Risks of obstructive genitourinary birth defects in relation to trihalomethane and haloacetic acid exposures: expanding disinfection byproduct mixtures analyses using relative potency factors

Abstract

BACKGROUND:

Some disinfection byproducts (DBPs) are teratogens based on toxicological evidence. Conventional use of predominant DBPs as proxies for complex mixtures may result in decreased ability to detect associations in epidemiological studies.

OBJECTIVE:

We assessed risks of obstructive genitourinary birth defects (OGDs) in relation to 12 DBP mixtures and 13 individual component DBPs.

METHODS:

We designed a nested registry-based case-control study (210 OGD cases; 2100 controls) in Massachusetts towns with complete quarterly 1999–2004 data on four trihalomethanes (THMs) and five haloacetic acids (HAAs). We estimated temporally-weighted average DBP exposures for the first trimester of pregnancy. We estimated adjusted odds ratios (aORs) and 95% confidence intervals (CIs) for OGD in relation to individual DBPs, unweighted mixtures, and weighted mixtures based on THM/HAA relative potency factors (RPF) from animal toxicology data for full-litter resorption, eye defects, and neural tube defects.

RESULTS:

We detected elevated aORs for OGDs for the highest of bromodichloromethane (aOR = 1.75; 95% CI: 1.15–2.65), dibromochloromethane (aOR = 1.71; 95% CI: 1.15–2.54), bromodichloroacetic acid (aOR = 1.56; 95%CI: 0.97–2.51), chlorodibromoacetic acid (aOR = 1.97, 95% CI: 1.23–3.15), and tribromoacetic acid (aOR = 1.90; 95%CI: 1.20–3.03). Across unweighted mixture sums, the highest aORs were for the sum of three brominated THMs (aOR = 1.74; 95% CI: 1.15–2.64), the sum of six brominated HAAs (aOR = 1.43; 95% CI: 0.89–2.31), and the sum of nine brominated DBPs (aOR = 1.80; 95% CI: 1.05–3.10). Comparing eight RPF-weighted to unweighted mixtures, the largest aOR differences were for two HAA metrics, which both were higher with RPF weighting; other metrics had reduced or minimally changed ORs in RPF-weighted models.

INTRODUCTION

Disinfection of drinking water systems is one of the most important public health interventions. Chlorination, chloramination, and other common forms of primary and secondary water disinfection intended to inactivate microorganisms and provide residual disinfectant to prevent growth in distribution systems, respectively, can result in the formation of potentially harmful disinfection by products (DBPs).

Toxicological studies provide evidence that some DBPs are teratogens; thus, there is concern that they may pose a risk to human health [1, 2]. Although not always consistent, developmental epidemiological studies have found some evidence of relationships between exposure to DBPs and birth outcomes such as fetal growth restriction [3–6]. This includes support for some birth defects, particularly cardiac defects, which are among the most commonly detected birth defects globally [7, 8]. Among prior epidemiological studies of DBPs and urinary tract defects, eight [9–16] out of nine [17] reported elevated associations for some DBP metrics or proxies (e.g., whether or not the mother’s water source was treated with chlorine or chloramine). Four of these studies [9, 10, 16, 17] examined individual DBP components (trihalomethanes (THMs), chlorite, and chlorate); none examined haloacetic acids (HAAs), and none examined mixtures beyond the sum of four trihalomethanes (THM4).

Obstructive genitourinary defects (OGDs) are a subset of urinary tract and kidney defects that are also among the most common birth defect groups, affecting approximately 0.5 to one percent of births worldwide [18]. Prevalence estimates are based on more severe cases, as milder manifestations are less likely to be detected but possibly more common [19]. Obstructive birth defects of the kidney and urinary tract can lead to buildup of urine in the urinary tract, increasing risks of urinary tract infections, kidney damage, and kidney failure, and are the leading cause of end-stage renal disease in children [20]. Genitourinary birth defects can have health impacts throughout the life-course, with detrimental effects on kidney function often increasingly severe with age [19].

A major challenge of conducting epidemiologic research on DBPs is that a small number of the predominant (i.e., occurring at the highest concentrations) organic DBPs, or sums of these DBPs, are used in exposure assessments, while over 700 DBPs have been identified in treated drinking water systems [21]. In the United States, DBPs in drinking water systems are regulated based on monitoring the sum of four THMs (THM4) and the sum of five haloacetic acids (HAA5) [22]. It remains unclear which DBP exposure metrics are the best surrogates to represent the most toxicologically relevant DBPs for specific health endpoints in humans; universal use of only a few DBP mixture surrogates (e.g., THM4 or HAA5) complicates interpretation of study results and can lead to decreased sensitivity to detect specific DBP-health endpoint associations. The existing DBP mixture surrogates are also concentration-based summaries that do not consider the potency of individual DBPs. Alternate approaches such as the use of relative potency factors can help create exposure assessments for mixtures that better reflect the relative toxicity of a mixture’s components.

To address these research needs, we conducted a case-control study of 210 OGD cases in relation to estimated first-trimester in utero DBP exposures within Massachusetts using data from 1999–2004. In addition to estimating risks from exposures to the four regulated THMs and five regulated HAAs individually and various mixture summary measures, we estimated risks from four unregulated brominated HAAs, and incorporated relative potency factors [23] based on animal toxicological evidence in an attempt to improve exposure estimates and consider previously unexamined DBP mixture metrics.

METHODS

Study design and population

We designed a registry-based case-control study nested within a retrospective birth cohort based on records data provided by the Massachusetts Birth Defects Monitoring Program. The inclusion criteria were OGD cases (n = 210) and non-OGD controls (n = 2100) born to mothers that resided in Massachusetts towns and cities with populations greater than 500 (n = 43 in our final sample) and with complete 1999–2004 quarterly data on four regulated THMs (THM4) and five regulated HAAs (HAA5). We restricted the study population to singleton live births born from 22–44 gestational weeks, weighing ≥350 grams, and without identified chromosomal birth defects (i.e., births where the birth certificate listed International Classification of Diseases 9th (ICD-9) code 758 for chromosomal anomalies were excluded). We included 210 OGD cases identified from the Massachusetts Birth Defects Monitoring Program and 2100 controls without known OGD from Massachusetts birth records provided by the Massachusetts Department of Public Health, matched 1:10 on week of conception. We restricted the study population to singleton live births born from 22–44 gestational weeks, weighing ≥350 grams, and without identified chromosomal birth defects (i.e., births where the birth certificate listed International Classification of Diseases 9th (ICD-9) code 758 for chromosomal anomalies were excluded).

Outcome Data

We obtained birth records for years 2000–2004 from the Massachusetts Department of Public Health, and records from the Massachusetts Birth Defects Monitoring Program that include identified and verified birth defect cases up to one year after birth using a variety of data sources (e.g., birth certificates, hospital discharge reports, hospital fetal and infant death certificates, hospital surgical and pathology departments, and nurseries and neonatal units). To monitor birth defects, the Massachusetts Center for Birth Defects Research and Prevention collects data from Massachusetts birthing hospitals, Rhode Island birthing hospitals near the Massachusetts border, hospital nurseries, tertiary care hospitals, and from vital records (birth certificates, fetal death reports, and infant death certificates) [24]. Birthing hospitals, other pediatric care centers, nurseries, and neonatal intensive care units report discharge lists and cases with birth defect diagnoses. Medical record abstractors go to reporting hospitals to collect additional data and verify birth defects cases, and a clinical geneticist reviews the cases. Surveillance data are then entered into the database.

