci/kftl90
Pauwels, W; Veulemans, H. (1998). Comparison of ethylene, propylene and styrene 7,8-oxide in
vitro adduct formation onN-terminal valine in human haemoglobin and on N-7-guanine
in human DNA. Mutat Res 418: 21-33. httpy/dx.doi.org/10.1016/S1383-5718(98)00106-
5
Pero, RW; Widegrea B; Hogstedt, B; Mitel man, F. (1981). In vivo and in vitro ethylene oxide
exposure of human lymphocytes assessed by chemical stimulation of unscheduled DNA
synthesis. Mutat Res 83: 271-289. httpy/dx.doi.org/10.1016/0027-5107(81)90011-7
Philippia G; Cadet, J; Gasparutto, D; Mazoa G; Fuchs, RP. (2014). Ethyleae oxide aad
propyleae oxide derived N7-alkylguaaiae adducts are bypassed accurately ia vivo. DNA
Repair 22: 133-136. http://dx.doi.Org/10.1016/i.daarep.2014.08.001
R-13
-------
Piloa D: Roberts. AE; Rickert. DE. (1988). Effect of glutathione depletion on the irreversible
association of acrylonitrile with tissue macromolecules after oral administration to rats.
Toxicol Appl Pharmacol 95: 311-320. httpy/dx.doi.org/10.1016/0041-008X(88)90167-6
Prentice. RL. (1982). Covariate measurement errors and parameter estimation in a failure time
regression model. Biometrika 69: 331-342. http://dx.doi.Org/10.1093/biomet/69.2.331
Prentice, RL. (1985). Relative risk regression-analysis of epidemiologic data. Environ Health
Perspect 63: 225-234. http://dx.doi.oru/10.2307/3430050
Preston. RJ. (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
Prestoa RJ: Fennell. TR; Leber. AP; Sielkea RL. Jr; Swenberg, JA. (1995). Reconsideration of
the genetic risk assessment for ethylene oxide exposures. Environ Mol Mutagen 26: 189-
202. http://dx.doi.org/10.1002/em.2850260303
Rapoport. LA. (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; Steea A. -M; Wong„ BA; James, A; Prestoa 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. http://dx.doi.org/10.1093/mutage/geh017
Ribeiro, LR; Rabello-Gay, MN; Salvadori, DMF; Pereira, CAB; Becak, W. (1987). Cytogenetic
effects of inhaled ethylene oxide in somatic and germ cells of mice. Arch Toxicol 59:
332-335. httpy/dx.doi.org/10.1007/BF00295085
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. http://dx.doi.org/10.1016/Q165-l 161(94)90035-3
Robins, JM; Blevins, D; Ritter, G; Wulfsoha 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
Rothmaa K; Greenland, S; Lash, T. (2008). Modern epidemiology. In Modern Epidemiology (3
ed.). Philadelphia, PA: Lippincott, Williams & Wilkins.
Rothmaa KJ. (1986). Modern epidemiology. In Modern Epidemiology (1 ed.). Boston, MA:
Little ,Brown & Co.
Rothmaa KJ; Greenland, S. (1998). Modern epidemiology. In Modern Epidemiology (2 ed.).
Philadelphia, PA: Lippincott-Raven.
Rusva I: Asakura, S; Li, Y; Kosyk, O; Koc, H; Nakamura, J: Uptoa 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.Org/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: U.S. Environmental Protection Agency, Science Advisory Board.
http//yosemite.epa.gov/sab/sabproduct.nsf/368203197al5308a852574ba005bbd01/5D66
1BC118B527A3852573B80068C97B/$File/EP A-SAB-08-004-unsigned.pdf
R-14
-------
SAB (Science Advisory Board). (2015). Science Advisory Board Review of the EPAs evaluation
of the inhalation carcinogenicity of ethylene oxide: Revised external review draft -
August 2014 [EPA Report], (EPA-SAB-15-012). Washington, DC: U.S. Environmental
Protection Agency, Science Advisory Board.
http sy/vo Semite.epa.gov/sab/sabproduct.nsf/fedrgstr activites/BD2B2DB4F84146A5852
57E9A0070E655/$File/EPA-SAB-15-012+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. http://dx.doi.org/10.1016/0021-
9673(95)00545-X
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 JInd Med 12: 625-637.
http//dx.doi.org/10.1002/aiim.4700120515
Sarto, F; Cominato, I; Pinton, AM; Brovedanj PG; Facciolj CM; Bianchi, V; Levis, AG.
