Drinking water disinfection byproducts and risk of orofacial clefts in the National Birth Defects Prevention Study

2.1 Study sample

The National Birth Defects Prevention Study (NBDPS) was a multisite, population-based case-control study funded by the Centers for Disease Control and Prevention. As detailed elsewhere (Reefhuis et al., 2015), interview reports were collected from mothers of cases and controls with estimated dates of delivery (EDD)s from October 1, 1997–December 31, 2011. Case deliveries were live births, stillbirths, or elective terminations diagnosed with one or more of over 30 major structural birth defects identified from the population-based birth defect surveillance program at each NBDPS site; cases with monogenic or chromosomal etiologies or whose OFCs were secondary to another defect were excluded. Eligible cases for the current project were those diagnosed with CL (modified British Paediatric Association [BPA] codes 749.101–749.103, 749.110, 749.120, 749.195); CP (BPA codes 749.001–749.003, 749.010, 749.020, 749.030, 749.041–749.043, 749.050, 749.060, 749.070, 749.090); or CLP (BPA codes 749.201–749.203, 749.210, 749.220, 749.290). OFC cases were further classified as isolated (no other major defects) or multiple (at least one additional major, unrelated defect) (Rasmussen et al., 2003). Eligible controls were live births without major birth defect diagnoses randomly selected from birth certificates or birth hospitals in the corresponding surveillance catchment areas for each NBDPS site. The NBDPS sites were Arkansas, California, Georgia, Iowa, Massachusetts, New Jersey, New York, North Carolina, Texas, and Utah. North Carolina and Utah joined the NBDPS in 2003. For this analysis, cases and controls from California and New Jersey were excluded because access to water quality data from individual public water systems at these sites was not available, and controls from Utah delivered in 2003 were excluded because Utah did not contribute OFC cases during that year.

2.2 Data collection

Mothers of eligible OFC cases and controls were asked to complete the NBDPS interview by telephone in either English- or Spanish-language (a copy of the interview is available from the corresponding author upon request). Briefly, the interview included items regarding parental sociodemographics, family history of birth defects, and occupation as well as maternal medical history, prenatal care, diet, lifestyle, and residence history beginning three months prior to conception (B3) through the EDD or end of pregnancy. Interviews conducted with mothers of cases and controls delivered between January 1, 2000 and December 31, 2005 included a detailed drinking water module that asked about maternal water sources, residential water treatment, drinking water consumption, and additional water use activities. Participation rates for the interview for all study years (1997–2011) were 71% for mothers of OFC cases and 64% for mothers of all controls. Our analyses included 1582 mothers of OFC cases and 3962 mothers of controls who completed an interview with the detailed drinking water module.

2.3 DBP concentration estimates

The critical period for lip and palate development is the first three months of gestation (Burdi, 2006). The estimated date of conception (EDC) for each mother was calculated by subtracting 280 days (40 weeks) from the EDD to obtain the last menstrual period date, then adding 14 days. Using the EDC and assigning 30 days per month, we estimated maternal exposure to DBPs from tap water for the preconception month (B1) immediately before the EDC through the first three months of pregnancy (P1, P2, P3) by applying a structured algorithm that combined interview data and DBP water sample results. The preconception month was included as part of the critical exposure period due to the potential for maternal prepregnancy behaviors to extend into early pregnancy.

In the interview, mothers were asked to report the full address, along with the residency start date (month, year) and end date (month, year), for each residence reported. Addresses were geocoded using Centrus (Group 1 Software; Lanham, MD); 97% of addresses were matched at any level, and 89% were matched at the address level. Public water systems were identified by linking the geocoded maternal residence to digitized public water system service area maps; 2010 census place shapes were used to approximate service boundaries if the public water system service area map was not available. For a city served by multiple public water systems that lacked information on the public water system boundaries within the district, the public water system that served the largest number of residences in the city was linked to the maternal geocoded residence.

An attempt was made to obtain DBP sampling results (contaminant levels, sampling date, location) from linked public water systems. DBP data are available on monitoring schedules for all public water systems in accordance with Safe Drinking Water Act regulations and guidelines (U.S. EPA, 2010). Nationally, public water systems are required to monitor THM and HAA levels; however, the frequency of mandatory reporting varies depending on the size of the public water system and the population served. Larger systems are required to monitor quarterly—compliance is based on a running annual average of quarterly samples. Small surface water and small groundwater systems are required to monitor annually. As such, for some cases and controls, TTHM or HAA5 levels were not available from the linked public water systems that served the respective maternal geocoded residences from B1 through P3.

