Study Population

This study was part of our ongoing research on the consequences of environmental exposures early in life and was carried out in Matlab, a rural area of Bangladesh [15]. The International Center for Diarrhoeal Disease Research in Bangladesh (icddr,b) provides health services to the residents of the study area. It also has a well-established, population-based Health and Demographic Surveillance System (HDSS) that has been ongoing since the 1960s [16]. Community health workers visit households on a monthly basis to collect information on vital events such as pregnancies, births, deaths, and morbidities. Based on this information, we identified a population-based cohort of 3,971 women, who were pregnant between February 2002 and April 2003 (Figure 1). Pregnancy was identified by a urine test, most often in gestational week (GW) 9 (5^th–95^th percentiles=5–15 weeks). Exclusion criteria for the study included migration out of the area (n=42) and induced abortion (n=110). If women became pregnant twice during the study period (n=50), only the first pregnancy was included in the analysis. We also excluded women who were drinking surface water (n=219), as it is prone to microbial contamination and this could contribute to study outcomes. Thus, 3,550 pregnant women of the original 3,971 identified were eligible for the study.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow chart depicting selection of pregnancy cohort (February 2002 to April 2003).

https://doi.org/10.1371/journal.pone.0074119.g001

Ethics Statement

This study was approved by the ethical review committees of icddr,b, Bangladesh and the Karolinska Institutet, Sweden. Written consent was obtained from all subjects prior to enrollment in the initial study. Subjects were free to leave the study at any time.

Assessment of Manganese and Other Elements in Drinking Water

Because there is no suitable biomarker of Mn exposure [14], drinking water Mn was used as the measure of exposure. The drinking water samples for this study were available from a parallel, population-based study (AsMat), which evaluated the association between well water arsenic (As) and the presence of skin lesions [17]. Information on lifetime drinking water sources had been collected for all inhabitants over 4 years of age via household interviews. This information was used to identify water samples for the women in our cohort.

Drinking water samples were available for 2,264 (64%) of the 3,550 eligible pregnant women (Figure 1), and they were drawn from a total of 1,480 wells. The two primary reasons for missing drinking water samples (n=1,286) were (1) a lack of information on the relevant water source, because the pregnancy was identified after completion of the household water survey; or (2) that water samples were inadvertently discarded after the initial analysis of As. Changes in water source during pregnancy were common in this cohort due to a Bangladeshi tradition in which particularly primiparous women move to their parents’ homes for delivery. In addition, some families may have changed their water source in response to the identification of elevated As levels in their wells [18]. To minimize exposure misclassification stemming from well switching, we compared previously measured concentrations of drinking water As with the concentrations of As metabolites in urine, a useful biomarker of ongoing As exposure, at approximately GWs 8 and 30 [19]. We excluded women from our study if water and urinary As concentrations differed by more than 100 µg/L, which left 1,875 and 1,887 pregnancies with reliable exposure data for the assessment of association with fetal and infant survival, respectively (Figure 1). The correlation between As concentrations in drinking water and urine was strong in the women included in the study, both at GW 8 (Spearman’s correlation, r_s=0.85) and GW 30 (r_s=0.73).

Water samples were collected in 20-mL polyethylene vials after flushing the water with approximately 30 strokes of the pump. The vials contained 30 µL concentrated nitric acid (69% HNO_3, Suprapur, MERCK, Germany) to prevent precipitation of iron (Fe), Mn, and As. The samples were stored at -20°C in Matlab prior to being transported to the Karolinska Institutet, Sweden, for analysis of Mn and other elements, including As, Fe, magnesium (Mg), calcium (Ca), and zinc (Zn). Element concentrations were measured using inductively coupled plasma mass spectrometry (ICP-MS; Agilent 7500ce, Agilent Technologies, Tokyo, Japan) with the collision cell in helium mode [14]. Before analysis, 1% HNO₃ was added to the water samples at a ratio of 1:10. Analytical performance was ascertained by analysis of certified reference material (NIST 1643e, Trace Elements in Water, National Institute of Standards and Technology, Gaithersburg, MD), the results of which showed good agreement with the reference value.

Assessment of Outcomes

Information on all pregnancy outcomes (spontaneous abortion, induced abortion, stillbirth, and live birth) and infant deaths was obtained from the HDSS. We evaluated the potential effect of water Mn concentrations on spontaneous abortion and perinatal mortality. Spontaneous abortion was defined as unintended loss of the fetus before GW 28. Thus, no form of induced abortion was included. Perinatal deaths included both stillbirths and early neonatal deaths. Stillbirth was defined as the death of a fetus at 28 completed weeks of gestation or later. Early neonatal death was defined as the death of an infant within one week (seven completed days) after birth. The denominator used for calculating the rate of spontaneous abortion was the total number of pregnancies, whereas for perinatal mortality the denominator was the total number of live births and births of dead fetuses after GW 28.

