Arsenic Exposure and Blood Pressure Changes in Pregnancy
Arsenic Exposure and Blood Pressure Changes in Pregnancy
In January 2009, we began recruiting 18- to 45-year-old pregnant women receiving prenatal care at study clinics, as previously described (Gilbert-Diamond et al. 2011). Women were enrolled at 24–28 weeks gestation if they reported using water from a private well at their residence since their last menstrual period and were not planning to move prior to delivery. Only singleton births are included in the study. All protocols were approved by the Dartmouth College Institutional Review Board. All participants provided written, informed consent upon enrollment.
Participants completed a detailed medical history and lifestyle questionnaire upon enrollment and a follow-up questionnaire at 2 weeks postpartum to provide updated information about changes in key exposures and prenatal complications. After delivery, participants' medical records were reviewed to abstract pre- and post-delivery health information, including all clinically measured maternal BP levels, diagnoses of gestational diabetes, hypertension, preeclampsia and eclampsia. Other clinical information was recorded to verify self-reported medical and reproductive history. Maternal systolic (SBP) and diastolic (DBP) BP was measured in the study clinics, using either automated or mercury sphygmomanometers, throughout pregnancy and was generally recorded at each prenatal visit.
Women provided a spot urine sample upon enrollment, which was collected and stored as previously described (Farzan et al. 2013; Gilbert-Diamond et al. 2011). Urine samples were analyzed for levels of arsenite (iAs), arsenate (iAs), monomethylarsonic acid (MMA), dimethylarsinic acid (DMA), and arsenobetaine by high-performance liquid chromatography (HPLC) inductively coupled plasma mass spectrometry (ICP-MS) at the University of Arizona Hazard Identification Core (Larsen et al. 1993; Le et al. 2000; Wei et al. 2001). Samples that registered below the detection limit (ranging from 0.10 to 0.15 μg/L for individual species; 0.6%, 16.5%, and 37.0% of the study population were below the detection limit for DMA, MMA, and iAs, respectively) were assigned a value equal to the detection limit divided by the square root of 2. Urinary creatinine levels (milligrams per deciliter) were determined using Cayman's creatinine assay kit, according to the manufacturer's instructions. Our primary exposure measure was total urinary arsenic at 24–28 weeks gestation, calculated by summing inorganic (iAs = iAs + iAs) and organic (DMA, MMA) metabolites (Farzan et al. 2013; Gilbert-Diamond et al. 2011). Arsenobetaine, an unmetabolized form of arsenic found in seafood, was excluded because it is considered nontoxic (Tseng 2009). As secondary exposure measures, we examined the absolute values of urinary metabolites (MMA, DMA, and iAs). We also constructed primary (PMI) and secondary methylation indices (SMI) from ratios of MMA to iAs and DMA to MMA in urine, respectively, because these are considered indicators of methylation capacity that may impact individual variability in health effects of arsenic exposure (Chen et al. 2013). Upon enrollment, participants also were given instructions and prepaid mailing materials to collect samples of their home tap water and return the samples to the study office; these samples were analyzed by ICP-MS at the Dartmouth Trace Element Analysis Core, as previously described (Gilbert-Diamond et al. 2011). Maternal toenail samples were collected at 2 weeks postpartum, washed five times by sonication in a solution of Triton X-100 and acetone, followed by deionized water, and then dried before low-pressure microwave digestion. Samples were analyzed for trace elements previously related to BP (i.e., selenium, cadmium, iron, mercury, and lead) (Houston 2007; Kennedy et al. 2012; Wells et al. 2012) using ICP-MS as previously described for arsenic (Davis et al. 2014).
