The chemical mixture of disinfection by-products
(DBPs) has not been fully characterized but is
known to contain trihalomethanes (THMs), haloacetic
acids (HAAs), haloacetonitriles, and other classes
of chemicals, some of which are mutagenic or
carcinogenic in laboratory animals (Nieuwenhuijsen
et al. 2000). Total THMs (TTHMs) are the sum
of the concentrations of the THM species chloroform,
bromodichloromethane (BDCM), dibromochloromethane
(DBCM), and bromoform. The five regulated HAAs
(HAA5) include monochloroacetic acid (MCAA),
dichloroacetic acid (DCAA), trichloroacetic acid
(TCAA), monobromoacetic acid (MBAA), and dibromoacetic
acid (DBAA). Concerns have been raised regarding
the potential effects of by-products on reproductive
outcomes, supported in part by the findings that
some by-products cause reproductive and developmental
toxicity in laboratory animals, albeit at doses
much higher than those encountered by humans.
In addition, exposure to DBPs has been associated
with an increased risk of impaired fetal growth
in several epidemiologic studies (Bove et al.
1995; Dodds et al. 1999; Gallagher et al. 1998;
Kramer et al. 1992; Savitz et al. 1995; Wright
et al. 2003).
The third trimester of pregnancy is considered
the period of human fetal development during
which fetal growth and birth weight are maximally
sensitive to environmental influences (Kline
et al. 1989). The third trimester lasts from
approximately the 26th week of gestation to parturition,
with the actual length of time dependent on the
individual pregnancy. However, only a few prior
investigations of DBPs have evaluated exposure
during this period (Dodds et al. 1999; Gallagher
et al. 1998; Savitz et al. 1995; Wright et al.
2003). Risks for adverse birth outcomes depend
on the magnitude of exposure over critical time
windows. Therefore, analyses over exposure windows
that are too wide may bias risk estimates. Because
the critical time period for the potential effects
of DBP exposure on fetal growth is uncertain,
the use of multiple, shorter exposure windows
may provide less biased risk estimates (Hertz-Picciotto
et al. 1996). The purpose of this study was to
examine the effects of exposure to THMs and HAAs
during the third trimester and during individual
weeks and months of late gestation on the risks
for term low birth weight, intrauterine growth
retardation, and very preterm and preterm births.
We conducted a retrospective cohort study in
a large Arizona community served by three water
treatment facilities. This community of more
than half a million residents living in 24 ZIP
codes is located adjacent to a major metropolitan
area. Most water used by this community originates
from surface water sources by means of the Salt
River and Central Arizona projects.
Table
1
|
The community was selected from the U.S. Environmental
Protection Agency (EPA) Information Collection
Rule database (U.S. EPA 1999) because the distribution
systems displayed large temporal fluctuations
(range, 7-81 µg/L) and low spatial
variability in TTHM levels that permitted a natural
experiment through intracommunity comparisons
of exposures and outcomes. We determined spatial
variability using the methods described by Hinckley
et al. (2005). Briefly, we classified a facility
as having low spatial variability if TTHM values
measured at four points in the facility’s
distribution system consistently fell, each season,
within established boundaries for low, medium,
and high exposure as based on concentration cut-points
for TTHMs derived from prior epidemiologic studies
of birth outcomes. The study population included
all live births and fetal deaths for women whose
residence was provided water by one of three
facilities serving the community from January
1998 through December 2002. Table 1 summarizes
the annual frequency of births in this population,
as well as the distribution of TTHM and HAA5
concentrations by year over the period of the
study. This study was approved by the human subjects
institutional review boards of Colorado State
University and the Arizona Department of Health
Services.
Subjects were identified from Arizona birth
records (n = 48,119) and were matched
to a facility service area by residential ZIP
code. In cases where two facilities shared the
same distribution system, treatment facility
employees identified service boundaries. Subjects
who lived in ZIP codes that received water from
more than one facility were excluded from the
analysis. Maternal residence at birth was assumed
to be the same as residence during the third
trimester.
We estimated exposure from data obtained from
each facility (facilities A-C) for the
years 1998-2002. Total and individual THMs
were measured quarterly during each of the 5
years, for each of the facilities. Facility A
provided quarterly THM and HAA data for the entire
study period, with monthly data available for
2001 and 2002. At facilities B and C, HAA5 data
were available only for 2000 and 2002. Supplemental
monthly and biweekly TTHM and HAA5 data were
provided by facility B for 2000 and 2002, respectively.