Exposure measurement error can occur when sampling results from a public water system are assigned to a single residence due to temporal (e.g., monthly or seasonal) and spatial fluctuations in DBP concentrations throughout the system. To account for these fluctuations, we estimated an inverse-time weighted mean using all sample measurement days (up to a maximum of 10) for each available THM and HAA during the critical exposure period for each case and control mother, giving a higher weight to those measurements that occurred closest to the EDC. For mothers served by systems with multiple DBP sample measurements taken in a single day at different locations throughout the system during the critical exposure period, we used the mean concentration for each THM and HAA to estimate exposure for that day. Additionally, we explored a weighting factor to reduce the statistical contribution of mothers whose public water systems had high spatial variability in DBP concentrations throughout the system within individual measurement days (Waller, Swan, Windham, & Fenster, 2001); this factor was omitted from our final analyses due to a large proportion of mothers in our analytic sample whose public water systems provided only a single measurement for each DBP per measurement day.

2.4 Maternal water consumption estimates

During the interview, mothers were asked whether: the drinking water at the residence closest to the EDC was from a private well or public water system; the water was chemically disinfected (private well only); and the water used for drinking or cooking was filtered (none, whole house filter, faucet filter, etc.), and if filtered, the type of filter (membrane, charcoal, etc.) and frequency of filter changes per year. For each reported residence, mothers were also asked about the water source(s) (unfiltered tap, filtered tap, bottled, other) used for drinking, number of 8 oz. glasses of water consumed per day from each source, sources used to make hot drinks and for cooking, and details about changes in drinking water consumption from B3 through the end of the pregnancy (the month of a change in amount, number of 8 oz. glasses of water consumed per day after a change in amount, water sources used after a change in source). Additionally, mothers were asked about the water source(s) (unfiltered tap, filtered tap, drinking fountain [coded as unfiltered tap], bottled/cooler, brought from home, other) used for drinking at school (if enrolled) or at each job (if employed), and the average number of 8 oz. glasses of water consumed per day from each source at school and work. We reviewed the “other” responses for water source at home, work, and school and recoded them into one of the predefined sources, where possible. Additional interview items asked about water use activities, including washing dishes and clothes, showering and bathing, and swimming.

Total water consumption during the critical exposure period from each water source was estimated using the number of 8 oz. glasses of water consumed per day at home and while at each job and school, and the estimated number of days spent at each job and school during the critical exposure period (for an example, see Box 1). We did not use the responses about water sources used to make hot drinks and for cooking, as associated consumption amounts were not reported. A mother's estimated average daily consumption from a water source was calculated by dividing her total estimated consumption by 120 days.

For mothers who reported changing their daily amount of water consumption or starting or stopping living at a residence, working at a job, or attending school, the date of each event was collected in the interview at the level of calendar month and year. If the year was reported but not the month, the change was assumed to occur in July. To determine if and when any of these events occurred during the critical exposure period, we converted the calendar month and year to the 30-day period (B1, P1, P2, P3) that contained the largest number of days within the identified calendar month. For example, January would be assigned P1 for a mother with an EDC of January 10th but assigned B1 for a mother with an EDC of January 20th. If two 30-day periods contained the same number of days for a given calendar month, the earlier 30-day period was assigned. Changes to consumption amount were assumed to apply to one-half of the 30-day period during which they were reported. Mothers were assumed to be at a residence, job, or school during one-half of the 30-day period during which they started or stopped being at that location.

Although the interview asked about the timing of any change in the amount of drinking water consumed at home, it did not ask about the distribution of the change by water source. If more than one water source was reported, we estimated the source distribution of total home drinking water consumption after the change using unweighted and weighted approaches (Figure 1). Our unweighted approach distributed total consumption across reported water sources according to the proportions before the change in consumption amount. Our weighted approach used a ranking (lowest to highest) of water sources by DBP concentration (bottled water, filtered tap water, unfiltered tap water). Arbitrary weights of 3, 2, and 1 were used if all three water sources were reported, and arbitrary weights of 3 and 1 were used if only two water sources were reported. Our low-DBP-weighted approach assigned the highest weight value to the water source with the assumed lowest DBP concentration. Conversely, our high-DBP-weighted approach assigned the highest weight value to the water source with the assumed highest DBP concentration (for an example, see Box 2).