Information on the cause of infant death was based on a verbal autopsy questionnaire developed by the World Health Organization (WHO) and adapted for the Matlab HDSS area [20]. Interviews were conducted 2–6 weeks after the death of an infant, and they were administered to a caretaker or relative who had lived with the infant during the terminal stage of illness or death. Three physicians independently reviewed the questionnaires to assign cause of death. Cause of death was coded based on the tenth revision of the International Classification of Disease, Disease Injuries and Cause of Death (ICD-10).

Assessment of Covariates

Information on women’s demographic characteristics (age, weight, height, date of birth, gravidity, education, and socioeconomic status) as well as place of delivery (home/health facility), infants’ sex and birth weight were available from the HDSS and from a parallel, population-based, randomized food and micronutrient supplementation trial (MINIMat) [21]. Maternal body-mass index (BMI), calculated based on women’s weight and height (kg/m²) at approximately GW 9, was used as a continuous variable. Gravidity was defined as the number of pregnancies a woman had experienced, including the present one. Maternal education was defined as the number of years of formal schooling. Socioeconomic status (SES) was based on a wealth index constructed from information on family ownership of a number of consumer items, construction type of the house, and dwelling characteristics (range: -5 to +5) [22]. It was used as a continuous variable. Gestational age in weeks was calculated by subtracting the date of the last menstrual period from the date of the pregnancy outcome. In the MINIMat trial, women were randomly allocated to two food groups and three micronutrient groups, the latter being initiated at GW 14 [21]. Because most of the women in the current sample participated in this trial, we tested if the supplementation influenced the analysis of perinatal outcome. Because marked seasonal differences in neonatal mortality have been reported in the study area, peaking in November [23], we included season of birth as a covariate, categorized as pre-monsoon (January–May), monsoon (June–September), or post-monsoon (October–December).

We assessed water concentrations of other essential or toxic elements (As, Fe, Zn, Ca, and Mg). Arsenic exposure has previously been shown to be associated with pregnancy outcomes in this area [15,24]. Water concentrations of the essential elements Fe, Zn, Ca, and Mg have also been associated with fetal growth and infant survival [25,26]. Finally, Fe in drinking water has been found to affect Fe status in another rural area in Bangladesh [27].

Statistical Analysis

We used analysis of variance (ANOVA) and χ² tests to assess differences in continuous and categorical variables, respectively, and Spearman’s rank correlation coefficients (r_s) to assess associations between continuous variables. We used multivariable-adjusted logistic regression models to estimate odds ratios (ORs) and 95% confidence intervals (CIs) of spontaneous abortion and perinatal mortality in relation to tertiles of water Mn concentrations. The lowest tertile of exposure was used as the reference. In addition, we estimated ORs after assessing Mn concentrations (mg/L) as a continuous variable. Potential trends were evaluated using the slope of the appropriate adjusted logistic regression model.

Models were adjusted for variables that were known or proposed risk factors for pregnancy outcomes, were significantly associated with both exposure and outcome (P <0.05), or changed effect size estimates by 10% or more. We assessed collinearity between covariates, and variables that were strongly correlated (r_s >0.70) were not included in the same model. In addition, we stratified all analyses by BMI 18.5 kg/m², the WHO cut-off for undernutrition [28]. Missing data on covariates were handled by complete subject analysis.

All tests were two-sided. Statistical significance was defined as a 95% CI that did not include zero or a P value < 0.05. Statistical analyses were conducted using SPSS (version 20.0, IBM Corporation, USA) and STATA (version 11, STATA Corp, College Station, TX, USA).

Demographic Characteristics

There were no major differences between the spontaneous abortion (Table 1) and perinatal death (Table 2) cohorts in terms of demographic characteristics or drinking water concentrations of elements. Approximately 25% of the women in both cohorts were primigravida, and 50% of the women had 1–2 children. About 31% of the women delivered at health facilities. Only two women reported smoking during pregnancy, and none used alcohol. In both cohorts, women with missing water samples were slightly younger (mean ±SD age: 26±6 years), were more educated (6±4 years at school), had fewer children (1.2±1.4 children), and were more often pregnant for the first time (36%) than women with water samples.

