Association of Arsenic with Adverse Pregnancy Outcomes/Infant Mortality: A Systematic Review and Meta-Analysis

1Arsenic contamination of drinking water, air, food, and beverages is one of the major global health problems that affect > 300 million people worldwide. This includes an estimated 13 million people in the United States and about 70 million people in Bangladesh. At concentrations > 50 μg/L, inorganic arsenic (iAs) has been associated with elevated risk of cancer (e.g., bladder, kidney, liver, lung, skin, prostate), cardiovascular diseases, high blood pressure, anemia in pregnancy, mortality from respiratory diseases in both adults and children, diabetes in adults, and neurodevelopment problems. At concentrations around 10 μg/L, which is considered safe by the World Health Organization’s (WHO) provisional guideline, iAs may still cause cancer in approximately 0.1–0.3% and increased systolic blood pressure in women 6 weeks postpartum. iAs easily crosses human and animal placenta and has been demonstrated to increase the risk of impaired fetal growth and infant mortality in laboratory animal studies. Several epidemiologic studies (e.g.,;; ) have examined the relation between arsenic and adverse pregnancy outcomes/infant mortality, and the findings are equivocal. Our understanding of arsenic exposure and adverse pregnancy outcomes is limited and, at best, fragmented. To our knowledge, no systematic review and/or meta-analysis has reported on the effect of arsenic on human pregnancy and infant health. Given the widespread low, moderate, and high arsenic exposure in the general population, an understanding of the impact of iAs on maternal and fetal health is relevant for public health policy.

2To fill this gap, we conducted a systematic review and meta-analysis of epidemiologic studies to examine the association between arsenic exposure and the risk of spontaneous abortion, stillbirth, preterm delivery, birth weight, and neonatal/infant mortality.

3Search strategy and study selection. This study was conducted according to the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) group. We searched PubMed (http://www.ncbi.nlm.nih.gov/pubmed) and Ovid MEDLINE (http://ovidsp.tx.ovid.com) (from 1946 through July 2013) and EMBASE (http://www.embase.com/login) (from 1988 through July 2013) databases (Figure 1), using the terms “arsenic,” “arsenicals,” “arsenite,” “arsenate” and “abortion, spontaneous,” “fetal mortality,” “preterm delivery,” “low birthweight,” “birthweight,” “infant mortality,” “neonatal mortality” (see Supplemental Material, “Search Strategy”). In addition, we searched the reference lists of reviews and potentially relevant articles. Two authors (R.Q. and F.A.A.) independently evaluated the articles. Studies that fulfilled the following a priori eligibility criteria were included if the study a) was an original study; b) was a cross-sectional, or a case–control or a cohort design; c) reported on any one or more of the following outcomes: spontaneous abortion, stillbirth, preterm delivery, birth weight, and neonatal/infant mortality; and d) presented data on arsenic exposure determined using environmental measures (arsenic in drinking water or airborne arsenic, or arsenic in soil), biomarkers, or indirect measures (e.g., residing in arsenic endemic area). Our exclusion criteria were a) a study was an experimental or a case report or a case series or a letter, b) a study was of arsenic compounds for which human exposure was unlikely [e.g., arsenic in roots of plants], c) a study used job title or living close to a smelter house as surrogate for arsenic exposure, and d) a study did not include our relationships of interest.

4If more than one report was published from the same study, the most recent study or the study using the best assessment of arsenic and/or outcome was included. For studies that reported estimates for more than one biomarker, the estimate for the most appropriate biomarker was preferred. The order of preference was as follows: nail > hair > urine. If a study provided estimates for water and a biomarker, the estimate from the latter was used.

5Data extraction and quality assessment. Most relevant characteristics of eligible studies including study design, study size, location and country of study, method of arsenic assessment, exposure marker for arsenic, exposure contrast, exposure dose, type of adverse pregnancy/infant mortality and their definitions, year of publication, year of data collection, adjustment for adequate confounders, and study results (i.e., measures of association) were recorded in a standard data extraction form independently by two authors (R.Q. and F.A.A.). Any discrepancies were resolved by consensus. R.Q. and F.A.A. applied the Newcastle–Ottawa Scale for observational studies to assess quality of eligible studies, with the maximum score of 9. Studies scoring ≥ 7 were categorized as high quality (see Supplemental Material, Tables S1 and S2).