(1984a). Cytogenetic damage in workers exposed to ethylene oxide. Mutat Res 138: 185-
195. http://dx.doi.org/10.1016/0165-1218(84)90043-0
Sarto, F; Cominato, I; Pinton, AM; Brovedanj PG; Facciolj CM; Bianchi, V; Levis, AG.
(1984b). Workers exposed to ethylene oxide have increased incidence of sister chromatid
exchange. IARC SciPubl 59:413-419.
Sarto, F; Tomaniq R; Giacomellj L; Iannini, 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.
http y/dx. doi. or g/10.1016/0165-7992(90)90083 -V
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. httpy/dx.doi.org/10.5271/sjweh. 1733
Schulte, PA; Boeniger, M; Walker, JT; Schober, SE; Pereira, MA; Gulatj 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.
httpy/dx.doi.org/10.1016/S0165-1218(10)80003-5
Segerback, D. (1983). Alkylation of DNA and hemoglobin in the mouse following exposure to
ethene and ethene oxide. Chem Biol Interact 45: 139-151.
http y/dx. doi. or el 10.1016/0009-2797(83)90064-9
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 alkylation by ethylene oxide and some mono-substituted epoxides
[Review], IARC SciPubl 125:37-47.
Shea J; Kessler, W; Denk, B; Filser, JG. (1989). Metabolism and endogenous production of
ethylene in rat and man. Arch Toxicol Suppl 13: 237-239. httpy/dx.doi.org/10.1007/978-
3-642-74117-3 39
Shj Q; Wang„ L, -E; Bondy, ML; Brewster, A; Singletary, SE; Wej Q. (2004). Reduced DNA
repair of benzo[a]pyrene diol epoxide-induced adducts and common XPD
polymorphisms in breast cancer patients. Carcinogenesis 25:1695-1700.
http y/dx. doi. org/10.1093/carcin/bghl 67
R-15
-------
Shore. RE; Gardner. MJ; Pannett. B. (1993). Ethylene oxide: An assessment of the
epidemiological evidence on carcinogenicity. Occup Environ Med 50: 971-997.
http://dx.doi.Org/10.l 136/oem.50.11.971
Sielkea RL. Jr: Valdez-Flores. C. (2009a). Life-table calculations of excess risk for incidence
versus mortality: Ethylene oxide case study. Regul Toxicol Pharmacol 55: 82-89.
http://dx.doi.Org/10.1016/i.vrtph.2009.06.003
Sielkea RL; Valdez-Flores, C. (2009b). Calculating excess risk with age-dependent adjustment
factors and cumulative doses: Ethylene oxide case study. Regul Toxicol Pharmacol 55:
76-81. http//dx. do i. org/10.1016/i.yrtph.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 ofB6C3Fl lacl transgenic mice following
inhalation exposure. MutatRes 391: 153-164. http://dx.doi.org/10.lQ16/S1383-
5718(97)00063-6
Smith-Bindmaa 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.
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. http//dx.doi.org/10.1016/0041-008X(84)90081-4
StHelea G; Jacob, P, III; Peng„ M; Dempsey, DA; Hammond, SK; Benowitz, NL. (2014).
