A cross-sectional study of brownfields and birth defects
All funding for this study was provided by the US Environmental Protection Agency. The funder had no role in the study design, data collection, analysis, interpretation, or manuscript writing. This research was supported, in part, by an appointment to the Research Participation Program for the US Environmental Protection Agency, Office of Research and Development, administered by the Oak Ridge Associated Universities through an interagency agreement between the US Department of Energy and Environmental Protection Agency. This work was performed as part of normal duties by EPA employees and contractors.
Funding information: US Environmental Protection Agency
Abstract
Background
Brownfields are a multitude of abandoned and disused sites, spanning many former purposes. Brownfields represent a heterogenous yet ubiquitous exposure for many Americans, which may contain hazardous wastes and represent urban blight. Neonates and pregnant individuals are often sensitive to subtle environmental exposures. We evaluate whether residential brownfield exposure is associated with birth defects.
Methods
Using North Carolina birth records from 2003 to 2015, we sampled 753,195 births with 39,495 defects identified. We examined defect groups and 30 distinct phenotypes. Number of brownfields within 2,000 m of the residential address at birth was summed. We utilized mixed effects multivariable logistic regression models adjusted for demographic and environmental covariates available from birth records, 2010 Census, and EPA's Environmental Quality Index to estimate odds ratios (OR) and 95% confidence intervals (CI).
Results
We observed positive associations between cardiovascular and external defect groups (OR [95% CI]: 1.07 [1.02–1.13] and 1.17 [1.01–1.35], respectively) and any brownfield exposure. We also observed positive associations with atrial septal and ventricular septal defects (1.08 [1.01–1.16] and 1.15 [1.03–1.28], respectively), congenital cataracts (1.38 [0.98–1.96]), and an inverse association with gastroschisis (0.74 [0.58–0.94]). Effect estimates for several additional defects were positive, though we observed null associations for most group and individual defects. Additional analyses indicated an exposure–response relationship for several defects across levels of brownfield exposure.
Conclusions
Our results indicate that residential proximity to brownfields is associated with birth defects, especially cardiovascular and external defects. In-depth analyses of individual defects and specific contaminants or brownfield sites may reveal additional novel associations.
1 INTRODUCTION
Birth defects remain the leading cause of infant mortality and major contributors to infant morbidity and disability in the United States (Ely & Driscoll, 2020). Birth defects impose serious costs on health networks and communities in which they occur (Waitzman, Romano, & Scheffler, 1994). Much of the etiology of birth defects remains unknown; yet often environmental in utero exposures are associated with increased risk of various defects (Hansen, Barnett, Jalaludin, & Morgan, 2009; L. Wang et al., 2019; Yoon et al., 2001). One such exposure is air pollution, with many studies finding positive associations (Girguis et al., 2016; Ritz et al., 2002; Vinikoor-Imler, Davis, Meyer, & Luben, 2013; L. Wang et al., 2019). Water contamination by agricultural and industrial runoff has also been linked to increased risk for various birth defects and outcomes (Bove et al., 1995; Ward et al., 2018; Winchester, Huskins, Ying, & p., 2009). Further, birth defects have been linked to environmental exposures associated with hazardous materials like heavy metals (Al-Hadithi, Al-Diwan, Saleh, & Shabila, 2012; Sanders et al., 2014; Vinceti et al., 2001). Researchers have observed mixed associations with exposure to hazardous waste sites, such as Superfund sites and other large-scale contaminated sites, with some studies reporting positive associations and others null associations (Budnick, Sokal, Falk, Logue, & Fox, 1984; Persico, Figlio, & Roth, 2020; Suarez et al., 2007). Superfund sites are polluted sites across the United States that require long-term responses to clean up and were designated under the 1980 Comprehensive Environmental Response, Compensation, and Liability Act (EPA, U. S, 2022). Akin to Superfund sites, brownfields are ubiquitous sites in the United States that often contain hazardous materials but have not previously been examined in respect to birth defects.
