Prenatal Exposure to Toxic Metals and Neural Tube Defects: A Systematic Review of the Epidemiologic Evidence

Abstract

Background:

Neural tube defects (NTDs) affect $>\mathrm{300,000}$ pregnancies worldwide annually. Few nongenetic factors, other than folate deficiency, have been identified that may provide intervenable solutions to reduce the burden of NTDs. Prenatal exposure to toxic metals [arsenic (As), cadmium (Cd), mercury (Hg), manganese (Mn) and lead (Pb)] may increase the risk of NTDs. Although a growing epidemiologic literature has examined associations, to our knowledge no systematic review has been conducted to date.

Objective:

Through adaptation of the Navigation Guide systematic review methodology, we aimed to answer the question “does exposure to As, Cd, Hg, Mn, or Pb during gestation increase the risk of NTDs?” and to assess challenges to evaluating this question given the current evidence.

Methods:

We selected available evidence on prenatal As, Cd, Hg, Mn, or Pb exposure and risk of specific NTDs (e.g., spina bifida, anencephaly) or all NTDs via a comprehensive search across MEDLINE, Embase, Web of Science, and TOXLINE databases and applied inclusion/exclusion criteria. We rated the quality and strength of the evidence for each metal. We applied a customized risk of bias protocol and evaluated the sufficiency of evidence of an effect of each metal on NTDs.

Results:

We identified 30 studies that met our criteria. Risk of bias for confounding and selection was high in most studies, but low for missing data. We determined that, although the evidence was limited, the literature supported an association between prenatal exposure to Hg or Mn and increased risk of NTDs. For the remaining metals, the evidence was inadequate to establish or rule out an effect.

Conclusion:

The role of gestational As, Cd, or Pb exposure in the etiology of NTDs remains unclear and warrants further investigation in high-quality studies, with a particular focus on controlling confounding, mitigating selection bias, and improving exposure assessment. https://doi.org/10.1289/EHP11872

Introduction

Neural tube defects (NTDs), including myelomeningocele (commonly referred to as spina bifida), anencephaly, and encephalocele, occur in early pregnancy owing to improper closure of the embryonic neural tube.¹ Anencephaly is nearly uniformly fatal and spina bifida and encephalocele are often accompanied by serious disability in affected infants.^2,3 In the United States, the prevalence of NTDs decreased in the years following the introduction of folic acid fortification of cereal grain products in 1996–1998.² However, the prevalence of NTDs in the United States has remained stable in the years post-fortification, with an estimated 2,469 NTDs occurring annually, often termed folate-resistant NTDs.^1,2,4 Globally, where folic acid supplementation is not universal, it is estimated that $>\mathrm{300,000}$ babies are born with NTDs annually.⁵

NTD etiology is multifactorial, with genetic and environmental contributions.¹ Known mechanisms of genetic vulnerability to NTDs relate to one-carbon metabolism or planal cell polarity genes, with certain polymorphisms known to increase susceptibility of pregnancies to the effects of nongenetic risk factors, such as maternal diabetes and folate status.³ Aside from folate deficiency and diabetes, few nongenetic factors have been consistently linked to NTDs, although in general, they have not been widely studied.³ Because these birth defects remain a substantial cause of morbidity and mortality, even in regions with folic acid supplementation, there is an urgent need to identify modifiable factors that may reduce the burden of NTDs. Environmental contaminants, such as toxic metals, may feasibly be involved in known vulnerability pathways and represent a promising area of research for preventable causes of NTDs.³

Arsenic (As), cadmium (Cd), mercury (Hg), manganese (Mn), and lead (Pb) are toxic metals/metalloids (herein all referred to as “metals”) of global concern that have been most commonly implicated in the etiology of NTDs. Toxic metals are ubiquitous contaminants in the United States. In a nationally representative survey between 2003 and 2014, As, Cd, Mn, and Pb were detected in women of reproductive age in 97.8%, 87.9%, 86.9%, and 99.0% of biological samples, respectively.⁶ Exposures occur predominantly through ingestion of contaminated foods and drinking water, as well as through inhalation of polluted air.⁷ As, Cd, Hg, and Pb are known developmental toxicants and can cross from mother to fetus through transplacental transfer.^8–10 One of the proposed biological mechanisms through which environmental factors, including As, Cd, Hg, Mn, Pb, and other metals, influence NTD etiology is increased levels of oxidative stress, providing biological plausibility to the hypothesis that prenatal metal exposure increases the risk of NTDs.^11–14

Although the literature describing epidemiologic studies of toxic metals and NTDs has grown in recent years, to our knowledge no systematic reviews have been performed to summarize strengths and weaknesses of the evidence, nor have any meta-analyses been conducted to assess whether results may be suitable to deriving summary effect estimates or whether heterogeneity or evidence quality prevents useful summarization. Thus, the primary purpose of this review was to summarize and evaluate the quality of the epidemiologic evidence investigating associations of As, Cd, Hg, Mn, and Pb exposures during gestation and the prevalence of NTDs, with reference to the Navigation Guide methodology.^15,16 As a secondary goal, if a particular association was found to have evidence that supports a meta-analytic summary estimate (sufficient number of studies with consistent study characteristics and high-quality methods across the literature such that random error rather than systematic bias is the primary concern), then our goal was to also provide such an estimate. We additionally aimed to provide a discussion of key methodological considerations and knowledge gaps to inform future research.

Methods

Prior to starting the process of the systematic review, we developed a thorough initial protocol detailing all the steps that we anticipated would be undertaken (available in the Supplemental Material, “REVIEW PROTOCOL”). We followed, with adaptations to best meet our question’s needs, the Navigation Guide’s systematic review methodology’s first three steps, namely 1) specifying the study question; 2) selecting the evidence; 3) rating the quality and strength of the evidence.^15,16

Study Question

Our objective was to answer the question: “does exposure to As, Cd, Hg, Mn, or Pb during gestation increase the risk of NTDs?” We evaluated all articles for evidence in relation to this question even if the study question of the article itself was different. The PECO statement that we used was as follows:

• Population: Humans who were studied during pregnancy or after birth/termination

• Exposure: Prenatal exposure to metals of interest (As, Cd, Hg, Mn, Pb)

• Comparator: Humans exposed to lower levels of metals than the more highly exposed humans

• Outcome: All NTDs (i.e., the outcome assessed includes multiple NTDs considered together) or specific NTDs considered separately (e.g., spina bifida, anencephaly, hydrocephaly, encephalocele).

Note that the PECO statement was adapted following the second search (see the section “Search strategy” for search dates). The initial PECO statement that guided the search strategy listed the outcome as “All, subgroups or specific birth defects.” However, following extraction of data, we generated the updated PECO statement in which the outcome included only NTDs based on availability of studies to summarize (see the section “Data extraction and visualization”).

Selection of Evidence

Search strategy.

