METHODS

Study design

We conducted a cross sectional analysis of the relation between adverse birth outcomes and TTHMs in Massachusetts drinking water. We used a semi-individual study design,³⁴ estimating maternal exposure to DBPs from city specific aggregate monitoring data while utilising individual information on potential confounding variables and the outcomes of interest. Data extracted from birth records and hospital worksheets during 1990 (n = 94 407) were supplied by the Massachusetts Department of Public Health. The study population included singleton infants born to residents of communities that routinely monitored trihalomethanes (n = 57 378). We restricted the analyses to infants between 22 and 45 gestational weeks weighing more than 200 g at birth (n = 56 513).

Birth data

Infant birth weight and gestational age were abstracted from birth records. The recorded gestational age was determined from maternal recall of the date of last menses. Prematurity was defined as less than 37 gestational weeks and low birth weight as less than 2500 g. Infants were considered small for gestational age if they were in the lowest 10th centile of birth weight for each gestational week stratified by infant gender and maternal race (African American versus all other races combined). Gender and race specific deciles were determined from the complete dataset of all births in Massachusetts during 1990. The dichotomous outcomes (low birth weight, small for gestational age, and preterm delivery) were not mutually exclusive. For example, an infant could be classified as both preterm and small for gestational age. Term normal weight infants (>37 gestational weeks and >2500 g) served as the comparison group when calculating exposure odds ratios in the logistic regression models.

Exposure data

Drinking water is supplied by local water systems in Massachusetts, although multiple towns may share a common source. Utilities that disinfect their water and serve populations greater than 10 000 are required to monitor trihalomethane concentrations every quarter. Some towns with historically low trihalomethane levels are only required to monitor DBPs annually. We abstracted trihalomethane data (1987–93) from the Massachusetts Department of Environmental Protection records for 99 communities. Recorded trihalomethane concentrations were measured using gas chromatography (Environmental Protection Agency (EPA) Method 502.2) and gas chromatography/mass spectrometry (EPA Method 524.2).³⁵ Trihalomethane samples were analysed at the Massachusetts state laboratory (William Wall Experimental Station) until the third quarter of 1990 and at individual Massachusetts or EPA certified laboratories thereafter.

Eighty nine of the 99 towns collected quarterly samples, while the remaining 10 collected annual measurements. Groundwater was the primary water source for the towns with annual monitoring requirements. Groundwater that is not under the influence of surface water has minimal levels of organic matter and typically requires less disinfection than surface water. This results in fewer DBPs and limited seasonal fluctuations in communities which rely on groundwater.

Table 2 presents the quarterly TTHM concentrations in the study population. In communities with multiple sampling sites in the distribution system (n = 64), we calculated a quarterly average across locations to assign maternal exposure scores. Three of the original 99 towns were excluded from the study because of considerable missing data. Twenty eight of the remaining 96 communities were missing some quarterly TTHM measurements. We imputed data for the missing values based on the city specific quarterly averages from data available from adjacent years (1987–93).

Exposure assessment

Maternal trihalomethane exposure was estimated from the average TTHM concentration for the town of residence. Maternal residence reported on birth certificates was assumed to be constant for the entire pregnancy. Towns with annual trihalomethane measurements were assumed to have constant levels throughout the year, therefore residents were assigned the annual value for each trimester. Trimester specific and pregnancy average exposures were assigned based on the month of birth. Third trimester TTHM exposure was determined from the current quarterly average if the infant was born in the second or third month of that quarter. If the infant was born in the first month of a quarter, the third trimester exposure was assigned the value from the previous quarter. First and second trimester exposures were determined from the two quarters preceding the third trimester designation. We calculated an average maternal pregnancy exposure from the trimester specific values. Premature infants less than 29 gestational weeks were not assigned a third trimester value, so pregnancy average scores were based on exposures during the first and second trimesters.