DBP concentrations were monitored at two to four
locations within the distribution system of each
facility. The quarterly and monthly data indicated
the presence of very low levels of bromoform,
MBAA, and MCAA; therefore, these chemicals were
not included in the analyses. To estimate DBP
values for specific study periods corresponding
to months when no data were available, we performed
a spline regression (Greenland 1998) for each
water facility to impute the missing values using
procedures similar to those used by investigators
in Nova Scotia (Dodds et al. 1999; Dodds and
King 2001; King et al. 2000). This nonlinear
smoothing technique was applied to impute missing
exposure data from existing data by generating
a joined series of parabolic curve segments.
HAA exposure data were not estimated before the
year 2000 for facilities B and C.
Infant outcomes were identified through vital
records. The date of last menstrual period was
used to define the duration of gestation. We
identified infants born at ≥ 37
completed weeks of gestation and weighing < 2,500
g as being term low birth weight. We evaluated
this outcome only among term births to separate
children with true growth retardation from babies
that are small because of birth at a young gestational
age. We identified case infants with intrauterine
growth retardation as term or preterm babies
that fell below the published value for the lowest
10th percentile of birth weights by race, ethnicity,
and gestation age (Alexander et al. 1999). In
this investigation, term low birth weight and
intrauterine growth retardation were not mutually
exclusive, and cases born at term may have been
included in both outcome groups. Because published
values for the lowest 10th percentile of birth
weights were not available for extreme gestational
ages, births before 23 weeks’ gestation
were excluded from intrauterine growth retardation
analyses for Caucasians, African Americans, and
Hispanics, and births before 29 weeks’ gestation
were excluded for Native Americans (Alexander
et al. 1999). For intrauterine growth retardation
and term low birth weight, estimated monthly
DBP exposures were averaged over the third trimester.
Additionally, for DBPs associated with intrauterine
growth retardation or term low birth weight,
we averaged and evaluated exposure during specific
time windows, corresponding to gestation weeks
25-28, 29-32, 33-36, 37-40,
and 41-44, using monthly DBP concentrations.
Preterm births were defined as infants born
at < 37 completed weeks of gestation. Very
preterm births were defined as a birth occurring
before 32 completed weeks of gestation (Martin
et al. 2002). Because preterm birth outcomes
are defined by time length of gestation, it was
inappropriate to evaluate exposure averaged over
the third trimester. Preterm births have shorter
gestation lengths than the typical comparison
group (term births), increasing the potential
for bias (Hertz-Picciotto et al. 1996; Hinckley
2003; Hinckley et al. 2002). Therefore, for preterm
and very preterm births, we evaluated exposure
to DBPs only for the specific gestation week
intervals mentioned above.
We abstracted information on potential confounders
from birth records. These variables included
maternal age, race, ethnicity, education, parity,
smoking, and the Kessner index (a measure of
prenatal care adequacy) (Kotelchuck 1994). By
comparing the outcomes over different exposure
time windows within a single community, we attempted
to control for potential residual confounders
that could not be evaluated individually.
We calculated tertiles of DBP concentrations
using the species-specific data from all three
water treatment regions in analyses for potential
associations with each birth outcome. We used
stratified chi-square and logistic regression
analyses to evaluate the associations among demographic
variables, exposure variables, and adverse birth
outcomes. In addition, all covariates significantly
associated with growth outcomes at the < 0.20
level in univariate analyses were retained for
inclusion in multivariable analyses. After adjustment
for potential confounders, we calculated odds
ratios (ORs) and 95% confidence intervals (CIs)
for the relationships between all individual
THM species and growth and preterm birth outcomes.
For gestation week and third-trimester analyses,
a multivariate logistic regression model containing
all individual HAAs as continuous variables was
used to evaluate the possible relationship between
individual HAAs in increasing the risk of growth-related
outcomes. A similar model was not created for
individual THMs because there was no evidence
of any associations with growth-related outcomes.
Table 2 summarizes characteristics of subjects
and frequency of intrauterine growth retardation,
term low birth weight, and preterm and very preterm
births. Most mothers were white, non-Hispanic,
nulliparous women with some college education.