We searched for all literature types in MEDLINE (via PubMed), Embase, Web of Science, and TOXLINE from date of database inception through 11 November 2022, using terms for As, Cd, Hg, Mn, and Pb combined with terms for birth defects and studies with human subjects. We used either medical subject headings (MeSH) or Embase subject headings (Emtree) where available or keywords when applicable. The full search string can be found in Table S1. We conducted the first search on 10 July 2019, a second search on 20 March 2021, and a final search on 11 November 2022. Toxline was decommissioned by the National Library of Medicine on 29 January 2021, but all of the content is still available in MEDLINE. Toxline was therefore only searched in 2019. In addition, UNC-Chapel Hill stopped its subscription to Web of Science in December 2021; therefore, it was not included in the 2022 search.

Study selection criteria.

First, we screened articles by reviewing the title and abstract. Then, for articles not excluded based on title and abstract, we conducted a full-text review to assess for further eligibility. We used Covidence systematic review software for the screening process (Veritas Health Innovation; https://www.covidence.org). Authors L.A.E., G.C., E.H., F.C.M.S., J.P.B., and A.P.K. conducted the screening. In both the title and abstract screening and full-text review, all studies were reviewed independently by two authors. If there was a discrepancy, it was resolved by a third author or brought to the larger team to discuss and resolve the discrepancy through consensus. To be included, studies needed to meet all the following criteria:

1. The study was an epidemiologic study of human exposures. All animal or other non-epidemiologic studies were excluded. There were no criteria for the type of epidemiologic study design.

2. The study measured prenatal exposure to As, Cd, Hg, Mn, or Pb. For biomarker-based exposure assessment, we defined prenatal exposure as biomarkers measured after conception and before or at birth/termination from either the mother or offspring, including measurements immediately after birth/termination. For environmental or occupational exposure in which the exposure level could feasibly be considered constant over time (e.g., exposure via occupation or soil), no time restriction was enforced. Studies were excluded if they considered only paternal exposures or were conducted as part of assessments of diet or supplement interventions.

3. The study undertook a quantitative exposure assessment or for ecologic study designs, comparisons of exposed/unexposed (or less-exposed) groups were considered.

4. The study evaluated the outcome of all, subgroups, or specific birth defects, either at birth/termination or during pregnancy.

5. The article was published in English in a peer-reviewed journal after 1 January 1995, to limit the search to the past 25 y of literature.

Data extraction and visualization.

For each included study, we extracted data, including study sample characteristics, estimates of association, exposure assessment, and outcome assessment. When applicable, we extracted unadjusted and adjusted estimates of association, as well as sex-stratified estimates and dose–response assessments. However, other stratified estimates were not extracted. Authors L.A.E., G.C., E.H., J.P.B., and A.P.K. conducted data extraction using a custom Excel spreadsheet. The custom Excel spreadsheet, complete with extracted data for an example study, can be found in the Supplemental Material. For specific details regarding data extraction steps, please refer to the review protocol (available in the Supplemental Material, “REVIEW PROTOCOL”).

We performed three rounds of extractions with two authors per study and resolved discrepancies with a third author and discussed as an author group. During this process, the data extraction protocol was refined and once we were consistently reaching consensus as a group and extracting repeatable data in these early rounds, subsequent rounds of extraction were performed by single authors. Although the Cochrane Handbook for Systematic Reviews of Interventions describes the extraction of study characteristics by two authors as “desirable” and study outcomes by two authors as “mandatory,” we found that study outcomes were consistently abstracted in the first rounds of review and did not warrant duplicated efforts in further rounds.¹⁷

For comparison across all articles extracted, we grouped studies by birth defect category assessed, using the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD10) classification system, which is generally organized by biological system affected.¹⁸ Following this step, we evaluated the extracted data for the number of studies investigating associations between each metal–birth defect category pairing (Table S2). From this assessment, we decided to prioritize a review of NTDs, given the preponderance of studies for this category, and we then modified our PECO statement accordingly. Hydrocephaly, although not usually considered to be an NTD, was included in this review because two papers on NTDs included it and we wanted to be as comprehensive as possible. A Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) diagram summarizing the search, screen, and extraction process is provided in Figure S1.

For studies with information that was required in our data extraction process and not included in the published manuscript, we contacted the study authors. Specifically, we sent two email requests to the corresponding authors of two papers with incomplete information^19,20 in February and March of 2022; however, we received no response.

Quality and Strength of Evidence Evaluation

We assessed the quality and strength of the evidence in three sequential stages: 1) for each included study, we undertook a risk of bias assessment to evaluate study characteristics that may have introduced bias; 2) for each metal included, we evaluated the quality of the evidence across all studies; 3) for each metal included, we assessed the strength, or certainty, of the evidence across all studies. These steps were informed by the process detailed in the Navigation Guide; however, they were adapted to our study question in ways that are detailed below. Note that for steps 2 and 3, we focused on studies that provided estimates of association with any NTDs as the outcome, given how few studies evaluated specific NTDs. The majority of studies that did provide specific NTDs estimates, however, also provided estimates of association for any NTDs as the outcome.

Assessing the risk of bias for each study.

To differentiate between studies in terms of the scale of threats to validity for estimates of association, we performed a risk of bias assessment adapted from the Navigation Guide’s protocol.^15,16 Our protocol included five domains of potential biases: a) exposure measurement, b) outcome ascertainment, c) selection bias: recruitment and loss-to-follow-up, d) confounding, and e) missing data. These domains were selected based on considering the Navigation Guide’s proposed domains in the context of this specific study question and the authors’ understanding of the most concerning threats to internal validity in this literature. We note that any studies of birth defects would be subject to bias owing to the inability to fully enumerate conceptuses or fetuses at an early stage of pregnancy and, for NTDs specifically, it cannot be determined when incomplete closure of neural tubes “starts” (i.e., an incident case), thus prevalent cases are used as proxies for incident cases throughout the literature. Therefore, we did not consider these potential sources of bias when evaluating the literature.

Compared with the Navigation Guide, we omitted the domains of Conflict of Interest, Blinding, and Selective Outcome Reporting. We deemed conflict of interest to be less of a threat to the validity of this literature compared with studies of man-made pharmaceutical or industrial chemicals, common in other applications of the Navigation Guide. Therefore, we noted any declared conflicts of interest and funding sources for all included studies but did not formally assess conflict of interest in the risk of bias. Blinding was captured within the broader exposure assessment domain, which required exposure assessment to be independent of the outcome. Last, selective outcome reporting was removed because we decided that without preregistration of the studies, not common in the observational studies setting, we would not be able to accurately judge whether outcomes were being selectively reported. Study design features (e.g., whether cases derived from a population-based registry or hospital-based study) were considered in evaluating risk of bias. For example, we specifically noted potential for Berkson’s bias when using hospital-based studies depending on how controls were selected.