We confined our analysis to women without a history of hypertension prior to pregnancy with at least two pregnancy BP measurements. Our outcomes of interest were temporal changes in SBP, DBP, and pulse pressure (PP; SBP minus DBP) during pregnancy, which were analyzed as continuous variables with repeated measurements. For each measurement, we calculated the trimester and gestational week, based on the participant's last menstrual period. We restricted our analysis to measurements taken after 13 weeks gestation due to the low number of measurements recorded before this time. Measurements outside of a reasonable range (i.e., SBP: < 40 or > 250 mmHg, DBP: < 35 or > 180 mmHg) (Lee et al. 2012) were likely incorrectly recorded at time of measurement or incorrectly extracted from the medical record. All values that were excluded were well outside of the physiologically plausible range and were coded as missing (< 1% of measurements, n = 9). All other values recorded for these women were within a physiologically reasonable range. There were few cases of diagnosed pregnancy-induced hypertension (n = 15) or preeclampsia (n = 9) in our study population; thus, it was not possible to analyze these outcomes separately.
We fitted mixed-effect models (Demidenko 2004) of the repeated BP measurements to examine whether maternal urinary total arsenic or arsenic metabolite concentrations influenced SBP, DBP, and PP over the course of pregnancy, as follows:
where BPij represents BP at time i for subject j, As0j is urinary arsenic (total, DMA, MMA, or iAs) at baseline (time 0 represents baseline; i.e., the gestational month of each woman's first BP measurement after 13 weeks gestation) for subject j; TIME is gestational month of BP measurement; β1 is the coefficient for the association between TIME and BP when arsenic is held constant; β2 is the difference in BP for every unit increase in arsenic at baseline; β12 is the difference in monthly BP change over pregnancy per unit increase in arsenic (i.e., the estimated effect of arsenic levels on monthly BP change); α is a row vector of regression coefficients for covariates at baseline (T denotes vector transpose); Z0j is a vector of covariates at baseline. The random intercept μ0j and slope μ1j estimated the within-subject correlation among repeated measurements and between-subject heterogeneity, and rij is the error that cannot be accounted for by other covariates and random effects. The terms in the first and second brackets are the fixed and random parts of the model, respectively. We assessed nonlinear trends in the data using the same modeling strategy described above, including model terms to examine the interaction between TIME and categories of arsenic exposure variables (e.g., dummy variables for arsenic tertiles), as well as linearity of the time effect by including an additional interaction term between As0j and TIME. Neither test provided evidence of a nonlinear association (p > 0.05). For ease of interpretation, 5 μg/L (~ 1 SD) was used as the unit to report effect estimates for total urinary arsenic and metabolite levels.
Our models were adjusted for available covariates that could potentially influence BP based on a priori considerations, including age at enrollment, prepregnancy body mass index (BMI), smoking during pregnancy, marital status, educational attainment, gestational diabetes, parity, and number of BP measurements. As described above, we included the month of gestation during which each BP measurement was obtained in our models. We considered pregnancy BP measurements after 13 weeks gestation (our baseline) in our models because few subjects received BP measurements prior to that time point. Urinary arsenic concentrations were used as a measure of gestational arsenic exposure because urine samples earlier in pregnancy were not available and prior studies suggest that total arsenic concentrations remain relatively stable (Ahmed et al. 2011; Gamble et al. 2006). Because there is some debate as to whether creatinine adjustment is appropriate for urinary arsenic measures, we also tested models with and without urinary creatinine adjustment. We also tested inclusion of arsenobetaine levels as a covariate in our models. We found that neither creatinine nor arsenobetaine adjustment altered our estimates: Results were unchanged with or without creatinine adjustment {i.e., SBP β12 0.15 [95% confidence interval (CI): 0.02, 0.29] with creatinine adjustment}, as well as with or without arsenobetaine adjustment [i.e., SBP β12 0.15 (95% CI: 0.02, 0.29) with arsenobetaine adjustment] (data not shown). For individuals with missing covariate data (Table 1), we used multiple imputation to estimate missing covariate values (Little and Rubin 2002). We examined the missing data patterns; in our models we assumed that the data were missing at random with a monotone structure. We used the regression method within the SAS PROC MI procedure to generate five imputed data sets, then used the PROC MIANALYZE procedure to generate inferences for both the mixed and linear regression models. We also performed sensitivity analyses by excluding participants who smoked during pregnancy or those who developed gestational diabetes to evaluate the impact on our results, because BP may be altered in these groups (Bakker et al. 2010; Bryson et al. 2003; Carpenter 2007; Matkin et al. 1999). We also assessed other exposures from toenail levels as potential confounders. Toenail elements that have been associated with BP in the literature, such as selenium, cadmium, iron, mercury, and lead (Houston 2007; Kennedy et al. 2012; Wells et al. 2012), all had little to very weak correlations with toenail arsenic (r < 0.20) (data not shown) and thus were not adjusted for in our analysis.
