peak exposures in epidemiologic studies and …...peak exposures in epidemiologic studies and cancer...
TRANSCRIPT
Risk Analysis, Vol. 39, No. 7, 2019 DOI: 10.1111/risa.13294
Peak Exposures in Epidemiologic Studies and Cancer Risks:Considerations for Regulatory Risk Assessment
Harvey Checkoway,1,∗ Peter S. J. Lees,2 Linda D. Dell,3 P. Robinan Gentry,3
and Kenneth A. Mundt4
We review approaches for characterizing “peak” exposures in epidemiologic studies andmethods for incorporating peak exposure metrics in dose–response assessments that con-tribute to risk assessment. The focus was on potential etiologic relations between environ-mental chemical exposures and cancer risks. We searched the epidemiologic literature onenvironmental chemicals classified as carcinogens in which cancer risks were described inrelation to “peak” exposures. These articles were evaluated to identify some of the chal-lenges associated with defining and describing cancer risks in relation to peak exposures. Wefound that definitions of peak exposure varied considerably across studies. Of nine chemi-cal agents included in our review of peak exposure, six had epidemiologic data used by theU.S. Environmental Protection Agency (US EPA) in dose–response assessments to deriveinhalation unit risk values. These were benzene, formaldehyde, styrene, trichloroethylene,acrylonitrile, and ethylene oxide. All derived unit risks relied on cumulative exposure fordose–response estimation and none, to our knowledge, considered peak exposure metrics.This is not surprising, given the historical linear no-threshold default model (generally basedon cumulative exposure) used in regulatory risk assessments. With newly proposed US EPArule language, fuller consideration of alternative exposure and dose–response metrics will besupported. “Peak” exposure has not been consistently defined and rarely has been evaluatedin epidemiologic studies of cancer risks. We recommend developing uniform definitions of“peak” exposure to facilitate fuller evaluation of dose response for environmental chemicalsand cancer risks, especially where mechanistic understanding indicates that the dose responseis unlikely linear and that short-term high-intensity exposures increase risk.
KEY WORDS: Cancer epidemiology; peak exposure; risk assessment
1. INTRODUCTION
Regulatory risk assessment is a process thatfocuses on identifying potentially adverse health ef-fects associated with agents of concern, and for eachadverse effect, characterizing a dose–response func-
1Department of Family Medicine & Public Health, San DiegoSchool of Medicine, University of California, La Jolla, CA, USA.
2Johns Hopkins Bloomberg School of Public Health, Baltimore,MD, USA.
3Ramboll, Amherst, MA (LDD) and Monroe, LA, USA.4Cardno Chemrisk, Boston, MA, USA.∗Address correspondence to Harvey Checkoway, UC San DiegoSchool of Medicine, 9500 Gilman Drive, La Jolla, CA 92093,USA; [email protected].
tion and unit risk numbers. Toxicology studies thatcan characterize biological uptake and pathogenesismechanisms are important components of hazardidentification and risk assessment. Epidemiologicstudies, especially those of well-characterized occu-pational groups, often identify associations betweenexposures and increased disease risk. Depending onavailable detail and chemical specificity of exposureassessment, dose–response estimation may be basedon quantitative (usually cumulative) or qualitative(yes/no or high/medium/low) relative exposureestimates. In some instances, cruder surrogates ofexposure are based only on overall duration ofemployment or duration of employment in jobs
1441 0272-4332/19/0100-1441$22.00/1 C© 2019 The Authors. Risk Analysis pub-lished by Wiley Periodicals, Inc. on behalf of Society for Risk Analysis
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,provided the original work is properly cited.
1442 Checkoway et al.
classified by exposure potential, such as “presumedexposed/unexposed.” When quantitative exposuredata of sufficient validity and specificity are avail-able, epidemiologic studies are preferred over animalstudies for quantitative dose–response assessmentsthat can be applied to derive relative toxicity values,such as inhalation unit risk (IUR) values. Con-sequently, scientifically rigorous investigations ofdose–response relations between exposures to envi-ronmental chemicals and cancer risks are dependenton valid quantitative characterization of exposure.
In conducting quantitative risk assessments,individual-level exposure estimates are required,even if derived using ecological exposure assessment(i.e., group level, such as a similar exposure groupin occupational studies). However, several differentquantitative exposure metrics can be derived, andthose that most accurately reflect underlying biolog-ical mechanisms should more accurately reflect thetrue underlying dose–response gradient. Typically,in epidemiologic research, actual doses of chemicalcompounds delivered to a target tissue rarely canbe determined. A notable exception is the measure-ment of valid biomarkers of exposure, some of whichreflect recent exposures (e.g., urinary chromium),whereas others reflect long-term (e.g., polychlori-nated biphenyls or PCBs in blood sera) or even cu-mulative (e.g., tibial lead) exposure. Given that manycancers known or presumed to be caused by envi-ronmental agents have long latency periods, epidemi-ologists can improve individual-level exposure esti-mates by considering time windows of relevant expo-sure. In theory, estimating time-dependent exposureresponses for various exposure metrics would leadto more sophisticated characterizations of risk, andtherefore more valid risk assessments for policy andregulatory decision making.
For some carcinogens, disease processes conceiv-ably depend on the dose rate and the concentrationof the chemical (or its metabolites) that reach thetarget tissue, possibly overwhelming metabolicpathways that can eliminate or detoxify lower con-centrations. In this case, an appropriate exposuremetric might be the maximum concentration encoun-tered (i.e., “peak” exposure), or the amount of timeexposure exceeds some threshold. The identificationof peak exposures (or simply the exceedance of acritical threshold) is common in evaluating acuteeffects; however, for nearly all cancers and othernoninfectious diseases that require considerable timeto elapse between exposure and disease detection,
characterizing the role of historical peak exposuresremains challenging.
In this review, our primary objective is to con-sider “peak exposure” as variously defined in thecontext of epidemiologic studies of recognized car-cinogens and related epidemiology-based cancer riskassessments. We limit our review to occupationalrather than general nonworkplace environmentalexposures (e.g., outdoor air pollution) because theworkplace often represents the “high dose” scenario.Occupational exposure measures are frequentlymore detailed and quantitative than exposures thatoccur in the general environment. Moreover, occu-pational epidemiologic research findings often haveimportant risk assessment and policy implicationsfor ambient environmental exposures. Althoughsome low-concentration exposures likely contributeto disease risk, we focus here on ways that occupa-tional epidemiology might improve understandingof cancer risks associated with brief or intermit-tent exposures to high concentrations that may becharacterized as peak exposure.
Our research question was initially motivatedby a practical issue raised by the U.S. NationalResearch Council (NRC, 2011) regarding the hazardcharacterization of formaldehyde as potentially car-cinogenic to humans (Baan et al., 2009; InternationalAgency for Research on Cancer [IARC], 2006).While the epidemiologic evidence suggested someassociation between risks for nasopharyngeal cancer(NPC) and myeloid leukemia and peak formalde-hyde exposure metrics, there remains limited andinconsistent evidence for increased risks of thesecancers associated with cumulative exposure, theconventional dose metric. Nevertheless, the U.S.Environmental Protection Agency (US EPA, 2010)draft assessment of formaldehyde was based on thecumulative exposure metric from selected relevantoccupational epidemiologic studies (Beane Freemanet al., 2009; Hauptmann, Lubin, Stewart, Hayes, &Blair, 2003), and not peak exposure, to derive anIUR (US EPA, 2010). In 2018, US EPA proposedregulation “designed to increase transparency ofthe assumptions underlying dose–response models”given “growing empirical evidence of nonlinear-ity in the concentration-response function” forenvironmental exposures and health effects includ-ing cancers. The proposed rule cautions that “EPAshould also incorporate the concept of model uncer-tainty when needed as a default to optimize low doserisk estimation based on major competing models,
Peak Exposures in Epidemiologic Studies and Cancer Risks 1443
including linear, threshold, and U-shaped, J-shaped,and bell-shaped models” [40 CFR Part 30 RIN 2080-AA14]. Promulgation of this rule language likely willrevive interest in understanding alternative exposuremetrics as well, including peak exposure.
2. METHODS
We conducted a literature search in PubMed(including MEDLINE, 1966–present) to identifyepidemiologic studies in which results includedestimates of relative risks in relation to some form ofpeak chemical exposure metric. Initial search termsincluded “peak,” “exposure,” and “epidemiology.”We screened titles and abstracts from this initialsearch and then refined the search to include termsfor chemicals identified in the abstracts: acryloni-trile, benzene, 1,3-butadiene (BD), ethylene oxide(EO), formaldehyde, methylene chloride, styrene,trichloroethylene (TCE), and tetrachlorodibenzo-p-dioxin (TCDD).
We separately reviewed the IARC Monographswebsite5 to identify chemical agents characterized ashaving “sufficient evidence in humans” or “limitedevidence in humans” of causing lymphohematopoi-etic malignancies (LHMs) based on epidemiologicstudies. Our focus on LHMs was based on threefactors: (1) knowledge of cancer hazard character-ization driven by epidemiologic studies reportingincreased myeloid leukemia associated with peakformaldehyde exposure categories (Beane Freemanet al., 2009; Hauptmann et al., 2009); (2) the as-sumption that LHMs have shorter disease inductionand latency intervals (i.e., 2 to <10 years) thansolid tumors (i.e., �15 or �20 years) and/or may bemore sensitive or responsive to “peak” exposuresas demonstrated by induction of leukemias amongindividuals administered high doses of chemother-apeutic agents (IARC, 2012a; US EPA, 2012); and(3) our assumption that LHM risks associated withpeak exposures should be more easily identified overshorter follow-up periods in cohort studies, whichadmittedly is speculative, but helpful in providinga focused literature review in which peak exposureconcepts and analytic methods can be explored.
Next, we reviewed the IARC Monographs foreach of the identified chemical agents to determinewhich epidemiologic studies were relied upon for thecancer hazard characterization, and if any provided
5See list of classifications by cancer site at https://monographs.iarc.fr/agents-classified-by-the-iarc/.
information about peak exposures. We separatelyreviewed the US EPA Integrated Risk Informa-tion System (IRIS) database to identify whichepidemiologic studies, if any, were relied upon fordose–response assessment, and which of these stud-ies also reported risks in relation to peak exposures.6
For any epidemiologic study that reported peakexposure information, we extracted information onthe exposure reconstruction approach, choice of ex-posure modeling approach, definitions of peak expo-sure, factors considered as part of the time depen-dency of cancer (e.g., estimation of cancer diagnosislatency), risk estimates in relation to peak exposuremetrics, whether sensitivity analyses were conductedto assess exposure misclassification (and results ofthe analysis, if any), and considerations or discussionof validity and reliability of the exposure assessment.
3. CASE STUDIES
For LHMs, IARC Working Groups have char-acterized the epidemiologic evidence as “sufficient”for benzene (acute myeloid leukemia [AML]/acutenonlymphocytic leukemia [ANLL]), formaldehyde(leukemia), and BD (all LHMs). The epidemio-logic evidence has been characterized as “limited”for dichloromethane (methylene chloride) (non-Hodgkin lymphoma [NHL]), ethylene oxide (NHL,multiple myeloma, chronic lymphocytic leukemia),styrene (LHMs), and TCE (NHL). Each of these hasbeen classified as Group 1 (carcinogenic to humans)by IARC, except for methylene chloride and styrene,which are classified as Group 2A (probably carcino-genic). The IARC also has identified sufficientevidence in humans of NRC for formaldehyde, andkidney cancer for TCE (as well as “limited” evidencein humans for liver cancer). The IARC also reportedlimited evidence in humans of breast cancer forEO, and EO was upgraded to a Group 1 carcinogenbased on mechanistic data and sufficient evidence inexperimental animals although the epidemiologicalevidence does not demonstrate an increased risk ofbreast cancers.
In addition to these substances, our literaturesearch also identified acrylonitrile and TCDD aschemicals for which peak exposure was evaluatedin epidemiologic studies, and which IARC has clas-sified as possibly carcinogenic to humans (Group2B) and carcinogenic to humans (Group 1), respec-tively. After reviewing full text articles, we included
6See https://www.epa.gov/iris.
1444 Checkoway et al.
acrylonitrile in our review. We excluded TCDD fromfurther consideration because peak TCDD expo-sures were reported as the result of an industrial acci-dent in Seveso, Italy (Bertazzi, Bernucci, Brambilla,Consonni, & Pesatori, 1998) and thus were unusualand not representative of peak exposures that de-scribe the higher exposure concentrations associatedwith day-to-day occupational exposure variability.
For each identified substance, Supporting In-formation Table SI summarizes its common uses,existing occupational exposure limits, the IARCcharacterization as human carcinogens, the US EPAIURs for carcinogenicity (i.e., the upper-boundexcess lifetime cancer risk estimate to result fromcontinuous exposure at a concentration of 1 µg/m3),and references for epidemiologic studies with quan-titative or semiquantitative exposure metrics thatinformed the characterization of the human cancerhazard by IARC or the dose–response assessmentby the US EPA. Supporting Information Table SIIdescribes the rationale for evaluating peak exposuresin relation to risk for each reference in which a peakexposure metric was used (where available).
3.1. Benzene and LHM
The US EPA (1998) relied upon epidemiologicstudies (Rinsky et al., 1987; Rinsky, Young, & Smith,1981) for dose–response assessment using cumulativeexposure to derive an IUR for benzene (SupportingInformation Table SI).
Collins, Ireland, Buckley, and Shepperly (2003)evaluated associations between peak exposure esti-mates and risk of LHMs in a study of workers at aU.S. benzene-derived chemicals manufacturing plantin a pooled study of petroleum industry workers inCanada, the United Kingdom, and Australia (Glasset al., 2010; Glass, Schnatter, Tang, Irons, & Rushton,2014; Rushton, Schnatter, Tang, & Glass, 2014;Schnatter, Glass, Tang, Irons & Rushton, 2012) anda study of Norwegian offshore oil drillers (Stene-hjem et al., 2015) (Table I). Peaks in these stud-ies were defined, respectively, as number of dayswith 15-minute excursions >100 parts per million(ppm) (Collins et al., 2003), at least one year injob with >3 ppm for 15–60 minutes at least weekly(yes/no) (Glass et al., 2010, 2014; Rushton et al.,2014; Schnatter et al., 2012), or a semiquantitativescore (STEL score) that reflected the frequency (of-ten/sometimes/never) of exceeding the Norwegianshort-term exposure limit (STEL) of 3 ppm for15 minutes (Stenehjem et al., 2015).
Risks for specific LHMs were not consistently in-creased in relation to any of the peak exposure met-rics in these studies (Table I). Schnatter et al. (2012)reported increased risk of myelodysplastic syndromein workers exposed to peaks >3 ppm (Table I) andin relation to cumulative exposure: OR 1.73 (95%CI 0.55–5.47) for 0.348–2.93 ppm-years and OR 4.33(95% CI 1.31–14.3) for >2.93 ppm-years (comparedto the referent group exposed to �0.348 ppm-years).