Intake of toxic and carcinogenic volatile organic compounds from secondhand smoke in
motor vehicles. Cancer Epidemiol Biomarkers Prev 23: 2774-2782.
http//dx.doi.org/10.1158/1055-9965.EP I-14-0548
Stayner, L; Steenland, K; Greife, A; Hornung„ R; Haves, RB; Nowlia S; Morawetz, J;
Ringenburg„ V; Elliot, L; Halperia W. (1993). Exposure-response analysis of cancer
mortality in a cohort of workers exposed to ethylene oxide. Am J Epidemiol 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.
http//dx.doi.org/10.1093/aie/154.5.451
Steenland, K; Deddens, J; Salvaa 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, 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/10.1097/0 l.ede. 0000100287.45004. e7
Steenland, K; Deddens, JA; 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. http//dx. doi.org/10.5271/siweh. 508
Steenland, K; Seals, R; Kleia M; Jinot, J; Kaha 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
R-16
-------
Steenland. K; Stavner. L. (1993). An epidemiological study of workers potentially exposed to
ethylene oxide [Letter], Br JIndMed 50: 1125-1126.
Steenland, K; Stavner, 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; Stavner, L; Greife, A; Halperin, W; Haves, R; Hornung„ R; Nowlin, S. (1991).
Mortality among workers exposed to ethylene oxide. N Engl J Med 324: 1402-1407.
http J/dx. doi. or g/10.1056/NEJM199105163242004
Steenland, K; Whelan, E; Deddens, J; Stavner, 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. http://dx.doi.Org/10.1023/A:1024891529592
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. Gefahrst Reinhalt Luft 72: 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, GMH; Burns, C; Teta, JM; Bodner, K; Keenan, 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/JQM.0b013e3181a2ca20
Swaea GMH: Slangeq JMM; Ott, MG: Kusters, E; Van Den Langenbergh. G: Arends, JW;
Zober, A. (1996). Investigation of a cluster of ten cases of Hodgkiris disease in an
occupational setting. Int Arch Occup Environ Health 68: 224-228.
http//dx.doi.org/10.1007/BF00381432
Swenberg„ JA; Fryar-Tita, E; Jeong„ Y, -C; Boysen, G; Starr, T; Walker, YE; 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/10.102 l/tx700408t
Swenberg„ JA; Ham, A; Koc, H; Morinello, E; Ranasinghe, A; Tretyakova, N; Upton, PB; Wu,
K, -Y. (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: S130-S145.
http//dx. doi.org/10.1093/toxsci/kfq371
Tateo, F; Bononj M. (2006). Determination of ethylene chlorohydrin as marker of spices
fumigation with ethylene oxide. J Food Compos Anal 19: 83-87.
http//dx. doi. org/10.1016/i.ifca.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. Mutat Res 329: 63-77. http//dx.doi.org/10.1016/0027-
5107(95)00018-E
Tates, AD: Grummt, T; Tornqvist, M; Farmer, PB: van Dam. FJ; van Mossei H; Schoemaker,
HM; Osterman-Golkar, S; Uebei C; Tang. YS; Zwinderman, AH; Natarajan, AT;
Ehrenberg. L. (1991). Biological and chemical monitoring of occupational exposure to
R-17
-------
ethylene oxide. Mutat Res 250: 483-497. http://dx.doi.org/10.1016/0027-5107(91)90205-
3
Tates, AD; van Dam, FJ; Natarajaa AT; van Teylingea CMM; de Zwart, FA; Zwindermaa AH;
van Sittert. NJ; Nilsea A; Nilsea OG; Zahlsea K; Magnussoa A. -L; Tornqvist, M.