For each article, we scored the five potential bias domains as either low risk of bias, probably low risk of bias, probably high risk of bias, high risk of bias, or unclear risk of bias. These scores are meant to reflect the perceived validity of a result as well as believability. That is, two studies may have equally valid results, but we defined a study as more believable (i.e., identifiable lower risk of bias) if it transparently reported factors across each of these domains according to our protocol. For example, a study that reports few missing data was considered more believable than a study that makes no mention of missing data. The full risk of bias assessment protocol is detailed in the review protocol (available in the Supplemental Material, “REVIEW PROTOCOL”). Specific instructions for classifying studies as “high risk of bias” or “low risk of bias” for each domain are provided in the protocol. We assigned “probably low risk of bias” when there was insufficient information to fully meet the criteria for “low risk of bias” but there was indirect evidence that suggested the criteria were met. We assigned “probably high risk of bias” when there was insufficient information to fully meet the criteria for “high risk of bias” but there was indirect evidence that suggested the criteria were met. Although the Navigation Guide does not include an “unclear risk of bias” ranking, we added this option to distinguish between not having enough information to make a judgment vs. adequate information to determine a high risk of bias. Bias was aggregated on a studywide scale, so some specific analyses may differ in risk of bias from the study overall. L.A.E. drafted the initial protocol prior to beginning the screening process, with input and revisions from G.C., E.H., J.P.B., and A.P.K. Authors L.A.E., G.C., E.H., J.P.B., and A.P.K. conducted the risk of bias assessment. To calibrate scores across different reviewers, we conducted three initial rounds in which two reviewers assigned risk of bias scores for one study, we discussed and compared scores as a group, and came to a consensus ranking for each domain for each study. Through this process, we refined the protocol and once independent reviewers were consistently reaching the same scores, for remaining studies, one reviewer was assigned per study.

Assessing the quality of evidence for each metal.

We adapted the Navigation Guide protocol for determining the strength and quality of the evidence to best meet the needs of our study question.^15,16 Authors L.A.E., G.C., J.P.B., and A.P.K. individually completed assessments for each metal prior to a group discussion in which ratings and rationales were compared to reach a group consensus. At baseline, we assigned the evidence as “moderate” quality status (or 0), and then we considered three downgrade ($-1$) and three upgrade ($+1$) factors. The downgrade factors that we considered were a) risk of bias, b) imprecision, and c) publication bias. For assessing publication bias, we did not construct funnel plots given the heterogeneity of studies included. The upgrade factors that we considered were a) large magnitude of effect, b) dose response, and c) residual confounding. The Navigation Guide does not give explicit directions for how to convert the upgrade and downgrade factors into an overall rating.^15,16 Thus, to be transparent, we derived an overall quality of evidence rating classification by summing the upgrade and downgrade factors. If the sum was $\le -1$, we considered the evidence to be of “low” quality. If the sum was $\ge 1$, we considered the evidence to be of “high” quality. Summed values of 0 were assigned “moderate” quality status. The labels of “high,” “low,” and “moderate” should not be considered objective judgments but, rather, as parsimonious descriptions of how the numeric scores corresponded to subjective assessments of the literature on metals and NTDs relative to other literatures with which we are familiar. The complete detailed criteria for scoring each of the factors for the strength and quality assessment can be seen in the review protocol (available in the Supplemental Material, “REVIEW PROTOCOL”).

Note that in comparison with the Navigation Guide, we did not consider indirectness because there are several valid study questions that fit our PECO statement. For instance, environmental exposures may provide direct evidence about effects of external exposure, whereas biomarker exposures may provide direct evidence about effects of dose to target tissues. We did not prioritize either of these valid study questions and, consequently, did not find that “indirectness” was a salient issue for the body of literature under consideration. Furthermore, any study not addressing the broad scope of the PECO statement would already have been removed in the screening phase of the review. In addition, we did not consider inconsistency given that this was captured in our consideration of the direction of the effect and because exposure assessment and levels, along with other study methods, varied widely enough from study to study that we were not able to a priori hypothesize the expected effect of heterogeneity in study methods collectively on the direction and magnitude of effect estimates. Thus, apparent inconsistency in findings above and beyond effect direction was a stronger indication of heterogeneity of study methods, which we could not with confidence estimate the effect of, rather than quality of the literature (see the section “Assessing the strength of evidence for each metal”).

Assessing the strength of evidence for each metal.

Next, for each metal, we rated the strength of the evidence across all relevant studies. Our goal was to ultimately provide a concise statement summarizing the current state of the evidence regarding each metal’s association with risk of NTDs. Adapting the Navigation Guide, we evaluated the strength of evidence based on three factors: a) quality of the evidence (from the previous step), b) the direction of effect, and c) our confidence in the estimates of association.^15,16 We determined that the direction of effect would be based on a meta-analytic summary estimate in the case of conducting a meta-analysis. We a priori decided that we would only conduct a meta-analysis when supported by a sufficient number of high-quality studies with consistent study methodologies such that random error would be the primary concern. Where no meta-analytic estimate was estimated, we judged the direction of effect based on visual assessment of all studies included in forest plots. We decided on our confidence in the estimates of association based on a judgment regarding the likelihood that additional studies would change the conclusion.

After consideration of these three factors, we determined the evidence of an association for each metal to be “sufficient,” “limited,” “inadequate,” or “evidence of lack of effect,” based on the definitions used in the Navigation Guide.¹⁵ In brief, sufficient evidence was defined as an observed positive relationship between the metal and risk of NTDs where “chance, bias, and confounding can be ruled out with reasonable confidence.” Limited evidence was defined as an observed positive relationship between the metal and risk of NTDs where “chance, bias, and confounding cannot be ruled out with reasonable confidence,” and confidence in the relationship is hindered by number, size, or quality of studies, as well as by inconsistencies in findings.¹⁵ Inadequate evidence was defined as the evidence being insufficient to assess the relationship between the metal and NTDs given low number, size, or quality of studies. Evidence of lack of effect was defined as an observed lack of association between the metal and NTDs where “chance, bias, and confounding can be ruled out with reasonable confidence.”

Results

Included Studies

We retrieved 1,184 articles. In the title and abstract screening process, we excluded 1,055 articles for not being relevant according to the PECO statement. We screened the remaining 129 articles for eligibility via the full-text screen. Of these, 66 articles met the inclusion criteria and 63 were excluded. Specific reasons for the exclusion of each article at the full-text stage are provided in Table S2. Next, we categorized each of the 66 articles into groups according to the body system affected by the birth defect evaluated (Table S3). Following this step, we identified NTDs as the defect category to focus on based on the greatest number of articles to evaluate. From the 66 articles eligible for inclusion, 30 evaluated NTDs and represented the final group of studies we evaluated in this review (Table 1). The PRISMA flow diagram is shown in Figure S1.