Stenehjem et al. (2015) reported risk of AMLincreased with cumulative exposure (trend test p =0.052) more strongly than with cumulative peak ex-posure (trend test p = 0.166) or with average peakexposure (trend test p = 0.056). Cumulative peak ex-posure was calculated by multiplying the STEL scoreby the duration of employment for each of the jobsand summing over all jobs while average peak ex-posure was calculated by dividing cumulative peakexposure by exposure duration. Similarly, multiplemyeloma increased with cumulative exposure (trendtest p = 0.024) and average peak exposure, but notwith cumulative peak exposure. The trend test wasnot significant for either peak exposure metric. Expo-sure was not lagged, and most workers were exposedto benzene for fewer than 15 years.
Collins et al. (2003) reported no increased risksfor all leukemias combined or ANLL associatedwith cumulative exposure, and no trends by peakexposure for any of several LHMs; however, thehighest standardized mortality ratios (SMRs) formultiple myeloma and ANLL were seen in the high-est peak exposure category (>40 days with peak ex-posure >100 ppm), although based on very few (3and 2, respectively) observed deaths.
3.2. Formaldehyde and NPC and LHM
IARC Working Groups have classifiedformaldehyde as carcinogenic to humans, caus-ing NPC (IARC, 2006, 2012b) and leukemia, inparticular myeloid leukemia (IARC, 2012b), basedon results generated when exposure was character-ized as a “peak exposure.” The US EPA Draft IRISassessment, however, used cumulative exposuremetrics from epidemiologic studies (Beane Freemanet al., 2009; Hauptmann et al., 2003) for the dose–response assessment to derive an IUR value for thecombined risks of leukemia, Hodgkin lymphoma,and NPC (Supporting Information Table SI).
Peak formaldehyde exposures were consid-ered in two studies initiated by the U.S. NationalCancer Institute, one of workers from multiple
Peak Exposures in Epidemiologic Studies and Cancer Risks 1445
Tab
leI.
Ben
zene
Exp
osur
ean
dL
ymph
ohem
atop
oeti
cM
alig
nanc
ies
(LH
M)
Stud
y de
sign
, Pop
ulat
ion,
Loc
atio
n, R
efer
ence
Expo
sure
Expo
sure
Cat
egor
ies
Obs
erve
d[C
ases
/Con
trols
]R
R (9
5% C
I)Ex
posu
re A
sses
smen
t Lim
itatio
nsan
d C
omm
ents
Stud
y de
sign
Coh
ort s
tudy
Stud
y po
pula
tion:
44
17 h
ourly
wor
kers
beg
inni
ng e
mpl
oym
ent b
etw
een
1940
and
197
7 at
the
Solu
tia p
lant
, pre
viou
sly
Mon
sant
o, in
Sau
get,
Illin
ois.
USA
Col
lins e
t al.,
200
3
Exp
osur
e A
sses
smen
tPe
rson
al e
xpos
ure
colle
ctio
n (b
egin
ning
in 1
980
for o
nly
certa
in d
epar
tmen
ts),
hist
ory
of p
roce
ss c
hang
es, a
rea
sam
plin
g le
vels
, ind
ivid
ual l
evel
exp
osur
es, a
nd th
e pl
ant’s
in
dustr
ial h
ygie
nist’
s jud
gem
ent w
ere
used
to e
stim
ate
benz
ene
expo
sure
.
Ana
lysi
s2
expo
sure
met
rics,
one
of w
hich
mea
sure
d “p
eak”
: -n
umbe
r of d
ays
with
pea
ks (c
ateg
ory)
Num
ber
of D
ays w
ith p
eak
expo
sure
>10
0 pp
m
HL
SMR
Lim
ited
expo
sure
info
rmat
ion
for t
he 1
940s
an
d 19
50s –
auth
ors n
ote
expo
sure
s cou
ld
have
bee
n m
uch
high
er “
give
n th
e gr
eate
r po
tent
ial f
or u
pset
con
ditio
ns.”
Non
e2
0.7
(0.1
–2.
4)<7
00.
0 (0
.0 –
22.8
)7-
400
0.0
(0.0
-14
.5)
>40
00.
0 (0
.0 –
13.1
)N
HL
Non
e20
1.4
(0.9
–2.
2)<7
11.
3 (0
.0 –
7.4)
7-40
10.
7 (0
.0 –
4.2)
>40
31.
8 (0
.4 –
5.1)
MM
Non
e9
1.2
(0.6
–2.
4)<7
00.
0 (0
.0 –
10.2
)7-
401
1.7
(0.0
–9.
4)>4
03
4.0
(0.8
–11
.7)
Leu
kem
iaN
one
151.
0 (0
.5 –
1.6)
<71
1.2
(0.0
–6.
8)7-
401
0.7
(0.0
–4.
0)>4
05
2.7
(0.8
–6.
4)C
LL
Non
e3
0.9
(0.2
–2.
5)<7
00.
0 (0
.0 –
20.6
)7-
400
0.0
(0.0
–12
.2)
>40
12.
5 (0
.0 –
13.8
)A
NL
LN
one
51.
3 (0
.4 –
3.0)
<70
0.0
(0.0
–10
.2)
7-40
00.
0 (0
.0 –
10.4
)>4
02
4.1
(0.5
–14
.9)
(Con
tinue
d)
1446 Checkoway et al.
Tab
leI
(Con
tinu
ed)
Stud
y de
sign
Pool
ed n
este
d ca
se-c
ontro
l stu
dy
Stud
y po
pula
tion
Thre
e co
hort
stud
ies i
nclu
ding
Can
adia
n pe
trole
um
distr
ibut
ion
wor
kers
from
196
4 th
roug
h 19
83, U
K
petro
leum
dis
tribu
tion
wor
kers
from
195
0 th
roug
h 19
89,
and
Aus
tralia
n pe
trole
um d
istri
butio
n, re
finin
g, a
nd
upstr
eam
wor
kers
from
198
1 th
roug
h 19
96, r
espe
ctiv
ely.
Can
ada
stud
y in
clud
ed 3
1 LH
can
cer c
ases
, UK
stud
y in
clud
ed 9
0 ca
se su
bjec
ts w
ith le
ukem
ia, a
nd A
ustra
lian
stud
y in
clud
ed 7
9 ca
se su
bjec
ts w
ith L
H c
ance
rs.
Schn
atte
ret
al.,
2012
Exp
osur
e A
sses
smen
tW
ork
histo
ry o
btai
ned
from
com
pany
reco
rds (
Can
ada
and
UK
) or b
y in
terv
iew
(Aus
tralia
). M
onito
ring
data
col
lect
ed
at st
udy
and
simila
r fac
ilitie
s wer
e as
signe
d fo
r spe
cific
job
task
s.
Ana
lysi
s6
expo
sure
met
rics,
one
of w
hich
mea
sure
d “p
eak”
:≥
1-ye
ar e
mpl
oym
ent i
n jo
bs li
kely
exp
erie
ncin
g >3
ppm
ex
posu
re fo
r 15–
60 m
inut
es a
t lea
st w
eekl
y
Expo
sure
val
idat
ion
and
sens
itivi
ty a
naly
sis p
erfo
rmed
.
Peak
exp
osur
e>3
ppm
AM
LO
RPo
tent
ial e
xpos
ure
unde
rest
imat
ion;
unab
le to
acc
ount
for i
nfre
quen
t and
in
divi
dual
ized
exp
osur
e ev
ents.
Sens
itivi
ty a
naly
ses:
Ri
sk o
f MD
S in
hig
h an
d m
ediu
m c
erta
inty
di
agno
ses (
peak
exp
osur
e vs
no
peak
ex
posu
re):
OR
6.32
(95%
CI 1
.32
-30.
2)
Risk
of M
DS
in w
orke
rs h
avin
g th
e hi
ghes
t ex
posu
re c
erta
inty
(pea
k ex
posu
re v
s no
peak
ex
posu
re):
OR
5.74
(95%
CI 1
.05
-31.
2)
Non
e[2
5/12
4]1.
00 (R
efer
ence
)Y
es[3
5/11
7]1.
50 (0
.82
-2.7
5)
MD
SN
one
[13/
82]
1.00
(Ref
eren
ce)
Yes
[16/
47]
2.48
(0.9
7-6.
35)
pgl
obal
=0.0
5
CL
LN
one
[50/
187]
1.00
(Ref
eren
t)Y
es[3
0/15
8]0.
69 (0
.41
–1.
14)
CM
LN
one
[17/
63]
1.00
(Ref
eren
ce)
Yes
[11/
59]
0.67
(0.2
9 –
1.54
)
MPD
Non
e[1
2/63
]1.
00 (R
efer
ence
)Y
es[1
8/61
]1.
47 (0
.68
–3.
19)
Stud
y de
sign
Pool
ed n
este
d ca
se-c
ontro
l stu
dy,
Stud
y po
pula
tion
28 c
hron
ic m
yelo
id le
ukem
ia(C
ML)
cas
es a
nd 1
22
mat
ched
cont
rols
and
30
mye
lopr
olife
rativ
e di
seas
e (M
PD) c
ases
w
ith 1
24 m
atch
edco
ntro
ls.
Dra
wn
from
sam
e po
pula
tion
as S
chna
tter e
t al.
2012
Gla
ss e
t al.
2014
Exp
osur
e A
sses
smen
tSa
me
as d
escr
ibed
in S
chna
tter e
t al.
2012
Ana
lysi
sEv
er >
3 pp
m b
enze
ne fo
r 15-
60 m
in a
t lea
st w
eekl
y fo
r 1+
year
s (ye
s or n
o)
Ana
lyze
d pe
ak e
xpos
ure
com
bine
d w
ith c
umul
ativ
e ex
posu
re
Peak
exp
osur
e>3
ppm
CM
LO
REx
posu
re v
alid
atio
n fo
cuse
d so
lely
on
cons
iste
ncy
of e
stim
ates
acr
oss t
hree
stud
ies;
expo
sure
est
imat
es n
ot v
alid
ated
aga
inst
mea
sure
d ex
posu
res.
Non
e[1
7/63
]1.
0 (R
efer
ence
)Y
es[1
1/59
]0.
67 (0
.29
–1.
54)
Peak
exp
osur
e>
3 pp
m, 2
-20
year
s exp
osur
e w
indo
wC
ML
Non
e[2
1/82
]1.
0 (R
efer
ence
)Y
es[7
/40]
0.66
(0.2
5 –
1.72
)Pe
ak e
xpos
ure
>3
ppm
MPD
Non
e[1
2/63
]1.
0 (R
efer
ence
)Y
es[1
8/61
]1.
47 (0
.68
–3.1
9)Pe
ak e
xpos
ure
> 3
ppm
, 2-2
0 ye
ars e
xpos
ure
win
dow
MPD
Non
e[1
7/97
]1.
0 (R
efer
ence
)Y
es[1
3/27
]3.
81 (1
.36
–10
.7)
Stud
y de
sign
, Pop
ulat
ion,
Loc
atio
n, R
efer
ence
Expo
sure
Expo
sure
Cat
egor
ies
Obs
erve
d[C
ases
/Con
trols
]R
R (9
5% C
I)Ex
posu
re A
sses
smen
t Lim
itatio
nsan
d C
omm
ents
(Con
tinue
d)
Peak Exposures in Epidemiologic Studies and Cancer Risks 1447
Tab
leI
(Con
tinu
ed)
Stud
y de
sign
Cas
e-co
hort
study
(coh
ort e
stab
lishe
d by
Can
cer
Regi
stry
of N
orw
ay v
ia su
rvey
in 1
998
(res
pons
e ra
te
estim
ated
as 6
9%)
Stud
y po
pula
tion:
24
,917
men
em
ploy
ed in
offs
hore
oil
indu
stry,
196
5-19
99
112
LHM
s dia
gnos
ed 1
999-
2011
and
1,6
61 n
on-L
HM
s sa
mpl
ed ra
ndom
ly fr
om 5
-yea
r birt
h co
horts
Nor
way
Sten
ehje
m 2
015
Exp
osur
e A
sses
smen
tJo
b ex
posu
re m
atrix
, sem
i-qua
ntita
tive
sum
mar
y m
easu
res
of b
enze
ne e
xpos
ure
Det
aile
d w
ork
hist
ory
obta
ined
from
que
stion
naire
s
Ana
lysi
s5
expo
sure
met
rics,
two
of w
hich
are
mea
sure
s of “
peak
”:
-C
umul
ativ
e pe
ak>
3 pp
m (S
TEL-
scor
e-ye
ars)
-Ave
rage
freq
uenc
y sc
ore
of p
eak
expo
sure
s(a
vera
ge p
eak
> 3
ppm
), ST
EL-s
core
)A
vera
ge p
eak
STEL
-sco
re h
ad a
rang
e of
0–1
57 a
nd S
TEL-
scor
e-ye
ars h
ad a
rang
e of
0-2
08.3
; how
ever
, cut
-poi
nts f
or
the
terti
les w
ere
not r
epor
ted
in th
e pa
per.
STE
L-s
core
-yea
rsB
-cel
l NH
LH
RN
o sh
ort-t
erm
indu
stria
l hyg
iene
m
easu
rem
ents
wer
e us
ed.
Expo
sure
ass
essm
ent n
ot v
alid
ated
Expo
sure
was
not
lagg
ed in
ana
lyse
s
Sens
itivi
ty a
naly
ses:
Excl
uded
wor
kers
who
wer
e m
ain
plat
form
ac
tivity
onl
y an
d ex
clud
ed w
orke
rs w
ith
repl
aced
stop
or s
tart
date
s (ex
clud
ed 5
6 ca
ses
and
713
non-
case
s);
Ana
lyze
d 45
7 w
orke
rs st
ill e
mpl
oyed
in 1
998
and
assu
med
to b
e ex
pose
d at
the
sam
e in
tens
ity le
vel u
ntil
cens
orin
g or
end
of
obse
rvat
ion
Une
xpos
ed28
1.00
(Ref
eren
ce)
Terti
le 1
181.
28 (0
.69
-2.3
9)Te
rtile
218
1.44
(0.7
8 -2
.64)
Terti
le 3
171.
30 (0
.72
-2.4
4)ST
EL
-sco
reU
nexp
osed
281.
00 (R
efer
ence
)Te
rtile
119
1.45
(0.8
0 -2
.63)
Terti
le 2
191.
46 (0
.80
-2.6
6)Te
rtile
315
1.13
(0.6
0 -2
.11)
STE
L-s
core
-yea
rsM
MU
nexp
osed
51.
00 (R
efer
ence
)Te
rtile
12
0.79
(0.1
4 –
4.41
)Te
rtile
26
2.82
(0.8
5 –
9.34
)Te
rtile
34
2.06
(0.6
0 –
7.10
)ST
EL
-sco
reU
nexp
osed
51.
00 (R
efer
ence
)Te
rtile
13
1.37
(0.3
2 –
5.78
)Te
rtile
24
1.74
(0.4
5 –
6.67
)Te
rtile
35
2.42
(0.7
4 –
8.00
)ST
EL
-sco
re-y
ears
ML
Une
xpos
ed6
1.00
(Ref
eren
ce)
Terti
le 1
51.
51 (0
.44
-5.1
8)Te
rtile
26
2.46
(0.7
5 -8
.10)
Terti
le 3
41.
97 (0
.51
-7.6
9)ST
EL
-sco
reU
nexp
osed
21.
00 (R
efer
ence
)Te
rtile
12
2.09
(0.6
3 -6
.86)
Terti
le 2
21.