Although a mother could report a change in water source, the interview did not ask about the date corresponding to the change. As such, we estimated consumption from each water source using two different assumptions for mothers who reported changing sources. The first assumption was that the change in water source occurred before the critical exposure period; the second assumption was that the change in water source occurred after the critical exposure period. For the first assumption, because the interview did not ask about the distribution of total water consumption by water source after a reported change in source, we estimated the distribution to be proportional to the weights of the respective consumption estimation approach (for an example, see Box 3). For the second assumption, the water source change did not affect consumption estimates because it did not apply to the critical exposure period. In calculating total consumption that included amounts from work or school, we assumed the distribution of water sources that a mother reported bringing from home to work or school was proportional to the estimated distribution of home water consumption.

2.5 Maternal DBP ingestion estimates from public water systems

In our analyses, a mother's exposure to DBPs in tap water depends on the levels of DBPs in the public water system(s) serving her residence, work, and school; her average daily water consumption from each water source; and the use of any water filters at her residence, work, and school that affected DBP levels. We assumed that private well water (due to minimal reported disinfection treatments) and bottled water had 0 µg/L of DBP exposure. We also assumed a mother's work and school were in the same water district as her residence. If a mother reported use of a filtration system at home, the reported brand names for filtration systems and filters provided were queried in the list of NSF International certified drinking water treatment units (http://info.nsf.org/Certified/DWTU/) to determine whether the system or filter could remove DBPs. If no brand name was reported, the effectiveness in DBP removal of the filter or system was determined by the description of the filtration method. Types of filters that are known to remove DBPs were estimated to reduce the DBP concentration to 10% of that measured in the public water system. Reported types of filters unable to remove DBPs or those with undetermined capacity to remove DBPs were assumed to reduce the DBP concentration to 90% of that measured in the public water system. The interview did not ask about information regarding the types of filters used at work or school, so these filters also were assumed to reduce the DBP concentration to 90% of that measured in the public water system.

2.6 Maternal dermal and inhalation DBP exposures

A positive association between longer showers taken by mothers (duration ≥15 min) and cleft lip ± cleft palate (CL/P) has been previously reported using NBDPS data (Agopian, Waller, Lupo, Canfield, & Mitchell, 2013); therefore, we included average shower duration in our analysis as a covariate. Because our focus was on DBP exposures from drinking water consumption, we did not consider additional reported dermal or inhalation exposures from washing dishes, washing clothes, bathing, and swimming in our analyses.

2.7 Statistical analysis

Case and control mothers were excluded from analysis if they reported a diagnosis of pregestational diabetes due to the strong association between pregestational diabetes and birth defects, particularly OFCs (Aberg, Westbom, & Kallen, 2001; Becerra, Khoury, Cordero, & Erickson, 1990; Correa et al., 2008; Spilson, Kim, & Chung, 2001). Mothers were eligible to be included in the analyses if they resided at the same residence throughout the critical exposure period, and they either did not drink tap water provided by a public water system during the critical exposure period, or they drank tap water provided by a public water system during the critical exposure period and their DBP ingestion could be estimated. Mothers’ DBP ingestion could be estimated if: (a) they reported the year(s) and number of days per week they attended every reported job and school outside the home; (b) they reported water sources and consumption amounts at their residence and at all jobs and schools outside the home during the critical exposure period; (c) all reported water sources could be categorized as “unfiltered tap,” “filtered tap,” or “bottled;” (d) their residence was successfully geocoded; (e) the public water system that served their residence was identified; and (f) the DBP measurements of that water system were known while they resided at that address during the critical exposure period.

Selected child and maternal characteristics and maternal exposures during the critical exposure period were assessed as covariates. To evaluate the representativeness of OFC cases and controls available for analysis, we compared their covariates to those of all OFC cases and all controls delivered during 2000–2005, respectively, using chi-square goodness-of-fit tests (calculating exact p-values if expected cell counts <5). We also compared covariates of all OFC cases and controls eligible for analysis using chi-square tests of independence or Fisher's exact tests (if expected cell counts <5) to determine statistically significant differences (p < .05).