We conducted analyses stratified by PMI or SMI, using the median values (0.89 and 9.66, respectively) as cut points, to assess whether the association between urinary arsenic and BP changes over time differed by these arsenic methylation indices. We also performed analyses stratified by age (below or at/above a median of 30.9 years), history of prior pregnancy (nulliparous or parous), prepregnancy BMI (< 25 or ≥ 25 kg/m).
Because BP increases over the latter part of pregnancy (Cunningham et al. 2010; Miller et al. 2007; Thompson et al. 2007), we further examined whether women with higher urinary arsenic had higher BP at the end of pregnancy, using linear regression models with the outcome, respectively defined as the average of the last three BP measurements (SBP, DBP, PP), adjusting for the same covariate variables. The equation generated from the multivariable linear regression model was also used to graphically represent the relationship between maternal urinary arsenic and SBP at the end of pregnancy, when all covariates are set equal to the median values (Figure 1). In all analyses, p-values < 0.05 were considered significant. All analyses were performed using SAS 9.3 (SAS Institute Inc.).
(Enlarge Image)
Figure 1.
Blood pressure measurements over pregnancy by gestational week. For each 2-week period, all systolic blood pressure (A), diastolic blood pressure (B), or pulse pressure (C) measurements during that time were averaged individually for each woman and then averaged across all women and plotted. Error bars represent the 95% CIs. Measurements prior to 6 weeks of gestation were excluded due to few available measurements.
Methods
The New Hampshire Birth Cohort
In January 2009, we began recruiting 18- to 45-year-old pregnant women receiving prenatal care at study clinics, as previously described (Gilbert-Diamond et al. 2011). Women were enrolled at 24–28 weeks gestation if they reported using water from a private well at their residence since their last menstrual period and were not planning to move prior to delivery. Only singleton births are included in the study. All protocols were approved by the Dartmouth College Institutional Review Board. All participants provided written, informed consent upon enrollment.
Participants completed a detailed medical history and lifestyle questionnaire upon enrollment and a follow-up questionnaire at 2 weeks postpartum to provide updated information about changes in key exposures and prenatal complications. After delivery, participants' medical records were reviewed to abstract pre- and post-delivery health information, including all clinically measured maternal BP levels, diagnoses of gestational diabetes, hypertension, preeclampsia and eclampsia. Other clinical information was recorded to verify self-reported medical and reproductive history. Maternal systolic (SBP) and diastolic (DBP) BP was measured in the study clinics, using either automated or mercury sphygmomanometers, throughout pregnancy and was generally recorded at each prenatal visit.