53 (0
.42
-5.5
8)Te
rtile
34
2.19
(0.6
5 -7
.36)
STE
L-s
core
-yea
rsA
ML
Une
xpos
ed2
1.00
(Ref
eren
ce)
Terti
le 1
21.
95 (0
.27
-14)
Terti
le 2
33.
52 (0
.61
-20)
Terti
le 3
33.
92 (0
.59
-26)
STE
L-s
core
Une
xpos
ed2
1.00
(Ref
eren
ce)
Terti
le 1
22.
17 (0
.31
-15)
Terti
le 2
22.
21 (0
.30
-16)
Terti
le 3
44.
87 (0
.90
-26)
Stud
y de
sign
, Pop
ulat
ion,
Loc
atio
n, R
efer
ence
Expo
sure
Expo
sure
Cat
egor
ies
Obs
erve
d[C
ases
/Con
trols
]R
R (9
5% C
I)Ex
posu
re A
sses
smen
t Lim
itatio
nsan
d C
omm
ents
AM
L,
acut
em
yelo
idle
ukem
ia;
AN
LL
,ac
ute
non-
lym
phoc
ytic
leuk
emia
;C
LL
,ch
roni
cly
mpo
cyti
cle
uekm
ia;
CM
L,
chro
nic
mye
loid
leuk
emia
;H
L,
Hod
gkin
lym
phom
as;
HR
,ha
zard
rati
o;M
DS,
mye
lody
spla
stic
synd
rom
e;M
M,
mul
tipl
em
yelo
ma;
MP
D,
mye
lopr
olif
erat
ive
diso
rder
;N
HL
,no
n-H
odgk
inly
mph
oma;
OR
,od
dsra
tio;
RR
,re
lati
veri
sk;
STE
L,s
hort
-ter
mex
posu
relim
it.
1448 Checkoway et al.
industries with formaldehyde exposure (e.g.,formaldehyde resin manufacturing) and the otheron funeral directors and embalmers. In the U.S.formaldehyde producers and users study (BeaneFreeman et al., 2009, 2013), peak exposure wasdefined as short-term exposures (generally lessthan 15 minutes) that exceeded job-specific 8-hourtime-weighted average (TWA) exposure estimates(Stewart et al., 1986). For workers classified asnot having held jobs with exposures exceedingthose 8-hour TWA levels, peak exposures weredefined as the job-specific 8-hour TWA. BeaneFreeman et al. (2009) reported associations withpeak exposure (Table II), but not for cumu-lative exposure, average intensity, duration ofexposure, or cumulative number of peaks for allLHM and for ML. In an independent reanalysis ofthe data, peaks were redefined as absolute peaks, �1continuous month of employment in jobs identifiedin the original exposure characterization as likelyhaving short-term exposure excursions of �2 ppmto <4 ppm or �4 ppm on a weekly or daily basisfor all workers in the cohort (Checkoway et al.,2015). Using an absolute peak metric, Checkowayet al. (2015) reported an attenuated relation forML and no association for AML, as only four ofthe 34 workers with AML as the cause of deathhad been classified as having peak exposures in the20 years preceding their deaths. Separately, BeaneFreeman et al. (2013) reported increased lung cancermortality (SMR 1.15; 95% CI 1.07–1.20) and excessNPC mortality (SMR 1.84, 95% CI 0.84–3.49) forworkers assigned to their highest peak exposure cat-egory; internal analyses by peak exposure, however,generated no positive associations with lung cancer,and consistently increased mortality from NPC forthe highest category of each exposure metric: peakexposure �4 ppm (RR 7.66, 95% CI 0.94–62.3),average intensity �1 ppm (RR 11.54, 95% CI 1.38–96.8), and cumulative exposure �5.5 ppm (RR 2.94,95% CI 0.65–13.3).
In the funeral industry workers study, lifetimepeak formaldehyde exposure was defined as themaximum 15-minute average exposure intensityestimate derived over the entire duration of employ-ment (Hauptmann et al., 2009). A predictive modelwas developed based on a literature review andwalk-through surveys at funeral homes, resulting inthree determinants of exposure: ventilation (low,moderate, high) in the embalming room, concen-tration of formaldehyde in the embalming solution(high vs. low solution strength), and whether the
embalming was performed on an intact or autopsiedbody (Hornung et al., 1996). The investigatorsconducted limited sampling of formaldehyde con-centrations in the breathing zone of the embalmerand three locations in the room, based on the combi-nations of the determinants of exposure. The modeldefined peak exposure as the maximum of movingaverages of any series of measurements covering15 minutes (90 measurements, i.e., one measurementevery 10 seconds). In addition, the investigatorscalculated the lifetime number of embalmings thatwere associated with predicted peaks exceedinga certain level. For the evaluation of the averageexposure model, the investigators compared modelpredictions to measurements from 15 independentembalmings and reported that the model overesti-mated measured (average, not peak) formaldehydeexposure by 35%. The authors reported that thepeak model could not be validated due to lack ofreal-time measurements. The authors describedincreasing risks of AML in relation to peak exposureonly in a model that included embalmers and direc-tors who had performed 500 or more embalmings(Table II).
Two other large industrial cohort studies exam-ined LHM risks associated with formaldehyde ex-posure; however, neither study used peak exposureor any reasonable surrogate. The NIOSH garmentworkers study reported a geometric mean concen-tration of 0.15 ppm measured from the 1980s andreported that exposure levels at the three facili-ties were likely to have been similar to formalde-hyde exposures experienced in the NCI and U.K. co-horts in earlier years (Meyers, Pinkerton & Hein,2013). This conclusion was based on average con-centrations of formaldehyde at eight garment facto-ries in 1966 (range of average concentrations 0.3–2.7 ppm), and measured formaldehyde concentra-tions in cutting rooms reported to be as high as 10ppm in 1968, and subsequently reduced to 2 ppm in1973 (as reported in IARC, 2006). The U.K. cohortstudy of formaldehyde producers and users reportedthat the high-exposure category represented averageconcentrations of 2 ppm or higher (Coggon, Ntani,Harris, & Palmer, 2014). Overall, workers experi-enced higher average exposures in the United King-dom compared to the United States: 3,991 (28%) ofthe U.K. workers were assigned to average intensityidentified as “high” (i.e., �2 ppm) and 3,927 (15%)of the NCI cohort were reported by Beane Free-man et al. (2009) to be assigned to average intensity�1 ppm.
Peak Exposures in Epidemiologic Studies and Cancer Risks 1449
Tab
leII
.F
orm
alde
hyde
Pea
kE
xpos
ure
and
LH
Ms
Stud
y de
sign
Cas
e-co
ntro
l stu
dy o
f leu
kem
ias n
este
d w
ithin
a d
eath
ce
rtific
ate
stud
y
Stud
y po
pula
tion:
6,
808
fune
ral d
irect
ors a
nd e
mba
lmer
s who
die
d 19
60-1
986
Dat
a so
urce
s: N
atio
nal F
uner
al D
irect
ors A
ssoc
iatio
n (n
=665
1, H
ayes
, Bla
ir, S
tew
art,
Her
rick,
& M
ahar
, 19
90)
New
Yor
k St
ate
Bure
au o
f Fun
eral
Dire
ctor
s (n
=167
8, W
alra
th a
ndFr
aum
eni,
1983
)Cal
iforn
ia
Stat
e D
epar
tmen
t of H
ealth
, Div
isio
n of
Fun
eral
D
irect
ors a
nd E
mba
lmer
s (n=
5,66
5, W
alra
th a
nd
Frau
men
i, 19
84)
Uni
ted
Stat
es
Hau
ptm
ann
et a
l. 20
09
Exp
osur
e as
sess
men
tPr
edic
tive
mod
el b
ased
on
expo
sure
ass
essm
ent
expe
rimen
t con
duct
ed d
urin
g em
balm
ing
(Ste
war
t et a
l. 19
92)
Mod
el in
clud
ed e
xpos
ure
dete
rmin
ants
(suc
h as
ve
ntila
tion
rate
, con
cent
ratio
n of
form
alde
hyde
sol
utio
n,
emba
lmin
g of
inta
ct o
r aut
opsi
ed c
orps
e, sp
ill o
r no
spill
)
Indi
vidu
al a
vera
ge a
nd p
eak
inte
nsity
bas
ed o
n in
form
atio
n ob
tain
ed fr
om q
uest
ionn
aire
dat
a
Peak
cal
cula
ted
as m
axim
um o
f mov
ing
aver
ages
of a
ny
serie
s of m
easu
rem
ents
cov
erin
g 15
min
utes
(90
mea
sure
men
ts, 1
mea
sure
men
t eve
ry 1
0 se
cond
s) fr
om
mod
el
Ana
lysi
s6
expo
sure
met
rics,
incl
udin
g 2
that
est
imat
ed “
peak
”:-L
ifetim
e pe
ak e
xpos
ure,
ppm
(max
imum
15-
min
av
erag
e in
tens
ity e
ver e
xper
ienc
edov
er a
ll em
balm
ings
ov
er a
ll ye
ars)
-L
ifetim
e nu
mbe
r of e
mba
lmin
gs >
500
with
pre
dict
ed
peak
s
Peak
ML
OR
Mod
el fo
r pea
ks e
xpla
ined
44%
of
varia
bilit
y
Peak
mod
el c
ould
not
be
valid
ated
; in
depe
nden
t rea
l tim
e m
easu
rem
ents
not
av
aila
ble
for v
alid
atio
n em
balm
ings
“Ris
k w
as n
ot a
ssoc
iate
d w
ith in
crea
sing
num
ber o
f em
balm
ings
durin
g w
hich
pea
ks
in th
e hi
ghes
t cat
egor
y of
pea
k in
tens
ity
occu
rred
(i.e
., ex
ceed
ing
9.3
ppm
; dat
a no
t sh
own)
.”
Relia
bilit
y: “
Surr
ogat
e re
spon
dent
s (i.e
., ne
xt o
f kin
and
cow
orke
rs) m
ay o
r may
not
ac
cura
tely
repo
rt ex
posu
re-r
elat
ed
info
rmat
ion,
dep
endi
ng o
n th
e ty
pe o
f in
form
atio
n an
d ty
pe o
f sur
roga
te.”
Acc
urac
y ac
cept
able
for n
umbe
r of y
ears
w
orke
d in
fune
ral i
ndus
try (9
2%
conc
orda
nce
betw
een
mul
tiple
re
spon
dent
s) a
nd n
umbe
r of y
ears
dec
ease
d en
gage
d in
em
balm
ing
(86%
con
cord
ance
be
twee
n m
ultip
le re
spon
dent
s)
“Exp
osur
e m
iscl
assi
ficat
ion
was
like
ly to
be
nond
iffer
entia
l, so
that
any
resu
lting
bi
as w
ould
be
tow
ard
the
null
and
thus
w
ould
tend
to u
nder
estim
ate
risk.
”
0pp
m[1
/55]
1.0
(Ref
eren
ce)
>0 to
7.0
ppm
[12/
67]
15.2
(1.6
-14
1.6)
>7.0
-9.
3 pp
m[9
/72]
8.0
(0.9
-74
.0)
>9.3
ppm
[12/
71]
13.0
(1.4
-116
.9)
Life
time
emba
lmin
gs &
peak
com
bine
d
<500
emba
lmin
gs[5
/83]
1.0
(Ref
eren
ce)
>500
em
balm
ings
& >
0 to
7.0
pp
m[9
/54]
2.9
(0.9
–9.8
)
>500
em
balm
ings
&>7
.0-9
.3
ppm
[9/6
6]2.
0 (0
.6–6
.6)
>500
em
balm
ings
& >
9.3
ppm
[11/
62]
2.9
(0.9
–9.5
)
Pfo
r tre
nd=0
.036
Peak
AM
L0
ppm
NR
NR
>0 to
7.0
ppm
NR
NR
>7.0
-9.
3 pp
mN
RN
R>9
.3 p
pmN
RN
R
Life
time
emba
lmin
gs &
pe
akco
mbi
ned
<500
em
balm
ings
[3/8
3]1.
0 (R
efer
ence
)>5
00 e
mba
lmin
gs &
>0
to 7
.0
ppm
[4/5
4]1.
8 (0
.4–9
.3)
>500
em
balm
ings
&>7
.0-9
.3
ppm
[5/6
6]2.
1 (0
.5–
9.2)
>500
em
bala
min
gs &
>9.3
pp
m[7
/62]
2.9
(0.7
–12.
5)
Pfo
r tre
nd=0
.035
Stud
y de
sign
, Pop
ulat
ion,
Ref
eren
ce
Expo
sure
Expo
sure
Cat
egor
ies
Obs
erve
d[C
ases
/Con
trols
]R
R (9
5% C
I)Ex
posu
re A
sses
smen
t Lim
itatio
nsan
d C
omm
ents
Stud
y de
sign
His
toric
al c
ohor
t stu
dy
Stud
y po
pula
tion
25,6
19 fo
rmal
dehy
de u
sers
and
pro
duce
rs e
mpl
oyed
be
fore
196
6 at
one
of 1
0 U
S fa
ctor
ies
Follo
wed
for m
orta
lity
1943
-200
4
Uni
ted
Stat
es
Bea
ne F
reem
anet
al.
2009
Exp
osur
e as
sess
men
tJo
b ex
posu
re m
atrix
Exp
osur
e m
etri
cs5
expo
sure
met
rics,
incl
udin
g 2
estim
atin
g pe
ak:
-Pea
k ex
posu
re (p
pm)
-Num
ber o
f pea
ks (f
requ
ency
of p
eaks
-ho
urly
, dai
ly,
wee
kly,
and
mon
thly
–co
nver
ted
to c
ount
s: 20
00, 2
50,
50 o
r 8.2
per
yea
r, re
spec
tivel
y)
Wor
kers
ass
igne
d pe
ak e
xpos
ure
of h
ighe
st TW
A if
pea
k di
d no
t occ
ur
Peak
exp
osur
eN
HL
RR
No
IH m
easu
rem
ents
of p
eak
expo
sure
Lim
ited
IH m
onito
ring
data
Lagg
ed e
xpos
ure
by 2
thro
ugh
25 y
ears
(r
epor
ted
18 y
ears
fit d
ata
best
)
6,25
5 as
signe
d pe
ak fo
rmal
dehy
de
expo
sure
s ≥ 4
.0 p
pm(2
5%, o
r app
rox.
156
4, e
xper
ienc
ed p
eaks
<
25 ti
mes
; 25%
exp
erie
nced
such
pea
ks >
70
0 tim
es)
Risk
s did
not
incr
ease
with
num
ber o
f pe
aks ≥
4.0
ppm
for a
ny c
ause
of d
eath
0 pp
m12
1.06
(0.5
3-2.
14)
>0-<
2.0
ppm
391.
0 (R
efer
ence
)2.
0 -<
4.0
ppm
271.
08 (0
.65
-1.7
8)≥4
.0 p
pm28
0.91
(0.5
5 -1
.49)
HL
0 pp
m2
0.67
(0.1
2-3.
60)
>0-<
2.0
ppm
61.
0 (R
efer
ence
)2.
0 -<
4.0
ppm
83.
30 (1
.04-
10.5
0)≥4
.0 p
pm11
3.96
(1.3
1-12
.02)
Leu
kem
ia0
ppm
70.