Obstructive genitourinary defect (OGD) cases were identified based on the ICD-9 revision in use during the study period. We included all cases of OGD identified using ICD-9 code 753.2 (753.20–753.29, atresia and stenosis of urethra and bladder neck, kidney, pelvis, and urethra). We matched each case individually to ten controls, randomly sampled without replacement based on week of conception in order to maintain statistical efficiency while adjusting for potential time-varying confounding by seasonality [25]. We estimated week of conception by taking the difference between birth date and clinical estimates of gestational age on birth records. Given the lack of established risk factors for OGDs which could potentially act as confounders, we did not match on additional factors.

Measured Exposure Data

Routine monitoring data from 1999–2004 for nine individual DBPs [chloroform (trichloromethane, TCM); bromodichloromethane (BDCM); dibromochloromethane (DBCM); bromoform (tribromomethane, TBM); dichloroacetic acid (DCAA); trichloroacetic acid (TCAA); monochloroacetic acid (MCAA); dibromoacetic acid (DBAA); and monobromoacetic acid (MBAA)] were obtained from the Massachusetts Department of Environmental Protection and individual public water systems. THM concentrations were quantified by certified laboratories using capillary column gas chromatography with EPA Method 502.2 [26], capillary column gas chromatography/mass spectrometry with EPA Method 524.2 [27] and gas chromatography with electron capture detection with EPA Method 551.1 [28]. HAA concentrations were quantified with EPA Methods 552.1 [29] and 552.2 [30] using gas chromatography and electron capture detection plus Standard Method 6251B [31] using micro liquid-liquid extraction gas chromatography. Detection limits varied across laboratories and time, ranging from 0.1 to 2.5 μg/L for THMs and 0.4 to 5.0 μg/L for HAAs. We imputed a value of zero for study participants with DBP levels below the limit of detection and for those on untreated groundwater sources.

Predicted exposure data

In order to extend the scope of exposures available for examination, we estimated concentration values of four unregulated HAAs (tribromoacetic acid [TBAA], bromochloroacetic acid [BCAA], bromodichloroacetic acid [BDCAA], dibromochloroacetic acid [DBCAA]) with a kinetic binomial prediction model that used measured values of TCM, BDCM, TCAA, and DCAA [32]. Unlike THM4 and HAA5, these four brominated HAAs are currently unregulated, and measured monitoring data are unavailable from public water utilities. The prediction model interprets DBP formation as competitive reactions of chlorine versus bromine for reactive sites on organic precursors, following a binomial pattern weighted by the available active bromine at each halogenation step. This model has previously been shown to have high accuracy, with high correlations between predicted and measured values for these four unregulated HAAs (R² > 0.98).

Exposure assessment for unweighted metrics

We estimated exposures for the nine measured individual DBPs listed above, the four predicted individual HAAs, and twelve summary metrics (THMBr [sum of BDCM, DBCM, and bromoform]; THM4 [sum of chloroform and THMBr]; HAA5 [sum of DCAA, TCAA, MCAA, DBAA, and MBAA]; HAA4 [sum of predicted TBAA, BCAA, BDCAA, and DBCAA]; HAA9 [sum of HAA5 and HAA4]; HAACl [sum of TCAA, DCAA, and MCAA]; HAABr [sum of TBAA, DBAA, MBAA, BCAA, BDCAA, and DBCAA]; DBP9 [sum of THM4 and HAA5]; DBP13 [sum of DBP9 and HAA4]; DBPBr [sum of THMBr and HAABr]; DBPCl [sum of chloroform, TCAA, DCAA, and MCAA]; and the sum of TCAA and DCAA. The components of each mixture metric are shown in Supplementary Table 1. We predominantly used quarterly DBP monitoring data from public drinking water systems to create temporally weighted average first-trimester exposure estimates for the nine regulated DBPs, assigned based on maternal residential ZIP codes at delivery. For example, an infant born at 38 gestational weeks in February of 2003 would have two first-trimester weeks that occurred in the second quarter of 2002 and the remaining 11 weeks occurring in the third quarter of 2002. Their temporally weighted exposure scores would be calculated as (2/13) times the average sampled DBP concentration for the first quarter plus (11/13) times the average sampled DBP concentration for the second quarter using data from the whole public water system serving their ZIP code. Individual and summed DBP metrics were categorized based on exposure percentile distributions among controls. The number of exposure categories ranged from two for the least prevalent DBPs to five for the most prevalent DBPs. We chose the number of categories to maximize exposure contrasts (i.e., to allow highly exposed to be compared to low or unexposed referent) while maintaining sufficient cell counts for identifiability concerns and to minimize the risk of sparse data bias [33].

Generation of relative potency factors (RPF) for weighted mixtures exposure assessment

We calculated RPF values for individual DBPs based on toxicological evidence from studies of full litter resorption, eye defects, or neural tube defects. In applying RPF values derived from toxicological studies of these endpoints to OGDs, we assume that the relative teratogenicity of specific DBPs follows a similar order and magnitude across birth defect types. Generally, RPFs are an approach for weighting components of a mixture based on their relative toxicity, typically at 50% effect levels, compared to the most potent or most representative mixture component [34]. Using toxicology data to derive a benchmark dose (BMD) at which 50% of the study subjects experience the outcome of interest (BMD50), the component with the lowest BMD50 (i.e., induced effects at the lowest concentration and is therefore most potent) is chosen as the index chemical, and the other mixture components are weighted relative to that index. BMD50 values were calculated for both in vivo rat data (Narotsky et al. [35], Narotsky et al. [36], Smith et al. [37], Smith et al. [38]) and for rodent whole-embryo culture data (Hunter et al. [2], Hunter et al. [1], Andrews et al. [39]). BMDs were calculated using Benchmark Dose Software (BMDS) Version 3.2 [40]. BMD50 values were calculated using dichotomous models with 0.5 benchmark risk (BMR) of “extra risk” type, 95% confidence level, and estimated background (i.e. estimated risk at dose of 0). Frequentist restricted models that were considered were dichotomous Hill, gamma, log-logistic, multistage, and Weibull; whereas logistic, log-probit, and probit models were unrestricted. The BMD corresponding to the best fitting model (generally the model with the lowest Akaike Information Criteria value) was selected as the BMD50.

Relative potency factor weighted summary metrics

We applied BMD50 values to create RPF-weighted mixture metrics for individual DBPs where appropriate data were available from the toxicology literature. We only examined DBP combinations for which RPF data were available for common endpoints (i.e., we did not combine RPF values for different developmental or birth defect endpoints into the same weighted mixture metric). For each DBP mixture, weights for each mixture component were equal to the BMD50 value for the referent chemical (i.e., the chemical with the lowest BMD50) divided by the BMD50 value for each respective component. We multiplied each component’s weight by the estimated first trimester exposure level for that chemical and summed the weighted components to create the weighted mixture estimates. The first trimester is the critical developmental window when congenital anomalies including OGDs occur in humans. The urinary, renal, and reproductive structures affected by OGDs begin development in the fourth gestational week, and functional kidneys typically develop by the tenth week [41].