Arsenic Exposure Assessment
Women provided a spot urine sample upon enrollment, which was collected and stored as previously described (Farzan et al. 2013; Gilbert-Diamond et al. 2011). Urine samples were analyzed for levels of arsenite (iAs), arsenate (iAs), monomethylarsonic acid (MMA), dimethylarsinic acid (DMA), and arsenobetaine by high-performance liquid chromatography (HPLC) inductively coupled plasma mass spectrometry (ICP-MS) at the University of Arizona Hazard Identification Core (Larsen et al. 1993; Le et al. 2000; Wei et al. 2001). Samples that registered below the detection limit (ranging from 0.10 to 0.15 μg/L for individual species; 0.6%, 16.5%, and 37.0% of the study population were below the detection limit for DMA, MMA, and iAs, respectively) were assigned a value equal to the detection limit divided by the square root of 2. Urinary creatinine levels (milligrams per deciliter) were determined using Cayman's creatinine assay kit, according to the manufacturer's instructions. Our primary exposure measure was total urinary arsenic at 24–28 weeks gestation, calculated by summing inorganic (iAs = iAs + iAs) and organic (DMA, MMA) metabolites (Farzan et al. 2013; Gilbert-Diamond et al. 2011). Arsenobetaine, an unmetabolized form of arsenic found in seafood, was excluded because it is considered nontoxic (Tseng 2009). As secondary exposure measures, we examined the absolute values of urinary metabolites (MMA, DMA, and iAs). We also constructed primary (PMI) and secondary methylation indices (SMI) from ratios of MMA to iAs and DMA to MMA in urine, respectively, because these are considered indicators of methylation capacity that may impact individual variability in health effects of arsenic exposure (Chen et al. 2013). Upon enrollment, participants also were given instructions and prepaid mailing materials to collect samples of their home tap water and return the samples to the study office; these samples were analyzed by ICP-MS at the Dartmouth Trace Element Analysis Core, as previously described (Gilbert-Diamond et al. 2011). Maternal toenail samples were collected at 2 weeks postpartum, washed five times by sonication in a solution of Triton X-100 and acetone, followed by deionized water, and then dried before low-pressure microwave digestion. Samples were analyzed for trace elements previously related to BP (i.e., selenium, cadmium, iron, mercury, and lead) (Houston 2007; Kennedy et al. 2012; Wells et al. 2012) using ICP-MS as previously described for arsenic (Davis et al. 2014).
Statistical Analysis
We confined our analysis to women without a history of hypertension prior to pregnancy with at least two pregnancy BP measurements. Our outcomes of interest were temporal changes in SBP, DBP, and pulse pressure (PP; SBP minus DBP) during pregnancy, which were analyzed as continuous variables with repeated measurements. For each measurement, we calculated the trimester and gestational week, based on the participant's last menstrual period. We restricted our analysis to measurements taken after 13 weeks gestation due to the low number of measurements recorded before this time. Measurements outside of a reasonable range (i.e., SBP: < 40 or > 250 mmHg, DBP: < 35 or > 180 mmHg) (Lee et al. 2012) were likely incorrectly recorded at time of measurement or incorrectly extracted from the medical record. All values that were excluded were well outside of the physiologically plausible range and were coded as missing (< 1% of measurements, n = 9). All other values recorded for these women were within a physiologically reasonable range. There were few cases of diagnosed pregnancy-induced hypertension (n = 15) or preeclampsia (n = 9) in our study population; thus, it was not possible to analyze these outcomes separately.
We fitted mixed-effect models (Demidenko 2004) of the repeated BP measurements to examine whether maternal urinary total arsenic or arsenic metabolite concentrations influenced SBP, DBP, and PP over the course of pregnancy, as follows:
where BPij represents BP at time i for subject j, As0j is urinary arsenic (total, DMA, MMA, or iAs) at baseline (time 0 represents baseline; i.e., the gestational month of each woman's first BP measurement after 13 weeks gestation) for subject j; TIME is gestational month of BP measurement; β1 is the coefficient for the association between TIME and BP when arsenic is held constant; β2 is the difference in BP for every unit increase in arsenic at baseline; β12 is the difference in monthly BP change over pregnancy per unit increase in arsenic (i.e., the estimated effect of arsenic levels on monthly BP change); α is a row vector of regression coefficients for covariates at baseline (T denotes vector transpose); Z0j is a vector of covariates at baseline. The random intercept μ0j and slope μ1j estimated the within-subject correlation among repeated measurements and between-subject heterogeneity, and rij is the error that cannot be accounted for by other covariates and random effects. The terms in the first and second brackets are the fixed and random parts of the model, respectively. We assessed nonlinear trends in the data using the same modeling strategy described above, including model terms to examine the interaction between TIME and categories of arsenic exposure variables (e.g., dummy variables for arsenic tertiles), as well as linearity of the time effect by including an additional interaction term between As0j and TIME. Neither test provided evidence of a nonlinear association (p > 0.05). For ease of interpretation, 5 μg/L (~ 1 SD) was used as the unit to report effect estimates for total urinary arsenic and metabolite levels.