59 (0
.25
-2.6
7)>0
-<2.
0 pp
m41
1.0
(Ref
eren
ce)
2.0
-<4.
0 pp
m27
0.98
(0.6
0-1.
62)
≥4.0
ppm
481.
42 (0
.92-
2.18
)M
L0
ppm
40.
82 (0
.25
-2.6
7)>0
-<2.
0 pp
m14
1.0
(Ref
eren
ce)
2.0
-<4.
0 pp
m11
1.30
(0.5
8 -2
.92)
≥4.0
ppm
191.
78 (0
.87
-3.6
4)
AM
L,a
cute
mye
loid
leuk
emia
;HL
,Hod
gkin
lym
phom
a;M
L,m
yelo
idle
ukem
ia;N
HL
,non
-Hod
gkin
lym
phom
a;pp
m,p
arts
per
mill
ion;
NR
,not
repo
rted
;RR
,rel
ativ
eri
sk;T
WA
,ti
me-
wei
ghte
dav
erag
e.
1450 Checkoway et al.
3.3. Styrene and/or BD and LHM
Styrene and butadiene are common co-exposures in the synthetic rubber industry and havebeen considered together and separately in analysesof cancer risks in a series of studies conducted byUniversity of Alabama Birmingham researchers(Cheng, Sathiakumar, Graff, Matthews & Delzell,2007; Delzell, Macaluso, Sathiakumar, & Matthews,2001; Delzell et al., 1996; Graff et al., 2005; Macalusoet al., 1996; Macaluso, Larson, Lynch, Lipton &Delzell, 2004; Sathiakumar, Brill, Leader, & Delzell,2015). IARC (2012b) has characterized BD as car-cinogenic to humans based on limited epidemiologicevidence for a causal association with NHL, multiplemyeloma, and chronic lymphocytic leukemia. Morerecently, IARC characterized styrene as probablycarcinogenic to humans based on limited evidencefrom epidemiologic studies of LHMs and myeloidleukemia, as well as sinonasal adenocarcinoma(Kogevinas et al., 2018). An IUR of 3 × 10−5 perµg/m3 was derived for butadiene, in part based onthese epidemiologic studies; however, an IUR wasnot derived for styrene (Supporting InformationTable SI).
Graff et al. (2005) evaluated relative risk ofleukemia mortality in relation to exposure to BD,styrene, and dimethyldithiocarbamate (DMDTC),an immune system depressant. The investigatorsderived several estimates of exposure for 16,579workers at synthetic rubber production facilities, in-cluding cumulative exposure (ppm-years for BD andstyrene), the annual number of peaks for BD (num-ber of occurrences in which BD exposure exceededthe peak level of 100 ppm), and annual number ofpeaks for styrene (number of occurrences in whichstyrene exposure exceeded the peak level of 50 ppm)(Macaluso et al., 2004). In models that included onlystyrene, relative risks of leukemia mortality werestatistically significantly increased for workers in thethree highest peak exposure categories after adjust-ing for age and years since hire (Table III). The rel-ative risk estimates remained increased after addingBD peaks and cumulative exposure to DMDTCto the model (Table III). Separately, the authorsreported that cumulative exposure to BD was asso-ciated with increased mortality from all leukemias,chronic myelogenous leukemia and chronic lympho-cytic leukemia, although mortality risks were attenu-ated after adjusting for styrene and DMDTC. A pos-itive exposure–response pattern was not seen with
either styrene or DMDTC after controlling for BDexposure.
Styrene exposures are common in the reinforcedplastics and composite industry, in which Collins,Bodner, and Bus (2013) examined mortality ratesin relation to cumulative exposure, duration ofexposure, peak exposures, and average exposure.The peak exposure metric reflected days with 15 ormore minutes above 100 ppm styrene (reportedly thelowest level at which irritation from styrene occurs).No patterns of increasing risks of all LHM combinedor any specific LHM were reported in relation todays with peak exposure above 100 ppm styrene(Table III). The Danish Cancer Registry linkagestudy of reinforced plastics industry workers evalu-ated cumulative exposures to styrene in relation toleukemias and lymphomas (Christensen et al., 2018)and sinonasal carcinoma (Nissen et al., 2018). Thesestudies were identified as the most informative bythe IARC working group; however, the studies ofthe Danish reinforced plastic industry did not reportinformation on risks related to peak exposures orshort-term exposures, despite a substantial number(2,207) of short term (<1 hour) samples collectedduring 1960–1996 (short-term exposures as highas 2,720 mg/m3) (Kolstad, Sonderskov, & Burstyn,2005).
3.4. TCE and Leukemia, Lymphoma, andKidney Cancer
IARC has classified TCE as carcinogenic tohumans, causing cancer of the kidney (IARC, 2014).The epidemiologic evidence was considered suffi-cient for a majority of the IARC Working Group,and a minority considered the evidence to be limited.The Working Group also reported limited epidemi-ologic evidence for NHL (IARC, 2014) and livercancer in relation to TCE. The US EPA based itsdose–response analysis on cumulative exposure fromepidemiologic studies (Charbotel, Fevotte, Hours,Martin & Bergeret, 2006; Raaschou-Nielsen et al.,2003) and derived an IUR for combined risks fromkidney cancer, liver cancer, and NHL (US EPA,2011b).
In addition to categorical cumulative exposureassignments (1–150 ppm-years, 155–335 ppm-years,and >335 ppm-years), Charbotel et al. (2006) alsoevaluated peak exposure in a case-control study of 86renal cancer cases and 316 hospital and clinic-basedcontrols in France. Peak exposure was not clearly
Peak Exposures in Epidemiologic Studies and Cancer Risks 1451
Tab
leII
I.St
yren
eP
eak
Exp
osur
ean
dL
HM
s
Stud
y D
esig
n, P
opul
atio
n, R
efer
ence
Ex
posu
reEx
posu
re c
ateg
orie
sO
bser
ved
[Cas
es/C
ontro
ls]
RR
(95%
CI)
Expo
sure
Ass
essm
ent L
imita
tions
and
Com
men
tsSt
udy
Des
ign
His
toric
al c
ohor
t stu
dy
Stud
y po
pula
tion:
16
,579
mal
e w
orke
rs e
mpl
oyed
194
3-19
91 a
t 5 U
S an
d 1
Can
ada
synt
hetic
rubb
er p
lant
s.Fo
llow
ed fo
r mor
talit
y 19
43-1
998
U.S
. and
Can
ada
Gra
ff et
al.
2005
Exp
osur
e A
sses
smen
tJo
b ex
posu
re m
atrix
,wor
k hi
stor
y da
ta w
as o
btai
ned
from
em
ploy
men
t rec
ords
Ana
lysi
sTw
oex
posu
re m
etric
s for
styr
ene,
one
of w
hich
mea
sure
d “p
eak”
:-N
umbe
r ofp
eaks
>50
ppm
9,46
3 (5
7%)h
ad n
umbe
r of s
tyre
ne p
eaks
> 0
(med
ian=
60)
No.
ofp
eaks
> 5
0 pp
mL
euke
mia
RR
Inco
mpl
ete
wor
k hi
stor
ies
Expo
sure
est
imat
ion
used
mod
el th
at
requ
ired
man
y as
sum
ptio
ns
Lack
of c
ompr
ehen
sive
val
idat
ion
No
indu
stria
l hyg
iene
mea
sure
men
ts
014
1.0
>0 –
<58.
216
1.5
(0.7
–3.0
)58
.2 –
<169
.717
3.6
(1.8
–7.
3)16
9.7
–<6
99.1
174.
6 (2
.3–9
.4)
Stud
y de
sign
His
toric
al c
ohor
t stu
dy
Stud
y po
pula
tion:
15
,826
wor
kers
em
ploy
ed ≥
6 m
onth
s bet
wee
n 19
48
and
1977
at 3
0 U
.S. r
einf
orce
d pl
astic
faci
litie
s
Mor
talit
y fo
llow
ed 1
948-
2008
.
Uni
ted
Stat
es
Col
lins e
t al.
2013
Exp
osur
e A
sses
smen
t
Det
aile
d in
dustr
ial h
ygie
ne su
rvey
was
con
duct
ed a
t eac
h si
te; w
ork
hist
ory
info
rmat
ion
was
obt
aine
d fr
om
empl
oyee
reco
rds
Ana
lysi
sFo
urex
posu
re m
etric
s, on
e of
whi
ch m
easu
red
“pea
k”:
-Cum
ulat
ive
peak
: wor
king
day
s with
pea
k ex
posu
re (1
5 or
mor
e m
inut
es a
bove
100
ppm
styr
ene)
Not
e: M
ore
than
hal
f of s
tudy
mem
bers
exp
erie
nced
no
peak
exp
osur
es; 6
% h
ad >
5 ye
ars o
f cum
ulat
ive
peak
ex
posu
res
No.
of d
ays w
ith p
eak
expo
sure
NH
LSM
REx
posu
re e
stim
ates
unc
erta
in
Num
ber a
nd ra
nge
of e
xpos
ure
sam
ples
not
av
aila
ble;
no
valid
atio
n of
exp
osur
e
Expo
sure
not
acc
rued
afte
r 197
7 (a
ffect
s 27
%);
how
ever
, exp
osur
es lo
w p
ast 1
977
mea
nex
posu
re=3
5 pp
min
196
7; m
ean
expo
sure
=25
ppm
in 1
977
Ave
rage
exp
osur
e co
ncen
tratio
ns d
eclin
ed
furth
er a
fter 1
977
Peak
mea
sure
men
t doe
s not
acc
ount
for
freq
uenc
y of
pea
ks (e
.g. f
ourp
eaks
dai
ly is
sa
me
as o
ne d
aily
)
Expo
sure
not
lagg
ed
Non
e20
0.70
(0.4
3-1.
08)
1–71
99
0.70
(0.3
2-1.
33)
720–
1,79
95
1.12
(0.3
7-2.
63)
≥1,8
002
0.49
(0.0
6-1.
76)
Leu
kem
iaN
one
210.
76 (0
.47–
1.17
)1–
719
120.
99 (0
.51–
1.74
)72
0–1,
799
30.
72 (0
.15–
2.09
)≥1
,800
41.
03 (0
.28
–2.6
3)L
LN
one
40.
57 (0
.16–
1.47
)1–
719
20.
68 (0
.08–
2.46
)72
0–1,
799
00.
0 (0
.0–3
.52)
≥1,8
002
1.95
(0.2
4 –7
.03)
ML
Non
e11
0.86
(0.4
3–1.
53)
1–71
97
1.15
(0.4
6–2.
38)
720–
1,79
92
0.96
(0.1
2–3.
48)
≥1,8
002
1.07
(0.1
3–3.
86)
HL
,Hod
gkin
lym
phom
a;L
L,l
ymph
atic
leuk
emia
;NH
L,n
on-H
odgk
inly
mph
oma;
ppm
,par
tspe
rm
illio
n;M
L,m
yelo
idle
ukem
ia;R
R,r
elat
ive
risk
;SM
R,s
tand
ardi
zed
mor
talit
yra
tio.
1452 Checkoway et al.
defined but peak effects were evaluated in combina-tion with cumulative exposure by assigning exposedcases and controls to one of four exposure categories:exposed to low or medium cumulative dose withoutpeaks, low or medium cumulative dose with peaks,high cumulative dose without peaks, and high cumu-lative dose with peaks. After adjusting for smokingand BMI, Charbotel et al. (2006) reported a signifi-cantly increased odds ratio for renal cell carcinomafor high cumulative dose without peaks and high cu-mulative dose with peaks (Table IV).
Two epidemiologic cohort studies of aircraftworkers examined peak exposure to TCE. Morgan,Kelsh, Zhao and Heringer (1998) studied 20,508workers employed at least six months between 1950and 1985 at an aircraft manufacturing facility in Ari-zona. The cohort was followed for mortality from1950 to 1992. A total of 4,733 workers were classi-fied as exposed to TCE during their work. Peak ex-posures were defined as the job with the highest TCEexposure rating. Peak exposures were analyzed bycomparing workers with high and medium TCE ex-posure jobs (where peak exposures were assumed tohave occurred) to workers with no and low TCE ex-posure. High TCE exposures, estimated to be above50 ppm, involved assignments to degreaser machines,while work in areas near the vapor degreasing unitand with more than occasional contact with TCEwere assigned medium TCE exposure. No statisti-cally significant increased risks of leukemia, lym-phoma, or kidney cancer were observed for workerswith peak exposures when compared to workers withlow (occasional contact with TCE) and no TCE ex-posure (Table IV).
Radican, Blair, Stewart, and Wartenberg (2008)followed mortality from 1953 through 2000 in acohort of 14,455 aircraft maintenance workers em-ployed for one year or more between 1952 and 1956as civilians at Hill Air Force Base. Workers wereassigned to categories of TCE exposure based onjob tasks that identified four patterns of frequencyand exposure: intermittent exposure (i.e., used TCEinfrequently during the day), continuous exposure(i.e., used TCE regularly throughout the day), lowexposure (i.e., used TCE for bench top work to cleansmall parts), or peak exposure (i.e., workers whoused vapor degreasers). Each worker was assigned toone of four categories of exposure: low intermittentexposure, low continuous exposure, infrequentpeak exposure, and frequent peak exposure. Thisexposure metric differed somewhat from an earlierreport on the same cohort (Blair, Hartge, Stewart,
McAdams, & Lubin, 1998) that described resultsaccording to three categories of TCE exposure:low-level intermittent exposure, low-level contin-uous exposure, and frequent peaks. Radican et al.(2008) reported NHL mortality risks did not differsubstantively between infrequent peaks or frequentpeaks. Leukemia mortality was not associated withinfrequent peak exposure or frequent peak exposure,although the number of deaths in each group (threeand nine, respectively) was small. Similarly, mortalityfrom kidney or liver cancers was not associated withinfrequent or frequent peak exposure (Table IV).
3.5. Acrylonitrile and Lung Cancer
In 1999, IARC reclassified acrylonitrile aspossibly carcinogenic to humans (IARC, 1999)based on its determination that an earlier charac-terization of acrylonitrile as probably carcinogenicto humans (IARC, 1987) was not supported byupdated epidemiologic evidence. Lung cancer andprostate cancer were not consistently associated withacrylonitrile exposure in the earliest epidemiologicstudies. More recently, the US EPA released a draftIRIS assessment that characterized the weight ofevidence for acrylonitrile as likely to be carcinogenicto humans. The risk assessment used an epidemi-ologic study of the association between cumulativeexposure and lung cancer mortality (Blair, Stewart,et al., 1998) for the dose–response assessment toderive an IUR for lung cancer (US EPA, 2011c).
In addition to cumulative exposure, Blair, Stew-art et al. (1998) evaluated peak exposure in rela-tion to site-specific cancer risks for 25,460 workersin eight facilities producing or using acrylonitrile inthe United States. Peak exposures were defined as15-minute excursions that averaged 20 ppm orgreater and the analysis of peaks was based on fre-quency of peaks. Stewart, Zaebst et al. (1998) con-ducted an exposure reconstruction for the eight fa-cilities and estimated frequency of peaks based on8-hour TWAs that allowed (mathematically) for thepossibility of a peak of 20 ppm,7 and by consideringwhether tasks were likely to produce a peak (suchas manually taking a process sample). Lung can-cer mortality did not increase with cumulative expo-sure, although lung cancer mortality was slightly in-creased among the group with the highest cumulativeexposure (RR = 1.50, 95% 0.9–2.4 for >8 ppm-years)
7That is, a minimum 8-hour TWA of about 0.6 ppm (20 ppm ×15 min/480 min) (Stewart, Zaebst et al., 1998).