Data analysis

The Massachusetts birth data included individual information on numerous potential confounding variables. The following maternal risk factors were included in the regression models as dichotomous variables: diabetes, eclampsia, hydramnios, chronic hypertension, hypertension during pregnancy, incompetent cervix, lung disease, renal disease, uterine bleeding, previous preterm delivery, and previous birth of an infant weighing greater than 4000 g. We also controlled for gestational age (cubic polynomial) in the birth weight analyses. Adequacy of prenatal care was determined by the Kessner Index, which incorporates gestational age, timing of the first prenatal visit, and total number of prenatal visits.³⁶ We included dummy variables for prenatal care (none, inadequate, intermediate, and adequate), maternal race (Caucasian, African American, Asian, Native American, and other) and parity (0, 1, 2 or 3, and 4 or more children). For maternal education, we included indicators for 9th grade or less, 10th grade, 11th grade, completion of high school, and completion of at least one year of college. The effect estimates (and standard errors) for 10th and 11th grade education were similar in the regression analyses, so we combined them to improve precision.

We relied on findings from a previous analysis to determine the best modelling approach for continuous maternal risk factors.³⁷ In that analysis, non-parametric smooth functions were used to model the potential non-linear dependence of birth weight on maternal age and smoking. The authors found an inverted U shaped relation of birth weight and maternal age, so we included a quadratic polynomial for maternal age in the regression models. They also found a strong inverse relation between smoking and infant birth weight. That relation became less steep with increasing cigarette use, so we grouped mothers into the following categories: non-smokers, light smokers (1–5 cigarettes per day), moderate smokers (6–10 cigarettes per day), and heavy smokers (11 or more cigarettes per day).

Alcohol consumption is another important risk factor for adverse developmental outcomes.³⁸ The Massachusetts Department of Public Health had advised that the birth certificate data on maternal alcohol use were considered of poor quality and validity. Preliminary analyses indicated that alcohol was not associated with birth weight, so we did not include maternal alcohol consumption in the analyses. We merged 1990 US census data with the birth data to provide further adjustment of socioeconomic status. We adjusted for median household income by census tract for every county but Barnstable County. The Massachusetts Department of Public Health did not code the birth data by census tract for Barnstable County, so we adjusted for median household income based on mother's zip code of residence.

Statistical methods

Linear regression was used to evaluate the association between TTHMs and gestational age for all births and the association between TTHMs and birth weight for term births. We used logistic regression to estimate odds ratios and 95% confidence intervals for the dichotomous outcomes. Pearson correlation coefficients were used to describe the relation between different periods of exposure. All linear and logistic regression analyses were performed using SAS PC statistical software.

RESULTS

There were 1325 low birth weight, 5310 SGA, and 3173 premature infants in the study population (table 3). Fifty one per cent of the infants in the study were male. Seventy four per cent of the mothers were Caucasian and 11% were African American. Racial differences were evident, with a lower frequency of adverse developmental endpoints occurring among Caucasian infants. The average age of mothers in the study population was 28 years, with 68% of the pregnancies occurring between 23 and 34 years of age. Younger mothers (aged 11–22) were more likely to give birth to premature, small for gestational age and low birth weight infants. Other determinants of adverse reproductive outcomes controlled for in the analyses included infant gender, parity, adequacy of prenatal care, maternal education, and maternal cigarette use.

Table 4 presents trimester specific and pregnancy average TTHM exposures. On average, 80% of the study population resided in towns with TTHM concentrations of 60 μg/l or less across the four exposure windows. Eight per cent of women had a pregnancy average TTHM exposure greater than 80 μg/l, with slightly higher frequencies for trimester specific exposures. Average TTHM exposure was 39 μg/l during the first and second trimesters and 38 μg/l during the third trimester. These exposure indices were correlated between successive trimesters (r = 0.50 to 0.66) and between the specific trimesters and the pregnancy average values (r = 0.81 to 0.91).

Birth weight

We observed associations between pregnancy average and trimester specific TTHM exposures and birth weight among term births (table 5). After adjusting for gestational age and other covariates, a 2.8 g (95% CI −5.5 to −0.2) decrease in birth weight was observed for every 20 μg/l increase in pregnancy average TTHM exposure. A similar effect was observed for second trimester TTHM exposure (−2.6 g; 95% CI −4.8 to −0.4 per 20 μg/l). Compared to the reference group (≤60 μg/l TTHM), we found a 32 g (95% CI −47 to −18) reduction in birth weight among infants whose mothers had pregnancy average exposures greater than 80 μg/l. Mothers in the highest exposure group during the second trimester had infants that weighed 23 g (95% CI −36 to −10) less than the comparison group.