Table 1.

Summary of metals and defects assessed, outcome ascertainment methods, study designs, conflicts of interest and funding sources, and risk of bias rankings for included studies.

Reference	Metal(s) assessed					Defects evaluated					Outcome ascertainmenta		Conflict of interest (COI) and fundingb		Risk of bias rankings
	As	Cd	Pb	Mn	Hg	NTDs	Ac	Hc	Ec	SB	Hospital/facility	Population-based	Declared no COI	National agencies or university	E	O	S	C	M
Liu et al.21	—	X	X	—	—	X	—	—	—	—	X	—	X	X	PL	PL	L	PL	L
Tian et al.22	—	X	X	X	X	X	—	—	—	—	—	X	X	X	PL	PL	PL	PL	L
Tong et al.23	—	—	—	—	X	X	X	X	—	—	—	X	X	X	L	L	H	H	L
Liu et al.24	—	—	—	X	—	X	—	—	—	—	X	—	X	X	PH	L	PH	H	L
Ovayolu et al.25	X	X	X	X	X	X	—	—	—	—	X	—	X	—	L	PL	L	PH	PH
Yin et al.26	—	—	—	X	—	X	—	—	—	—	—	X	X	X	L	PL	PH	PH	L
Demir et al.27	X	X	X	X	—	X	—	—	—	—	X	—	X	X	PH	H	H	H	PH
Özel et al.28	X	X	X	X	X	X	—	—	—	—	X	—	X	—	L	L	PL	H	L
Troisi et al.29	—	X	X	X	X	X	—	—	—	—	X	—	X	—	PH	PL	PL	H	L
Wang et al.30	X	—	—	—	—	X	X	—	—	X	X	—	X	X	PL	PH	PH	PL	L
Cim et al.31	—	X	X	X	—	X	—	—	—	—	X	—	—	—	PL	PL	PH	H	L
Marie et al.32,c	X	—	—	—	—	X	—	—	—	—	—	X	X	X	PH	L	PH	PL	L
Yan et al.20	—	—	—	X	—	X	X	—	X	X	X	—	X	—	PL	H	H	PH	PH
Jin et al.33	—	—	—	—	X	X	X	—	—	X	—	X	—	X	PH	PL	U	H	U
Manduca et al.34	—	X	X	—	X	X	—	—	—	—	X	—	X	—	PL	PL	PH	H	L
Sanders et al.35,d	X	X	X	X	—	—	—	—	—	X	—	X	X	X	PL	L	L	PL	L
Jin et al.19	X	X	X	—	X	X	X	—	—	X	—	X	X	X	L	L	H	PL	L
Liu et al.36	—	—	—	X	—	X	X	—	—	X	X	—	—	X	L	U	U	PL	U
Ramírez-Altamirano et al.37	—	—	X	X	—	X	—	—	—	—	X	—	X	—	L	PH	H	H	L
Gatti et al.38,c	—	—	X	—	—	X	—	—	—	—	X	—	—	X	L	PL	H	H	L
Zeyrek et al.39	—	—	X	—	—	X	—	—	—	—	X	—	X	—	PL	U	U	H	L
Friel et al.40	—	X	—	—	—	—	X	—	—	—	X	—	—	X	PL	PL	H	H	L
Carrillo-Ponce et al.41	—	X	X	—	—	X	—	—	—	—	—	—	—	—	PH	U	U	U	U
Cengiz et al.42	—	—	X	—	—	X	—	—	—	—	X	—	—	—	PL	L	PL	H	L
Macdonell et al.43,c	—	—	X	—	—	X	—	—	—	—	—	X	—	—	H	PL	PH	H	L
Dawson et al.44	—	—	X	—	—	X	—	—	—	—	X	—	—	—	PL	L	PH	H	L
Stoll et al.45	—	—	—	X	—	—	—	X	—	X	—	X	—	X	PL	PL	L	H	L
Romero et al.46	—	—	X	—	X	—	X	—	—	—	X	—	—	—	PL	L	H	H	L
Bound et al.47	—	—	X	—	—	X	X	—	—	X	X	—	—	—	PH	PH	H	PL	U
Tahan et al.48	—	—	X	—	X	—	—	—	—	—	X	—	—	—	PH	L	PH	H	U

Note: As noted in the text, hydrocephaly, although not usually considered to be an NTD, was included in this review because two papers included it and we wanted to be as comprehensive as possible. For additional information on the study country, sample size, and exposure assessment, see Figures 2–6. —, not applicable; Ac, anencephaly; As, arsenic; C, confounding; Cd, cadmium; E, exposure assessment; Ec, encephalocele; H, high risk of bias; Hc, hydrocephaly; Hg, mercury; L, low risk of bias; M, missing data; Mn, manganese; NTD, neural tube defect (grouping of $\ge 2$ individual NTDs); O, outcome ascertainment; Pb, lead; PH, probably high risk of bias; PL, probably low risk of bias; S, selection; SB, spina bifida; U, unclear risk of bias.

^a The following definitions were used for categorizing the outcome assessment methods: a) population-based active surveillance: abstractor actively reviews multiple data sources to identify all cases, b) population-based registry (e.g., birth records or passive surveillance): case ascertainment from medical record abstraction or based on registry data only, c) hospital/facility based: based on $\ge 1$ hospital/facility, (a) and (b) are combined in this table as “population-based.”

^b Only entries with an “X” have a section in the published manuscript describing conflicts of interest/funding sources, of lack thereof.

^c All studies are case–control studies unless indicated by a superscript c.

^d Referred to as “semi-ecologic” in paper, described herein as a retrospective cohort with ecologic exposure assessment.

Risk of Bias Assessment for Individual Studies

The risk of bias scores for each domain for each study, along with rationale, are detailed in Table S4. The risk of bias rankings varied substantially across studies (Table 1). When we assessed risk of bias across all studies grouped by metal assessed, we identified confounding as a major area of bias, followed by selection bias and exposure assessment, across all metals (Figure 1). For all metals, over half of all articles ranked as high risk of bias for confounding (Figure 1). Specifically, for As, 4/7 studies (57%) ranked high risk of bias; for Cd, 8/13 (62%) ranked high; for Hg, 8/12 (75%) ranked high; for Mn, 6/12 (50%) ranked high; and for Pb, 16/20 (80%) ranked high. Selection bias had over half of all articles ranked as high or probably high risk of bias for most metals, other Mn. Specifically, for As, 5/7 studies (71%) ranked high or probably high risk of bias; for Cd, 7/13 (54%) ranked high or probably high; for Hg, 6/12 (50%) ranked high or probably high; for Mn, 5/12 (42%) ranked high or probably high; and for Pb, 13/20 (65%) ranked high or probably high. Exposure assessment was deemed high or probably high risk of bias in 17%–43% of articles for each metal, with Mn having the lowest percentage of studies ranked as probably high or high risk of bias [2/12 (17%)] and As having the highest [3/7 (43%)]. Outcome measurement was ranked as low or probably low risk of bias in the majority of included articles for As (5/7, 72%), Cd (10/12, 77%), Hg (10/11, 83%), Mn (9/12, 75%), and Pb (15/20, 75%). Similarly, missing data was ranked as low or probably low risk of bias in most included articles for As (6/7, 86%), Cd (10/12, 77%), Hg (8/11, 67%), Mn (9/12, 75%), and Pb (16/20, 80%).