The U.S. Environmental Protection Agency (EPA) defines a brownfield as any “…real property, the expansion, redevelopment, or reuse of which may be complicated by the presence or potential presence of a hazardous substance, pollutant, or contaminant.” (EPA, U. S, 2021). As the definition suggests, brownfields often represent a possible exposure to various substances; yet brownfields may also represent a common esthetic nuisance even in the absence of any detectable contaminants. Exposures from brownfields may occur through multiple media, including air, water, soil, or esthetic nuisance. Much of the previous work surrounding brownfields has focused on important economic and social justice concerns (Essoka, 2010; Haninger, Ma, & Timmins, 2017; Leigh & Coffin, 2005). Additionally, some research has begun to look at brownfields as an important exposure for human health, including cancer, cardiovascular, and respiratory health outcomes (Li et al., 2017; Litt, Tran, & Burke, 2002). Brownfields have also been linked to birth outcomes, such as low birth weight and preterm birth (J. Wang, 2011).
To date no study has evaluated associations between residential exposure to brownfields and birth defects. This study aims to use available birth record data from the State of North Carolina and brownfields identified by the North Carolina Department of Environmental Quality and EPA to assess the relationship between residential brownfield exposure and the prevalence of birth defects.
2 METHODS
2.1 Study population
Data from North Carolina birth records and the North Carolina Birth Defects Monitoring Program (NCBDMP) were utilized to identify live-born singleton births from 2003 to 2015. Trained NCBDMP staff systematically review and abstract hospital medical records to identify infants born with a birth defect that is diagnosed within the first year of life. Staff supplement these data with information from hospital discharge data, vital records, and Medicaid claims to estimate the incidence of birth defects in North Carolina. This database has been utilized in prior studies and is considered representative for the state (Meyer & Siega-Riz, 2002; Nelson, Stebbins, Strassle, & Meyer, 2016; Rappazzo et al., 2019). Of the more than 1.5 million births available from 2003 to 2015, the NCBDMP identified 39,495 infants with birth defects across 30 phenotypes. Cases for this analysis included all identified births with at least one birth defect. Controls for this analysis were considered on the basis they did not have a recorded birth defect, low birth weight (<2,500 g), or preterm birth (less than 37 weeks gestational age) per standard guidelines (CDC, 2021). For computational efficiency, approximately half of available controls were randomly selected for inclusion in this analysis with a final control sample of 703,140 births. A subset of male only births (n = 450,238) was used to create a separate control population for hypospadias. Address recorded at delivery was considered the principal residence and residences were geolocated using address, city, state, and ZIP in ArcGIS (ESRI, 2011). Only births successfully geocoded to point location, parcel, street address, or street segment were included; those at higher levels (ZIP code) were excluded. Data on gestational parent age (i.e., age of birthing parent at time of birth), race/ethnicity, and other demographics were pulled from birth records. We augmented our data with Census 2010 to capture important social and demographic covariates like urbanicity and education at the census block level, and used EPA's Environmental Quality Index (EQI) from 2006 to 2010 to capture measures of county-level environmental quality (i.e., air and water pollution). This study (#09-0828) was approved by the University of North Carolina—Chapel Hill Institutional Review Board (IRB) (under 45 CFR 46.110).
2.2 Exposure
We acquired data on brownfields by combining data from the North Carolina Department of Environmental Quality (NC-DEQ), and EPA's Clean Ups in my Community (CIMC) (EPA, 2020; Quality, 2020). NC-DEQ maintains a database of all brownfields in the state. This database includes completed (i.e., remediated and restored sites) and ongoing brownfields. NC-DEQ also provides a shapefile of the spatial boundaries for all the brownfields. To ensure complete coverage, we also included the CIMC database that maintains similar data but includes all the brownfield and hazardous waste cleanups for the entire United States. We spatially joined these two data sources to create a complete dataset of all the brownfields identified as of January 2020 in North Carolina. We created radial buffers of 2,000 m around geocoded birth addresses for each individual, with 2,000 m noted in a prior review of physical health and greenspace as an ideal buffer size compared to smaller buffers (Browning & Lee, 2017). This buffer was selected as brownfields can be thought to some extent as the opposite of a greenspace. A points-in-polygon analysis was conducted to calculate the number of brownfields within each 2,000 m buffer. In brief, the boundary shapefiles of brownfields were converted to centroids for computational efficiency and the total number of centroids in each 2,000 m buffer was tabulated to determine the number of brownfields around each residence. We specified any brownfield exposure as having at least one brownfield within the 2,000 m buffer. Multilevel brownfield exposure was broken into zero brownfields in 2,000 m (referent), one to five brownfields in 2,000 m, and more than six brownfields within 2,000 m. These groupings were informed by the distribution of brownfield exposure in North Carolina. Most births do not have any exposure to brownfield and relatively few have more than 10. Groups of one to five and more than six were selected to ensure sufficient sample size in all exposure groups. All spatial analyses were conducted in QGIS 3.14 (QGIS, 2019).