For example, we calculated RPF-weighted THM4 based on full-litter resorption studies using BDCM as the referent chemical via the following formula, weighting each component mixture relative to BDCM based on relative toxicity:

$$\begin{gathered} {\text{RPF}}_{\text{THM4}}=({\text{BMD50}}_{\text{BDCM}}/{\text{BMD50}}_{\text{chloroform}}) \\ {\text{*(estimated 1}}^{\text{st}}\text{trimester chloroform exposure in μg/L)} \\ {\text{+(BMD50}}_{\text{BDCM}}{\text{/BMD50}}_{\text{bromoform}}) \\ {\text{*(estimated 1}}^{\text{st}}\text{trimester bromoform exposure in μg/L)} \\ {\text{+(BMD50}}_{\text{BDCM}}{\text{/BMD50}}_{\text{DBCM}}) \\ {\text{*(estimated 1}}^{\text{st}}\text{trimester DBCM exposure in μg/L)} \\ {\text{+(BMD50}}_{\text{BDCM}}{\text{/BMD50}}_{\text{BDCM}}) \\ {\text{*(estimated 1}}^{\text{st}}\text{trimester BDCM exposure in μg/L)} \\ =(362/644)*\text{chloroform}+(362/609)*\text{bromoform}+(362/691) \\ *\text{DBCM}+(362/362)*\text{BDCM} \end{gathered}$$

Statistical analysis

We performed statistical analysis using SAS software (version 9.4; SAS Institute, Inc., Cary, NC). We categorized estimated DBP exposures into tertiles, quartiles, or quintiles based on the exposure distribution among the controls. For less prevalent DBP species with a large proportion of participants assigned values of zero, we dichotomized exposures at >0 μg/L for bromoform, MCAA, MBAA, and DBAA, and dichotomized DBCM at the 75^th percentile to maximize available exposure contrasts. We used categorical exposure metrics in order to evaluate the possibility of non-linear relationships and, to the degree possible, examine heterogeneity across sub-groups based on stratified analyses by disinfection type, with those on untreated systems serving as the referent. We also examined models stratified by infant’s sex. Given that DBP exposure data are generally not normally distributed, we used Spearman correlation coefficients to calculate correlations between DBPs. We estimated adjusted odds ratios (aORs) and 95% confidence intervals (CIs) for each DBP metric using conditional logistic regression, which adjusts for seasonality by stratifying on week of conception (i.e., our matching criterion). Births in the lowest DBP exposure categories served as the referents in logistic regression analyses. We selected individual- and area-level covariates as potential confounders for inclusion in regression models based on a priori knowledge of the source population and available literature. We assessed relationships between variables based on causal structure assumptions using a directed acyclic graph (Supplementary Figure 1). We did not examine gestational age and birth weight as confounders as they might be influenced both by DBP exposures and by the presence of a birth defect [42], thus controlling for these factors could introduce collider stratification bias [43]. We adjusted regression models for maternal education (less than high school; high school graduate or GED; some college, associates, or technical degree; college degree), 2000 US Census tract income (continuous), number of prenatal care visits ( < 9; 9–11; 12; 13–15; >15), prenatal care payment source (private health insurance; Medicaid or CommonHealth; other including Medicare and Healthy Start), maternal smoking (yes; no), and maternal age ( < 20; 20- < 35; ≥35). We further adjusted for DBP co-exposures in an attempt to better isolate independent associations for DBP groups, as follows: models adjusted for HAA5 (THM4, THMBr, chloroform, BDCM, DBCM, bromoform), THM4 (HAA9, HAA5, TCAA, DCAA, MCAA, TBAA, DBAA, MBAA, BCAA, BDCAA, CDBAA), DBPCl (DBPBr), DBPBr (DBPCl), THM4 and HAACl (HAABr), or THM4 and HAABr (HAACl).

RESULTS

Study population characteristics

There were a total of 210 OGD cases identified among all reported births from 2000–2004 in the source population. As shown in Table 1, cases and controls were similar across many maternal and sociodemographic characteristics, though cases were more likely to be male.

Table 1.

Study characteristics of obstructive genital defect cases and controls.

Characteristic	Study population	Cases, n (%)	Controls, n (%)
Total births	2310	210	2100
Infant sex
Male	1212	139 (66)	1073 (51)
Female	1098	71 (34)	1027 (49)
Maternal age (years)
≤20	191	15 (7)	176 (8)
>20-< 35	1650	145 (69)	1505 (72)
≥35	469	50 (24)	419 (20)
Maternal race
White	1544	150 (71)	1394 (66)
African American	221	20 (10)	201 (10)
Asian American	196	9 (4)	187 (9)
Others	347	30 (14)	317 (15)
Maternal education
Below high school graduate/ GED	260	19 (9)	241 (11)
High school graduate/GED	609	43 (20)	566 (30)
Some college or associates/technical degree	520	59 (28)	461 (22)
College or higher	921	89 (42)	832 (40)
Marital status
Married	1644	150 (71)	1494 (71)
Unmarried	666	60 (29)	606 (29)
Number of previous births
0	1056	94 (45)	962 (46)
1	796	75 (36)	721 (34)
≥2	455	41 (20)	414 (20)
missing	3	0 (0)	3 (0)
Maternal weight gain during pregnancy (lb)
<0	26	4(2)	22 (1)
0–25	862	84 (40)	778 (37)
25–50	1323	116 (55)	1207 (57)
>50	83	6(3)	89 (4)
Missing	10	0 (0)	10 (0)
Number of cigarettes mother smoked per day during pregnancy
0	2147	191 (91)	1956 (93)
1–5	63	12 (6)	51 (2)
6–10	58	4(2)	54(3)
>10	42	3 (1)	39 (2)
Number of prenatal care visits
<9	234	17 (8)	217 (10)
9–11	613	53 (25)	560 (27)
12	616	55 (26)	561 (27)
13–15	625	58 (27)	567 (27)
>15	217	27 (13)	190 (9)
Missing	5	0 (0)	5 (0)
Prenatal care adequacy (Kotelchuck Index)
No prenatal care	8	0 (0)	8 (0)
Inadequate	190	15 (7)	175 (8)
Intermediate	185	12 (6)	173 (8)
Adequate	1018	83 (40)	935 (45)
Adequate plus	909	100 (48)	809 (39)
Prenatal care source of payment
Public	600	51 (24)	549 (26)
Private	1539	149 (71)	1390 (66)
Other	170	10 (5)	160 (8)
Missing	1	0 (0)	1 (0)
Median household income (based onyear 2000 US Census tracts)
$9,751–34,412	553	45 (21)	508 (24)
>$34,412–47,034	565	57 (27)	508 (24)
>$47,034–60,472	598	55 (26)	543 (26)
>$60,472–157,624	594	53 (25)	541 (26)

Column percentages rounded to nearest whole and may not sum to 100.

DBP ranges and correlations

Among the regulated DBPs, median and interquartile ranges in μg/Lfor the nine predominant metrics were as follows: DBP9 (70.0; 42.2–93.6), THM4 (45.0; 28.1–62.7), THMBr (6.9; 4.7–10.4), BDCM (6.2; 4.3–8.7), DBCM (0.6; 0.0–1.8), HAA5 (22.0; 10.4–31.1), TCAA (10.9; 4.8–16.3), and DCAA (10.4; 4.7–13.7). HAA4 (the sum of predicted brominated HAAs) had a median of 3.3 and interquartile range of 2.7–4.8 (Table 2). We detected strong (r ≥ 0.9) Spearman correlation coefficients between the following: DBP9 with THM4, TCM, and HAA5; DBPBr with THMBr, BDCM, and HAABr; THM4 with TCM; HAA4 with HAABr and BDCAA; HAA5 with HAA9; and CDBAA with TBAA. Among individual chlorinated species, the highest correlations were between TCAA and DCAA (r = 0.79), TCM and TCAA (r = 0.75), and TCM and DCAA (r = 0.74). Among individual brominated species, the highest correlations were the following: BDCM with BDCAA (r = 0.78) and DBCM (r = 0.58); and BCAA with BDCAA (r = 0.67) and CDBAA (r = 0.66) (Supplementary Figure 2). When stratifying by treatment type, chloraminated systems generally had higher levels of chloroform and HAACl, while chlorinated systems generally had higher levels of THMBr and HAABr (Supplementary Table 2). Comparing RPF-weighted DBP mixture metrics to their unweighted counterparts, we observed the following Spearman correlations: r = 0.99 for DBP9, r = 0.97 for HAA3, r = 0.87 for HAA5, r = 0.64 for HAA9, r = 0.98 for TCAA + DCAA, r = 0.99 for THM4, r= 0.98 for THM4 + TCAA, r = 0.99 for THMBr.