(1999). Measurement of HPRT mutations in splenic lymphocytes and haemoglobin
adducts in erythrocytes of Lewis rats exposed to ethylene oxide. Mutat Res 431: 397-415.
http J/dx. doi. or g/ 10.1016/S0027- 5107(99)00182-7
Teta, MJ; Bensoa LO; Vitale, JN. (1993). Mortality study of ethylene oxide workers in chemical
manufacturing: A 10 year update. Br J Ind Med 50: 704-709.
http//dx.doi.org/10.1136/oem.50.8.704
Teta. MJ; Sielkea R.L. Jr; Valdez-Flores. C.(1999). Ethylene oxide cancer risk assessment based
on epidemiological data: Application of revised regulatory guidelines. Risk Anal 19:
1135-1155. http//dx.doi.org/10.1111/i. 1539-6924.1999.tb01134.x
Thiess. AM; Frentzel-Bevme. R; Link. R; Stocker. WG. (1982). Mortality study on employees
exposed to alkylene oxides (ethylene oxide/propylene oxide) and their derivatives. In
Prevention of Occupational Cancer - International Symposium. Geneva, Switzerland:
International Labour Office.
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. http J/dx. doi. org/10.1053/paor. 1999.0182
Tompkins, EM; Jones, DJL; Lamb, JH; Marsdea DA; Farmer, PB; Browa K. (2008).
Simultaneous detection of five different 2-hydroxyethyl-DNA adducts formed by
ethylene oxide exposure, using a high-performance liquid chromatography/electro spray
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; Browa K. (2006). Weak mutagenicity
of DNA adducts derived from ethylene oxide exposure [Abstract], Mutagenesis 21: 292.
Tompkins, EM; McLuckie, KIE; Jones, DJ; Farmer, PB; Browa 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.org/10.1016/j.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. http//dx.doi.org/10.1007/978-1-4757-9480-
9 36
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. http://dx.doi.org/10.5271/sjweh.1829
Tretvakova, N; Goggja M; Sangaraiu, D; Janis, G. (2012). Quantitation of DNA adducts by
stable isotope dilution mass spectrometry. Chem Res Toxicol 25:2007-2035.
http//dx.doi.org/10.1021/tx3002548
R-18
-------
Tucker. JD; Xu. J; Stewart. J; Baciu PC; Ong. T. -M. (1986). Detection of sister chromatid
exchanges induced by volatile genotoxicants. Teratog Carcinog Mutagen 6: 15-21.
http y/dx. doi. or g/10.1002/tcm. 1770060103
U.S. EPA (U.S. Environmental Protection Agency). (1985). Health assessment document for
ethylene oxide: Final report. (EPA-600/8-84-009F). Research Triangle Park, NC:
Environmental Criteria and Assessment Office, Office of Health and Environmental
Assessement, Office of Research and Development, U.S. Environmental Protection
Agency.
https://nepis.epa.gov/Exe/ZvNET.exe/20009FCT. TXT?ZyActionD=ZyDocument&C Kent
=EPA&Index= 1981+Thru+1985 &Docs=&Query=&Time=&EndTime=&SearchMethod
=l&TocRestrict=n&Toc=&TocEntrv=&OField=&OFieldYear=&OFieldMonth=&QFiel
dDay=&IntOFieldOp=0&ExtQF ieldOp=0&XmlQuery=&F ile=D%3 A%5C zyfiles%5CIn
dex%20Data%5C81thru85%5CTxt%5C00000002%5C20009FCT.txt&User=ANONYM
OUS&Password=anonvmous&SortMethod=h%7C-
&MaximumDocume nts= 1 &F uz zyDe gree=0& Ima geQ uality= r75g8/r75g8/xl50yl50gl6/i
425&Displav=hpfr&DefSeekPage=x&SearchBack=ZvActionL&Back=ZyActionS&Bac
kDesc=Results%20page&MaximumPages=l &ZyEntry=l &SeekPage=x&ZyPURL
U.S. EPA (U.S. Environmental Protection Agency). (2005a). Guidelines for carcinogen risk
assessment [EPA Report] (pp. 1-166). (EPA/630/P-03/001F). Washington, DC: U.S.