Our models were adjusted for available covariates that could potentially influence BP based on a priori considerations, including age at enrollment, prepregnancy body mass index (BMI), smoking during pregnancy, marital status, educational attainment, gestational diabetes, parity, and number of BP measurements. As described above, we included the month of gestation during which each BP measurement was obtained in our models. We considered pregnancy BP measurements after 13 weeks gestation (our baseline) in our models because few subjects received BP measurements prior to that time point. Urinary arsenic concentrations were used as a measure of gestational arsenic exposure because urine samples earlier in pregnancy were not available and prior studies suggest that total arsenic concentrations remain relatively stable (Ahmed et al. 2011; Gamble et al. 2006). Because there is some debate as to whether creatinine adjustment is appropriate for urinary arsenic measures, we also tested models with and without urinary creatinine adjustment. We also tested inclusion of arsenobetaine levels as a covariate in our models. We found that neither creatinine nor arsenobetaine adjustment altered our estimates: Results were unchanged with or without creatinine adjustment {i.e., SBP β12 0.15 [95% confidence interval (CI): 0.02, 0.29] with creatinine adjustment}, as well as with or without arsenobetaine adjustment [i.e., SBP β12 0.15 (95% CI: 0.02, 0.29) with arsenobetaine adjustment] (data not shown). For individuals with missing covariate data (Table 1), we used multiple imputation to estimate missing covariate values (Little and Rubin 2002). We examined the missing data patterns; in our models we assumed that the data were missing at random with a monotone structure. We used the regression method within the SAS PROC MI procedure to generate five imputed data sets, then used the PROC MIANALYZE procedure to generate inferences for both the mixed and linear regression models. We also performed sensitivity analyses by excluding participants who smoked during pregnancy or those who developed gestational diabetes to evaluate the impact on our results, because BP may be altered in these groups (Bakker et al. 2010; Bryson et al. 2003; Carpenter 2007; Matkin et al. 1999). We also assessed other exposures from toenail levels as potential confounders. Toenail elements that have been associated with BP in the literature, such as selenium, cadmium, iron, mercury, and lead (Houston 2007; Kennedy et al. 2012; Wells et al. 2012), all had little to very weak correlations with toenail arsenic (r < 0.20) (data not shown) and thus were not adjusted for in our analysis.
We conducted analyses stratified by PMI or SMI, using the median values (0.89 and 9.66, respectively) as cut points, to assess whether the association between urinary arsenic and BP changes over time differed by these arsenic methylation indices. We also performed analyses stratified by age (below or at/above a median of 30.9 years), history of prior pregnancy (nulliparous or parous), prepregnancy BMI (< 25 or ≥ 25 kg/m).
Because BP increases over the latter part of pregnancy (Cunningham et al. 2010; Miller et al. 2007; Thompson et al. 2007), we further examined whether women with higher urinary arsenic had higher BP at the end of pregnancy, using linear regression models with the outcome, respectively defined as the average of the last three BP measurements (SBP, DBP, PP), adjusting for the same covariate variables. The equation generated from the multivariable linear regression model was also used to graphically represent the relationship between maternal urinary arsenic and SBP at the end of pregnancy, when all covariates are set equal to the median values (Figure 1). In all analyses, p-values < 0.05 were considered significant. All analyses were performed using SAS 9.3 (SAS Institute Inc.).
(Enlarge Image)
Figure 1.
Blood pressure measurements over pregnancy by gestational week. For each 2-week period, all systolic blood pressure (A), diastolic blood pressure (B), or pulse pressure (C) measurements during that time were averaged individually for each woman and then averaged across all women and plotted. Error bars represent the 95% CIs. Measurements prior to 6 weeks of gestation were excluded due to few available measurements.