Table 2.

First-trimester disinfection byproduct (DBP) (μg/L) exposure estimates for the study population.

DBP metric	n	missing	Mean	25th %	50th %	75th %	90th %	95th %	Max
Chloroform	2308	2	34.49	14.80	36.14	52.57	64.11	67.93	97.03
BDCM	2308	2	6.83	4.33	6.17	8.71	13.27	16.43	46.37
DBCM	2308	2	1.36	0.00	0.58	1.82	4.24	5.97	14.65
Bromoform	2308	2	0.11	0.00	0.00	0.00	0.28	0.86	6.83
THM4	2310	0	42.80	28.14	44.96	62.65	73.28	77.62	121.40
THMBr	2308	2	8.31	4.72	6.91	10.37	18.07	21.27	51.840
TCAA	2284	26	11.50	4.81	10.86	16.33	22.71	26.73	91.14
DCAA	2284	26	9.85	4.73	10.36	13.65	18.75	21.81	38.25
MCAA	2284	26	0.77	0.00	0.01	0.85	1.70	2.47	32.21
DBAA	2284	26	0.20	0.00	0.00	0.00	0.51	1.13	23.58
MBAA	2284	26	0.02	0.00	0.00	0.00	0.00	0.00	4.75
BCAA	2269	41	1.46	0.69	1.14	1.70	2.83	3.99	29.76
BDCAA	2269	41	2.18	1.06	1.78	2.52	4.57	5.97	55.85
CDBAA	2269	41	0.23	0.04	0.08	0.15	0.67	0.99	30.59
TBAA	2269	41	0.01	0.00	0.00	0.00	0.03	0.07	5.41
HAA9	2269	41	26.34	16.18	25.95	35.20	48.60	54.96	179.50
HAA5	2310	0	22.32	10.42	22.04	31.13	43.38	49.53	120.59
HAA4	2269	41	3.88	2.01	3.14	4.31	8.42	10.88	121.62
HAABr	2269	41	4.09	2.08	3.23	4.44	9.10	11.92	122.54
HAACl	2284	26	22.13	9.77	21.98	30.97	43.25	49.53	120.59
DBPBr	2269	41	12.34	7.36	10.16	14.43	27.29	31.88	174.39
DBPCl	2282	28	56.50	26.33	60.66	84.27	99.10	110.29	208.46
DBP9	2310	0	65.13	42.20	70.02	93.56	110.40	119.71	215.42
DBP13	2269	41	69.13	47.74	73.72	98.33	115.37	127.31	256.31

% percentile, DBP9 sum of chloroform, bromodichloromethane (BDCM), dibromochloromethane (DBCM), bromoform, monochloroacetic acid (MCAA), dichloroacetic acid (DCAA), trichloroacetic acid (TCAA), monobromoacetic acid (MBAA), dibromoacetic acid (DBAA), bromochloroacetic acid (BCAA), bromodichloroacetic acid (BDCAA), chlorodibromoacetic acid (CDBAA), and tribromoacetic acid (TBAA); HAA5, sum of MCAA, DCAA, TCAA, MBAA, and DBAA; HAA9, sum of all nine HAAs; THM4, sum of chloroform, BDCM, DBCM, and bromoform; THMBr, sum of BDCM, DBCM, and bromoform; HAA4, sum of BCAA, BDCAA, CDBAA, and TBAA; HAABr, sum of DBAA, MBAA, BCAA, BDCAA, CDBAA, and TBAA; DBPBr, sum of HAABr and THMBr.

Logistic regression

We detected consistently elevated ORs for the associations of OGDs with several brominated DBP metrics, including monotonic exposure-response relationships for tertiles of THMBr and BDCM. For example, we observed elevated aORs for tertiles 3 vs. 1 of THMBr (1.74; 95% CI: 1.15, 2.64) and BDCM (1.75; 95% CI: 1.15, 2.65), for any vs. zero DBCM (1.71; 95% CI: 1.15, 2.54) and DBAA (1.23; 95% CI: 0.88, 1.72), and for the quartile 4 vs. bottom 75% of BDCAA (1.56; 95% CI: 0.97, 2.51), CDBAA (1.97; 95% CI: 1.23, 3.15), and TBAA (1.90; 95% CI: 1.20, 3.03). We also detected elevated aORs for quintiles 5 vs. 1 of THM4 (1.54; 95% CI: 0.77, 3.09) and chloroform (1.52; 95% CI: 0.76, 3.06) (Table 3). For chloroform, adjustment for HAA5 slightly increased the estimates. For chlorinated HAAs, adjustment for THM4 generally reduced ORs towards the null. For brominated DBPs, adjustment of THM models for HAA5 and of HAA models for THM4 generally did not materially affect ORs. Specific concentration ranges for each DBP category are shown in Table 3.

Table 3.

Adjusted odds ratios (aORs) between disinfection by-product (DBP) exposures and obstructive genital defects (OGD).