2.3 Outcomes
This analysis focused on birth defects and included 30 different phenotypes identified by the NCBDMP. A complete list of the phenotypes can be referenced in Table S1 or Figure 2. In addition to these individual defects, we also examined seven defect groups. Central nervous system defects include anencephaly, hydrocephalus, and spina bifida. Cardiovascular system defects represent the largest group in terms of cases and include atrial septal defect, aortic valve stenosis, atrioventricular septal defect, coarctation of aorta, hypoplastic left heart syndrome, pulmonary valve atresia, transposition of the greater vessels, tetralogy of Fallot, tricuspid valve atresia, and ventricular septal defect. The orofacial defect group includes cleft lip and cleft palate. Digestive system defect group includes anorectal atresia/stenosis, congenital diaphragmatic hernia, esophageal atresia, gastroschisis, and pyloric stenosis. External defect group includes anophthalmia/microphthalmia, anotia/microtia, congenital cataract, lower limb reduction, and upper limb reduction. Urinary defect group includes obstructive genitourinary defect, omphalocele, and renal agenesis. Hypospadias was not included in the urinary group due to its restriction to males only. Chromosomal defect group consists of several syndromes such as Klinefelter, Down, Edwards, and other chromosomal anomalies.
2.4 Covariates
Potential confounders were selected a priori through a combination of expert knowledge and literature review. We included: gestational parent age at delivery (in continuous years), race/ethnicity (White, Black, Hispanic, Asian American/Pacific Islander, American Indian, Other), prenatal care (month prenatal care began), gestational parent smoking (current smoker yes/no) and diabetes status (includes both gestational and pregestational yes/no), census block group urbanicity (% of census block group designated as urban), and area level education (% of census block group with a bachelor's degree or more). Demographic and behavioral covariates were pulled from birth records. Age, race/ethnicity, smoking habits, and diabetes status have all been documented in prior literature as important factors in various birth defects (Canfield et al., 2014; Gill et al., 2012; Hackshaw et al., 2011; Reefhuis & Honein, 2004; Tinker et al., 2020). Race/ethnicity is not included as a biological variable but as proxy indicator for differential experiences of stressors related to race/ethnicity. Race/ethnicity categories include many populations and cultures. In North Carolina, roughly half of individuals identified as Hispanic identify as Mexican and roughly a quarter identify as of Central American background (i.e., Salvadoran, Honduran, Guatemalan; Ordonez, 2020). Census covariates were collected from Census 2010, via recorded address with both education and urbanicty noted in prior literature (Farley, Hambidge, & Daley, 2002; Langlois et al., 2010). We also augmented covariate data with the EPA's EQI. EQI provides an overview, at the county level, of environmental quality across five domains: air, water, land, built, and sociodemographic environments covering the years from 2006 to 2010, where higher values indicate worse environmental quality. By expanding beyond single routes of exposure, EQI better represents an approximated total exposome for many environmental exposures. EQI has been utilized in a variety of prior epidemiologic studies, including those looking at birth outcomes (Krajewski, Rappazzo, Langlois, Messer, & Lobdell, 2021). Details on the creation and specifics of the EQI can be found in the technical report (EPA, U. S, 2020). We included indices for each of the five domains matched by county to the birth record addresses from EQI 2006 to 2010 as an indicator of overall environmental quality.
2.5 Statistical analyses
Odds ratios (OR) and 95% confidence intervals (95% CI) for brownfield exposure and various birth defects were estimated using mixed effects multivariable logistic regression models with a random intercept for county, in part due to the use of county-level EQI. We utilized a model of any versus zero brownfields within 2,000 m of residence and a multilevel: zero versus one to five versus more than five brownfield exposure within 2,000 m of residence. For both models zero (i.e., no brownfield exposure) is the referent group. All models were adjusted for gestational parent age, month prenatal care began, race/ethnicity group, Census 2010 block group urbanicity, block group education (% with bachelor's degree or more), EQI air, water, land, sociodemographic, and built domains, gestational parent smoking status during pregnancy (yes/no), and gestational parent diabetes during pregnancy (includes both gestational and pregestational and is coded as yes/no).