6Statistical methods. First, odds ratio (OR) or relative risk (RR) and their 95% confidence intervals (CIs) were derived or abstracted from eligible studies. Almost all the studies presented ORs and their 95% CIs; therefore, we used ORs in our analysis. One study presented adjusted ORs and exact p-values but not the 95% CI, so we calculated the 95% CI from the p-values following. Two studies presented RRs, and these were converted to ORs. Some relevant studies presented ORs for more than two exposure levels for the outcomes; therefore, in the meta-analysis we calculated summary ORs comparing our outcomes of interest (adverse pregnancy outcomes/infant mortality) in the highest (exposed group) and lowest (reference group) arsenic exposure categories presented in the studies. The exposed groups were heterogeneous and consisted of populations exposed to arsenic dose above the WHO guideline (i.e., > 10 μg/L). We separated the groups into high dose (exposed to ≥ 50 μg/L) and low-to-moderate dose (exposed to < 50 μg/L) for further analysis (explained below).

7We applied the random-effects model because we anticipated heterogeneity in the study-specific estimates. In the forest plots, we presented summary ORs of the random-effects model. Heterogeneity was computed using the Q (p < 0.1 considered significant), and I2-statistics (I2-statistic > 50% indicates high, 25–50% moderate, and < 25% low heterogeneity). We examined the influence of various characteristics on the study-specific effect estimates by stratifying the analysis by a) arsenic dose [i.e., high arsenic dose (> 50 μg/L) vs. low-to-moderate arsenic dose (< 50 μg/L)] and b) arsenic measured using individual data versus group data. We also performed a series of sensitivity analyses. First, we investigated the relative influence of each study on the summary OR by omitting each study one at a time. None of the studies had substantial influence on the summary ORs for our relations of interest and this finding was not reported. Second, we restricted the analysis to high-quality studies, prospective cohort studies, and studies adjusting for potential adequate confounders (see Supplemental Material, Table S3) documented in the literature (e.g.,;;,;;; ). We also presented dose response for studies with at least three exposure levels graphically. Publication bias was explored with funnel plots. The trim and fill method was used to assess the potential impact of missing studies in the funnel plot. Statistical analysis was performed using STATA software version 9 (StataCorp, College Station, TX, USA).

8Study characteristics. Our systematic literature search strategy is shown in Figure 1. A total of 888 studies were retrieved, from which 56 studies were reviewed in depth. Twenty-three studies fulfilled our a priori inclusion criteria (Table 1) for this systematic review. Data from 16 studies were included in our quantitative analysis (see Supplemental Material, Tables S4 and S5). Thirty-three studies were excluded for various reasons (see Supplemental Material, Table S6). We had no data from the authors of two studies on birth weight to calculate 95% CIs, so we included the studies only in our qualitative analysis. Five studies did not control for potential confounders, and they were also included in our qualitative analysis. Five studies were ecological retrospective cohort designs. Two studies were ecological case–control designs. Only 1 of the 5 prospective cohort designs was an ecological study. Seven of the 10 cross-sectional designs were ecological studies. Nine studies were conducted in Bangladesh, 5 in India, 3 in China, 2 in Chile, 1 in Taiwan, and 3 in the United States. Twenty-two studies were conducted in populations exposed to arsenic in drinking water. Of these, 6 applied biomarkers including urine, maternal/cord/placenta blood, hair, and nail; and the remaining studies measured arsenic dose at the region/village/household level. measured arsenic dose in airborne emissions. measured maternal hair arsenic dose at first prenatal visit, maternal hair arsenic dose at birth, and maternal nail arsenic dose at first prenatal visit, but the estimates of the latter biomarker was considered suitable and included in the meta-analysis. There were 10 reports on spontaneous abortion, 14 on stillbirth, 3 on preterm delivery, 6 on birth weight, 5 on neonatal mortality, and 7 on infant mortality. Eligible studies applied either questionnaire/interview or hospital/medical records or national registers to ascertain information on the outcomes of interest. Most of the studies (Table 1) scored low on the Newcastle–Ottawa Scale for several reasons, including bias associated with selection of study population, measurement of arsenic exposure, lack of individual arsenic data, inappropriate definition of cases/controls, inappropriate comparable reference, and a lack of adequate adjustment for potential confounders (Table 1).