	Cases/ controlsa	Adjustedb		Further adjusted for DBP co-exposuresc
DBP metric (μg/L)		OR	95% CI	OR	95% CI
DBP9
0–28.15	36/420	1.00	Referent	--	--
>28.15–59.35	47/420	1.39	0.87, 2.25	--	--
>59.35–80.16	45/420	1.32	0.81, 2.14	--	--
>80.16–99.08	35/421	1.02	0.62, 1.69	--	--
>99.08–215.42	47/419	1.39	0.86, 2.25	--	--
THM4
0–18.81	36/420	1.00	Referent	1.00	Referent
>18.81–38.02	43/420	1.28	0.79, 2.08	1.32	0.77, 2.27
>38.02–51.65	46/420	1.36	0.84, 2.20	1.43	0.79, 2.58
>51.65–66.67	37/420	1.11	0.67, 1.83	1.18	0.61, 2.28
>66.67–121.40	48/420	1.43	0.88, 2.33	1.54	0.77, 3.09
THMBr
0–5.53	53/699	1.00	Referent	1.00	Referent
>5.53–8.64	74/700	1.52	1.01, 2.28	1.53	1.00, 2.34
>8.64–51.84	83/699	1.74	1.16, 2.59	1.74	1.15, 2.64
Chloroform
0–10.67	36/420	1.00	Referent	1.00	Referent
>10.67–29.66	44/421	1.31	0.81, 2.14	1.35	0.79, 2.30
>29.66–42.32	45/417	1.34	0.82, 2.18	1.40	0.76, 2.59
>42.32–55.98	37/422	1.11	0.67, 1.85	1.18	0.59, 2.33
>55.98–97.03	48/418	1.44	0.89, 2.33	1.52	0.76, 3.06
BDCM
0–5.05	54/701	1.00	Referent	1.00	Referent
>5.05–7.61	71/698	1.41	0.95, 2.11	1.46	0.94, 2.25
>7.61–46.37	85/699	1.70	1.16, 2.50	1.75	1.15, 2.65
DBCM
0	49/647	1.00	Referent	1.00	Referent
>0–14.65	161/1451	1.72	1.16, 2.54	1.71	1.15, 2.54
Bromoform
0	174/1754	1.00	Referent	1.00	Referent
>0–6.83	36/344	1.04	0.70, 1.55	1.08	0.72, 1.63
HAA9
0–10.10	36/413	1.00	Referent	1.00	Referent
>10.10–23.28	42/413	1.23	0.76, 1.99	1.05	0.58, 1.90
>23.28–29.40	48/413	1.43	0.88, 2.31	1.14	0.59, 2.24
>29.40–38.41	38/413	1.11	0.68, 1.83	0.86	0.41, 1.81
>38.41–179.50	40/413	1.18	0.72, 1.96	0.89	0.40, 1.97
HAA5
0–7.07	37/420	1.00	Referent	1.00	Referent
>7.07–18.84	46/420	1.31	0.82, 2.10	1.10	0.60, 2.02
>18.84–25.34	46/420	1.33	0.82, 2.14	1.06	0.53, 2.10
>25.34–33.86	40/420	1.17	0.72, 1.91	0.90	0.42, 1.92
>33.86–120.59	41/420	1.14	0.70, 1.87	0.86	0.38, 1.91
HAA4
0–2.52	52/689	1.00	Referent	1.00	Referent
>2.52–3.72	89/687	1.90	1.29, 2.78	2.11	1.35, 3.29
>3.72–121.62	63/689	1.26	0.83, 1.90	1.37	0.87, 2.15
TCAA
0–4.65	44/519	1.00	Referent	1.00	Referent
>4.65–10.86	57/519	1.39	0.90, 2.14	1.28	0.73, 2.24
>10.86–16.41	60/520	1.47	0.95, 2.25	1.30	0.67, 2.53
>16.41–91.14	45/520	1.06	0.67, 1.67	0.93	0.45, 1.92
DCAA
0–4.76	52/520	1.00	Referent	1.00	Referent
>4.76–10.38	55/520	1.10	0.73, 1.68	0.88	0.53, 1.44
>10.38–13.62	45/519	0.87	0.56, 1.35	0.63	0.35, 1.11
>13.62–38.25	54/519	1.12	0.73, 1.72	0.77	0.42, 1.41
MCAA
0	103/1033	1.00	Referent	1.00	Referent
>0–32.21	103/1045	0.99	0.74, 1.33	0.90	0.65, 1.25
DBAA
0	151/1597	1.00	Referent	1.00	Referent
>0–23.58	55/481	1.23	0.88, 1.72	1.23	0.88, 1.72
MBAA
0	201/2029	1.00	Referent	1.00	Referent
>0–4.75	5/49	0.95	0.37, 2.46	0.95	0.37, 2.45
BCAA 25th
0–0.66	43/516	1.00	Referent	1.00	Referent
>0.66–29.76	161/1549	1.31	0.91, 1.89	1.33	0.86, 2.06
BDCAA 25th
0–1.05	41/516	1.00	Referent	1.00	Referent
>1.05–55.85	163/1549	1.42	0.98, 2.05	1.56	0.97, 2.51
CDBAA 25th
0–0.04	36/517	1.00	Referent	1.00	Referent
>0.04–30.59	168/1548	1.74	1.17, 2.59	1.97	1.23, 3.15
TBAA (25th)
0.0003	36/518	1.00	Referent	1.00	Referent
>0.0003–5.41	168/1547	1.71	1.15, 2.54	1.90	1.20, 3.03
HAABr
0–2.06	44/516	1.00	Referent	1.00	Referent
>2.06–3.23	55/517	1.36	0.88, 2.09	1.31	0.79, 2.15
>3.23–4.41	43/514	1.03	0.64, 1.64	0.98	0.57, 1.71
>4.41–122.54	62/518	1.48	0.96, 2.27	1.43	0.89, 2.31
DBPBr
0–6.12	32/413	1.00	Referent	1.00	Referent
>6.12–9.21	35/413	1.12	0.67, 1.89	1.15	0.62, 2.12
>9.21–11.47	47/412	1.65	1.00, 2.70	1.68	0.94, 3.01
>11.47–17.02	38/415	1.25	0.75, 2.10	1.28	0.70, 2.33
>17.02–174.39	52/412	1.77	1.10, 2.87	1.80	1.05, 3.10
DBPCl
0–17.86	35/416	1.00	Referent	1.00	Referent
>17.86–50.18	52/416	1.57	0.98, 2.52	1.69	0.72, 3.94
>50.18–71.28	38/415	1.15	0.69, 1.92	1.23	0.52, 2.94
>71.28–89.60	36/414	1.11	0.67, 1.85	1.21	0.50, 2.91
>89.60–208.46	45/415	1.37	0.84, 2.24	1.46	0.61, 3.49
HAACl
0–6.53	38/416	1.00	Referent	1.00	Referent
>6.53–18.83	45/414	1.22	0.76, 1.96	0.94	0.51, 1.71
>18.83–25.19	44/417	1.21	0.75, 1.96	0.84	0.42, 1.69
>25.19–33.76	41/415	1.17	0.72, 1.90	0.76	0.36, 1.64
>33.76-max	38/416	1.03	0.62, 1.70	0.65	0.29, 1.46
DBP13
0–22.44	29/344	1.00	Referent	--	--
>22.44–57.30	34/346	1.23	0.72, 2.10	--	--
>57.30–73.68	37/342	1.33	0.77, 2.30	--	--
>73.68–89.68	34/343	1.29	0.75, 2.23	--	--
>89.68–105.30	34/346	1.25	0.73, 2.15	--	--
>105.30–256.31	36/344	1.30	0.76, 2.24	--	--
HAA3
0–6.58	37/416	1.00	Referent	1.00	Referent
>6.58–18.43	44/414	1.22	0.76, 1.97	0.99	0.54, 1.81
>18.43–24.83	46/415	1.31	0.81, 2.12	0.98	0.48, 1.97
>24.83–32.63	41/416	1.18	0.73, 1.92	0.84	0.39, 1.81
>32.63–120.06	38/411	1.07	0.65, 1.78	0.74	0.32, 1.68
TCAA + DCAA
0–6.23	38/416	1.00	Referent	1.00	Referent
>6.23–18.31	44/415	1.19	0.74, 1.91	0.91	0.50, 1.66
>18.31–24.78	46/413	1.28	0.79, 2.07	0.89	0.44, 1.78
>24.78–32.51	41/417	1.15	0.71, 1.86	0.75	0.35, 1.60
>32.51–120.06	37/411	1.01	0.61, 1.67	0.63	0.28, 1.43
THM4 + TCAA
0–23.21	37/416	1.00	Referent	1.00	Referent
>23.21–48.67	43/416	1.23	0.76, 1.99	1.26	0.76, 2.08
>48.67–66.84	47/415	1.38	0.85, 2.23	1.35	0.82, 2.22
>66.84–83.96	31/416	0.89	0.53, 1.49	0.88	0.52, 1.51
>83.96–185.97	48/415	1.41	0.87, 2.28	1.29	0.77, 2.16

^aCase and control distribution prior to modeling.