We also conducted sensitivity analyses stratifying by single (isolated) versus multiple defects and race/ethnicity to assess the potential for effect modification; all models were adjusted for listed potential confounders less the stratifying variables. Additional sensitivity analyses included running models without the EQI land domain, with a categorical age variable, census block group as the random intercept, exclusion of births with more than 20 km separation from a brownfield for a potentially more representative control group, and stratification by Medicaid status at delivery to examine potential modification by socioeconomic status. All analyses were conducted using R Version 3.5.3 (Core, R, 2019) with figures created using the package ggplot2 (Wickham, 2016). Maps were created using QGIS 3.14 (QGIS, 2019).
3 RESULTS
Our sample of 753,195 births from across the state of North Carolina included 39,495 birth defect cases. Table 1 provides breakdowns of key covariates between cases and controls. Cases, in general, had lower average birth weight and lower average gestational age but did not vary largely from controls by other covariates. While our sample includes births from all 100 counties in North Carolina, births do cluster around urban centers along with brownfields (Figure 1).
| Covariate | ||
|---|---|---|
| mean ± (SD) or n (%) | Case | Control |
| n | 39,495 | 703,140 |
| Gestational parent age | 27.5 ± (6.3) | 27.3 ± (6.0) |
| Race/ethnicity (%) | ||
| White non-Hispanic | 23,065 (58.4) | 405,009 (57.6) |
| Black non-Hispanic | 9,123 (23.1) | 154,691 (22.0) |
| Hispanic | 5,688 (14.4) | 111,096 (15.8) |
| Asian American/Pacific Islander | 987 (2.5) | 24,610 (3.5) |
| American Indian | 592 (1.5) | 6,328 (0.9) |
| Other/unknown | 40 (0.1) | 1,406 (0.2) |
| Smoking (% current) | ||
| Yes | 4,976 (12.6) | 74,533 (10.6) |
| No | 31,043 (78.6) | 571,653 (81.3) |
| Missing | 3,476 (8.8) | 56,954 (8.1) |
| Prenatal care (month prenatal care began) | 2.7 ± (1.6) | 2.7 ± (1.5) |
| Education (%)a | 25.9 ± (19.4) | 27.2 ± (19.9) |
| Urbanictya | 65.7 ± (41.1) | 67.1 ± (40.5) |
| Diabetes | ||
| Yes | 2,133 (5.4) | 25,313 (3.6) |
| No | 34,045 (86.2) | 622,279 (88.5) |
| Missing | 3,317 (8.4) | 55,548 (7.9) |
| EQI: Airb | 1.0 ± (0.5) | 1.1 ± (0.5) |
| EQI: Waterb | 1.0 ± (0.1) | 1.0 ± (0.1) |
| EQI: Landb | −0.1 ± (0.5) | 0.0 ± (0.5) |
| Average birthweight (g) | 2,958.5 ± (881.6) | 3,423.7 ± (441.9) |
| Average gestational age (weeks) | 37.1 ± (3.7) | 39.1 ± (1.1) |
| Marital status (%) | ||
| Married | 22,868 (57.9) | 426,806 (60.7) |
| Unmarried | 16,627 (42.1) | 276,334 (39.3) |
| Medicaid (%) | ||
| Yes | 21,604 (54.7) | 353,679 (50.3) |
| No | 17,891 (45.3) | 349,461 (49.7) |
- a From Census 2010 data (block group level) mean percent with Bachelor's degree or more or mean percent designated urban and (SD).
- b From U.S. EPA Environmental Quality Index (EQI) where higher values represent worse environmental domain quality.
Due to the relative rarity of some birth defects, we evaluated associations between any brownfield exposure and birth defects using the birth defect groupings from the NC Birth Defects Monitoring Program: central nervous, cardiovascular, orofacial, digestive, external, urinary, and chromosomal. Group results were mostly null. The cardiovascular group was positively associated with residential proximity to any brownfield (OR: 1.07; 95% CI: 1.02–1.13). The external defect group was also positively associated (OR: 1.17; 95% CI: 1.01–1.35) with residential proximity to any brownfield.