Most mothers received adequate prenatal care,
and < 10% smoked during pregnancy. Subjects
were excluded if there was no date for last menstrual
period or no estimated date of conception. The
estimated date of conception was used to estimate
the last menstrual period when data on last menstrual
period were missing or considered extreme (> 44
weeks before the birth date). The results were
not different when using last menstrual period,
estimated date of conception, or a combination
of both methods; therefore, we used the combined
method to minimize the number of subjects lost
for this reason (
n = 42).
The ORs and 95% CIs for intrauterine growth
retardation and term low birth weight and exposure
to DBPs during the third trimester are shown
in Table 3. We found no evidence of an association
with either outcome for exposure to TTHMs or
specific brominated and chlorinated THMs. We
also found no association between exposure to
HAA5 and intrauterine growth retardation. The
second and third tertiles of exposure to HAA5
showed evidence of a weak association with term
low birth weight [OR = 1.26 (95% CI, 0.96-1.65),
and OR = 1.25 (95% CI, 0.96-1.64), respectively]
compared with referent exposure levels.
Exposures to the highest tertiles of DCAA and
TCAA were associated with an increased risk of
intrauterine growth retardation [OR = 1.28 (95%
CI, 1.08-1.51), and OR = 1.19 (95% CI,
1.01-1.41), respectively]. DCAA and TCAA
were also associated with intrauterine growth
retardation when analyzed as continuous variables.
Weak associations were found for exposure to
the highest tertile of DBAA and DBAA analyzed
as a continuous variable, although the 95% CIs
for those results all included 1.0. Analyses
of intrauterine growth retardation were adjusted
for parity, smoking, maternal education, and
Kessner index.
The risk of term low birth weight was increased
(OR = 1.49; 95% CI, 1.09-2.04) among women
exposed to average DBAA concentrations of ≥ 5 µg/L
during the third trimester compared with those
who were exposed to the referent category of < 4 µg/L.
Continuous (unit) increases in average exposure
to DBAA also indicated a weak association with
term low birth weight (OR = 1.17; 95% CI 1.03-1.32).
Analyses of term low birth weight were adjusted
for maternal age, parity, education, race, ethnicity,
smoking, and Kessner index.
Table 4 presents ORs and 95% CIs for exposure
to HAA5 and individual HAAs over specific gestation
time windows for intrauterine growth retardation
and term low birth weight. Because the potential
for bias due to averaging was reduced when examining
shorter time intervals, exposure values were
generally slightly higher or slightly lower over
the specific gestation week intervals than over
the third trimester. In analyses for intrauterine
growth retardation, small increases in risk were
observed for DBAA concentrations ≥ 5 µg/L
(OR = 1.15; 95% CI, 0.98-1.35) and for
DBAA analyzed as a continuous variable (OR =
1.06; 95% CI, 1.01-1.12) over gestation
weeks 25-28. The largest risk was observed
with exposure to DCAA ≥ 8 µg/L
(OR = 1.27; 95% CI, 1.02-1.59) during gestation
weeks 37-40. In addition, an increased
risk was observed for exposure to moderate concentrations
of TCAA (OR = 1.58; 95% CI, 1.02-2.46)
and DCAA (OR = 1.51; 95% CI, 0.98-2.32)
during gestation weeks 41-44, but the risk
estimates were lower at higher levels of estimated
exposure.
Exposure to DBAA was associated with an increase
in risk for term low birth weight in analyses
by gestation week. Between gestation weeks 33
and 36, the second and third tertiles of exposure
to DBAA showed evidence of a dose-dependent trend
[OR = 1.29 (95% CI, 0.94-1.79), and OR
= 1.49 (95% CI, 1.10-2.02), respectively]
compared with referent exposure levels. Similarly,
moderate exposure to DBAA between gestation weeks
37 and 40 was associated with an increased risk
for term low birth weight (OR = 1.38; 95% CI,
1.02-1.86).
No associations were observed between preterm
or very preterm birth and exposure to TTHMs,
HAA5, or specific DBPs during any gestation week
interval. ORs for the associations between individual
HAAs and term low birth weight and intrauterine
growth retardation were not affected by inclusion
of other HAAs in the logistic regression model.