^bAdjusted for maternal education (less than high school; high school graduate or GED; some college, associates, or technical degree; college degree), 2000 US Census tract income (continuous), number of prenatal care visits ( < 9; 9–11; 12; 13–15; >15), prenatal care payment source (private health insurance; Medicaid or CommonHealth; other including Medicare and Healthy Start), maternal smoking (yes;no), maternal age ( < 20; 20- < 35; ≥35).

^cFurther adjusted for DBP co-exposures: adjusted for HAA5 (THM4, THMBr, chloroform, BDCM, DBCM, bromoform), THM4 (HAA9, HAA5, TCAA, DCAA, MCAA, TBAA, DBAA, MBAA, BCAA, BDCAA, CDBAA, HAA3, TCAA + DCAA), DBPCl (DBPBr), DBPBr (DBPCl), THM4 and HAACl (HAABr), THM4 and HAABr (HAACl), or HAABr only (THM4 + TCAA).

Stratified logistic regression models

In models stratified by infant’s sex as reported on birth certificates (male/female), ORs were more consistently elevated across exposure categories for males than females for DBP13, DBPBr, THM4, THMBr, HAA9, and HAABr (Supplementary Table 3). Some individual ORs were higher among females than males, e.g., for the highest exposure categories of the summed brominated DBP metrics HAABr (2.45 for females, 1.42 for males), THMBr (3.16 for females, 1.40 for males) and DBPBr (3.15 for females, 1.61 for males). In models stratified by disinfection treatment type (chlorinated vs. untreated, chloraminated vs. untreated, and other disinfection types vs. untreated), we observed exposure-response patterns and the highest aORs for chloraminated systems compared to untreated systems for THM4 (adjusted for HAA5), THMBr (adjusted for HAA5), HAABr (adjusted for HAACl and THM4), and DBPBr (adjusted for DBPCl). Across all three stratified treatment comparisons, aORs were consistently elevated for models comparing births in the highest exposure tertile to those with untreated water for THMBr, HAABr, and DBPBr (Supplementary Table 4).

Logistic regression with RPF-BMD50 weighting

When comparing results of models using RPF-weighted exposure mixtures created from BMD50 values (see Supplementary Table 5 for BMD modeling) extracted from toxicological studies to results of unweighted models, aORs were similar for DBP9, HAA9, HAA5, THM4, and THMBr (average changes across quantiles of −1.8%, −4.9%, −0.7%, −3.8%, and +1.8%, respectively) and higher for HAA3, TCAA + DCAA, and THM4 + TCAA (average changes across quintiles of +58%, +66%, and +29%, respectively) (Table 4). Most of the upper quantile changes were coherent with these averages, although the aOR for THM4 quintiles 5 vs. 1 decreased by 16% comparing the unweighted (OR = 1.54) to weighted (OR = 1.29) models. As with the unweighted models, the most consistently elevated aORs were for THMBr tertile 3 vs. 1 (aOR=1.79; 95% CI: 1.18, 2.71 for the weighted model). Similarities or differences between weighted and unweighted results can be explained by the mixture components’ relative potencies (Supplementary Table 5), the strength of their individual effects (Table 3), and their concentrations (Table 2). For example, the weighted and unweighted results for THMBr were similar, which is expected given that the aORs for the highest categories among the two most predominant brominated THMs (BDCM [the index chemical] and DBCM) were identical (OR = 1.7). Comparing the unweighted to weighted HAA9 results showed an average change of 9%. For the OR comparing quartiles 5 vs. 1, the weighted value was 9% higher, while the other three ORs were each approximately 9–10% lower. This likely resulted from a null result in the index chemical (MBAA) as well as for TCAA and DCAA, the two most predominant HAAs. In contrast, the largest component aORs for the nine HAAs were for TBAA and CDBAA, the eight and sixth most potent, and fifth and ninth lowest median concentration HAAs examined here, respectively.

Table 4.

Adjusted odds ratios (aORs) between relative potency factor weighted summary disinfection by-product (DBP) exposures and obstructive genital defects.

	Weighted models: cases/controls	Weighted models: aOR (95% CI)a	Unweighted models: cases/controls	Unweighted models: aOR (95% CI)a
DBP9 (BDCM>TBM>TCM>DBCM>HAA5)b
Quintile 1	36/420	1.00 (ref)	36/420	1.00 (Referent)
Quintile 2	47/419	1.38 (0.86, 2.22)	47/420	1.39 (0.87, 2.25)
Quintile 3	45/419	1.24 (0.76, 2.01)	45/420	1.32 (0.81, 2.14)
Quintile 4	35/419	1.06 (0.64, 1.73)	35/421	1.02 (0.62, 1.69)
Quintile 5	47/417	1.33 (0.82, 2.15)	47/419	1.39 (0.86, 2.25)
HAA9 (MBAA>BCAA>MCAA>DBAA>BDCAA>CDBAA>TCAA>TBAA>DCAA)c
Quintile 1	36/413	1.00 (ref)	36/413	1.00 (Referent)
Quintile 2	42/413	0.94 (0.56, 1.57)	42/413	1.05 (0.58, 1.90)
Quintile 3	48/413	1.05 (0.63, 1.75)	48/413	1.15 (0.59, 2.24)
Quintile 4	38/413	0.78 (0.45, 1.35)	38/413	0.86 (0.41, 1.81)
Quintile 5	40/413	0.97 (0.57, 1.64)	40/413	0.89 (0.40, 1.97)
HAA5 (MBAA>MCAA>DBAA>TCAA>DCAA)c
Quintile 1	37/420	1.00 (ref)	37/420	1.00 (Referent)
Quintile 2	46/420	1.35 (0.75, 2.42)	46/420	1.10 (0.60, 2.02)
Quintile 3	46/420	0.73 (0.37, 1.44)	46/420	1.06 (0.53, 2.10)
Quintile 4	40/420	0.94 (0.49, 1.81)	40/420	0.90 (0.42, 1.92)
Quintile 5	41/420	0.87 (0.43, 1.76)	41/420	0.86 (0.38, 1.91)
HAA3 (DBAA>TCAA>DCAA)c
Quintile 1	33/416	1.00 (ref)	37/416	1.00 (Referent)
Quintile 2	41/415	1.30 (0.71, 2.37)	44/414	0.99 (0.54, 1.81)
Quintile 3	51/414	1.64 (0.85, 3.18)	46/415	0.98 (0.48, 1.97)
Quintile 4	41/415	1.32 (0.63, 2.76)	41/416	0.84 (0.39, 1.81)
Quintile 5	40/412	1.31 (0.62, 2.80)	38/411	0.74 (0.32, 1.68)
TCAA+DCAA (TCAA>DCAA)c
Quintile 1	33/415	1.00 (ref)	38/416	1.00 (Referent)
Quintile 2	47/416	1.42 (0.75, 2.68)	44/415	0.91 (0.50, 1.66)
Quintile 3	48/413	1.48 (0.70, 3.14)	46/413	0.89 (0.44, 1.78)
Quintile 4	38/417	1.11 (0.49, 2.55)	41/417	0.75 (0.35, 1.60)
Quintile 5	40/411	1.21 (0.51, 2.89)	37/411	0.63 (0.28, 1.43)
THM4 (BDCM>TBM>TCM>DBCM)b
Quintile 1	36/419	1.00 (ref)	36/420	1.00 (Referent)
Quintile 2	45/419	1.35 (0.79, 2.31)	43/420	1.32 (0.77, 2.27)
Quintile 3	44/417	1.33 (0.73, 2.42)	46/420	1.43 (0.79, 2.58)
Quintile 4	41/419	1.25 (0.66, 2.34)	37/420	1.18 (0.61, 2.28)
Quintile 5	44/418	1.29 (0.65, 2.56)	48/420	1.54 (0.77, 3.09)
THM4+TCAA (BDCM>TBM>TCM>DBCM>TCAA)b
Quintile 1	37/416	1.00 (ref)	37/416	1.00 (Referent)
Quintile 2	43/415	1.66 (0.88, 3.13)	43/416	1.26 (0.76, 2.08)
Quintile 3	47/415	1.50 (0.71, 3.17)	47/415	1.35 (0.82, 2.22)
Quintile 4	31/413	1.45 (0.63, 3.32)	31/416	0.88 (0.52, 1.51)
Quintile 5	48/413	1.39 (0.58, 3.31)	48/415	1.29 (0.77, 2.16)
THMBr (BDCM>TBM>DBCM)b
Tertile 1	53/697	1.00 (ref)	53/699	1.00 (Referent)
Tertile 2	74/696	1.54 (1.00, 2.35)	74/700	1.53 (1.00, 2.34)
Tertile 3	83/699	1.79 (1.18, 2.71)	83/699	1.74 (1.15, 2.64)