Peak Exposures in Epidemiologic Studies and Cancer Risks 1453
Tab
leIV
.T
rich
loro
ethy
lene
and
Can
cers
Stud
y D
esig
n, P
opul
atio
n, R
efer
ence
Ex
posu
reEx
posu
re C
ateg
orie
sO
bser
ved
[Cas
es/C
ontro
ls]
RR
(95%
CI)
Expo
sure
Ass
essm
ent L
imita
tions
and
Com
men
tsSt
udy
Des
ign
Hos
pita
l-an
d cl
inic
-bas
ed c
ase-
cont
rol s
tudy
Stud
y po
pula
tion
86 c
ases
of r
enal
cel
l car
cino
ma
and
316
cont
rols
. C
ontro
ls re
crui
ted
from
pat
ient
s with
out R
CC
of t
he
sam
e ur
olog
ist.
Cas
es a
nd c
ontro
ls id
entif
ied
1993
-200
3
Fran
ce
Cha
rbot
el e
t al.
(200
6)
Exp
osur
e as
sess
men
tO
ccup
atio
nal h
isto
ry o
btai
ned
for c
ases
and
con
trols
us
ing
a te
leph
one
inte
rvie
w
Sem
i-qua
ntita
tive
expo
sure
ass
essm
ent f
or T
CE
Aco
nfid
ence
scor
e w
as g
iven
: 3=
certa
in2=
prob
able
1=po
ssib
le (f
or e
ach
of th
e ex
posu
res a
sses
sed)
Exp
osur
e an
alys
isTh
ree
expo
sure
met
rics,
one
of w
hich
incl
uded
pea
k (c
umul
ativ
e do
se p
lusp
eaks
)-e
xpos
ed to
low
or m
ediu
m c
umul
ativ
edo
se w
ithou
t pe
aks,
-low
or m
ediu
m c
umul
ativ
e do
se w
ith p
eaks
, -hi
gh
cum
ulat
ive
dose
with
out p
eaks
-h
igh
cum
ulat
ive
dose
with
pea
ks.
Cum
ulat
ive
dose
plu
s pea
ksR
enal
cel
l car
cino
ma
OR
OR
adju
sted
for t
obac
co sm
okin
g (n
on-
smok
er, 1
-20,
20-
40, a
nd 4
0 or
mor
e pa
ck-
year
s) a
nd B
MI (
≤25,
25
-29.
9, ≥
30)
Stat
istic
al si
gnifi
canc
e se
en o
nly
for h
igh
cum
ulat
ive
expo
sure
plu
s pea
ks.
OR
for
expo
sed
with
pea
ks w
as n
ot si
gnifi
cant
ly
elev
ated
whe
n co
mpa
red
to e
xpos
ed
with
out p
eaks
(ris
k es
timat
e no
t rep
orte
d)
Sens
itivi
ty a
naly
sis:
incl
uded
onl
y al
ive
subj
ects
<80
yea
rs o
f age
and
onl
y jo
b pe
riods
des
crib
ed w
ith sc
rew
-cut
ting
ques
tionn
aire
and
hav
ing
high
con
fiden
ce
in T
CE
expo
sure
Low
/med
ium
,w
ith p
eaks
[8/3
]1.
61 (0
.36
-7.3
0)
Hig
h, w
ith p
eaks
[8/1
4]2.
73 (1
.06
-7.0
7)
Stud
y de
sign
His
toric
al c
ohor
t stu
dy
Stud
y po
pula
tion
20,5
08 a
eros
pace
wor
kers
incl
udin
g 4,
733
wor
kers
ex
pose
d to
tric
hlor
oeth
ylen
e em
ploy
ed ≥
6 m
onth
s be
twee
n 19
50 a
nd 1
985
at o
ne fa
cilit
y.
Mor
talit
y fo
llow
ed 1
950-
1993
.
Uni
ted
Stat
es
Mor
gan
et a
l. (1
998)
Exp
osur
e A
sses
smen
t
Wor
k hi
story
info
rmat
ion
was
obt
aine
d fro
m e
mpl
oyee
re
cord
s
Job-
expo
sure
mat
rix w
as c
reat
ed: T
CE
expo
sure
was
ra
ted
high
bas
ed o
n w
ork
on d
egre
aser
mac
hine
s(IH
es
timat
e w
as e
quiv
alen
t to
50 p
pm);
med
ium
for j
obs
near
the
degr
easin
g ar
ea w
ith m
ore
than
occ
asio
nal
expo
sure
; and
low
for j
obs
with
occ
asio
nal c
onta
ct w
ith
TCE
Ana
lysi
sTw
oex
posu
re m
etric
s, on
e of
whi
ch m
easu
red
“pea
k”:
-pea
k: jo
b w
ith th
e hi
ghes
t TCE
exp
osur
e ra
ting.
Peak
LH
MH
RLi
mite
d in
dustr
ial h
ygie
ne d
ata
befo
re
1975
; TCE
exp
osur
e w
as ra
ted
by
empl
oyee
s w
ith m
ore
than
30
year
s of
expe
rienc
e
Smal
l num
bers
of c
ance
rs in
the
TCE-
expo
sed
coho
rt
Low
& n
one
901.
00 (R
efer
ence
)M
ediu
m&
hig
h17
1.08
(0.6
4–1.
82)
Leu
kem
iaLo
w &
non
e35
1.00
(Ref
eren
ce)
Med
ium
& h
igh
71.
10 (0
.49–
2.49
)L
ung
Low
& n
one
324
1.00
(Ref
eren
ce)
Med
ium
& h
igh
641.
07 (0
.82–
1.40
)L
iver
Low
& n
one
171.
00 (R
efer
ence
)M
ediu
m&
hig
h3
0.98
(0.2
9–3.
35)
Kid
ney
Low
& n
one
241.
00 (R
efer
ence
)M
ediu
m&
hig
h8
1.89
(0.8
5–4.
23)
Bla
dder
Low
& n
one
181.
00 (R
efer
ence
)M
ediu
m&
hig
h5
1.41
(0.5
2–3.
81)
Pros
tate
Low
& n
one
601.
00 (R
efer
ence
)M
ediu
m&
hig
h16
1.47
(0.8
5–2.
55)
Ova
rian
Low
& n
one
91.
00 (R
efer
ence
)M
ediu
m&
hig
h4
2.74
(0.8
4–8.
99)
(Con
tinue
d)
1454 Checkoway et al.
Tab
leIV
(Con
tinu
ed)
Stud
y D
esig
nH
isto
rical
coh
ort s
tudy
Stud
y po
pula
tion
14,4
55 (1
0,73
0 m
en, 3
725
wom
en) c
ivili
an
empl
oyee
s of H
ill A
ir Fo
rce
Base
em
ploy
ed ≥
1 y
ear,
1952
-195
6
Follo
wed
for m
orta
lity
1950
-200
0
Uni
ted
Stat
es
Rad
ican
et a
l. (2
008)
Exp
osur
e A
sses
smen
tW
ork
histo
ry d
ata
was
obt
aine
d fr
om e
mpl
oym
ent
reco
rds.
Each
job-
depa
rtmen
t com
bina
tion
was
ass
igne
d di
chot
omou
s exp
osur
e (y
es/n
o) to
21
solv
ents
and
ch
emic
als
For T
CE, f
requ
ency
and
pat
tern
of e
xpos
ure
was
as
sign
ed b
ased
on
job
task
s. In
term
itten
t or c
ontin
uous
ex
posu
re w
as a
ssig
ned
to e
mpl
oyee
s who
use
d TC
E in
freq
uent
ly o
r reg
ular
ly d
urin
g th
e da
y. L
ow e
xpos
ure
assi
gned
to su
bjec
ts w
ho u
sed
TCE
for b
ench
top
wor
k (c
lean
ing
smal
l par
ts) a
nd p
eak
expo
sure
ass
igne
d to
in
divi
dual
s who
wor
ked
with
vap
or d
egre
aser
s.
Exp
osur
e an
alys
isSi
x ex
posu
re m
etric
s, tw
o of
whi
ch in
clud
ed “
peak
.”-P
eak
infr
eque
nt
-Pea
k fre
quen
t
Peak
ME
NL
HM
HR
Expo
sure
was
est
imat
ed b
ased
on w
ork
hist
orie
s; a
utho
rs a
ssum
ed n
on-d
iffer
entia
l m
iscl
assi
ficat
ion
of e
xpos
ure
and
expe
cted
bi
as to
war
d th
e nu
ll.
Smal
l num
bers
of d
eath
s,es
peci
ally
am
ong
wom
en.
No
wom
en w
ith p
eak
expo
sure
die
d fr
om
Hod
gkin
lym
phom
a or
live
r dis
ease
.
Mos
t wor
kers
wer
e al
so e
xpos
ed to
oth
er
chem
ical
s; h
owev
er, r
efer
ence
gro
up w
as
com
pris
ed o
f ind
ivid
uals
who
wer
e ne
ver
expo
sed
to a
ny c
hem
ical
s in
the
wor
kpla
ce.
No
expo
sure
1836
1.0
(Ref
eren
ce)
Infre
quen
t17
1.39
(0.7
6–2.
57)
Freq
uent
421.
24 (0
.76–
2.03
)H
LN
o ex
posu
re18
361.
0 (R
efer
ence
)In
frequ
ent
12.
00 (0
.13–
31.9
8)Fr
eque
nt4
2.70
(0.3
0–24
.15)
NH
LN
o ex
posu
re18
361.
0 (R
efer
ence
)In
frequ
ent
71.
90 (0
.69
–5.2
4)Fr
eque
nt16
1.57
(0.6
7–3.
69)
MM
No
expo
sure
1836
1.0
(Ref
eren
ce)
Infre
quen
t5
1.78
(0.5
4 –5
.84)
Freq
uent
101.
31 (0
.48
–3.6
3)L
euke
mia
No
expo
sure
1836
1.0
(Ref
eren
ce)
Infre
quen
t3
0.63
(0.1
7–2.
30)
Freq
uent
90.
69 (0
.28–
1.69
)L
iver
No
expo
sure
1836
1.0
(Ref
eren
ce)
Infre
quen
t3
6.42
(0.6
7–61
.87)
Freq
uent
32.
13 (0
.22–
20.4
3)K
idne
yN
o ex
posu
re18
361.
0 (R
efer
ence
)In
frequ
ent
21.
04 (0
.19–
5.70
)Fr
eque
nt6
1.11
(0.3
1–3.
96)
WO
ME
NL
HM
No
expo
sure
1983
1.0
(Ref
eren
ce)
Infre
quen
t4
1.98
(0.6
9–5.
68)
Freq
uent
111.
24 (0
.76–
2.03
)N
HL
No
expo
sure
1983
1.0
(Ref
eren
ce)
Infre
quen
t3
3.45
(0.9
6–12
.37)
Freq
uent
61.
27 (0
.47–
3.45
)M
MN
o ex
posu
re19
831.
0 (R
efer
ence
)In
frequ
ent
13.
20(0
.36–
28.6
9)Fr
eque
nt3
1.93
(0.4
3–8.
65)
Leu
kem
iaN
o ex
posu
re19
831.
0 (R
efer
ence
)In
frequ
ent
0--
Freq
uent
20.
38(0
.08–
1.76
)K
idne
yN
o ex
posu
re19
831.
0 (R
efer
ence
)In
frequ
ent
0--
Freq
uent
21.
50(0
.24–
9.40
)
Stud
y D
esig
n, P
opul
atio
n, R
efer
ence
Ex
posu
reEx
posu
re C
ateg
orie
sO
bser
ved
[Cas
es/C
ontro
ls]
RR
(95%
CI)
Expo
sure
Ass
essm
ent L
imita
tions
and
Com
men
ts
HL
,Hod
gkin
lym
phom
a;H
R,h
azar
dra
tio;
MM
,mul
tipl
em
yelo
ma;
NH
L,n
on-H
odgk
inly
mph
oma;
ppm
,par
tspe
rm
illio
n;O
R,o
dds
rati
o.
Peak Exposures in Epidemiologic Studies and Cancer Risks 1455
when compared to the group with cumulative expo-sure <0.13 ppm-years. Lung cancer mortality did notincrease with frequency of peaks, and results of thepeak analysis for other cancer sites were not pre-sented.
Swaen et al. (2004) evaluated a cohort of 2,842workers in the Netherlands producing or usingacrylonitrile employed at eight companies and 3,961workers at a nitrogen fixation facility for fertilizerproduction that were not exposed to acrylonitrile.In earlier years, industrial hygienists had performedexposure monitoring to identify locations withpotential for peak acrylonitrile concentrations (seeSwaen et al., 1992). The investigators defined peakexposures as “intervals with elevated exposure con-centrations in ranges of <10, 10–20, and over 20 ppmwhich occurred regularly on at least a weekly basis.”The duration of intervals was not specified. Mortalityfrom lung cancer, prostate cancer, brain cancer,and leukemia were analyzed in relation to peakexposure, but no associations were seen.
3.6. Ethylene Oxide and LHMs and Breast Cancer
IARC classified ethylene oxide as carcinogenicto humans, based on limited epidemiologic evidencefor a causal association with lymphoid tumors(NHL, multiple myeloma, and chronic lymphocyticleukemia) and breast cancer, sufficient evidencein experimental animals, and strong evidence thatEO acts by a genotoxic mechanism (IARC, 2008,2012b). The US EPA (2016) derived an IUR for thecombined risk of lymphoid cancer and female breastcancer using an epidemiologic study of mortalityrisks (and breast cancer incidence) associated withcumulative exposure (Steenland, Stayner, & Ded-dens, 2004; Steenland, Whelan, Deddens, Stayner,Ward, 2003).
Although Steenland et al. (2004) was one ofthree epidemiologic studies that evaluated peakexposure metrics in models, the investigators re-ported that models based on peak exposure didnot predict NHL, Hodgkin disease, leukemia, ormultiple myeloma, and therefore did not present theresults for these models. Steenland et al. (2004, 2003)defined peak exposure as the “highest one-time ex-posure.” In an earlier analysis of the cohort, Stayneret al. (1993) reported that the highest one-time expo-sure for each employee was the maximum recordedTWA exposure over the worker’s job history.Stayner et al. (1993) reported “it was not possibleto test the influence of short-term exposure peaks
experienced during the course of the day, since real-time data on ethylene oxide exposure levels were notavailable for each individual in the study.” Approxi-mately 2,400 TWA measurements from 1976 to 1985were used to inform the exposure assessment model(Greife, Hornung, Stayner, & Steenland, 1988).(This highlights the need to review earlier studies toidentify a clear definition of “peak” exposure.)