Environmental Protection Agency, Risk Assessment Forum.
http://www2.epa.gov/osa/guidelines-carcinogen-risk-assessment
U.S. EPA (U.S. Environmental Protection Agency). (2005b). Supplemental guidance for
assessing susceptibility from early-life exposure to carcinogens (pp. 1-125). (EPA/630/R-
03/003F). Washington, DC: U.S. Environmental Protection Agency, Risk Assessment
Forum. https://www3.epa.gov/airtoxics/childrens supplement final.pdf
U.S. EPA (U.S. Environmental Protection Agency). (2006a). Evaluation of the carcinogenicity
of ethylene oxide: External review draft [EPA Report], (EPA/63 5/R-06/003).
Washington, DC: U.S. Environmental Protection Agency, Office of Research and
Development, National Center for Environmental Assessment.
http://cfpub.epa. gov/ncea/cfm/recordisplay.cfm?deid=l 57664
U.S. EPA (U.S. Environmental Protection Agency). (2006b). U.S. Environmental Protection
Agency peer review handbook 3rd edition (3 ed.). (EPA/100/B-06/002). Washington,
DC: U.S. Environmental Protection Agency, Science Policy Council.
https://www.epa.gov/sites/production/files/2015-
09/documents/peer review handbook 2006 3rd editioapdf
U.S. EPA (U.S. Environmental Protection Agency). (2012). Benchmark dose technical guidance.
(EPA/100/R-12/001). Washington, DC: U.S. Environmental Protection Agency, Risk
Assessment Forum. https/Avww. epa. gov/risk/benchmark-dose-technical- guidance
U.S. EPA (U.S. Environmental Protection Agency). (2013 a). 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) [EPA Report],
(EPA/635/R-13/128b). Washington, DC: U.S. Environmental Protection Agency, Office
of Research and Development, National Center for Environmental Assessment.
https://nepis.epa. gov/Exe/ZvPURL.cgi9Dockev=Pl 00ICKB.txt
U.S. EPA (U.S. Environmental Protection Agency). (2013b). Evaluation of the inhalation
carcinogenicity of ethylene oxide (CASRN 75-21-8): In support of summary information
R-19
-------
on the Integrated Risk Information System (IRIS) [EPA Report], (EPA/635/R-13/128a).
Washington, DC: U.S. Environmental Protection Agency, Office of Research and
Development, National Center for Environmental Assessment.
https//nepis.epa.gov/Exe/ZyPURL.cgi?Dockev=P100ICFH.txt
U.S. EPA (U.S. Environmental Protection Agency). (2014a). Evaluation of the inhalation
carcinogenicity of ethylene oxide (Revised Aug 2014 external review draft) [EPA
Report], (EPA/635/R-14/194A). Washington, DC: U.S. Environmental Protection
Agency, Office of Research and Development, National Center for Environmental
Assessment, https://cfbub.epa. gov/ncea/iris drafts/recordisplay.cfm?deid=282012
U.S. EPA (U.S. Environmental Protection Agency). (2014b). Evaluation of the inhalation
carcinogenicity of ethylene oxide (Revised Aug 2014 external review draft): Appendices
[EPA Report], (EPA/635/R-14/194B). Washington, DC: U.S. Environmental Protection
Agency, Office of Research and Development, National Center for Environmental
Assessment, https://cfbub.epa. gov/ncea/iris drafts/recordisplay.cfm?deid=282012
Uziej M; Munro, NB; Katz, DS; Vo-Dinh, T; Zeighami, EA; Waters, MP; Griffith, JD. (1992).
DNA adduct formation by 12 chemicals with populations potentially suitable for
molecular epidemiological studies [Review], Mutat Res 277: 35-90.
http//dx.doi.org/10.1016/0165-1110(92)90025-5
Valdez-Flores, C; Sielken, RL, Jr. (2013). Misinterpretation of categorical rate ratios and
inappropriate exposure-response model fitting can lead to biased estimates of risk:
Ethylene oxide case study. Regul Toxicol Pharmacol 67: 206-214.
http//dx.doi.org/10.1016/i.yrtph.2013.07.011
Valdez-Flores, C; Sielken, RL, Jr; 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.Org/10.1016/i.vrtph.2009.10.001
Valdez-Flores, C; Sielken, RL, Jr; Teta, MJ. (2011). Quantitative cancer risk assessment for
ethylene oxide inhalation in occupational settings. Arch Toxicol 85: 1189-1193.