^aAll models adjusted for maternal education (less than high school; high school graduate or GED; some college, associates, or technical degree; college degree), 2000 US Census tract income (continuous), number of prenatal care visits ( < 9; 9–11; 12; 13–15; >15), prenatal care payment source (private health insurance; Medicaid or CommonHealth; other including Medicare and Healthy Start), maternal smoking (yes; no), maternal age ( < 20; 20- < 35; ≥35). Further adjusted for DBP co-exposures: THM4 and THMBr models adjusted for HAA5; HAA9, HAA5, HAA3, and TCAA + DCAA models adjusted for THM4; THM4 + TCAA model adjusted for HAABr.

^bRelative potency factors weights based on BMD50 values from four studies: (1) Smith, et al., 1989; (2) Smith, et al., 1992; (3) Narotsky et al., 2011; (4) Narotsky et al., Submitted.

^cRelative potency factors weights based on BMD50 values from two studies: (1) Hunter, et al., 2006.; (2) Hunter et al., 1996.

DISCUSSION

In unweighted and RPF-weighted analyses, we detected elevated risks of OGDs with higher exposures to most brominated DBPs examined, including for several brominated HAAs previously unexamined in epidemiologic studies. Toxicological research supports the concept that brominated DBPs are more potent teratogens than their chlorinated analogs, including brominated HAAs [2, 44]. Although teratogenicity data far more limited, existing research generally agrees that brominated DBPs are more mutagenic, cytotoxic, and genotoxic than their chlorinated analogs [45–47]. Our results add novel and unique data to the epidemiologic literature of urogenital birth defects and DBPs, including for several DBP mixture measures. Our RPF-weighted application provides a first attempt in the epidemiologic literature on DBP exposures to use toxicologically-informed mixture metrics.

A 2008 meta-analysis [8] reported summary odds ratio estimates for “high vs. low” chlorination by-product exposures of 1.33 (95% CI: 0.92, 1.92) for urinary tract defects based on four studies [11, 15, 17, 48], and 1.07 (95% CI: 0.87, 1.30) for obstructive urinary defects based on three studies [15–17]. Of the individual studies included, two based exposure estimates on THM values (THM4, THMBr, or bromoform) [16, 17] while the others used more indirect proxies such as a binary variable for whether the public water supply was chlorinated. Using data from Taiwan with a low exposure range, Hwang et al. [16] reported an aOR for obstructive urinary tract defects for exposures to ≥20 vs 0–4 μg/L of THM4 of 1.44 (95% CI: 0.66, 3.14). Contrary to what we saw for some different THM metrics, Nieuwenhuijsen et al. [17]. did not report elevated aORs for obstructive urinary tract defects for THM4 (aOR ≈ 0.98 > 60 vs. <30 μg/L), THMBr (aOR ≈ 1.1 for >20 vs. <10 μg/L), or bromoform (aOR = 1.2 for >4 vs. <2 μg/L). This compares to an aOR for bromoform in our analysis of 1.08 (95% CI: 0.72, 1.63) for >0 vs 0 μg/L, the lowest aOR for any individual brominated species we examined. Their study population was from England and Wales and had generally higher exposure levels than in our study based in Massachusetts.

Two additional case-control studies performed after this meta-analysis reported elevated adjusted ORs for urinary tract defects. One study reported an aOR of 3.01 (95% CI: 1.11, 8.16) for urogenital anomalies with high (21.9 ug/L; 18 cases) vs low (1.3 ug/L; 5 cases) first trimester THM4 exposures; associations for chloroform, BDCM, and DBCM did not exhibit monotonic exposure-response relationships. [9]. Another study with very low THM levels in chlorine dioxide-treated waters by Righi et al. [10]. reported an adjusted OR of 2.88 (95% CI: 1.09, 7.63) for obstructive urinary defects with exposure to high ( > 200 ug/L) vs. low (=<200 ug/L) chlorate, but not chlorite ( > 700 vs =<200 ug/L; aOR=1.58; 95% CI: 0.65, 3.83). They also observed null results for THM4 exposures, but their very low THM4 levels likely precluded sufficient contrasts and sample size to examine associations with obstructive urinary defects [10]. This study used ICD 753 to identify urinary tract defects, further grouped into renal defects (753.0, 753.1, 753.3) and obstructive urinary defects (753.2, 753.4, 753.5–753.8). In general, examination of specific birth defects (e.g., OGDs alone) rather than grouped outcomes (e.g., grouping OGD with hypospadias or other urinary tract defects) can increase study sensitivity by representing a more etiologically homogeneous outcome group, though at the cost of decreased sample size and statistical power. We are unaware of any epidemiological studies of urinary tract or genital defects to examine HAA exposures; we observed elevated aORs for OGD with all brominated HAA metrics with the exception of MBAA (OR Range: 1.23–1.97 for the highest quantiles in multi-DBP adjusted models). While we focus our analyses on a specific birth defect outcome, we acknowledge that study sensitivity and statistical power may be particularly limited for the analyses stratified by sex and treatment type (see Supplementary Tables 3 and 4).

Our study benefited from the availability of a large number of cases of a specific birth defect type, along with data on nine measured individual DBPs and four additional estimated DBPs. Having these data allowed us to examine thirteen individual DBP-outcome relationships with better specificity, as well as several mixture combinations. Another study strength was the ability to differentiate and compare a range of exposure levels given wide exposure contrasts for most measured DBPs and the large number of study participants on untreated drinking water systems. Availability of data on these two DBP classes also allowed us to adjust for co-occurring HAAs and THMs in an attempt to improve specificity and reduce potential confounding by correlated DBP classes. THMs and HAAs often correlate in chlorinated water systems. However, correlations between individual DBPs can vary across public water systems, making this potential confounding hard to control and quantify. Although collinearity and co-exposure amplification bias can be analytical challenges, we assume that the likelihood of this type of bias in our multipollutant models is low given that they share a common source (e.g., water treatment) and are, therefore, not likely to have different confounding factors [49].