Low birth weight

No consistent associations were detected for TTHM exposure and low birth weight among term births (table 6). Increased risks were detected for intermediate first trimester (OR 1.13; 95% CI 0.93 to 1.38) and high second trimester exposures (OR 1.14; 95% CI 0.95 to 1.38). Similar effects were observed for intermediate and high third trimester TTHM exposures.

Small for gestational age

We found a higher frequency of SGA births among mothers with pregnancy average TTHM exposures greater than 80 μg/l (OR 1.14; 95% CI 1.02 to 1.26) compared to the reference group (table 7). Increased risk of SGA was also detected for first (OR 1.09; 95% CI 0.98 to 1.21) and second (OR 1.13; 95% CI 1.03 to 1.24) trimester TTHM exposures.

Gestational duration

We observed small increases in gestational age per 20 μg/l TTHM during the entire pregnancy and for each trimester (table 8). Statistically significant associations were detected for mothers in the highest TTHM exposure group (>80 μg/l) during the second trimester (0.09 gestational weeks; 95% CI 0.02 to 0.15) and over the entire pregnancy (0.08 gestational weeks; 95% CI 0.01 to 0.14). There was little evidence of an association between TTHM exposure and preterm delivery. In fact, the data indicate that mothers with higher exposures during the second trimester (OR 0.90; 95% CI 0.79 to 1.03) and the entire pregnancy (OR 0.90; 95% CI 0.77 to 1.04) were less likely to deliver prematurely (table 9).

DISCUSSION

Our findings indicate that exposure to high trihalomethane levels during pregnancy is associated with reductions in birth weight among term births. After adjusting for gestational age and other covariates, we observed reductions of 2.8 g for each 20 μg/l increase in pregnancy average TTHM concentration and 2.6 g for each 20 μg/l increase in second trimester TTHM concentration (table 5). Maternal exposure to increased TTHMs (>80 μg/l) was associated with lower birth weight when compared to the reference group (≤60 μg/l). The largest effects were observed for pregnancy average (−32 g; 95% CI −47 to −18) and second trimester (−23 g; 95% CI −36 to −10) exposures. Bove²⁷ was the only investigator to examine the effect of TTHM on birth weight as a continuous measure, reporting a 70 g (95% CI −106 to −35) decrease in birth weight for each 100 μg/l increase in TTHM. This is considerably larger than the pregnancy average estimate from this study, which would have predicted a 14 g decrease if extrapolated to 100 μg/l. However, birth weight deficits were only found in the highest exposure category, suggesting a non-linear response in our data.

Overall, we found weaker associations between TTHMs and adverse developmental outcomes than previously reported. Our data indicate that exposure to TTHMs may shift the birth weight distribution downward among term births without significantly impacting the prevalence of low birth weight infancy. We found a small increased risk for term low birth weight among mothers with increased second trimester TTHM exposures (OR 1.14; 95% CI 0.95 to 1.38). In contrast to findings from Gallagher and colleagues,²⁹ we found a similar effect of TTHM on low birth weight among all births (data not presented).

Since low birth weight is a function of fetal growth rate and gestational duration, we evaluated the impact of TTHM on both small for gestational age and premature infants. We observed an increased risk of SGA among mothers with high second trimester (OR 1.13; 95% CI 1.03 to 1.24) and pregnancy average (OR 1.14; 95% CI 1.02 to 1.26) TTHM exposures. These relative risks are similar in magnitude to the study by Dodds and colleagues,³⁰ but are substantially smaller than reports by Kramer and colleagues²⁸ and Bove.²⁷ Consistent with other studies, we did not detect an association between TTHM exposure and preterm delivery. Therefore, DBPs do not appear to adversely impact length of gestation. In fact, our data indicate that TTHMs are associated with small increases in gestational duration. The clinical significance of such a small increase in gestational duration is unclear, especially given the poor precision of gestational age estimates.³⁹