Two additional epidemiologic studies reported“peak” exposures in descriptions of cohorts ofethylene oxide manufacturing workers, but thesestudies did not evaluate risk estimates using peakexposure metrics (Coggon, Harris, Poole, & Palmer,2004; Greenberg, Ott, & Shore, 1990). Greenberget al. (1990) did not define peak exposures but notedthat there were no deaths from leukemia “amongthe subcohort of men who worked where bothaverage and peak exposure levels were probablyhighest” (likely before 1978 in production unitswith continuously operating EO converters andrecovery systems). Coggon et al. (2004) reported“environmental and personal monitoring carried outsince 1977 indicated TWA exposures of less than5 ppm in almost all jobs, but with occasional peaksof up to several hundred ppm because of operatingdifficulties in the chemical plants and when sterilizerswere loaded and unloaded in hospitals. In earlieryears, exposures were probably somewhat higher,and peak exposures above the odor threshold of700 ppm were reported at both factories andhospitals.” Breast cancer risk was not increasedamong women hospital workers with continual orintermittent EO exposure (Coggon et al., 2004).
3.7. Methylene Chloride and NHL, Biliary Tract,Liver Cancers
An IARC Working Group classified methylenechloride as probably carcinogenic to humans and de-scribed the epidemiologic evidence as limited, not-ing positive associations with cancer of the biliarytract and NHL (IARC, 2017). The US EPA dose–response assessment derived an IUR for the com-bined risk of liver and lung tumors using a PBPKmodel of glutathione S-transferase metabolism dosemetrics for mice (US EPA, 2011a).
We identified only one study that discussed peakexposure to methylene chloride. Hearne and Pifer(1999) reported historical ambient measurementsranging between 5 and 100 ppm and short-term peakconcentrations of 1,000 to 10,000 ppm measuredduring manual adjustments of roll coating machines
1456 Checkoway et al.
in cellulose triacetate film manufacture. The inves-tigators also reported peak concentrations greaterthan 1,000 ppm several times per day when filtermedia were changed or during the loading of batchmixers in the Dope department; however, there wereno risk analyses by a peak exposure metric. There-fore, the literature on methylene chloride contributesto this review only as an example of descriptionsof the occurrence of peak exposure (similar to thatreported for formaldehyde in several studies).
4. ISSUES WITH ESTIMATING ANDINTERPRETING ASSOCIATIONS WITHPEAK EXPOSURES IN EPIDEMIOLOGICSTUDIES
There appear to be sound scientific reasons citedfor considering peak exposures in epidemiologicevaluations of chemicals carcinogens (Supporting In-formation Table SII). Not the least of these is the bi-ological understanding that peak exposures that ex-ceed certain levels or thresholds—even for brief pe-riods of time—likely trigger metabolic processes ordirect genetic, epigenetic, or cytotoxic damage thatlead to cancers (Kriebel, Checkoway, & Pearce, 2007;Macaluso et al., 2004; Morgan et al., 1998; Rappa-port, 1991; Stewart et al., 1986, 1992). In theory, accu-rate characterization of these peak exposures wouldenhance the identification of exposed groups withincreased cancer risks and their differentiation fromgroups at lower or no increased risk. Identificationof risks associated with peak exposures also wouldimpact the risk assessment process and derivation ofexposure limits—especially occupational STELs.
However, we identified surprisingly few, if any,good examples where efforts to define and evaluatepeak exposures systematically and rigorously haveinformed valid risk assessment. Our review pointsto several limitations and issues that might, at leastin part, explain this dearth of epidemiologic studieseffectively identifying increased cancer risks amongthose exposed to peaks.
4.1. Lack of Consistent Definitions ofPeak Exposures
Epidemiologic studies sometimes rely on data—largely 8-hour TWA measurements—routinelycollected to demonstrate compliance with regula-tory standards. “Peak exposure” is not a standardindustrial hygiene term, however, and may not bemeasured even when a STEL exists for a chemical.In the epidemiologic literature we reviewed, peak
exposure was defined in different ways, including butnot limited to the following: the highest short-term(15-minute) exposure concentration sustained byan individual (Hauptmann et al., 2009), the highestTWA experienced by an exposed individual (Steen-land et al., 2003, 2004), the number of days withpeaks that exceed an estimated short-term value(Collins et al., 2003), the estimated relative peakand estimated number of peaks (Beane Freemanet al., 2009), and the estimated number of totalpeaks exceeding a threshold of exposure (>100ppm butadiene or >50 ppm styrene), regardlessof duration (Cheng, et al., 2007; Graff et al., 2005;Macaluso et al., 2004).
Others relied upon derived semiquantitativepeak scores or qualitative rating (Blair, Hartge, et al.,1998; Morgan et al., 1998; Stenehjem et al., 2015).One study evaluated effect of peaks by combining“peak” with cumulative exposure: exposed to lowor medium cumulative dose without peaks, low ormedium cumulative dose with peaks, high cumulativedose without peaks, and high cumulative dose withpeaks (Charbotel et al., 2006). Finally, Graff et al.(2005) also analyzed leukemia risks separately bybutadiene ppm-years that had been partitioned intoexposure concentrations above and below the peakthreshold (i.e., >100 ppm butadiene) (Graff et al.,2005; Macaluso et al., 2004).
Some definitions of peak exposure incorporateda time dimension, such as a 15–60 minute excursionabove an average or some threshold for a specifiedfrequency of occurrence (e.g., peaks occur at leastweekly) (Checkoway et al., 2015; Glass et al., 2014;Schnatter et al., 2012). Others included “short-termexposure levels” (without specifying the time in-terval). The definitions of peak varied across theepidemiologic literature, which complicates interpre-tation of data from studies applying different peakmetrics. Thus, uniform definitions for peak, relevantto research on environmental risk factors for cancer,clearly are needed before their ability to improverisk assessments can be determined.
4.2. Lack of Reliable Exposure Data or ValidatedPeak Metric
In the studies that analyzed risks in relation tosome form of peak exposure, nearly all peak expo-sure estimates were based on expert judgment andnot on measurements. Although study investigatorsfrequently discussed the validity and reliability of thesummary exposure estimate (e.g., average estimates
Peak Exposures in Epidemiologic Studies and Cancer Risks 1457
in the JEM cells), the validity and reliability of peakexposure metrics were rarely considered. Validationof peak exposures was not reported in any study,although the U.S. formaldehyde industrial workerscohort reported an evaluation of the TWA measure-ments. Blair and Stewart (1990) also reported corre-lations among various formaldehyde exposure met-rics; however, this would not necessarily validate anyof the metrics. Short-term exposure measurementsrarely were available, and in none of the studies wereactual peaks measured for individual workers or forjob-defined groups of workers applied to a JEM. Thisis not surprising, as most studies relied on average ex-posure or qualitative exposure metrics based on jobtitle/work area and calendar year(s). Although somestudies had limited historical industrial hygiene mea-surements, few can be considered good indicatorsof individual worker or group-level peak exposure.Average exposures assigned to workers with higherand lower actual exposures leads to misclassification.Assuming a true relationship with peak exposures,this random misclassification would be expected toreduce the relative risk estimates associated withpeak exposures and the associated risk numbers.
In many cases, the examples we found in re-lation to peak exposure did not report sensitivityanalyses of the potential magnitude and direction ofbias associated with the use of the peak exposuremetric. However, some studies conducted analysesthat evaluated two measures of peak exposure, suchpeak intensity and frequency (hourly, daily, weekly,monthly) or number of peaks, or hybrid measureswith a peak component (e.g., peak frequency, aver-age or cumulative peaks). For example, Beane Free-man et al. (2009) evaluated the frequency of peaksand reported relative risks of any LHM did not in-crease with the number of peaks � 4 ppm, althoughthe data were not shown. Even if this approach rea-sonably describes the exposure of groups of employ-ees working in a given job or area, it is not knownwhether any specific individual (including those de-veloping cancers) actually had been exposed to anypeaks. In contrast, there was fairly good agreementbetween risk estimates for leukemia associated withppm-years above exposure intensities of 100 ppmbutadiene and number of total peaks of butadieneabove 100 ppm (Graff et al., 2005).
In the benzene literature, risks were elevatedfor peak and for cumulative exposure. In contrast,the styrene example did not find relations betweenpeak exposure and any of the LHMs. Beane Free-man et al. (2009) reported associations between their
peak formaldehyde exposure metric and LHM andML, but not with cumulative, average, or duration ofexposure—or with frequency (hourly, daily, weekly,monthly, no peaks) or total number of peaks. Haupt-mann et al. (2009) found no association betweenpeak exposure and ML; however, when the analysisof “peak exposure” was restricted to embalmers, astatistically significant trend was seen with peak ex-posure among embalmers for whom a lifetime historyof performing more than 500 embalmings was re-ported by next of kin (many decades later). The oddsratios for the different peak exposure categories werebased on few myeloid leukemias and even fewerAMLs, and none were statistically significant.
Further complicating this is the likelihood thatpeak exposures occurred more frequently in work-places in the distant past, and their frequency (andpossibly magnitude) has declined over time due toimproved controls and regulations. For example,although increased AML risk was reported amongembalmers, 76% of those whose cause of death wasclassified as AML were first employed in a funeralhome prior to WWII, and 44% were first employedbefore 1932, requiring speculative approaches toexposure assessment (in which peak exposuresultimately were reported in categories defined bythe following narrow ppm ranges: <7; >7–9.3; >9.3)and reduced comparability with the control group(Hauptmann et al., 2009).
Therefore, while epidemiologially evaluatingcancer risks assocated with peak exposure remainsconceptually atrractive, there are too few exampleswhere studies demonstrate the contribution of peakexposure to risk beyond that estimated—even ifpoorly—by other exposure metrics.
4.3. Exposure Misclassification
Peaks and their duration are also componentsof average and cumulative exposure. However, asillustrated in Fig. 1, variability of exposure inten-sity (including short-term excursions or exceedancesthat may be considered exposure peaks) and fre-quency over time can result in distinctly different ex-posure profiles for workers with the same cumulativeexposure. Epidemiologically, this would be consid-ered exposure misclassification. Sufficient exposuresampling is necessary to capture relatively rare oc-currences of peaks; otherwise, they will be missedand contribute only trivially to cumulative expo-sure estimates. When time-varying exposures fluctu-ate from hour to hour and day to day between cohort
1458 Checkoway et al.
0
5
10
15
20
2519
9019
9119
9219
9319
9419
9519
9619
9719
9819
9920
0020
0120
0220
0320
0420
0520
0620
0720
0820
09
Expo
sure
con
cent
ra�o
n
Year
Worker AWorker BWorker C
Fig. 1. Peak exposure profiles for three workers with identicalcumulative exposure after 20 years. Worker A encounters onepeak, Worker B encounters five peaks, and Worker C does notencounter peaks.
members in the same job setting, infrequent sam-pling of exposure risks missing the measurement ofpeak exposures because these peak exposures mayfall outside of rare sampling occasions. These peakexposures may constitute a large fraction of cumula-tive exposure, in theory. In addition, historical sam-pling data were often collected to demonstrate com-pliance with a regulatory standard rather than toidentify and characterize heterogeneity in exposurescenarios. This practice likely results in exposure mis-classification even when the exposure metric is cumu-lative exposure.
Many studies offered that nondifferentialmisclassification of cumulative exposure resultsin risk estimates that are biased toward the null(Beane Freeman et al., 2009; Hauptmann et al.,2009; Radican et al., 2008; Schnatter et al., 2012;Stenehjem et al., 2015). Although seemingly true,Jurek, Greenland, Maldonado, and Church (2005)and Lash et al. (2014) suggest that nondifferentialexposure misclassification only guarantees an expec-tation that on average, the bias will be toward thenull when exposure is assessed using a dichotomousvariable (exposed/not exposed). It is well known thatwhen risks are evaluated using multiple exposurecategories (as they generally are in historical recon-structions of average and peak exposure intensity),exposure misclassification results in biased riskestimates in the middle categories that are awayfrom the null (Dosemeci, Wacholder & Lubin, 1990;Jurek et al., 2005; Weinberg, Umbach, & Green-land, 1994). Furthermore, the bias in any particularstudy can be away from the null because the actual
misclassification can result in a data arrangementfar from what is expected. We raise one particularconcern with respect to this issue: in studies that relyupon expert judgment to estimate average exposureintensity and peak exposure (especially semiquanti-tative and quantitative measurements) based on jobtasks and industrial processes, it is conceivable (evenrational) that experts are subject to an “anchoringeffect,” a cognitive bias in which initial information(estimates of or measurements of TWA concentra-tions) may influence the estimate of peak exposure.For example, experts may be more likely to assignprobability and/or frequency of experiencing peakexposures when the 8-hour TWA value (measuredor estimated) is high and may be less likely to assignthe probability and/or frequency of experiencingpeak exposures when the 8-hour TWA value is low.Conversely, experts may be more likely to estimatehigher 8-hour TWA values when information aboutprocesses and job tasks suggests higher probabilityand/or frequency of peak exposure. When peak isdefined by multiple categories, which are themselvesbased on consideration of multiple categories ofaverage concentrations, we hypothesize that esti-mated peak exposure in particular is susceptible toincreased variability and bias that is not predictableas to its direction or magnitude. In addition, analysesby multiple exposure metrics also increase theopportunity for false-positive associations to arisesimply by chance (Blair, Stewart, et al., 1998).
4.4. Consideration of Disease Latency in Relationto Peak Exposure Metric
Specific types of most cancers, including theLHMs, are relatively rare, and all but the largest oc-cupational epidemiologic studies have too few casesto allow a meaningful analysis across peak expo-sure categories. If peak exposures also are rare, thenumber of cases with peak exposures will be evensmaller, precluding any meaningful estimate of rela-tive risk. For example, Checkoway et al. (2015) notedthat the updated NCI formaldehyde industrial work-ers study (Beane Freeman et al., 2009) identified34 AML deaths, the type of leukemia most plau-sibly related to chemical exposures. However, lessthan one-third (n = 13) of these were classified ashaving had any peak exposure (defined as working injobs expected to have peak formaldehyde exposures>2 ppm).
Furthermore, when reasonable latency timewindows are considered, even fewer observed cases
Peak Exposures in Epidemiologic Studies and Cancer Risks 1459
remain plausibly linked with exposure. Leukemiaslikely have shorter latencies than solid tumors. Stud-ies of atomic bomb survivors in Japan reported thegreatest increases in AML incidence occurred 5–7years after exposure and declined thereafter (Darby,Nakashima, & Kato, 1985; Moloney & Lange 1954).Incidence of AML following chemotherapy is high-est within 5–10 years after treatment (Deschler &Lubbert, 2006) but could be as short as 2–3 yearsfollowing treatment with topoisomerase inhibitors(IARC, 2012a). Considering latency in the BeaneFreeman et al. study (2009) of 13 AML deathsamong workers classified as having a peak exposure,only four had jobs associated with peak exposurewithin the 20 years preceding death, and only oneof these occurred within a conservative latencywindow of 2 to 15 years after the last peak exposure.Therefore, it is not possible to draw any conclusionregarding peak exposure and AML from these studydata (Checkoway et al., 2015).
