Abstract
Goal:
The goal of this study was to evaluate the association between groundwater arsenic and fasting blood glucose in the context of other groundwater chemicals, in Bangladesh.
Methods:
Fasting blood glucose, gender, body mass index, sociodemographic variables, and diabetes medication use were measured among adults ≥ 35 years of age (n=6,587) participating in the Bangladesh Demographic and Health Survey (BDHS) 2011. Groundwater chemicals in 3,534 well water samples were measured in the British Geological Survey (BGS) and Department of Public Health Engineering (DPHE) 1998–99 survey. We assigned the nearest BGS-DPHE well’s chemical exposure to each BDHS participant. We used survey-estimation linear regression methods to model natural log-transformed fasting blood glucose, among those using groundwater as their primary drinking-water source, as a function of groundwater arsenic. We considered possible interactions between categorical arsenic exposure and each of 14 other groundwater chemicals dichotomized at their medians. The chemicals considered as possible effect modifiers included: aluminum, barium, calcium, iron, potassium, lithium, magnesium, manganese, sodium, phosphorous, silicon, sulfate, strontium, and zinc.
Results:
Compared to persons exposed to groundwater arsenic ≤ 10 μg/L, the adjusted geometric mean ratio (GMR) of fasting blood glucose was 1.01 (95% confidence interval: 0.98, 1.04) for individuals exposed to groundwater arsenic concentrations >10 μg/L and ≤50 μg/L, and was 1.01 (0.97, 1.03) for those with > 50 μg/L arsenic. There were no Bonferroni-significant interactions with other chemicals, after accounting for the large number of chemicals tested as modifiers.
Conclusions:
In our analysis of groundwater chemistry data from 1998/99 and fasting blood glucose outcomes measured in nearby populations approximately a decade later, there was no overall association of fasting blood glucose with nearby historical groundwater arsenic. This null association was not significantly modified by the historical levels of other groundwater chemicals. These null results are inconclusive regarding shorter-term potential toxicity of arsenic for glucose regulation, if there are differences between the historical concentrations measured in nearby groundwater and the actual drinking water chemical exposures in the population during the etiologically relevant period for more acute phenotypes like fasting blood glucose. Drinking water supply-relevant, longitudinal exposure assessment with less measurement error is needed to more precisely evaluate the joint impacts of drinking water chemicals and establish if there is a sensitive time window for glycemic outcomes.
Keywords: cumulative risk assessment, exposure mixtures, medical geology, epidemiology, diabetes, Exposome
1. Introduction
Efforts to reduce water-borne infectious diseases caused by drinking contaminated surface water led to millions of shallow wells to be built in Bangladesh during the last 30 years, triggering mass population exposures to geogenic arsenic [1, 2]. This has caused nearly 35 million people to be exposed to arsenic exceeding Bangladesh’s 50 μg/L standard, and 57 million people exposed to a concentrations exceeding the World Health Organizations standard of 10 μg/L [3], and has led to considerable arsenic poisoning and related diseases [1]. It is not yet firmly established whether a diabetes epidemic is one of the diseases driven by the mass arsenic exposure in Bangladesh. The 2012 United States National Toxicology Program workshop review on arsenic and diabetes concluded that there was limited-to-sufficient support for an association between arsenic and diabetes in highly exposed populations[4], but the evidence from Bangladesh reviewed in that study came from four non-representative samples, the largest of which relied on self-reported diabetes status prior to baseline and reported a null association [5]. There have been studies of urine arsenic biomarkers and biological measures of diabetes pathophysiology in nationally representative survey samples of the United States [6, 7], Canada [8], and Korea [9], but to our knowledge there has not yet been a nationally representative survey sample-based investigation of arsenic exposure and biological measures of diabetes pathophysiology in Bangladesh.