Figure 1.

Summary of risk of bias analysis for each domain and for each metal, summarized across all studies. For detailed criteria for classifying studies as “high risk of bias” or “low risk of bias,” see the review protocol (available in the Supplemental Material, “REVIEW PROTOCOL”). “Probably low risk of bias” was assigned when there was insufficient information to fully meet the criteria for “low risk of bias” but there was indirect evidence that suggested the criteria were met. “Probably high risk of bias” was assigned when there was insufficient information to fully meet the criteria for “high risk of bias” but there was indirect evidence that suggested the criteria were met. “Unclear risk of bias” was assigned when there was not enough information in the study to determine the risk of bias.

Although conflict of interest was not evaluated formally in our risk of bias assessment, we did note declared conflicts of interests and funding sources for each study (Table 1). Nearly all studies published since 2005 either declared no conflict of interest or were funded by universities or national agencies. For most studies published before 2005, conflicts of interest and funding sources were not described, which is likely reflective of publishing conventions changing over time.

Summary of Results across Studies

We generated forest plots for each metal to aid with the visual assessment of the direction, consistency, and precision of the estimates of association (Figures 2–6). We were unable to retrieve needed additional or corrected data for two studies.^19,20 To allow for comparison of methodologies across articles that did not report an estimate of association, we reported study characteristics alongside the forest plots for visual comparison of methodologies across studies. There were too few studies within each of the outcome categories (any NTDs, or each of the specific NTDs) to formally assess heterogeneity in study effect estimates via methods like meta-regression. However, various orderings of the forest plots (e.g., by exposure assessment, which we felt to be the most salient domain by which study quality varied), did not produce discernible patterns.

Figure 2.

Summary of data extracted from all studies of arsenic exposure and risk of NTDs. For each study, we selected the estimate to plot based on the following prioritization: 1) Dose response: plot quartile estimates, then tertile, then above/below median or other dichotomous split. We plotted continuous estimates in addition if presented, as well as nonlinear dose response. 2) Adjustment: fully adjusted estimates were preferentially plotted. Studies that did not present point estimates are summarized in Table S5. Given the range of scales and media of these measurements, they could not be adequately visually summarized. Under the “OR (95% CI)” column, the squares represent the odds ratio values and the bars represent the 95% confidence intervals. Risk of bias (ROB) rankings are indicated by the color coding in the far right column. ROB rankings can also be found in Table 1. The single asterisk indicates studies that did not present point estimates and rather presented comparisons of concentrations between cases and controls are listed without an OR; summary results from these studies are listed in Table S5. The double asterisk indicates that the article provided the $N$ for the full study cohort but only the male-stratified effect estimate. Note: %ile, percentile; C, confounding; CI, confidence interval; E, exposure measurement; edu, maternal education; fa, folic acid use; flufev, flu or fever during pregnancy; geosize, size of municipality; hxBD, history of birth defect; M, missing data; ma, maternal age; Med, median; NTD, neural tube defect; O, outcome measurement; occup, occupation or employment; OR, odds ratio; par, parity; pass smoke, passive smoking; Q, quartile; raceeth, maternal race or ethnicity; S, selection; yr, year of birth.

Figure 6.

Summary of data extracted from all studies of lead exposure and risk of NTDs. For each study, we selected the estimate to plot based on the following prioritization: 1) Dose response: plot quartile estimates, then tertile, then above/below median or other dichotomous split. We plotted continuous estimates in addition if presented, as well as nonlinear dose response. 2) Adjustment: fully adjusted estimates were preferentially plotted. Studies that did not present point estimates are summarized in Table S5. Given the range of scales and media of these measurements, they could not be adequately visually summarized. Under the “OR (95% CI)” column, the squares represent the odds ratio values and the bars represent the 95% confidence intervals. Risk of bias (ROB) rankings are indicated by the color coding in the far right column. ROB rankings can also be found in Table 1. The single asterisk indicates that the studies did not present point estimates and, rather, presented comparisons of concentrations between cases and controls and are listed without an OR; summary results from these studies are listed in Table S5. The double asterisk indicates the studies provided no confidence limits. Note: %ile, percentile; bmi, body mass index; C, confounding; CI, confidence interval; deprivation, area-level deprivation score; E, exposure measurement; edu, maternal education; fa, folic acid use; flufev, flu or fever during pregnancy; ga, gestational age; M, missing data; ma, maternal age; Med, median; n/a, not available; NTD, neural tube defect; O, outcome measurement; OR, odds ratio; par, parity; raceeth, maternal race or ethnicity; ROC, receiver operating characteristic; S, selection; sex, infant sex; T, tertile.

Figure 3.

Summary of data extracted from all studies of cadmium exposure and risk of NTDs. For each study, we selected the estimate to plot based on the following prioritization: 1) Dose response: plot quartile estimates, then tertile, then above/below median or other dichotomous split. We plotted continuous estimates in addition if presented, as well as nonlinear dose response. 2) Adjustment: fully adjusted estimates were preferentially plotted. Studies that did not present point estimates are summarized in Table S5. Given the range of scales and media of these measurements, they could not be adequately visually summarized. Under the “OR (95% CI)” column, the squares represent the odds ratio values and the bars represent the 95% confidence intervals. Risk of bias (ROB) rankings are indicated by the color coding in the far right column. ROB rankings can also be found in Table 1. The single asterisk indicates the study was not plotted because the reported OR was outside of the reported 95% CI. The double asterisk indicates that the studies did not present point estimates and, rather, presented comparisons of concentrations between cases and controls and are listed without an OR; summary results from these studies are listed in Table S5. Note: %ile, percentile; bmi, body mass index; C, confounding; CI, confidence interval; E, exposure measurement; edu, maternal education; fa, folic acid use; flufev, flu or fever during pregnancy; ga, gestational age; hxBD, history of birth defect; M, missing data; ma, maternal age; Med, median; n/a, not available; NTD, neural tube defect; O, outcome measurement; occup, occupation or employment; OR, odds ratio; pass smoke, passive smoking; raceeth, maternal race or ethnicity; S, selection; sex, infant sex; T, tertile.

Figure 4.