The third trimester is considered the most critical period of exposure for a developing fetus. While fetal growth tends to peak between 43 and 44 weeks, second trimester exposures may be important as well. Ninety five per cent of fetal growth occurs after the 20th gestational week,⁴⁰ with increases in fetal growth rate beginning after 34 gestational weeks.⁴¹ Compared to the third trimester, we saw slightly stronger effects of second trimester TTHM exposure on fetal growth indices. Our estimates of maternal exposure between successive trimesters are highly correlated (r = 0.50 to 0.65), so we may not be able to differentiate the effect of DBPs between the different trimesters.

One of the strengths of our study was the comprehensive information on individual pregnancy risk factors available from birth records and hospital worksheets. The large study population allowed us to control for important maternal risk factors not considered in previous studies. Confounding by gestational age and other maternal risk factors had the largest impact on the association of TTHMs and birth weight (tables 5–9) but did not have much of an effect on the relation between TTHMs and the other outcomes. Nevertheless, the potential for residual confounding remains, since we were not able to control for other reproductive risk factors (marital status, weight gain during pregnancy, maternal nutrition, alcohol use, etc). These risk factors would have to be associated with TTHMs as well as the reproductive outcomes to be confounders.

A potential limitation of our study is exposure misclassification, since we relied on the address of residence at time of birth to assign exposure status. Previous reports indicate that a considerable number of women move during pregnancy (10–37%).²⁸ An additional limitation of our study is the inability to distinguish the impact of specific routes of exposure (ingestion, inhalation, and dermal absorption) to volatile compounds such as the trihalomethanes.⁴² Although the findings are inconsistent, a few studies have been able to evaluate the impact of water consumption^24,43 and showering.⁴³ While these sources of exposures are important, any resulting misclassification errors in our study are likely to be independent of the outcome measures. Non-differential exposure misclassification typically biases effect estimates towards the null, which may explain the smaller relative risks in this study compared to previous reports. Despite the potential for non-differential misclassification, the aggregate measures of exposure should reflect the gross temporal and spatial differences in the exposure experience of our study population.

Variability in DBP formation has been noted in distribution systems given adequate chlorine residual. Trihalomethanes can increase further out in the distribution system, with more variation observed in the brominated compounds (compared to chloroform).⁴⁴ In contrast, haloacetic acids may undergo biological degradation since maximum values have been found at the closest sampling locations.⁴⁵ Understanding these relations is important for exposure assessment strategies that rely on limited DBP monitoring data. While we were able to capture some of the temporal variability associated with DBP formation through the use of routinely monitored exposure data, our inability to account for spatial variability may also result in misclassification. Gallagher and colleagues²⁹ were able to address seasonal and spatial variation in trihalomethane concentrations by using specific sampling dates and incorporating hydraulic characteristics in a computer simulation model of residential exposures. They reported the largest relative risk for low birth weight (OR 2.1; 95% CI 1.0 to 4.8) and term low birth weight (OR 5.9; 95% CI 2.0 to 17.0) despite low TTHM concentrations.

Only one study has explored the role of other DBPs on adverse reproductive outcomes. Klotz and colleagues⁴⁶ reported an OR of 1.2 (95% CI 0.5 to 2.6) for total haloacetic acids >35 μg/l compared to <3 μg/l and an OR of 1.3 (95% CI 0.6 to 2.5) for haloacetonitriles >3.0 μg/l compared to <0.5 μg/l. Kramer and colleagues²⁸ reported an 80% (OR 1.8; 95% CI 0.9 to 3.4) increased risk of SGA for exposure to total organic halides >100 μg/l. Total organic halides were highly correlated with trihalomethane concentrations, so the authors were not able to determine the independent effect of each DBP marker. More research is needed to identify the specific compounds that DBP surrogates represent. If TTHM concentration (or another surrogate marker) is a poor proxy for specific DBPs, even the most elaborate exposure modelling will not be able to fully elucidate the potentially harmful effects of DBPs on fetal development. Our results add to the body of literature indicating that there are such effects, but further research is needed to better delineate them.