Overall mean and median concentrations of drinking water Mn were 615 µg/L and 217 µg/L, respectively, with the 5^th and 95^th percentiles being 28 µg/L and 2,326 µg/L (total range: 3–6,550 µg/L). Water Mn concentrations were inversely correlated with water concentrations of As (r_s=-0.46) and Fe (r_s=-0.25) and positively correlated with water Ca concentrations (r_s =0.49), but they were not significantly correlated with water Zn (r_s=0.15) or Mg (r_s=0.07) concentrations. Water Mn concentrations were higher in deeper wells (those >50 m; median=734 µg/L) than in shallow wells (those <50 m; median=126 µg/L). By contrast, As and Fe concentrations were higher in shallow wells, which had medians of 193 µg/L and 3.9 mg/L, respectively, than in deeper wells, which had medians of 2.6 µg/L and 1.2 mg/L, respectively. The concentrations of Ca, Zn, and Mg did not vary by the depth of the well; overall median concentrations were 42 mg/L, 18µg/L, and 20 mg/L, respectively.

Spontaneous Abortion

Of the 1,875 pregnancies included in the analysis, 158 (8.4%) resulted in spontaneous abortion, which occurred, on average, in GW 12 (range: GWs 5–27). Compared with the women included in the cohort, women with missing water samples or discordant water and urine As concentrations had a slightly higher frequency of spontaneous abortion (8.4% vs. 11%; P=0.005).

As shown in Table 1, women in the highest water Mn tertile had higher BMI, SES, and water Ca concentrations; were more educated; and had lower water As and Fe concentrations than women in the lowest Mn tertile. The adjusted OR for spontaneous abortion (Table 3) was significantly lower for women in the highest Mn tertile (median Mn=1,292 µg/L, OR=0.65, 95% CI 0.43–0.99; model I) compared with the women in the lowest tertile (median Mn=56 µg/L). Additional adjustment for BMI, SES, and mother’s education (model II), which reduced the sample size to 1,642, did not change the estimate (OR=0.66, 95% CI 0.40–1.10) for the highest Mn tertile. Using water Mn concentrations as a continuous variable, the adjusted OR was 0.75 (95% CI 0.58–0.98, model I) per 1 mg/L (1,000 µg/L) increase in Mn concentration, with a significant trend (P=0.034). Additional adjustment (model II) changed the OR only marginally (0.78, 95% CI 0.57–1.06) per 1 mg/L increase in Mn concentration; P for trend=0.109.

We did not adjust for maternal age due to its collinearity with gravidity; however, including the age of the mother instead of gravidity in the adjusted model gave similar results. Stratifying by BMI gave a model II-adjusted OR of 0.61 (95% CI 0.27–1.35) per 1 mg/L increase in water Mn for undernourished women (BMI <18.5; n=460) and an OR of 0.83 (95% CI 0.60–1.15) for women with BMI ≥18.5 kg/m² (n=1,182). We performed a separate analysis of spontaneous abortion in relation to gestational age (with a median split at GW 12) and obtained a model II-adjusted OR of 0.64 (95% CI 0.37–1.09) per 1 mg/L increase in drinking water Mn before GW 12 and an OR of 0.88 (95% CI 0.61–1.27) thereafter.

Perinatal Mortality

The 1,887 pregnancies included in the perinatal cohort resulted in 37 (2%) stillbirths and 1,850 (98%) live births. The mean gestational age at stillbirth and live birth was 36 weeks (range: 29–42) and 39 weeks (range: 23–47), respectively. Of the total 47 infant deaths during the first year of life, two-thirds (n=33) were early neonatal deaths, in accordance with another study conducted in the same geographical area [20]. Causes of early neonatal death included birth asphyxia (n=14), infections (n=7), complications of prematurity or low birth weight (n=8), congenital defects (n=3), and unspecified causes (n=1). Hence, 70 perinatal deaths, consisting of 37 stillbirths and 33 early neonatal deaths, were included in the analysis. Compared with the women included in the cohort, women with missing water samples or discordant water and urine As concentrations had similar rates of perinatal mortality (3.7% vs. 4.5%, respectively; P=0.233). As expected, the risk of perinatal mortality was significantly higher in women with preterm delivery (GW <37; n=242) than those who delivered at term (OR 7.72, 95% CI 4.73–12.61).

As shown in Table 2, women in the highest water Mn tertile had higher BMI, were more educated with higher SES, had higher water Ca concentrations, and had lower As and Fe concentrations than women in the lowest water Mn tertile. In the lowest Mn tertile, there were more deliveries after the monsoon than in other seasons or in other Mn tertiles (P=0.017). No significant association was observed between Mn concentrations in drinking water and perinatal mortality in an adjusted regression analysis (Table 4). Although lower odds of perinatal mortality (OR=0.69, 95% CI 0.28–1.71; model I) were observed among women in the highest water Mn tertile (median Mn=1,341 µg/L) than women in the lowest tertile (median Mn=55 µg/L), the OR estimate changed by 13% (from 0.69 to 0.78) after adjustment for BMI and place of delivery (model II; n=1,648). We did not include gravidity as a covariate because of collinearity with maternal age; however, we found similar results when we included gravidity rather than maternal age in the model.