Summary of data extracted from all studies of mercury exposure and risk of NTDs. For each study, we selected the estimate to plot based on the following prioritization: 1) Dose response: plot quartile estimates, then tertile, then above/below median or other dichotomous split. We plotted continuous estimates in addition if presented, as well as nonlinear dose response. 2) Adjustment: fully adjusted estimates were preferentially plotted. Studies that did not present point estimates are summarized in Table S5. Given the range of scales and media of these measurements, they could not be adequately visually summarized. Under the “OR (95% CI)” column, the squares represent the odds ratio values and the bars represent the 95% confidence intervals. Risk of bias (ROB) rankings are indicated by the color coding in the far right column. ROB rankings can also be found in Table 1. Note: bmi, body mass index; C, confounding; CI, confidence interval; E, exposure measurement; edu, maternal education; fa, folic acid use; flufev, flu or fever during pregnancy; ga, gestational age; hxBD, history of birth defect; M, missing data; ma, maternal age; Med, median; n/a, not available; NTD, neural tube defect; O, outcome measurement; OR, odds ratio; pass smoke, passive smoking; S, selection; sex, infant sex; T, tertile.

Figure 5.

Summary of data extracted from all studies of manganese exposure and risk of NTDs. For each study, we selected the estimate to plot based on the following prioritization: 1) Dose response: plot quartile estimates, then tertile, then above/below median or other dichotomous split. We plotted continuous estimates in addition if presented, as well as nonlinear dose response. 2) Adjustment: fully adjusted estimates were preferentially plotted. Studies that did not present point estimates are summarized in Table S5. Given the range of scales and media of these measurements, they could not be adequately visually summarized. Under the “OR (95% CI)” column, the squares represent the odds ratio values and the bars represent the 95% confidence intervals. Risk of bias (ROB) rankings are indicated by the color coding in the far right column. ROB rankings can also be found in Table 1. The single asterisk indicates the study had Q3 vs. Q1 OR listed as 0.14 (0.02–1.18) and is not plotted given that this is likely an error. The double asterisk indicates that the studies did not present point estimates and, rather, presented comparisons of concentrations between cases and controls and are listed without an OR; summary results from these studies are listed in Table S5. Note: %ile, percentile; bmi, body mass index; C, confounding; CI, confidence interval; E, exposure measurement; edu, maternal education; fa, folic acid use; flufev, flu or fever during pregnancy; ga, gestational age; hxBD, history of birth defect; M, missing data; ma, maternal age; NTD, neural tube defect; O, outcome measurement; occup, occupation or employment; OR, odds ratio; par, parity; Q, quartile; raceeth, maternal race or ethnicity; ROC, receiver operating characteristic; S, selection; sex, infant sex; smoke, smoking; T, tertile.

Quality of Body of Evidence

We assessed the quality of the evidence across all included articles for each metal (Table 2). The quality $<0.66$ of the evidence for all metals was downgraded owing to numerous studies with notable risk of bias. Specifically, for all metals at least 50% of studies were ranked as having high risk of bias for confounding and high or probably high risk of bias for selection bias. No body of evidence was downgraded for imprecision because none of them had numerous studies with a sample size of $\le 15$ cases. Although we did not formally assess publication bias because of the heterogeneity of studies, based on the other considerations outlined in the review protocol (available in the Supplemental Material, “REVIEW PROTOCOL”), we deemed substantial publication bias to be unlikely for all metals. Specifically, no body of evidence was dominated by early, small studies with negative associations; studies were not uniformly small across any of the bodies of evidence; and a comprehensive search was conducted. We upgraded the evidence for a large magnitude of effect for Hg and Mn (most studies had effect estimates in the same direction and $>1.5$) as well as Cd (most studies had effect estimates in the same direction and $<0.66$). We also upgraded the evidence based on an observed dose response for Hg in three studies,^19,22,33 Mn in three studies,^22,24,26 Cd in one study,¹⁹ and Pb in one study.²² No other upgrade factors were applied. Overall, we assigned the evidence for Cd, Hg, and Mn a high quality rating, Pb a moderate quality rating, and As a low quality rating.

Table 2.

Summary of quality and strength of evidence for metals and neural tube defects.

	Mn	Pb	Cd	Hg	As
Quality factor
Downgrade factors
Risk of bias across studies	$-1$	$-1$	$-1$	$-1$	$-1$
Imprecision	0	0	0	0	0
Publication bias	0	0	0	0	0
Upgrade factors
Large magnitude of effect	1	0	1	1	0
Dose response	1	1	1	1	0
Confounding minimizes effect	0	0	0	0	0
Overall quality of evidence	1	0	1	1	$-1$
Strength considerations
Quality of body of evidence	High	Moderate	High	High	Low
Direction of effect	Increase	Increase	Decrease	Increase	Null
Confidence in estimates of association: “Is it likely or unlikely that a new study would change the trend/pattern observed in the high quality literature at the moment?”	Likely	Likely	Likely	Likely	Likely
Other compelling attributes of the data that may influence certainty	There are three studies that demonstrate a monotonic dose response. If you remove the study using hair as the exposure, the studies are very consistent.	None	Although one study did demonstrate a monotonic dose response, this is not consistent across studies.	There are three studies that demonstrate a monotonic dose response.	None
Overall strength of evidence (“sufficient,” “limited,” “inadequate,” or “evidence of lack of effect”)	Limited	Inadequate	Inadequate	Limited	Inadequate

Note: decrease, increasing exposure levels/decreasing risk; high, sum of quality factors is $\ge 1$; increase, increasing exposure levels/increasing risk; low, sum of quality factors is $\le -1$; moderate, sum of quality factors is 0.

Strength of Body of Evidence

We concluded that none of the metals had a literature supportive of conducting a meta-analysis. Specifically, the literature for all metals was such that we could not consider random error as the primary concern. In addition, heterogeneous exposure assessment and outcome assessment methodologies were used across studies.

We concluded that the overall direction of association was positive or, in other words, increasing levels of the metal increased the risk of any NTDs, for Hg, Mn, and Pb. We found that the direction of association was consistently negative for Cd and was null for As. For considering confidence in estimates of association, across the evidence for all metals, we deemed it likely that a hypothetical, well-designed new study that disagrees with the current trend would shift our assessment of the quality of the literature. For considering the strength of the evidence for Cd, whereas a high quality rating was given based on the review protocol, our confidence in the upgrade of the evidence for the dose response was weakened, compared with for Hg and Mn, given that the dose response was observed only in one study and was not consistent across studies. Taken together, we concluded that the current epidemiologic literature provides inadequate evidence for an association of exposure to As, Cd, or Pb with NTD prevalence. We determined that the current body of literature provides a limited evidence base for an adverse association of prenatal exposure to Hg or Mn on NTD prevalence. It is worth noting that studies on Hg included those assessing total Hg, and one specifically evaluating methyl-Hg (MeHg),³³ generally considered to be more toxic.⁴⁹ Total Hg includes inorganic Hg and MeHg, and can act as a good biomarker for MeHg in populations with high fish consumption. However, only two studies that assessed total Hg noted that the population was not a fish-consuming population^19,23; therefore, it was difficult to distinguish the effects of MeHg vs. inorganic Hg from the literature currently.