Child characteristics examined were sex (male, female), gestational age (<37, 37–45 weeks), family history of OFC (first-degree relative, other relative, none), and plurality (single birth, twin, other multiple birth). Maternal characteristics examined were age at delivery (<20, 20–24, 25–29, 30–34, ≥35 years), education at delivery (<12, 12, 13–15, ≥16 years), race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, other), gravidity (first pregnancy, second pregnancy, third or higher pregnancy), prepregnancy body mass index (BMI) (<18.5, 18.5–25.0, 25.0–30.0, ≥30.0 kg/m²), and study site (Arkansas, Georgia, Iowa, Massachusetts, New York, North Carolina, Texas, Utah). Maternal exposures during the critical exposure period examined were cigarette smoking (no active or passive smoking, active smoking only, passive smoking only [exposed to cigarette smoking at home or workplace], active and passive smoking), alcohol consumption (no drinking, binge drinking [≥4 drinks on one occasion], drinking but no binge drinking), use of a folic acid-containing supplement (yes, no), and average shower duration (<15, ≥15 min).

Crude odds ratios (cORs), adjusted odds ratios (aORs), and 95% confidence intervals (CI)s were estimated using logistic regression analysis. Associations between OFCs and DBPs were analyzed for all OFCs, isolated OFCs, and two subtypes of isolated OFCs: CP and CL/P. We examined potential associations between each outcome group and maternal exposure to TTHMs and HAA5 in µg per liter of water consumed. Categories of exposure were based on one-half the MCLs (40 µg/L for TTHMs and 30 µg/L for HAA5), and mothers unexposed to TTHMs or HAA5 were used as the referent groups for the respective analyses. Crude analyses were conducted only for categories of TTHM and HAA5 exposures that included at least three exposed and three referent case mothers.

Adjusted analyses were conducted only for categories of TTHM and HAA5 exposures that included at least five exposed and five referent case mothers. Covariates that were found to be statistically associated (p < .05) with the outcome group and exposure using chi-square tests of independence or Fisher's exact tests (if expected cell counts <5) were added separately to the exposure-only model; covariates that changed the cOR estimate by more than 10% were included in the adjusted model. Because a previous NBDPS study reported that increases in maternal water consumption were inversely associated with CL (Alman, Coffman, Siega-Riz, Luben, & National Birth Defects Prevention Study, 2017), we considered total maternal water consumption during the critical exposure period as a potential covariate in adjusted analyses. We observed that total maternal water consumption was not associated with any OFC outcome group using logistic regression analysis, and thus, it was not included in any adjusted models.

We also examined potential associations between the outcome groups and individual THM and HAA contaminants in µg per liter of water consumed. Categories of exposure were based on one-half the Maximum Contaminant Level Goals (MCLG)s, if available and non-zero (35 µg/L for chloroform, 30 µg/L for dibromochloromethane, 35 µg/L for monochloroacetic acid, and 10 µg/L for trichloroacetic acid) (U.S. EPA, 2010). Contaminants with MCLGs of 0 µg/L (bromoform, bromodichloromethane, and dichloroacetic acid) or no MCLGs (bromoacetic acid and dibromoacetic acid) (U.S. EPA, 2010) were analyzed as dichotomous indicators of any maternal exposure from drinking water. Mothers unexposed to a specific contaminant were used as the referent group for the analyses of that contaminant. Mothers from Massachusetts and Utah were excluded from our analysis of individual contaminants because these sites did not report individual THM and HAA contaminant concentrations.

As a subanalysis, we examined potential associations between the same outcome groups and concentrations of TTHMs, HAA5, and individual THM and HAA contaminants in the public water systems linked to maternal residences without accounting for filtration and consumption. These results are more directly comparable to previous studies that lacked this information (Dodds et al., 1999; Hwang et al., 2008; Kaufman et al., 2018; Nieuwenhuijsen et al., 2008; Righi et al., 2012; Shaw et al., 2003). Concentrations less than one-half the MCLs or MCLGs were used as the referent groups, except for contaminants with MCLGs of 0 µg/L or no MCLGs for which concentrations of 0 µg/L were used.

To assess the magnitude of improvement of exposure classification in our study compared to ecological studies, we compared mothers’ classifications of TTHM and HAA5 exposures per liter of water consumed with their classifications of TTHM and HAA5 concentrations in public water systems. All analyses were conducted using the Statistical Analysis System (SAS) version 9.4 statistical software (SAS institute, Cary, NC).