Many epidemiologic studies address cancer la-tency by evaluating different time windows of expo-sure; for example, Glass et al. (2014) used a 2–20-yearexposure window and Schnatter et al. (2012) used a2–15-year exposure window. Relevant time windowsof risk may be evaluated by lagging cumulative (orother) exposures, and only exposures that occurredin specified time intervals before diagnosis or deathare considered. Only a few studies that reported can-cer risks in relation to peak exposure metrics alsoconducted analyses in which the investigators laggedpeak exposures (Beane Freeman et al., 2009; Check-oway et al., 2015; Cheng et al., 2007; Hauptmannet al., 2009; Steenland et al., 2003, 2004).
Furthermore, even for agents classified as leuke-mogens (benzene, formaldehyde, BD), which we as-sume have shorter latencies, we found little com-pelling evidence that any peak exposure metric waspredictive of increased leukemia (or LHM) risk, andespecially in the absence of demonstrable increasedrisks with cumulative exposure. This likely reflectsweaknesses and inconsistencies in the exposure met-rics employed, even if an association with peak expo-sure exists.
Some modern exposure monitoring appro-aches—even if not yet commonly used—wouldallow rapid or even real-time measurement of airconcentrations. In theory this would validly identifythe exact time and magnitude of peak exposures;however, detection of increased cancer risk as aconsequence of these exposure(s) would require (1)large numbers of workers exposed to peaks and (2)
the workers to be followed for up to 10 or 15 yearsdue to latency issues. Therefore, it is unlikely thatstudies using measured peak exposures to examinecancer risks will be common if conducted at all, anduniform and transparent approaches will need to berefined for retrospectively deriving peak exposuremetrics for the relevant latency time window.
4.5. Considerations of Biological Processes andMechanisms Associated with Peak Exposure
We postulate that it is plausible that one or moremechanisms for carcinogenesis may be initiated by asudden upward spike in the intensity of exposure, tothe degree that such an exposure reflects an under-lying “dose rate” that suddenly accelerates. Do con-tributions to risk differ if such a “dose rate” plateausand moves to a higher average exposure or, alterna-tively, decreases?
Specific hypotheses regarding the importance ofpatterns of exposure are rarely discussed explicitly(i.e., how does the pattern of exposure factor intothe increased risk?). For example, alternative expla-nations for increased risks of neoplasms at low lev-els of exposure include: (1) average exposures in thepast were underestimated based on recent lower lev-els of exposure and past exposures were actually con-siderably higher than estimated; and (2) occasionalhigh exposures (peaks) factor disproportionately intothe risks associated with lower average exposures. In-stead, pattern of exposure is viewed as part of theblack box—and, therefore, risk estimates should becalculated using multiple exposure metrics becausethe underlying disease process is unknown. Interest-ingly, cumulative exposure has emerged as the de-fault metric in a majority of epidemiologic studiesthat quantify exposures, most likely due to simplic-ity (e.g., “more exposure should result in more risk”)rather than a biological understanding of underlyingdisease pathways and what type of exposures (e.g.,peaks or other exposures exceeding some threshold)increase risk.
5. CHALLENGES FOR IMPLEMENTATIONOF UNIFORM PEAK EXPOSURE METRICSFOR CANCER RISK EVALUATION
For several chemicals classified as known orprobable carcinogens, we attempted to identifythe main epidemiologic studies that reported rel-ative risks in relation to some form of peakexposure. However, we found “peak” exposure
1460 Checkoway et al.
characterization to be highly idiosyncratic. There-fore, perhaps it is not surprising that we did not findany clear or convincing examples in the epidemio-logic literature of increased cancer risks in relation topeak exposure estimates for chemicals, even wherethe chemicals had been classified as known carcino-gens based on or strongly influenced by the epidemi-ologic evidence. To our knowledge, no regulatorycarcinogenic risk assessment has been based on epi-demiologic findings using peak exposure. The draftIRIS assessment for formaldehyde reported an as-sociation between peak exposure and LHM, espe-cially myeloid leukemia, but relied on cumulative ex-posure metrics in the dose–response assessment. Allof the studies we reviewed were limited by the sim-ple fact that industrial hygiene measurements of ex-cursions in short-term exposures are relatively rare,and there are no standard sampling schemes to mea-sure them. Most studies classified workers as havingpeak exposures based on expert opinion or an under-standing of the working environment and not specificexposure measurements. Therefore, real-time moni-toring of workplace air quality may shed some lighton this conundrum; however, dose–response model-ing based on peak exposure has not been defined andwill require development.
Standardized exposure definitions for “peak ex-posure” do not exist currently in epidemiology butwarrant separate consideration given possibly uniquecontributions to risk. Unlike cumulative exposuremeasures, which can be compared readily acrossstudies, peak exposure has not been defined suffi-ciently similarly across studies measuring the sameoutcomes. Therefore, it is not clear how the con-tribution to risk from peak exposure can be evalu-ated separately from cumulative exposure in eluci-dating exposure responses for risk assessment andrisk assessors will not be able to evaluate con-tributions to risk specifically associated with peakexposure.
Although it is not within the scope of this arti-cle to derive standard definitions for peak exposure,a number of parameters and desirable qualities canbe offered. First, quantified exposure measurementsand estimates (e.g., ppm) are preferable over qualita-tive estimates (e.g., exposed/not exposed, or rankedon an ordinal scale, such as “low/background,”“moderate,” and “high”), provided that they are rea-sonably valid and some estimate of variability (in-cluding measurement error) is included. Second, thetime period and frequency for which a peak expo-sure estimate is likely to be valid should be deter-
mined, as the time dependency of exposure to car-cinogens can be important (and not limited to latencyissues). Third, potential for individual workers’ peakexposures to differ from group values should be ad-dressed, as it is conceivable that environments likelyto confer peak exposures are not uniformly likelyto do so across work locations, times, and workers.Fourth, and especially where estimates are less dataderived and more assumption laden, sensitivity test-ing should be used to determine the influence of vari-ous components to the peak metric. In circumstancesof limited data or poor understanding of the rele-vant underlying carcinogenic mechanisms, present-ing more detailed documentation and explanation ofrationale are helpful, and will clarify situations thatare hypothesis driven from those that are more ex-ploratory.
Choice of peak metric likely matters. There aremultiple variations of time-varying exposure patternsconsistent with peak exposures. One potentially use-ful way to consider peak is under two exposure cir-cumstances: (1) peaks are commonly encountered(extreme values are frequent) or, conversely, (2) fewpeaks are encountered such that extreme values arerare.
Given that occupational exposure concentra-tion(s) often vary considerably over time, a basicprobabilistic approach to defining peak exposurein these terms might serve as a reasonable start-ing point. The simplest definition of peak exposure,therefore, might be an exposure event in which themeasured or estimated concentration exceeds a spec-ified exposure threshold value. This alone, however,does not address the duration of time over which thisconcentration level is exceeded, the magnitude of thepeak, or the number of exceedances. Alternatively,as the pharmacokinetics of a potential carcinogen arebetter understood, an effective dose can be estimatedand from this—and somewhat ideally—an equiva-lent peak exposure derived. The pharmacokineticallyderived peak conceptually is not limited to the sin-gle dimension of exposure concentration (expressedabove as exceedance of an arbitrary percentile) butcan combine concentration and time to define a tar-get dose. By extension, an effective dose rate mightalso be derived.
A thorough analysis of potential etiologic rela-tions would involve estimation of risks associatedwith: peaks based on various thresholds; magni-tude of peaks above thresholds; number of peaks;and timing of peaks in relation to disease onset ordiagnosis. These approaches can take the form of
Peak Exposures in Epidemiologic Studies and Cancer Risks 1461
sensitivity analyses comparing findings based onvarious peak exposure metrics. Assessing the sensi-tivity analysis findings in relation to chemical-specificcarcinogenesis models derived from toxicological re-search can enhance data interpretation.
Assuming uniform approaches for developingstandard peak exposure metrics are successful, thedose–response assessment still requires careful con-sideration. In particular, when peak exposure metrics(and not cumulative exposure) drive epidemiologiccancer hazard determination, the dose–responseassessment should not be based on cumulativeexposures. IURs inherently characterize cancer risksas linear dose–response relations over the entireexposure spectrum, typically reflecting cumulativeexposure. New risk metrics are needed to describerisks that are not monotonic, including risks thatincrease in response to peak exposures. Regardlessof how peaks are defined, they inherently argueagainst the “additivity to background” or “everymolecule counts” perspective that is synonymouswith risk as a direct function of cumulative exposure.This acknowledges that risk will be more accuratelyquantified when the most biologically relevantcharacterization of exposure is captured.
Contrasts of associations observed for peaksversus those observed for cumulative exposureand exposure duration are highly recommended,especially as there are clear implications for riskassessment. Although peak exposure correlates withother exposure metrics, selection of peak exposureas the primary exposure metric of interest shouldbe accompanied by judicial consideration of othermetrics in attempting to isolate and characterizecausal exposure characteristics. It should be appre-ciated that, by definition, peak exposure contributesto cumulative exposure. As such, a correlationbetween dose–response models derived from peakand cumulative exposures should be anticipated inmany instances, and the biological rationale—as wellas results of sensitivity analyses—should guide theselection and interpretation of the best exposuremetric and subsequently improve the derivationof risk numbers including the IUR and others notassumed to be linear. Future epidemiologic research,including studies of new cohorts and reanalyses ofdata from existing cohorts, ideally will address andcontrast dose–response associations for cumulativeand peak exposures, where data permit suitableexposure assessment. However, investigators shouldbear in mind that exploring associations using mul-tiple exposure metrics will increase the number of
chance (i.e., false-positive) results, and unless thereis an underlying biological basis for interpretingthem as causal, they should be viewed with caution.
ACKNOWLEDGMENTS
This research was sponsored in part by the Foun-dation for Chemistry Research and Initiatives, atax-exempt public foundation described in Section501(c)(3) of the Internal Revenue Code, and theElectric Power Research Institute (EPRI). The spon-sors had no role in the design, conduct, analysis, orreporting of results.
REFERENCES
Baan, R., Grosse, Y., Straif, K., Secretan, B., El Ghissassi, F., Bou-vard, V., . . . Cogliano, V.; WHO International Agency for Re-search on Cancer Monograph Working Group. (2009). A re-view of human carcinogens—Part F: Chemical agents and re-lated occupations. Lancet Oncology, 10, 1143–1144.
Beane Freeman, L. E., Blair, A., Lubin, J. H., Stewart, P. A.,Hayes, R. B., Hoover, R. N., & Hauptmann, M. (2009). Mor-tality from lymphohematopoietic malignancies among workersin formaldehyde industries: The National Cancer Institute Co-hort. Journal of the National Cancer Institute, 101(10), 751–761.
Beane Freeman, L. E., Blair, A., Lubin, J. H., Stewart, P. A.,Hayes, R. B., Hoover, R. N., & Hauptmann, M. (2013). Mor-tality from solid tumors among workers in formaldehyde indus-tries: An update of the National Cancer Institute (NCI) cohort.American Journal of Industrial Medicine, 56(9), 1015–1026.
Bertazzi, P. A., Bernucci, I., Brambilla, G., Consonni, D., &Pesatori, A. C. (1998). The Seveso studies on early and long-term effects of dioxin exposure: A review. EnvironmentalHealth Perspective, 106(Suppl 2), 625–633.
Blair, A., Hartge, P., Stewart, P. A., McAdams, M., & Lubin, J.(1998). Mortality and cancer incidence of aircraft maintenanceworkers exposed to trichloroethylene and other organic sol-vents and chemicals: Extended follow up. Occupational & En-vironmental Medicine, 55, 161–171.
Blair, A., & Stewart, P. A. (1990). Correlation between differentmeasures of occupational exposure to formaldehyde. AmericanJournal of Epidemiology, 131(3), 510–516.
Blair, A., Stewart, P. A., Zaebst, D. D., Pottern, L. M., Zey, J. N.,Bloom, T. F., . . . Lubin, J. (1998). Mortality of industrial work-ers exposed to acrylonitrile. Scandinavian Journal of Work, En-vironment & Health, 24(Suppl 2), 25–41.
Charbotel, B., Fevotte, J., Hours, M., Martin, J. L., & Bergeret,A. (2006). Case-control study on renal cell cancer and occupa-tional exposure to trichloroethylene. Part II: Epidemiologicalaspects. Annals of Occupational Hygiene, 50(8), 777–787.
Checkoway, H., Dell, L. D., Boffetta, P., Gallagher, A. E., Craw-ford, L., Lees, P. S., & Mundt, K. A. (2015). Formaldehyde ex-posure and mortality risks from acute myeloid leukemia andother lymphohematopoietic malignancies in the US NationalCancer Institute Cohort Study of workers in formaldehyde in-dustries. Journal of Occupational & Environmental Medicine,57(7), 785–794.
Cheng, H., Sathiakumar, N., Graff, J., Matthews, R., & Delzell,E. (2007). 1,3-Butadiene and leukemia among synthetic rubberindustry workers: Exposure-response relationships. Chemico-Biological Interactions, 166(1-3), 15–24.
Christensen, M. S., Vestergaard, J. M., d’Amore, F., Gorlov,J. S., Toft, G., Ramlau-Hansen, C. H., . . . Kolstad, H. A.
1462 Checkoway et al.
(2018). Styrene exposure and risk of lymphohematopoieticmalignancies in 73,036 reinforced plastics workers. Epidemiol-ogy, 29(3), 342–351.
Coggon, D., Harris, E. C., Poole, J., & Palmer, K. T. (2004). Mor-tality of workers exposed to ethylene oxide: Extended follow upof a British cohort. Occupational & Environmental Medicine,61(4), 358–362.
Coggon, D., Ntani, G., Harris, E. C., & Palmer, K. T. (2014).Upper airway cancer, myeloid leukemia, and other cancers ina cohort of British chemical workers exposed to formalde-hyde. American Journal of Epidemiology, 179(11), 1301–1311.
Collins, J. J., Bodner, K. M., & Bus, J. S. (2013). Cancer mortalityof workers exposed to styrene in the U.S. reinforced plasticsand composite industry. Epidemiology, 24(2), 195–203.
Collins, J. J., Ireland, B., Buckley, C. F., & Shepperly, D. (2003).Lymphohaematopoeitic cancer mortality among workers withbenzene exposure. Occupational & Environmental Medicine,60(9), 676–679.
Darby, S. C., Nakashima, E., & Kato, H. (1985). A parallel analysisof cancer mortality among atomic bomb survivors and patientswith ankylosing spondylitis given X-ray therapy. Journal of theNational Cancer Institute, 75(1), 1–21.
Delzell, E., Macaluso, M., Sathiakumar, N., & Matthews, R.(2001). Leukemia and exposure to 1,3-butadiene, styrene anddimethyldithiocarbamate among workers in the synthetic rub-ber industry. Chemico-Biological Interactions, 135–136, 515–534.
Delzell, E., Sathiakumar, N., Hovinga, M., Macaluso, M., Julian,J., Larson, R., . . . Muir, D. C. (1996). A follow-up study of syn-thetic rubber workers. Toxicology, 113(1-3), 182–189.
Deschler, B., & Lubbert, M. (2006). Acute myeloid leukemia: Epi-demiology and etiology. Cancer, 107(9), 2099–2107.
Dosemeci, M., Wacholder, S., & Lubin, J. H. (1990). Does non-differential misclassification of exposure always bias a true ef-fect toward the null value? American Journal of Epidemiology,132(4), 746–748.