Literature on Metal Mixtures

There was only one included study that assessed the effect of exposure to metal mixtures and risk of NTDs.²⁴ That study evaluated essential metals, including only Mn in its mixture among the metals of interest in this review. Thus, no included study evaluated mixture effects among As, Cd, Hg, Mn, or Pb.

Discussion

Many regulatory agencies, including the U.S. Food and Drug Administration (FDA),⁵⁰ the U.S. Environmental Protection Agency (EPA),⁵¹ the National Institute of Environmental Health Sciences,⁵² the World Health Organization,⁵³ and the International Agency for Research on Cancer⁵⁴ make use of systematic reviews of epidemiologic studies to evaluate the safety of chemical exposures within the framework of regulatory decision-making. Because regulatory actions may stem from systematic reviews, it is important that robust approaches be applied to such studies, including prespecification of and transparency in methodology as well as bias identification and possible reduction. With regulatory consequences in mind, we adapted the Navigation Guide methodology, a systematic review approach for use in observational environmental health research, and summarized 30 epidemiologic studies of As, Cd, Hg, Mn, and Pb and NTDs.^15,16 There were two major findings. First, we determined that there was limited evidence of an adverse association between Hg or Mn exposure during pregnancy and prevalence of NTDs. Second, we concluded that there was inadequate evidence to assess a relationship between prenatal exposure to As, Cd, or Pb and prevalence of NTDs, thus highlighting the need for additional high-quality studies to make any regulatory decision based on the risk of NTDs upon As, Cd, or Pb exposure.

Our conclusion of limited evidence of an adverse association for prenatal Hg and Mn exposures during pregnancy and prevalence of NTDs reflects consistently positive associations observed across the current bodies of evidence for these metals, which were each deemed high quality. This denoting implies that we believe the evidence suggests there is an effect, but this judgment of the evidence could be convincingly overturned by contradictory, high-quality studies. These bodies of literature consisted of four (Hg)^19,22,23,33 to six (Mn)^{20,22,24,26,28,36} case–control studies using biomarker-based exposure assessment, the majority of which reported large magnitudes of adverse associations. In addition, for both Hg and Mn, three studies demonstrated monotonic dose responses.^{19,22,24,26,33} There was particularly strong agreement across high-quality studies of Hg measured in placenta^19,23,33 and Mn biomarkers other than hair,^22,24,26 although all these studies were limited to exposure assessment at birth, which may not accurately capture early pregnancy exposure, the period of most concern for NTDs.^55,56 Further and Mn on NTDs. In addition, assessments of effect measure modification are needed to assist with identifying population subgroups that are most vulnerable to these metals during pregnancy.

A special consideration is needed in interpreting findings for Hg given that most studies measured total Hg, which includes exposure to inorganic Hg (e.g., dental amalgams and preservatives) that do not cross the placenta as easily as MeHg.^33,57 Further, all our reviewed Hg studies were based in China, where placental Hg is generally higher than that in other countries,⁵⁸ so internationally relevant low-dose studies should be conducted to assess toxicity at current levels elsewhere. Regardless, we did note consistent, positive associations in the majority of studies assessing prenatal Hg and NTD prevalence. Putting this finding in the context of regulatory action, the FDA and the U.S. EPA issued joint statements on Hg in 2004 and 2019, detailing fish consumption guidelines based on MeHg exposure in the context of in utero exposure and neurodevelopment risk.^59,60 The statement encourages women who are or might become pregnant, breastfeeding mothers, and young children to limit fish consumption to no more than 12 oz ($340\text{g}$) of commercially caught fish per week and no more than 6 oz ($170\text{g}$) of locally caught fish per week.^59,60 Even though the evidence of an adverse association between prenatal Hg exposure and NTDs observed in this review is limited, it adds to knowledge of the prenatal neurotoxic potential of this metal and thus further emphasizes the need to avoid exposures as advised by the FDA and U.S. EPA.

One notable challenge for regulatory decision-making is that most chemical exposures do not occur in isolation. The Agency for Toxic Substances and Disease Registry (ATSDR) has identified the following priority mixtures: As, Cd, Cr, and Pb, as well as chlorpyrifos, Pb, Hg, and MeHg.⁶¹ Notably, even marginal evidence for single metal association with NTDs, as reported herein for Hg and Mn, could be highly relevant if synergistic effects occur in the presence of exposure to these common mixtures, or such evidence may be relevant without synergism in the case where small independent effects of multiple exposures can additively lead to larger joint effects. However, to date, the literature evaluating metal mixture effects on NTDs is extremely limited.

For all other metals examined (As, Cd, and Pb), the literature was deemed inadequate to determine the association with NTDs. This classification means that the literature is insufficient to assess the effect of the exposure, and that more high-quality studies are needed for any determination. The major area of concern generated from our assessment of the strength and quality of the literature was confounding bias. Notably, many studies included in this review adjusted for no confounders, and among those that did adjust for confounders, no study explained their approach to selection of confounders. This was especially true for studies that evaluated Pb: 12 (60%) of the 20 included studies simply presented a comparison of median or mean concentrations in cases or controls with accompanying $t$-tests or equivalents.

Crucially, with respect to metals, environmental measurements and biomarker measurements can differ substantially with respect to potential for confounding bias. In studies with environmental matrix, where exposures are attributed to individuals based on area-level concentrations detected in soil or water samples, a key potential confounder would be exposure sources (e.g., pollution sources) emitting copollutants that may truly be driving an effect. There is also potential confounding by factors that vary geographically along with exposure sources, such as socioeconomic characteristics.⁶² For instance, it is documented that populations living closer to Superfund sites in the United States are more likely to be low-income and racial minorities, vulnerable also to exposures connected to income inequality and systemic racism that may influence the risk of NTDs as well.^63,64 In contrast, when exposure measures derive from biomarker measurements (e.g., maternal urine), the potential for “physiologic confounding” becomes an additional concern, where (sometimes hard-to-capture) factors, such as metabolism, urine dilution, and blood volume, can influence biomarker measurements and reflect underlying physiologic differences that relate to the subsequent risk of a NTD, such as diabetes status and nutritional status (including folate levels).⁶⁵ Both biomarker and environmental measures are common in assessing health effects of metals exposure, emphasizing a need to address the potential for confounding differently for these two types of exposure assessment methods.