http Z/dx. doi. or g/10.1007/s00204-011 -0669-2
van Balen, 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; 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. http//dx.doi.org/10.2307/3431453
van Delft, JHM; van Winden, MJM; 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.
http//dx.doi.org/10.1093/carcin/15.9.1867
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/S0Q27-
5107(99)00208-0
R-20
-------
van Sittert. NJ; de Jong. G. (1985). Biomonitoring of exposure to potential mutagens and
carcinogens in industrial populations. Food Chem Toxicol 23:23-31.
http//dx.doi.org/10.1016/0278-6915(85)90216-9
van Wijngaardeq 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.ore/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: 65-77.
http://dx.doi.org/10.1002/em.2850110108
Vogei EW; Natarajan, AT. (1995). DNA damage and repair in somatic and germ cells in vivo
[Review], Mutat Res 330: 183-208. httpZ/dx.doi.org/10.1016/0027-5107(95)00040-P
Vogei 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. http//dx.doi.org/10.1002/(SICI)1098-
2280(1997)29:2<124::AID-EM3>3.0.CQ;2-E
Vogei EW; Nivard, MJ. (1998). Genotoxic effects of inhaled ethylene oxide, propylene oxide
and butylene oxide on germ cells: Sensitivity of genetic endpoints in relation to dose and
repair status. Mutat Res 405: 259-271. http://dx.doi.org/10.1016/S0027-5107(98)00143-2
Walker, VE; Fennelj TR Boucheron, JA; Fedtke, N; Ciroussei 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], DNARepair 233: 151-164. http//dx.doi.org/10.1016/0027-5107(90)90159-2
Walker, VE; Fennelj 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. http//dx.doi.org/10.2307/3431451
Walker, VE; Fennelj TR Upton, PB; Skopek, TR Prevost, V; Shuker, PEG; Swenberg. JA.
(1992). 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; Sisk, SC; Upton, PB; Wong„ BA; Recio, L. (1997). In vivo mutagenicity of
ethylene oxide at the hprt locus in T-lymphocytes ofB6C3Fl lacl transgenic mice
following inhalation exposure. Mutat Res 392: 211-222. http://dx.doi.org/10.lQ16/S1383-
5718(97)00062-4
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 Z/dx. doi. or g/10.1016/0027- 5107(93)90216- 3
Walker, VE; Wu, K, -Y; Upton, PB; Ranasinghe, A; Scheller, N; Cho, M, -H; 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. http://dx.doi.Org/10.1093/carcin/21.9.1661
Warwick, GP. (1963). The mechanism of action of alkylating agents [Review], Cancer Res 23:
1315-1333.
Waters, MP; Stack, HF; Jackson, MA. (1999). Genetic toxicology data in the evaluation of
potential human environmental carcinogens [Review], Mutat Res 437: 21-49.
http Z/dx. doi. org/10.1016/S1383-5742(99)00037-X
R-21
-------
Weiderpass. E; Meo. M; Vainio. H. (2011). Risk factors for breast cancer, including
occupational exposures. Saf Health Work 2: 1-8.
httpy/dx. doi.org/10.5491/SHAW.2011.2.1.1
WHO (World Health Organization). (2003). Concise international chemical assessment
document: Ethylene oxide. In Concise International Chemical Assessment: Ethylene
Oxide. Geneva, Switzerland: World Health Organization, International Programme on
Chemical Safety (IPCS), http://www.who.int/ipcs/publications/cicad/en/cicad54.pdf
Won, JU. (2010). Health effects of chemicals used in hospitals among healthcare workers. J
Korean Am Med Assoc 53: 474-482. http://dx.doi.Org/10.5124/ikma.2010.53.6.474
Won» O. (1991), Mortality among workers exposed to ethylene oxide [Letter], NEngl J Med
325: 1254. httpJ/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. http://dx.doi.org/10.2307/27727609
Wu, K. -Y; Chiang. S. -Y; Shih. W. -C: Huang. C. -CJ: Chea M. -F; 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.20299
Wu, K, -Y; Ranasinghe, A; Upton. PB; Walker, YE; Swenberg„ JA. (1999a). Molecular
dosimetry of endogenous and ethylene oxide-induced N7-(2-hydroxyethyl) guanine
formation in tissues of rodents. Carcinogenesis 20: 1787-1792.