Glass, D. C., Armstrong, T. W., Pearlman, E. D., Verma, D. K.,Schnatter, A. R., & Rushton, L. (2010). Ensuring comparabilityof benzene exposure estimates across three nested case-controlstudies in the petroleum industry in support of a pooled epi-demiological analysis. Chemico-Biological Interactions, 184(1-2), 100–111.
Glass, D. C., Schnatter, A. R., Tang, G., Irons, R. D., & Rush-ton, L. (2014). Risk of myeloproliferative disease and chronicmyeloid leukaemia following exposure to low-level benzene ina nested case-control study of petroleum workers. Occupational& Environmental Medicine, 71(4), 266–274.
Graff, J. J., Sathiakumar, N., Macaluso, M., Maldonado, G.,Matthews, R., & Delzell, E. (2005). Chemical exposures inthe synthetic rubber industry and lymphohematopoietic cancermortality. Journal of Occupational & Environmental Medicine,47(9), 916–932.
Greenberg, H. L., Ott, M. G., & Shore, R. E. (1990). Men assignedto ethylene oxide production or other ethylene oxide relatedchemical manufacturing: A mortality study. British Journal ofIndustrial Medicine, 47(4), 221–230.
Greife, A. L., Hornung, R. W., Stayner, L. G., & Steenland, K.N. (1988). Development of a model for use in estimating ex-posure to ethylene oxide in a retrospective cohort mortalitystudy. Scandinavian Journal of Work, Environment & Health,14(Suppl 1), 29–30.
Hauptmann, M., Lubin, J. H., Stewart, P. A., Hayes, R. B., & Blair,A. (2003). Mortality from lymphohematopoietic malignanciesamong workers in formaldehyde industries. Journal of the Na-tional Cancer Institute, 95(21), 1615–1623.
Hauptmann, M., Stewart, P. A., Lubin, J. H., Beane Freeman, L.E., Hornung, R. W., Herrick, R. F., . . . Hayes, R. B. (2009).Mortality from lymphohematopoietic malignancies and brain
cancer among embalmers exposed to formaldehyde. Journal ofthe National Cancer Institute, 101(24), 1696–1708.
Hayes, R. B., Blair, A., Stewart, P. A., Herrick, R. F., & Ma-har, H. (1990). Mortality of U.S. embalmers and funeral di-rectors. American Journal of Industrial Medicine, 18(6), 641–652.
Hearne, F. T., & Pifer, J. W. (1999). Mortality study oftwo overlapping cohorts of photographic film base manu-facturing employees exposed to methylene chloride. Journalof Occupational & Environmental Medicine, 41(12), 1154–1169.
Hornung, R. W., Herrick, R. F., Stewart, P. A., Utterback, D. F.,Feigley, C. E., Wall, D. K., . . . Hayes, R. B. (1996). An exper-imental design approach to retrospective exposure assessment.American Industrial Hygiene Association Journal, 57(3), 251–256.
International Agency for Research on Cancer (IARC). (1987).Overall evaluations of carcinogenicity: An updating of IARCmonographs (Vol. 1–42, 7th ed.). Lyon, France: Author.
International Agency for Research on Cancer (IARC). (1999). Re-evaluation of some organic chemicals, hydrazine and hydrogenperoxide (Vol. 71). IARC Monographs on the Evaluation ofCarcinogenic Risk to Humans. Lyon, France: World Health Or-ganization.
International Agency for Research on Cancer (IARC). (2006).Formaldehyde, 2-butoxyethanol and 1-tert-butoxypropan-2-ol(Vol. 88). IARC Monographs on the Evaluation of Carcino-genic Risk to Humans. Lyon, France: World Health Organiza-tion.
International Agency for Research on Cancer (IARC). (2008).1,3-Butadiene, ethylene oxide and vinyl halides (vinyl fluoride,vinyl chloride and vinyl bromide) (Vol. 97). IARC Monographson the Evaluation of Carcinogenic Risk to Humans. Lyon,France: World Health Organization.
International Agency for Research on Cancer (IARC). (2012a).A review of human carcinogens Part A: Pharmaceuticals. A re-view of human cracinogens (Vol. 100). IARC Monographs onthe Evaluation of Carcinogenic Risk to Humans. Lyon, France:World Health Organization.
International Agency for Research on Cancer (IARC). (2012b). Areview of human carcinogens Part F: chemical agents and relatedoccupations (Vol. 100). IARC Monographs on the Evaluationof Carcinogenic Risk to Humans. Lyon, France: World HealthOrganization.
International Agency for Research on Cancer (IARC). (2014).Trichloroethylene, tetrachloroethylene, and some other chlori-nated agents (Vol. 106). IARC Monographs on the Evaluationof Carcinogenic Risk to Humans. Lyon, France: World HealthOrganization.
International Agency for Research on Cancer (IARC). (2017).Some chemicals used as solvents and in polymer manufacture(Vol. 110). IARC Monographs on the Evaluation of Carcino-genic Risk to Humans. Lyon, France: World Health Organiza-tion.
Jurek, A. M., Greenland, S., Maldonado, G., & Church, T. R.(2005). Proper interpretation of non-differential misclassifica-tion effects: Expectations vs observations. International Journalof Epidemiology, 34(3), 680–687.
Kogevinas, M., Gwinn, W. M., Kriebel, D., Phillips, D. H., Sim, M.,Bertke, S. J., . . . Straif, K.; IARC Monographs Vol 121 Work-ing Group. (2018). Carcinogenicity of quinoline, styrene, andstyrene-7,8-oxide. Lancet Oncology, 19(6), 728–729.
Kolstad, H. A., Sonderskov, J., & Burstyn, I. (2005). Company-level, semi-quantitative assessment of occupational styrene ex-posure when individual data are not available. Annals of Occu-pational Hygiene, 49(2), 155–165.
Kriebel, D., Checkoway, H., & Pearce, N. (2007). Exposure anddose modelling in occupational epidemiology. Occupationaland Environmental Medicine, 64(7), 492–498.
Peak Exposures in Epidemiologic Studies and Cancer Risks 1463
Lash, T. L., Fox, M. P., MacLehose, R. F., Maldonado, G.,McCandless, L. C., & Greenland, S. (2014). Good practices forquantitative bias analysis. International Journal of Epidemiol-ogy, 43(6), 1969–1985.
Macaluso, M., Larson, R., Delzell, E., Sathiakumar, N., Hovinga,M., Julian, J., . . . Cole, P. (1996). Leukemia and cumulativeexposure to butadiene, styrene and benzene among workersin the synthetic rubber industry. Toxicology, 113(1-3), 190–202.
Macaluso, M., Larson, R., Lynch, J., Lipton, S., & Delzell, E.(2004). Historical estimation of exposure to 1,3-butadiene,styrene, and dimethyldithiocarbamate among synthetic rubberworkers. Journal of Occupational & Environmental Hygiene,1(6), 371–390.
Meyers, A. R., Pinkerton, L. E., & Hein, M. J. (2013). Cohort mor-tality study of garment industry workers exposed to formalde-hyde: Update and internal comparisons. American Journal ofIndustrial Medicine, 59(9), 1027–1039.
Moloney, W. C., & Lange, R. D. (1954). Leukemia in atomic bombsurvivors. II. Observations on early phases of leukemia. Blood,9(7), 663–685.
Morgan, R. W., Kelsh, M. A., Zhao, K., & Heringer, S. (1998).Mortality of aerospace workers exposed to trichloroethylene.Epidemiology, 9(4), 424–431.
National Research Council (NRC). (2011). Review of the Environ-mental Protection Agency’s draft IRIS assessment of formalde-hyde. Washington, DC: National Academies Press.
Nissen, M. S., Stokholm, Z. A., Christensen, M. S., Schlunssen,V., Vestergaard, J. M., Iversen, I. B., & Kolstad, H. A. (2018).Sinonasal adenocarcinoma following styrene exposure in thereinforced plastics industry. Occupational & EnvironmentalMedicine, 75(6), 412–414.
Raaschou-Nielsen, O., Hansen, J., McLaughlin, J. K., Kolstad, H.,Christensen, J. M., Tarone, R. E., & Olsen, J. H. (2003). Cancerrisk among workers at Danish companies using trichloroethy-lene: A cohort study. American Journal of Epidemiology,158(12), 1182–1192.
Radican, L., Blair, A., Stewart, P., & Wartenberg, D. (2008). Mor-tality of aircraft maintenance workers exposed to trichloroethy-lene and other hydrocarbons and chemicals: Extended follow-up. Journal of Occupational & Environmental Medicine, 50(11),1306–1319.
Rappaport, S. M. (1991). Selection of the measures of exposure forepidemiology studies. Applied Occupational & EnvironmentalHygiene, 6(6), 448–457.
Rinsky, R. A., Smith, A. B., Hornung, R., Filloon, T. G., Young,R. J., Okun, A. H., & Landrigan, P. J. (1987). Benzene andleukemia. An epidemiologic risk assessment. New EnglandJournal of Medicine, 316(17), 1044–1050.
Rinsky, R. A., Young, R. J., & Smith, A. B. (1981). Leukemiain benzene workers. American Journal of Industrial Medicine,2(3), 217–245.
Rushton, L., Schnatter, A. R., Tang, G., & Glass, D. C. (2014).Acute myeloid and chronic lymphoid leukaemias and exposureto low-level benzene among petroleum workers. British Journalof Cancer, 110(3), 783–787.
Sathiakumar, N., Brill, I., Leader, M., & Delzell, E. (2015).1,3-Butadiene, styrene and lymphohematopoietic canceramong male synthetic rubber industry workers—Preliminaryexposure-response analyses. Chemico-Biological Interactions,241, 40–49.
Schnatter, A. R., Glass, D. C., Tang, G., Irons, R. D., & Rush-ton, L. (2012). Myelodysplastic syndrome and benzene expo-sure among petroleum workers: An international pooled anal-ysis. Journal of the National Cancer Institute, 104(22), 1724–1737.
Stayner, L., Steenland, K., Greife, A., Hornung, R., Hayes, R. B.,Nowlin, S., . . . Halperin, W. (1993). Exposure-response analy-
sis of cancer mortality in a cohort of workers exposed to ethy-lene oxide. American Journal of Epidemiology, 138(10), 787–798.
Steenland, K., Stayner, L., & Deddens, J. (2004). Mortality analy-ses in a cohort of 18, 235 ethylene oxide exposed workers: Fol-low up extended from 1987 to 1998. Occupational & Environ-mental Medicine, 61(1), 2–7.
Steenland, K., Whelan, E., Deddens, J., Stayner, L., & Ward, E.(2003). Ethylene oxide and breast cancer incidence in a cohortstudy of 7576 women (United States). Cancer Causes & Con-trol, 14(6), 531–539.
Stenehjem, J. S., Kjaerheim, K., Bratveit, M., Samuelsen, S. O.,Barone-Adesi, F., Rothman, N., . . . Grimsrud, T. K. (2015).Benzene exposure and risk of lymphohaematopoietic cancersin 25,000 offshore oil industry workers. British Journal of Can-cer, 112(9), 1603–1612.
Stewart, P. A., Blair, A., Cubit, D., Bales, R., Kaplan, S. A., Ward,J., . . . Walrath, J. (1986). Estimating historical exposures toformaldehyde in a retrospective mortality study. Applied Indus-trial Hygiene, 1(1), 34–41.
Stewart, P. A., Herrick, R. F., Feigley, C. E., Utterback, D.,Homung, R., Mahar, H., & Blair, A. (1992). Study design forassessing exposures of embalmers for a case-control study. PartI. Monitoring results. Applied Occupational & EnvironmentalHygiene, 7(8), 532–540.
Stewart, P. A., Zaebst, D. D., Zey, J. N., Herrick, R., Dosemeci,M., Hornung, R., . . . Blair, A. (1998). Exposure assessmentfor a study of workers exposed to acrylonitrile. ScandinavianJournal for Work, Environment & Health, 24(Suppl 2), 42–53.
Swaen, G. M., Bloemen, L. J., Twisk, J., Scheffers, T., Slangen,J. J., Collins, J. J., & Berge, W. F. (2004). Mortality updateof workers exposed to acrylonitrile in the Netherlands. Jour-nal of Occupational & Environmental Medicine, 46(7), 691–698.
Swaen, G. M., Bloemen, L. J., Twisk, J., Scheffers, T., Slangen,J. J., & Sturmans, F. (1992). Mortality of workers exposedto acrylonitrile. Journal of Occupational Medicine, 34(8), 801–809.
U.S. Environmental Protection Agency (US EPA). (1998). Car-cinogenic effects of benzene: An update. Prepared by the Na-tional Center for Environmental Health, Office of Researchand Development. EPA/600/P-97/001F. Washington, DC:Author.
U.S. Environmental Protection Agency (US EPA). (2010). IRIStoxicological review of formaldehyde (inhalation) (externalreview draft 2010). EPA/635/R-10/002A. Washington, DC:Author.
U.S. Environmental Protection Agency (US EPA). (2011a). Toxi-cological review of dichloromethane (CAS No. 75-09-2) in sup-port of summary information on the Integrated Risk Informa-tion System (IRIS). Integrated Risk Information System (IRIS),National Center for Environmental Assessment, EPA/635/R-10/003F. Washington, DC: Author.
U.S. Environmental Protection Agency (US EPA). (2011b). Toxi-cological review of trichloroethylene (CASRN 79-01-6) in sup-port of summary information on the Integrated Risk Infor-mation System (IRIS). EPA/635/R-09/011F. Washington, DC:Author.
U.S. Environmental Protection Agency (US EPA). (2011c). Tox-icological review of acrylonitrile (CAS No. 107-13-1) in supportof summary information on the Integrated Risk Information Sys-tem (IRIS) (external review draft). EPA/635/R-08/013A. Wash-ington, DC: Author.
U.S. Environmental Protection Agency (US EPA). (2012).Lymphohematopoietic cancers induced by chemicals and
1464 Checkoway et al.
other agents: Overview and implications for risk assessment.EPA/600/R-10/095F. Washington, DC: Author.
U.S. Environmental Protection Agency (US EPA). (2016). Eval-uation of the inhalation carcinogenicity of ethylene oxide(CASRN 75-21-8) in support of summary information on the In-tegrated Risk Information System (IRIS). EPA/635/R-16/350Fa.Washington, DC: Author.
Walrath, J., & Fraumeni, J. F., Jr. (1983). Mortality patternsamong embalmers (case reports). International Journal of Can-cer, 31(4), 407–411.
Walrath, J., & Fraumeni, J. F. (1984). Cancer and other causes ofdeath among embalmers. Cancer Research, 44(10), 4638–4641.
Weinberg, C. R., Umbach, D. M., & Greenland, S. (1994). Whenwill nondifferential misclassification of an exposure preservethe direction of a trend? American Journal of Epidemiology,140(6), 565–571.
SUPPORTING INFORMATION
Additional supporting information may be found on-line in the Supporting Information section at the endof the article.
Table SI. Uses, Exposure Limits, Cancer Char-acterization, Inhalation Unit Risk, and Epidemio-logic Studies with Quantitative Exposure Assess-ment Metric by SubstanceTable SII. Rationale for Using Peak Exposure orReason for Considering Peak Exposure as Reportedby Study Investigators