When evaluating associations between any brownfield exposure (any brownfield within 2,000 m of residential address) and individual birth defects our results were mostly null. However, we observed positive associations with atrial septal defect (OR 1.08; 95% CI: 1.01–1.16) and ventricular septal defect (OR: 1.15; 95% CI: 1.03–1.28). We also observed an inverse association with gastroschisis (OR: 0.74; 95% CI: 0.58–0.94). Several cardiovascular and external individual defects (coarctation of aorta, tetralogy of Fallot, anophthalmia/microphthalmia, anotia/microtia, congenital cataract, lower limb reduction, and to a lesser extent upper limb reduction) had positive associations as well. Figure 2 plots the individual and group effects.
The second multilevel model had similar findings to the any versus zero model above. Group defects were mostly null with the notable exception of the cardiovascular group. Individual defects also remained mostly null and are consistent with the any versus zero model above with positive associations for atrial septal defect and ventricular septal defect. However, in addition to the inverse gastroschisis association for the one to five exposure group, we also observed an inverse association indicated for anencephaly in the one to five exposure group. Some defects had mixed directions between exposure groups, the most obvious being anencephaly, pulmonary valve atresia, and omphalocele. The main finding from the multilevel model is that many of the point estimates for most defect phenotypes show an exposure–response relationship, with higher effect estimates with increasing density of brownfields within 2,000 m of the residential address at date of birth (Figure 3). While this relationship is not uniform across all phenotypes, it is indicated across several of them and generally in a positive direction.
In sensitivity analyses with effects stratified by single (isolated) versus multiple defects, and race/ethnicity, we did not observe any indication of potential effect modification. Table 2 provides odds ratios and confidence intervals for each stratified analysis. Additional sensitivity analyses examined the models with and without the inclusion of the EQI land domain, as it incorporates an aggregated county level measure of brownfields. The differences in estimates were negligible with these changes, and model estimates are reported in Table S3. Inclusion of EQI land domain was kept in analysis to account for possible additional differences between counties captured in the land domain (i.e., Superfund sites, agriculture, mining, radon, etc.). An additional sensitivity analysis included running a model with categorical age. Again, changes to the estimates were negligible. To help address concerns around spatial autocorrelation we ran a sensitivity analysis with any births with 20 km or greater distance from a brownfield. This produced results nearly identical to the model as reported, with only slight changes to confidence intervals. To further address autocorrelation, we also ran the models with census block group as our random intercept, with again only minor changes to the reported estimates. Lastly, to try to capture some level of possible individual socioeconomic stratification, we conducted a sensitivity analysis looking at gestational parent Medicaid status at time of delivery. This analysis showed attenuation of the association, but consistency of the exposure-response relationship as indicated by the point estimates, and consistent direction of increased odds. Additional sensitivity analyses are presented in Table S3.
| Any brownfields | ||
|---|---|---|
| OR | CI | |
| Single (isolated) versus multiple defectsa | ||
| Single defect | 1.02 | 0.99–1.06 |
| Multiple defects | 1.04 | 1.01–1.06 |
| Race/ethnicityb | ||
| White | 1.07 | 0.99–1.15 |
| Black | 1.10 | 1.01–1.20 |
| Hispanic | 1.07 | 0.94–1.21 |
| Asian American/Pacific islander | 0.97 | 0.73–1.27 |
| American Indian | 0.87 | 0.39–1.74 |
| Other/unknown | 2.28 | 0.88–6.12 |
- a For all defects.
- b For cardiovascular defect group.