http://dx.doi.Org/10.1093/carcin/20.9.1787
Wu, K, -Y; Scheller, N: Ranasinghe, A: Yea T, -Y; 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.
Chem Res 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.
http J/dx. doi. org/ 10.1002/em.2860040204
Yager, JW; Hines, CJ; Spear, RC. (1983). Exposure to ethylene oxide at work increases sister
chromatid exchanges inhuman peripheral lymphocytes. Science 219: 1221-1223.
httpy/dx. doi.org/10.1126/science.6828851
Yong„ LC; Schulte, PA; Kao, C, -Y; Giese, RW; Boeniger, MF; Strauss, GHS; Petersea MR;
Wiencke, JK. (2007). DNA adducts in granulocytes of hospital workers exposed to
ethylene oxide. Am J Ind Med 50: 293-302. httpy/dx.doi.org/10.1002/aiim.20443
Yong„ LC; Schulte, PA; Wiencke, JK; Boeniger, MF; Connally, LB; Walker, JT; Whelaa EA;
Ward, EM. (2001). Hemoglobin adducts and sister chromatid exchanges in hospital
workers exposed to ethylene oxide: Effects of glutathione S-transferase T1 and Ml
genotypes. Cancer Epidemiol Biomarkers Prev 10: 539-550.
Yuaa J, -M; Butler, LM; Gao, Y, -T; Murphy, SE; Carmella, SG; Wang„ R; Nelsoa HH; Hecht,
SS. (2014). Urinary metabolites of a polycyclic aromatic hydrocarbon and volatile
organic compounds in relation to lung cancer development in lifelong never smokers in
the Shanghai Cohort Study. Carcinogenesis 35: 339-345.
httpy/dx. doi.org/10.1093/carcin/bgt352
Zhang, F; Battels, MJ; LeBaroa MJ; Schisler, MR; Gollapudi BB; Moore, NP. (2015a). A novel
approach for concurrent quantitation of glutathione, glutathione disulfide, and 2-
hydroxyethylated glutathione in lungs of mice exposed to ethylene oxide, using liquid
R-22
-------
chromatography-positive electrospray tandem mass spectrometry. Biomed Chromatogr
29: 1364-1374. http://dx.doi.org/10.1002/bmc.3432
Zhang„ F; Battels, MJ; LeBaron, MJ; Schisler, MR; Jeong„ Y, -C; Gollapudi, BB; Moore, NP.
(2015b). LC-MS/MS simultaneous quantitation of 2-hydroxyethylated, oxidative, and
unmodified DNA nucleosides in DNA isolated from tissues of mice after exposure to
ethylene oxide. J Chromatogr B Analyt Technol Biomed Life Sci 976-977: 33-48.
http://dx.doi.ors/10.1016/i.jchromb.2014.10.042
Zhang„ J; Wang„ C; Ji, L; Liu, W. (2016). Modeling of toxicity-relevant electrophilic reactivity
for guanine with epoxides: estimating the hard and soft acids and bases (HSAB)
parameter as a predictor. Chem Res Toxicol 29: 841-850.
http://dx.doi.Org/10.1021/acs.chemrestox.6b00018
Zhao, C; Kumar, R; Zahlsen, 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. http://dx.doi.org/10.1080/135475097231445
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.
http Z/dx. doi. or g/10.1016/S0027- 5107(99)00013 - 5
R-23
------- |