9Arsenic exposure in the general population. Spontaneous abortion. In all 10 studies that examined the association with spontaneous abortion, 4 were excluded from our quantitative analysis because the authors did not control for potential confounders. studied outcome of pregnancy in 240 married women. In women with the highest concentrations of arsenic in drinking water (501–1,200 μg/L), there was an increase in spontaneous abortion. A similar observation was noted by,, and. Six studies provided data for our quantitative analysis (see Supplemental Material, Table S4). All the studies reported ORs. Summary OR in populations exposed to high arsenic dose (> 50 μg/L) in groundwater showed increased association (OR = 1.98; 95% CI: 1.27, 3.10) (Figure 2A). Our finding in populations exposed to low to moderate arsenic in groundwater or in public tap water was inconclusive (Figure 2B). Overall, the summary OR was 2.02 (95% CI: 1.40, 2.91). The direction and magnitude of the association persisted in studies applying biomarkers/individual arsenic data, prospective studies, studies adjusting for adequate potential confounders, and high-quality studies (Table 2). Figure 3A shows the dose–response relation of arsenic in drinking water and spontaneous abortion. The risk trend was not consistent across the studies. A funnel plot suggested influence of publication bias (see Supplemental Material, Figure S1A), and an adjustment with the trim and fill method did not change the strength of the overall summary OR (Table 2).

10Stillbirth. Of the 14 studies reporting association with stillbirth, 5 were excluded from our quantitative analysis because the authors did not control for potential confounders. All 5 studies observed an increase in stillbirth in women who had the highest concentrations of arsenic in their drinking water. Nine studies examining the association between environmental arsenic and stillbirth provided data for our quantitative analysis (see Supplemental Material, Table S4). Two studies reported RRs. Arsenic was measured in groundwater in 8 studies and in air in 1 study. Summary OR in populations exposed to high arsenic dose (> 50 μg/L) in groundwater was increased (OR = 1.77; 95% CI: 1.32, 2.36; Figure 4A). Only 1 study was conducted in a population exposed to low-to-moderate arsenic dose (Figure 4B). The overall summary OR for environmental arsenic was 1.84 (95% CI: 1.38, 2.45). In subgroup/sensitivity analyses, the risk of stillbirth was increased in studies applying biomarkers/individual arsenic data, studies using group data on arsenic, studies adjusting for adequate potential confounders, prospective studies, and high-quality studies (Table 2). Five studies reported a dose–response relationship between environmental arsenic and stillbirth (Figure 3B). Risk trend was consistent in 2 studies in high arsenic dose area, but this trend was not obvious in the other studies. A funnel plot suggested influence of small positive studies (see Supplemental Material, Figure S1B). The trim and fill method for adjustment of publication bias imputed 4 studies, and, as expected, the strength of the summary OR was attenuated but remained statistically significant (Table 2).

11Preterm delivery. In all, three studies investigated the relation between arsenic exposure and preterm delivery (see Supplemental Material, Table S4). They all reported ORs and were conducted in populations exposed to high arsenic dose in groundwater. The finding of the summary OR was inconclusive (OR = 1.41; 95% CI: 0.83, 2.41) (Figure 5A).

12Birth weight. The association between arsenic and birth weight was examined in six studies. Two studies did not provide sufficient quantitative data for the meta-analysis. measured arsenic dose in maternal urine (UAs) and observed an inverse dose response (coefficient: β = –1.3) between UAs and birth weight. also observed that newborns of mothers whose UAs was > 5.30 μg/L weighed on average 0.22 kg less than those of mothers whose UAs was ≤ 5.30 μg/L. Four studies provided regression coefficient and standard errors for our quantitative analysis (see Supplemental Material, Table S4). Environmental arsenic shows a significant reduction in birth weight (–53.2 g; 95% CI: –94.9, –11.4) (Figure 5B).