Reduced fetal weight is one of the most consistent
developmental effects observed with exposure
to high concentrations of DBPs in laboratory
animals (Nieuwenhuijsen et al. 2000). The biologic
mechanisms for DBP-induced growth retardation
are not well understood. In animal studies, reductions
in birth weight have been commonly described
after exposure to THMs, especially chloroform
(Murray et al. 1979; Ruddick et al. 1983; Schwetz
et al. 1974; Thompson et al. 1974). Two studies
by Smith et al. (1989, 1992) found reductions
in rat pup body weight after exposure to DCAA
and TCAA. Recently, Christian et al. (2001) found
that DBAA administration (of 250, 500, and 1,000
mg/L) was associated with exposure-related decreases
in rat pup body weight. This effect, however,
was thought to be due to reduced parental water
consumption.
The epidemiologic evidence for an association
between exposure to THMs and indicators of fetal
growth is relatively sparse and inconsistent,
and few studies have investigated this relationship
with respect to HAAs. Four prior epidemiologic
studies have evaluated exposure to total and
individual THMs in relation to intrauterine growth
retardation. In a study by Kramer et al. (1992),
a dose-related trend was observed for intrauterine
growth retardation at the 5th percentile for
exposure to chloroform ≥ 10 µg/L
and BDCM ≥ 4 µg/L,
with ORs of 1.8 (95% CI, 1.1-2.9) and 1.7
(95% CI, 0.9-2.9), respectively. Bove et
al. (1995) also found an increased risk of intrauterine
growth retardation (adjusted OR = 1.50; 90% CI,
1.19-1.86) with exposure to TTHMs > 100 µg/L
during pregnancy. In a Massachusetts cohort,
Wright et al. (2003) found increased risk of
intrauterine growth retardation (10th percentile)
for mean exposures to TTHMs > 80 µg/L
throughout pregnancy (adjusted OR = 1.14; 95%
CI, 1.02-1.26) and during the second trimester
(adjusted OR = 1.13; 95% CI, 1.03-1.24).
However, Dodds et al. (1999) found no association
between intrauterine growth retardation (10th
percentile) and TTHM exposure ≥ 100 µg/L
in a large cohort of Nova Scotia women.
Three studies have evaluated term low birth
weight and exposure to TTHMs. Gallagher et al.
(1998) found an adjusted OR of 5.9 (95% CI, 2.0-17.0)
for term births, although only six cases were
analyzed. Bove et al. (1995) also observed a
positive, but smaller, association between TTHM
exposures averaged over the entire pregnancy
and term low birth weight with an OR of 1.42
(50% CI, 1.22-1.65). In a study by Wright
et al. (2003), no associations were reported
between term low birth weight and trimester-specific
exposures or entire pregnancy exposures to TTHMs.
Six studies have evaluated preterm birth or very
preterm birth; none found a significant relationship
with DBPs (Bove et al. 1995; Dodds et al. 1999;
Gallagher et al. 1998; Kramer et al. 1992; Savitz
et al. 1995; Wright et al. 2003).
As a group, these studies differed in their
selection of a referent group for exposure, in
their ability to control for potential confounding,
and in their assessment of exposure during the
third trimester or late stages of pregnancy.
To evaluate the relationship between TTHMs and
growth-related birth outcomes, Bove et al. (1995)
averaged quarterly TTHM concentrations over each
subject’s entire pregnancy. In a study
of miscarriage, low birth weight and preterm
delivery in North Carolina, Savitz et al. (1995)
assigned exposure by using the quarterly value
nearest the 28th week of pregnancy. Gallagher
et al. (1998) used the median of all quarterly
measurements taken during the third trimester.
For children born in the second or third month
of the quarter, Wright et al. (2003) used the
average quarterly values for the third trimester;
children born in the first month of the quarter
were assigned the preceding quarterly averages.
In Nova Scotia, Dodds et al. (1999) used linear
regression of quarterly data to estimate average
exposures during the last 3 months of pregnancy.
Our method of assigning exposure included estimating
some periodic study time exposures using a spline
regression based on quarterly sampling values.
Further, all data were interpolated from month
midpoint and converted to ordinal study time,
to better align with gestation time (Yang et
al. 2005). This regression method, which is similar
to that used in the Nova Scotia studies, permitted
estimation of exposure for time periods when
data were missing or when sampling was not performed.
We performed a sensitivity analysis by systematically
repeating the spline regression with varying
subsets of exposure data. By this method, we
found that the model consistently predicted existing
data points to within ± 5%. However, the
spline regression technique requires additional
validation in other distribution systems.
Our study is the first to examine associations
between exposures to specific HAAs and impaired
fetal growth. We found evidence of associations
between exposure to specific HAAs and term low
birth weight and intrauterine growth retardation.