In addition to confounding bias, we identified selection bias as a major area of concern in the current literature evaluating metals and NTDs. One salient concern in studies of birth outcomes such as NTDs arises when the target population is the population of conceptuses, or very early stage fetuses. Often, studies of birth defects are restricted to live births or pregnancies that progress past some period of gestation (e.g., 20 wk). In such cases, the study population consists of pregnancies that last at least to the minimum gestational age, so that processes occurring prior to that time can lead to informative loss of fetuses from the target population. This phenomenon, which has been termed live birth bias, is illustrated in the directed acyclic graph (DAG) shown in Figure 7.^66,67 In this figure, restricting the study population to fetuses that survive past a certain gestational age, or to birth, means that exposures measured before fetal death can have spurious relationships with NTDs if there are unaccounted for factors (U) that can result in either fetal death or NTDs. This bias can potentially be reduced by analytic approaches to selection bias, such as inverse probability weighting, or by generating different target parameters, such as survivor average causal effects.^68,69 However, in many cases U must be known and well measured to do so, or the full population of conceptuses must be enumerated, and these challenges can be substantial. To some extent, all studies examined in this review were subject to potential bias from missing information on early pregnancy losses, although the potential magnitude of this bias is not well understood. Thus, rather than conclude that every study was subject to a high risk of bias for selection bias, we assessed studies relative to what could be feasibly done in studies of birth defects. If exposure to metals does increase the risk of early loss and if other factors also jointly increase both early loss and NTDs, then we would expect this bias to be toward the null such that prevalence ratios and odds ratios would underestimate true risk ratios and cumulative incidence ratios that would have been estimated in the absence of early losses.

Figure 7.

Directed acyclic graph describing live birth bias in the context of prenatal metal exposure and neural tube defects (NTDs). Note: U, unknown or unaccounted for factor.

The final major area of needed improvement identified was exposure assessment. In general, for all metals, articles included in this review collected samples after the critical window for NTDs, which predominantly develop in the first 4 wk of pregnancy.^55,56 It is challenging to capture biomarkers of exposure during this window given that many women do not yet know they are pregnant. Future studies could use maternal toenails as a biomarker of exposure, which are noninvasive, have been shown for As and Mn to be valid biomarkers of maternal–fetal transfer and have the potential to be time stamped to a window of exposure, such as early pregnancy.^70–72 In addition, studies could consider the use of children’s teeth, which can also be time stamped, although this would not be feasible for life-limiting birth defects.^73,74

The studies we reviewed also varied in sample collection, preparation, and analytical methods used for biomarker measurements, with some studies failing to provide details on laboratory procedures at all. This is critical because some of the biomarkers used are particularly susceptible to misclassification from sample collection and preparation methods. For example, past studies have shown that Cd concentrations vary throughout the placenta, and the way in which samples are collected and processed has an important impact on results.⁷⁵ In addition, hair samples can reflect both internal dose and external deposition of metals. Consistent methods for washing hair samples prior to processing are necessary to ensure externally deposited contaminants are removed.⁷⁰

Moreover, regardless of laboratory procedures, the matrices used to estimate exposure concentrations in many of the studies reviewed may not be optimal for the specific metals of interest. For instance, Cd is known to accumulate in the placenta.^58,76 Therefore, measured placental concentrations of these metals likely represent cumulative maternal exposures during gestation. Given that NTD cases have a high risk of preterm birth, the time period for accumulation to occur (i.e., gestation), can be much shorter in cases vs. controls.⁷⁷ This may lead to placental Cd being lower in cases vs. controls as an artifact of gestational age, rather than this metal truly being protective of NTDs. In the one study included that assessed placental Cd and NTD prevalence, indeed a negative relationship was found for both spina bifida and any NTDs.¹⁹ This accumulation of Cd in placenta may represent a bias in exposure assessment that explains the counterintuitive findings finding that Cd demonstrated a negative relationship with NTDs prevalence. This negative relationship was, however, also observed for infant serum collected at birth and census tract well-water concentrations; therefore, it is also possible that the accumulation of Cd in placenta tissue represents a true barrier to fetal exposure, thus protecting it from Cd toxicity.^35,41 Overall, although a negative relationship was observed for Cd, the strength of the evidence was deemed inadequate to make a formal judgment.

Although outcome assessment was judged to be less of an urgent concern for improving this literature, there are several relevant notes for considerations of future study designs. The articles included in this review varied in their outcome assessment methods. Some relied on a single trained health care worker to visually assess outcomes and identify cases, whereas others involved a team of physicians who reviewed and cross-validated outcomes, potentially using imaging and laboratory studies to assist with case identification. Many did not provide clear details about the outcome assessment methods, limiting our ability to compare across studies.^73,74 Misdiagnosis of less-severe, nonvisually observable birth defects (e.g., congenital heart defects) is a well-known issue.^78–82 This could lead to missed cases of less-severe NTDs, such as spina bifida. However, several studies have found that most NTDs (e.g., spina bifida, anencephaly, encephalocele), which are severe and easy to see, are less prone to misclassification and diagnostic variability than other birth defects.^78,81,82 Thus, although issues related to outcome assessment methods could have led to some misclassification or missed cases, this is less likely with NTDs, and misclassification would not be expected to vary by exposure status.

Despite the use of a transparent and thorough protocol, there are some limitations of this approach. Although we aimed to conduct a comprehensive review of prior literature, it is possible that we did not identify every relevant study. For instance, we did not search the gray literature for possibly relevant studies. For some studies, point estimates or precision measures were not available, and we were not able to obtain this information despite our attempts to contact study authors. We did not conduct a formal meta-analysis of study effect estimates nor a formal assessment of publication bias for any of the metals we assessed owing to the lack of a sufficient number of studies and the heterogeneity of the study methodologies. We did not review toxicologic studies to inform potential biological mechanisms and plausibility of relations under study because we considered this literature to be outside the scope of our review. Finally, our risk of bias decisions were based on our author team’s knowledge, expertise, and judgments, and it is possible that our conclusions may differ from those of another research team. As an example, there are some other potentially important confounders that could have also been considered in the risk of bias, including maternal obesity and diabetes, as well as folate status; however, we did not consider these based on our DAG analysis. Although not a limitation of this review, per se, it is worthy of note that many other categories of birth defects, including urinary system defects, genital organ defects, and digestive system effects, did not have an adequate number of studies to summarize the literature, thus pointing to literature gaps in need of further research.

Overall, we found that the literature examining relationships between prenatal exposure to As, Cd, Hg, Mn, and Pb and NTDs is sparse and generally in need of higher-quality studies. Future studies can improve the literature substantially by employing robust exposure assessment methods in relevant time windows and consider modern approaches to reducing or controlling confounding and selection bias. Although the overall strength of the evidence was inadequate to decide for other heavy metals, we found limited evidence of an adverse association between prenatal Hg or Mn and NTD prevalence, which speaks to the consistency of the results despite a multitude of potential biases. Overall, owing to limitations of the existing evidence base, the role of prenatal As, Cd, and Pb exposure in etiology of NTDs remains unclear and warrants further investigation in high-quality studies.