13Neonatal mortality. Five studies examined neonatal mortality (see Supplemental Material, Table S5), of which two reported RRs. All the studies were conducted in populations exposed to high arsenic dose in groundwater. The overall summary OR was 1.51 (95% CI: 1.28, 1.78) (Figure 5C). The direction of association did not change in studies applying biomarkers/individual arsenic data and in studies adjusting for adequate potential confounders (Table 2). reported on this relation. Dose response was examined in three studies (Figure 3C.) A consistent dose–response trend was observed by, but the risk trend was inconsistent in studies by and. Evidence of publication bias was observed in the funnel plot (see Supplemental Material, Figure S1C). The trim and fill method imputed two studies, and the strength of association was reduced marginally (Table 2).

14Infant mortality. Arsenic and infant mortality was investigated in seven studies (see Supplemental Material, Table S5), with two reporting RRs. The studies were conducted in populations exposed to high arsenic dose in groundwater. Summary OR was 1.35 (95% CI: 1.12, 1.62) (Figure 5D). Compared with that of the overall summary OR, the association was slightly elevated in studies applying biomarkers/individual arsenic data, studies adjusting for adequate potential confounders, and high-quality studies, but marginally reduced in studies using group data on arsenic (Table 2). Our findings from two prospective studies were inconclusive. Among three studies examining dose response (Figure 3D), a consistent risk trend was observed by, but the risk trend was not consistent in studies by and. A funnel plot showed evidence of asymmetry, suggesting influence of small positive studies (see Supplemental Material, Figure S1D). As expected, the strength of association was attenuated with the trim and fill method, and three missing studies were imputed (Table 2).

15This is the first systematic review and meta-analysis on the association between iAs exposure and adverse pregnancy outcomes/infant mortality. We found positive associations of arsenic with spontaneous abortion, stillbirth, birth weight, and neonatal and infant mortality. These findings are important to many countries around the globe where pregnant women and infants continue to be exposed to low through moderate to high arsenic dose in different media (e.g., drinking water, air, food, beverages).

16Validity issues. Our study has a number of strengths. We searched several databases including reference lists of reviews and relevant studies. Two authors independently checked the eligibility of the studies according to a predefined set of criteria. We followed systematically the guideline of PRISMA.

17Because the upper limits of arsenic exposure differed among studies, we studied the effect in populations in low-to-moderate arsenic areas (i.e., < 50 μg/L) separately from the effect in populations in high arsenic areas (i.e., ≥ 50 μg/L). In studies of spontaneous abortion and stillbirth, the findings were inconclusive.

18We excluded from our quantitative analysis small ecological studies that did not adjust for potential confounders. Also, in considering our core and additional confounders various studies should have adjusted for, we followed recommendations in the literature (e.g.,;;,;;;; ).

19While acknowledging the importance of our findings, we note a number of limitations. First, the use of summary scores to identify high-quality studies in Newcastle–Ottawa Scale is a bit problematic. A risk of bias tool that applies a domain-based evaluation may allow one to explore the influence of each domain on the overall summary effect estimate. Second, a well-designed study may be categorized as low quality because the authors failed to provide detail information in the publication. Finally, some items of the Newcastle–Ottawa Scale such as representativeness of study cohort with respect to community and duration of follow-up do not belong to the risk of bias tools. Thus, the interpretation of how well a study does on the Newcastle–Ottawa Scale in our study should be done with caution. Inclusion of ecological studies in our review may lead to underestimation of our observed associations. Also the studies incorporated in this meta-analysis were different with regard to exposure levels in the reference groups. However, in computing the overall summary OR from the different studies, we made an implicit assumption that any differences in exposure levels in the reference groups will not have much influence on our summary OR.

20We observed substantial heterogeneity in the study-specific estimates for studies on spontaneous abortion and stillbirth, and moderate heterogeneity for studies on neonatal and infant mortality. In stratified analysis, heterogeneity persisted in studies applying biomarkers for the association with spontaneous abortion, stillbirth, and infant mortality. In sensitivity analyses, heterogeneity persisted in prospective studies on infant mortality, studies on stillbirth and infant mortality that have controlled for adequate potential confounders, and high-quality studies on neonatal and infant mortality. The original studies also applied different exposure assessment methods and incorporated different exposure contrasts, thus making it difficult to relate any exposure increase to change in birth weight. Differences in responses to arsenic exposure may also exist across study populations, and these could be potential sources of the observed heterogeneity. We lacked data on these factors and we also did not have sufficient data from the original studies to elaborate further the reasons for the heterogeneity. We applied the trim and fill method to examine the impact of publication bias on our overall summary OR, and the summary OR was slightly reduced for stillbirth, neonatal mortality, and infant mortality, suggesting that publication bias is not an explanation of our observed associations. Nonetheless, the trim and fill method performs poorly in the presence of substantial heterogeneity, so the influence of publication bias on the observed associations cannot be ruled out.