4 DISCUSSION
The primary aim of this analysis was to establish what, if any, association(s) exists between residential proximity to brownfields and birth defects. Using the robust data available from the NC Birth Defects Monitoring Program, CIMC, and NC-DEQ, we were able to assess residential exposure to brownfields across many different birth defect phenotypes. While associations with many of the group and individual birth defects were null, possibly due to relatively small numbers of individual birth defect phenotypes, we observed some positive associations for cardiovascular and external defect groups. For some larger sample size individual cardiovascular defects (atrial septal defect, ventricular septal defect), we also detected positive associations and one inverse association for gastroschisis with exposure to any brownfields. Additionally, we observed several positive associations for other cardiovascular and external defects (coarctation of aorta, tetralogy of Fallot, anophthalmia/microphthalmia, anotia/microtia, congenital cataract, lower limb reduction, upper limb reduction) and a few inverse or mixed associations (gastroschisis, anencephaly, omphalocele). The mixed associations between exposure groups can be interpreted as inconclusive and in some cases are likely due to the small sample size (i.e., anencephaly). Still, the multilevel brownfield exposure models indicated an exposure–response relationship. These results suggest that increasing brownfield exposure may increase odds for various birth defects. While this relationship was not uniform across all phenotypes, it was indicated across several of them and generally in a positive direction.
Our results indicate that brownfield exposure may be linked to birth defects. Two possible mechanisms by which brownfields may be impacting birth defects are contamination and/or esthetic nuisance. Contamination of air and water from nearby hazardous waste has previously been linked to various birth outcomes including birth defects (Al-Sabbak et al., 2012; Aschengrau et al., 2018; Bove et al., 1995; Orr, Bove, Kaye, & Stone, 2002). The specific route depends on the contaminant (i.e., lead [Pb] and neurodevelopment) but several contaminants have been linked to various birth defects (Rappazzo et al., 2019; Sanders et al., 2014; Vinceti et al., 2001; Winchester et al., 2009). Brownfields can be a source for exposure to a variety of contaminants and so residential proximity to and density of brownfields may interrupt, or otherwise harm, fetal development through several different routes. Yet, brownfields might also contain no contaminant or pollutant but still represent a blight and stressor to the neighborhood and community they are in. Prior literature has documented that vacant or abandoned properties pose serious health risks to communities (i.e., increasing injury and fire risk) (Garvin, Branas, Keddem, Sellman, & Cannuscio, 2013; Schachterle, Bishai, Shields, Stepnitz, & Gielen, 2012). Living near these properties might impact gestational parent health and well-being, through inflammatory stress response, mental health, or heart rate variability, among others, which in turn might impact fetal development (Carmichael, Shaw, Yang, Abrams, & Lammer, 2007; South, Kondo, Cheney, & Branas, 2015; Webb et al., 2008). While the direct biological link to birth defects is less obvious, the social stress and conditions associated with living near vacant properties could negatively impact fetal development. As a corollary, these sites may also indicate lower access to services and resources that add stress during the gestational period. This mechanism of increased stress and reduced access are closely tied to issues of environmental justice and could be another route by which proximity to brownfields can impact birth defects (Carmichael et al., 2007; Landrigan, Rauh, & Galvez, 2010; Yang, Carmichael, Canfield, Song, & Shaw, 2008). These proposed mechanisms are not exhaustive of the possibilities but offer initial explanations for our observations.
We are not aware of any previous studies that look at brownfields and birth defects directly. One previous study dealt with birth weight and brownfield proximity in Charlotte, North Carolina and reported positive associations with low birth weight (J. Wang, 2011). Increases in birth defect rates and exposure to hazardous waste and lead (Pb) have also been noted in prior literature (Orr et al., 2002; Sanders et al., 2014; Vinceti et al., 2001). As mentioned, brownfields may represent a source of exposure to hazardous waste and heavy metals like Pb and mercury (Hg). While we cannot confirm that exposure to these heavy metals occurred, it is a possibility as several brownfield sites in North Carolina have been assessed for Pb, Hg, and other hazardous wastes. Further, the exposure–response findings indicate that the odds increase with increasing exposure, consistent with heavy metal contamination studies (Al-Sabbak et al., 2012; Jin et al., 2013). Other studies have linked neighborhood distress, which includes vacant buildings, and adverse birth outcomes (Zuberi, Duck, Gradeck, & Hopkinson, 2016). Studies have also examined remediation efforts of vacant lots, properties, and other esthetic nuisances and birth outcomes, finding that successful conversion of these sites into usable urban greenspace lowers risk of various birth outcomes (i.e., low birth weight and preterm birth) (Margerison, Pearson, Lin, & Sanciangco, 2021). Our results suggest that brownfield exposure may be a factor in risk of birth defects, but further investigation may help to better understand the possible mechanisms (i.e., hazardous waste exposure vs. urban blight).