The second and third tertiles of exposure to
HAA5 were also associated with a small increase
in risk for term low birth weight when evaluated
over the third trimester (Table 3). The increased
risk in the second tertile did not seem to be
due to a higher risk from DBAA, DCAA, or TCAA.
HAA5 concentration is currently regulated in
the United States, but concentrations of specific
HAAs are not. Our findings suggest a critical
window of exposure during weeks 33-40 for
the effects of DBAA on fetal development. To
our knowledge, this is the first time that DBAA
has been investigated in an epidemiologic study
of developmental outcomes. Studies of exposure
to HAAs are relatively new, and none have been
performed in communities where DBAA concentrations
in drinking water were above detection (King
et al. 2005; Wright et al. 2004). In this investigation,
the levels of DBAA were well above the 90th percentile
concentrations reported by the U.S. EPA (1998).
We also observed evidence of an association
between intrauterine growth retardation and exposure
to chlorinated HAAs during specific critical
time windows of gestation, with modest increases
in risk for third-trimester exposure to DCAA
and slightly lower estimates for TCAA. When analyzed
as continuous variables, exposure to DCAA and
TCAA also showed slight increases in risk of
intrauterine growth retardation between weeks
29 and 40 of gestation. The risk estimates remained
consistent during the gestation week windows
comprising this time period.
Our study is the first to examine exposure
to DBPs during specific gestation week intervals
of exposure. In previous studies, exposure for
fetal development was usually averaged over the
longer third-trimester window. Averaging a variable
exposure over longer time periods such as the
third trimester is likely to introduce misclassification
over the critical time periods and lead to biased
risk estimates (Hertz-Picciotto et al. 1996).
However, for the highest level of exposure to
DBAA, we observed the same OR for exposure averaged
over the third trimester as for exposure averaged
over gestation weeks 33-36. The CIs were
narrower for DBAA exposure during weeks 33-36
than for the entire third trimester, reflecting
increased precision due to the slightly larger
sample population retained for analysis of gestation
week intervals. Because the third trimester is
a longer time period, it is more likely to fall
outside of the study initiation and termination
(or beginning and end) date than are single week-long
periods (Hinckley 2003; Hinckley et al. 2002).
Windows of exposure have been historically
important in epidemiologic investigations of
thalidomide, retinoic acid (vitamin A), maternal
rubella, and radiation (O’Rahilly and Muller
2001). For exposures during the first 2 weeks
of gestation, few congenital abnormalities are
observed because the teratogen either damages
most cells, resulting in cell and embryonic death,
or affects only a few cells that can be repaired
without resultant birth defects (Moore and Persaud
1998). After the first 2 weeks, the tissue or
organ that is most susceptible to malformation
is the part undergoing critical development when
the teratogen is active. Exposures that occur
later in gestation have a less drastic effect
and are thought to primarily affect fetal growth.
The strengths of this study include the large
number of birth records, high quantity of exposure
data (including some biweekly data), and the
ability to evaluate multiple time periods of
exposure to specific THMs and HAAs. By comparing
subjects within the same community with respect
to exposure levels, we may have reduced potential
residual confounding. We also selected this community
to minimize misclassification due to spatial
variability within the distribution systems (Hinckley
et al. 2005).
Our study was limited by the use of birth records
to ascertain individual exposure information.
Maternal residence was identified from birth
records to assign the appropriate water service,
but residential mobility during pregnancy may
have introduced exposure misclassification. Potential
exposure misclassification could also have resulted
from lack of information regarding exposures
from inhalation or dermal exposure from showering,
bathing, and washing. Exposure estimates were
based on distribution system DBP concentrations
and did not account for variability in personal
habits affecting ingestion, such as the use of
bottled water (Zender et al. 2001). Finally,
exposure misclassification could have resulted
from exposures outside the service area (e.g.,
at work) of the designated water treatment system.
In summary, despite toxicologic evidence of
growth retardation after exposure to DBPs, few
human studies have been conducted on this relationship.
The pervasive nature of the exposure suggests
that even small effects may be important. This
work explored this relationship using seasonal
variability and intracommunity comparisons to
define a natural experiment. We improved on previous
exposure assessments by considering total and
individual THMs and HAAs, and multiple time periods
of exposure in late gestation. Further studies
are needed to confirm our observations for DBAA,
TCAA, and DCAA as well as other relationships
between DBPs and growth outcomes.