21Our findings, however, should be interpreted in the light of limitations inherent in the original studies. Some studies failed to adjust for appropriate potential confounders of adverse pregnancy outcomes/infant mortality and could not establish the independent role of arsenic. Although few studies adjusted for proxies of socioeconomic status, only one study considered access and utilization of prenatal care. This is an important socioeconomic factor to be considered in the studies of stillbirth and neonatal/infant mortality. Exposure assessment was also a major challenge in the studies. Three studies measured arsenic contents in urine. One study measured arsenic content in blood, and measured arsenic content in hair. Arsenic content in urine/blood is a marker of current exposure, whereas information on chronic exposure can be obtained from arsenic content in hair or finger/toe nails. The remaining studies applied ecological measures. Questionnaires were administered in most studies, but data on water consumption pattern (i.e., the frequency and quantity of water intake) were not reported. Lack of individual data may result in measurement error with underestimation of the true effect. Many of the studies were cross-sectional in design, precluding temporality. Although few studies have collected data on our outcomes of interest from medical records/established registers, most studies have relied on maternal recall. Methods applied in collecting data on spontaneous abortion were not sensitive enough to detect events occurring in early pregnancy. Thus, the fetal and infant health effect of arsenic observed in our study may have been substantially underestimated.

22Comparison with previous studies. Only two qualitative reviews were available on this subject. In the first study, examined the effects of arsenic and chromium in drinking water on low birth weight and infant mortality. The authors identified 10 studies and failed to reach any conclusion. In the second study, examined the association of spontaneous abortion with arsenic in drinking water. The authors also identified 9 studies and concluded that chronic exposure to arsenic was associated with spontaneous abortion. In the present study, we observed excess risk of 102% for spontaneous abortion, 84% for stillbirth, 51% for neonatal mortality, 35% for infant mortality, and a 53-g reduction in birth weight. The magnitude of association persisted in studies applying biomarkers, studies using aggregate data on arsenic exposure, studies adjusting for adequate potential confounders, and high-quality studies. From the global public health point of view, the observed association is relevant considering the magnitude of the estimated effect and the extent of exposure to arsenic globally.

23The precise biologic window of susceptibility of arsenic for adverse pregnancy outcomes is unknown. But arsenic exposure at different periods before or during pregnancy could cause a wide range of adverse pregnancy outcomes. In laboratory animals, prenatal arsenic exposure causes spontaneous abortion by defective implantation or zygote development and aneuploidy, or through aberrant placental vasculogenesis and placental insufficiency. Epidemiologic studies have also shown that arsenic causes oxidative stress, lipid peroxidation, interference of hormonal activities, and perturbation of DNA methylation, which may be associated with a wide range of adverse pregnancy outcomes through defective placentation and preeclampsia.

24Our findings suggest that the effect of arsenic is strongest for spontaneous abortion. Although methylation is expected to have improved dramatically in the second trimester, at high arsenic dose (≥ 50 μg/L) observed in the populations included in our study, methylation is inhibited and the fetus blood plasma may essentially contain unmethylated arsenic and monomethylarsonic acid, which could threatened fetal survival and growth. Exposure to arsenic in utero and in early life may also pose a threat to infant survival. This observation has been noted in series of cohort studies conducted in the developing countries.

25Studies with the greatest weight in the meta-analyses did not provide data for the evaluation of dose–response trend. However, in the few studies that provided data, we observed inconsistent dose–response trend at high arsenic dose. The evidence was scarce for low to moderate arsenic dose and for studies evaluating preterm delivery.

26Our systematic review and meta-analysis found positive associations of arsenic exposure with spontaneous abortion, stillbirth, birth weight, and neonatal and infant mortality. However, the interpretation of causal association of high arsenic dose in drinking water is limited by methodological problems in the original studies and limited studies on dose response.