Although confounding by correlated co-exposures can be a concern for epidemiologic studies focused on assessing health risks of individual exposures, researchers should also address the potential for interactive effects between components within complex exposure mixtures. This is especially relevant for epidemiologic studies examining health effects of DBPs, as DBP exposures ubiquitously occur as complex mixtures. To this end, our incorporation of toxicology-based RPF values into our epidemiologic analyses is novel in the field of DBP research. However, it is possible that our use of data from rat gavage studies of full-litter resorption and eye defects and whole embryo culture studies may decrease study sensitivity of our weighted analysis if these endpoints are not coherent with the OGD outcomes we examined. To our knowledge, toxicological studies of DBPs have not reported OGDs. The overall purpose of developing an RPF-based approach was to produce estimates of exposure mixtures that up-weight more toxicologically relevant components, given that the most prevalent DBPs (i.e., chloroform, TCAA, and DCAA in our data) may not be the most potent based on current developmental toxicity knowledge. This RPF approach is also only useful for mixtures of measured DBPs and does not address the impact on any unmeasured DBPs, especially those that may be uncorrelated with the predominant DBPs that are often measured and studied. A further limitation of our preliminary RPF approach is that few toxicology studies of DBPs examined common endpoints for both THMs and HAAs, so that we could not examine combinations of THMs and HAAs beyond THM4 with TCAA or HAA5. We adjusted for classes of DBPs (i.e., THM4, HAA5, HAACl, HAABr) in models of individual DBPs (Table 3), summed unweighted DBP mixtures (Table 3), and RPF-weighted summed mixtures (Supplementary Table 6), largely resulting in little impact on OR estimates. Finally, our unweighted and weighted summation approach for mixtures did not specifically examine interactions, which could possibly be addressed with other data-based mixtures approaches and larger sample sizes. Nonetheless, it is our opinion that the examination of combined toxicity and various summary mixture groupings within and across volatility sub-groups help advance the field.

Our study had both strengths and limitations. Similar to other study designs that rely on aggregate exposure assessments, exposure misclassification is likely present in our analyses given that we assigned time-weighted area-level estimates to individuals based on their home address and created estimates for the first pregnancy trimester based on quarterly DBP sampling. Our exposure estimates do not capture the full spatial or temporal (seasonal or diurnal) variability of DBP exposures [50, 51], and do not account for differences in individual water use such as recreational swimming or use of bottled drinking water, as such data were unavailable in this population. Further, it is possible that some individuals had lower DBP exposures than we assigned, if they lived within an area served by a public water system but actually drew water from private wells. Overall, the DBP levels used in our exposure assignment are based on measured monitoring data and represent a range of common exposure levels. Within our sample population of 2310, exposure levels were above the US EPA maximum contaminant levels of 80 ppb for THM4 and 60 ppb for HAA5 for 3.3% and 1.3% of participants, respectively. We found evidence for elevated OGD risks for brominated DBPs in particular, which tend to occur at comparatively low levels relative to chlorinated DBPs, and which are not individually regulated.

Assigning exposures based on residential addresses on birth records may result in exposure misclassification if residential mobility occurs during pregnancy, and if significant DBP exposure occurs in settings outside of the home. In a study of neural tube defects and DBPs, stronger associations were reported among births to mothers with confirmed residences at the start of the first trimester than among the overall sample [52]. In a published review of epidemiology studies with residential mobility data, 9%–32% of women moved during pregnancy across the 14 included studies, with the median distance moved <10 km [53]. This review and another both reported that most moves occurred during the second trimester [53, 54], which could add some information bias uncertainty from measurement error in our first trimester exposure estimates if participants in our study population moved across water systems. Our temporally weighted average of multiple DBP measures across sampling periods should better help characterize average exposures during the first trimester, a critical window for OGDs and other birth defects. Therefore, while some misclassification of first trimester DBP exposures is likely present in our study given our indirect exposure data, our categorical approach should generally provide a relative ranking of the exposures experienced by the study population. More frequent DBP sampling data could further enhance study sensitivity to examine peak exposures across more specific developmental windows.

An additional study strength is that our data included many potential confounders through birth records data which allowed us to conduct an extensive confounding analysis. However, reliance on mothers’ self-reported information, especially for smoking and alcohol use, is a potential limitation. Due to validity concerns from the Massachusetts Department of Public Health, which provided the data, we did not use data on alcohol drinking during pregnancy. Maternal alcohol consumption is not a strong established risk factor for OGDs; one study identified a moderate association with renal anomalies [55], and another study did not identify associations with obstructive gastrointestinal defects [56]. Thus, alcohol is considered to be an unlikely confounder in our analysis. In prior papers [42, 57] using a larger sample of birth records from Massachusetts, we reported strong associations between fetal growth measures and cigarette smoking during pregnancy. This evidence and other validated data from cotinine studies [58] indicate that birth records may accurately describe smoking behaviors during pregnancy. Therefore, we believe that the impact of any mismeasurement of maternal smoking is likely minimal in our study. Given the lack of strong evidence for confounders of the DBP-OGD relationship, we limited matching to week of conception as a marker for seasonality to allow more flexibility with regression adjustment sets. Future research could benefit from additional clarification on which OGD risk factors are most probable confounders for adjustment or other approaches, while maximizing exposure contrasts and sample size.

Given that all OGD cases and the controls were randomly selected from the same source population, selection bias is unlikely in our analysis. We did restrict the analysis to public water systems with complete quarterly THM4/HAA5 data, which decreased the sample size and therefore decreased our statistical power. However, it is unlikely that case ascertainment is related to public water systems, and, therefore, would not be expected to result in bias. Some studies of birth defects may be subject to selection bias by conditioning on survival in cases where only birth defects among live births are included (i.e., birth defects among stillbirths or elective terminations are not counted). OGDs are generally neither fatal to fetuses nor a common reason for elective terminations, thus the potential for “livebirth bias” may be less here than in studies examining other types of birth defects [59]. The Massachusetts Birth Defects Monitoring Program tracks incidence of birth defects among livebirths, stillbirths, and other fetal losses including elective abortions. According to their reporting, the occurrence of OGDs among non-livebirths is low. For the most recent (2015–2018) report, the incidence of OGD among stillbirths and other fetal losses was much lower than for musculoskeletal, chromosomal, or syndromic birth defects. In this 2015–2018 report, 1151 OGD cases were detected among livebirths statewide, while 5 and 34 were detected among stillbirths and other fetal losses, respectively [60].

We examined nine HAAs and four THMs and included THM models adjusted for HAA5 and HAA models adjusted for THM4, providing greater specificity than prior studies on OGDs and DBPs. We noted consistently elevated ORs for OGDs with higher exposures to brominated THMs and HAAs in weighted and unweighted exposure models and across various adjustment sets. These results were quite robust across regression models with different adjustment sets and warrant further investigation in other populations. Given that >700 DBPs have been detected in treated water systems, and the evidence for elevated OGD risks with chlorate reported by Righi et al. [10] toxicological and epidemiological research on a wider variety of specific DBPs would be informative. Chlorine dioxide use in Massachusetts and the United States is rare in general, therefore we do not anticipate these types of DBPs to be confounders here.

In summary, we saw stronger associations for brominated DBPs in relation to OGD risk than for the other DBPs we examined. While our analyses of weighted mixture metrics did not show dramatic differences compared to unweighted mixture sums, this was limited to a few summary measures that were based on the most relevant available developmental toxicity studies. Although our consideration of exposures beyond regulated DBPs and using toxicologically informed weighting approaches to assess mixture risks may more accurately reflect population health impacts of widespread DBP exposures, more specific developmental effects data across all key DBPs and individual-level water use data among pregnant mothers may be needed to further articulate key differences.

DATA AVAILABILITY

Data used in this analysis are based on birth records and birth defects registry data and are not publicly available.