Using a widespread cross-sectional approach allowed us to evaluate multiple defect phenotypes for associations across all types and categories of brownfields. Although several phenotypes are extremely rare, the large birth records dataset included thousands of cases across North Carolina in a representative population. By examining many defect types, we were able to evaluate multiple possible associations between brownfield exposure and birth defects and establish a baseline for observed effects from residential brownfield exposure. The use of group defects allowed us to pool cases to increase power. Presenting these group defects next to individual defects allowed comparisons and to see how the individual phenotypes may contribute to the overall group effect, aiding in hypothesis generation. Among the strengths were the various sensitivity analyses that looked at gestational parent age as a categorical rather than continuous variable, differential impacts from race/ethnicity group, isolated versus multiple defects, and running models with and without the EQI Land domain all of which did not indicate potential modification or had otherwise negligible differences compared to the reported models.
Our study has some limitations. By using residential proximity, we are assuming that the individual had some exposure to the nearby brownfields, but this cannot be confirmed without more information on individual time-activity patterns, lacking this information may increase the risk of exposure misclassification. Additionally, not all brownfields contain hazardous wastes or meaningful levels of contamination. Yet, even nonhazardous brownfields may represent a substantial urban blight, which has been documented to have negative health effects (Semenza, 2003; South et al., 2015). An additional concern with this buffer analysis is that all brownfields inside that buffer are weighted equally regardless of the distance from birth residence, site type, or contaminant presence, which may lead to exposure misclassification. Future studies may want to examine brownfields assessed for specific contaminants or specific former functions (i.e., former gas station) to better understand possible mechanisms. Lastly, this analysis was widespread in its evaluation of brownfields and birth defects. The adjustment scheme applied to all defect phenotypes may not be well suited for individual defects, or defect groups that have varying etiologies. Misestimation (over or under) may occur due to missing important covariate adjustment, and risk of residual confounding remains. It should also be noted that brownfields themselves may be proxies for other causative factors. Residential proximity to a brownfield may be indicative of lower socioeconomic status or additional environmental justice concerns that may impact birth defects. Further research will be useful to parse out the specific impacts of brownfields on birth defects adjusting for covariables tailored to specific birth defect phenotypes or groups and possibly evaluating brownfields assessed for specific pollutants and contaminants. These steps may continue to elucidate the impact brownfields have on birth defects.
5 CONCLUSIONS
This study of residential brownfield exposure and prevalence of birth defects found evidence that proximity to and density of brownfields may impact the risk for various birth defects and birth defect groups, especially cardiovascular defects. Brownfields may represent an important element to the built environment's contribution to birth defect risk. Future research is needed to identify the specific brownfield site types and contaminants that may be conferring risk as well as understanding the important environmental justice concerns surrounding brownfields.
ACKNOWLEDGMENTS
The authors would like to thank Dr. Alison Krajewski, Ms. Nina Forestieri, and Dr. Monica Jimenez. All funding for this study was provided by the US Environmental Protection Agency. The funder had no role in the study design, data collection, analysis, interpretation, or manuscript writing. This research was supported, in part, by an appointment to the Research Participation Program for the US Environmental Protection Agency, Office of Research and Development, administered by the Oak Ridge Associated Universities through an interagency agreement between the US Department of Energy and Environmental Protection Agency. This work was performed as part of normal duties by EPA employees and contractors.
CONFLICT OF INTEREST
The authors declare that they have no conflict of interest.
DISCLAIMER
The research described in this article has been reviewed by the Center for Public Health and Environmental Assessment, US EPA, and approved for publication. Approval does not signify that the contents necessarily reflect the views and policies of the Agency, nor does the mention of trade names of commercial products constitute endorsement or recommendation for use.
Open Research
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available for public download. https://deq.nc.gov/about/divisions/waste-management/brownfields-program; https://www.epa.gov/cleanups/cleanups-my-community; https://schs.dph.ncdhhs.gov/units/bdmp/; https://www.epa.gov/healthresearch/environmental-quality-index-eqi; https://www.census.gov/programs-surveys/decennial-census/decade.2010.html