Much of the evidence from Bangladesh regarding an association of drinking water arsenic exposure with biological measures of diabetes pathophysiology is consistent with a possible effect but is limited by non-representative samples and limited control for potential confounders. A cross-sectional study examining n=1,014 persons (n=43 with diabetes, n=971 without) drinking tube-well water in western Bangladesh showed monotonically increasing, significant associations of estimated cumulative arsenic intake from groundwater (based on concurrent well water measures and self-reported history of use of wells) with the prevalence of diabetes, after controlling for age and sex considering body mass index (BMI) as a potential confounder, and defining diabetes using self-reported history of symptoms, self-report of previous diagnosis, glucosuria, and oral glucose tolerance test findings [10]. Another cross-sectional study in four arsenic-endemic villages used concurrent well-water arsenic measures and history of well use to construct a cumulative exposure index and found a significant, monotonically increasing prevalence of glucosuria with increasing well-water arsenic exposure, among n=430 persons with skin lesions and again separately among n=1,165 persons without skin lesions, and controlling for age and sex and considering BMI as a possible confounder [11]. A case-control study in northwestern Bangladesh of n=115 arsenicosis patients from four rural villages with high arsenic, n=80 geographically matched controls with safe drinking water, and n=40 additional controls from cities, estimated that the prevalence of having random serum glucose > 140 mg/dL among arsenicosis patients was 2.8 times higher than controls [12].
A much more rigorous study focused on arsenic and diabetes in Bangladesh was conducted in 2009 among n=1,004 persons age ≥ 30 years old without chronic diseases other than diabetes, recruited in a stratified cluster sample (participation rate 59.7%) drawing 100 households from each of seven either high or low arsenic districts in Bangladesh to ensure an arsenic exposure gradient, but not necessarily to be reflective of the overall Bangladeshi population [13], This study estimated the association between a cumulative exposure index (based on the household’s water supply tube well arsenic measurements and the duration of well use) and prevalent diabetes (defined by fasting glucose ≥126 mg/dL or self-reported previous diagnosis) to be a prevalence odds ratio of 1.9 (95% confidence interval [CI]: 1.1 – 3.5), controlling for age, sex, BMI, education, family history of diabetes, and a random effect for tube well, and using robust standard errors. Other measured potential confounders were also considered in models but deemed not significant.
Biomarkers measuring arsenic toxicokinetics are known to be affected by many environmental factors [14] but drinking water co-exposures have not been extensively studied as predictors of arsenic toxicokinetic differences between people that could moderate arsenic toxicity. For specific health outcomes (e.g., glucose regulation) it is possible that there also could be toxicodynamic interactions of arsenic with other drinking water chemicals if the co-exposures act on similar pathways, which has not been extensively studied, although it is possible to speculate about possible toxicodynamic mechanisms based on previous toxicological studies. There are many possible pairwise chemical interactions that could be investigated through toxicological experiments, so an agnostic screen for interactions using existing epidemiological data might help prioritize specific targets for future toxicological experimental studies.
The goal of this hypothesis-generating study is to assess the cross-sectional association of groundwater arsenic with fasting blood glucose among individuals 35 or older near wells with moderate (As ≤ 10 μg/L), high (10 < As ≤ 50 μg/L), or very high (As > 50 μg/L) elevated environmental groundwater arsenic and other groundwater chemicals for glycemic outcomes in a sample representative of Bangladesh.
2. Methods
2.1. Source Population
The Demographic and Health Surveys (DHS) are nationally-representative household surveys that collect data on a wide range of indicators pertaining to population demographics, health status, and nutrition [15]. The 2011 Bangladesh Demographic Health Survey (BDHS) was a two-stage stratified cluster sample designed to allow for valid inferences at national scale; regional scale (for the seven administrative districts); or for urban (including sub-strata of city-corporations and urban areas that were not city-corporations) or rural subpopulations of Bangladesh in 2011 [15]. The seven administrative divisions of Bangladesh were sub-divided into sampling strata of city-corporations, urban areas that are not city corporations, and rural areas; a total of 600 clusters were sampled to include clusters within each of these urban/rural administrative divisions (except Rangpur has no city corporations). Then, from each of the 600 selected clusters, 30 households were sampled, for a total of ~18,000 residential households sampled. The survey weights provided for BDHS data analysis are normalized inverse-probability-of-selection design weights corrected for nonresponse.
2.2. Clinical and Demographic Data
Data collection for the BDHS began July 2011 and finished in December 2011 [15]. The survey was conducted by 16 interviewing teams, each with one supervisor, one field editor, and female and male interviewers [15]. Information on clinical and demographic factors relevant to fasting blood glucose were collected by the BDHS from 18,000 residential households, and included: age, sex, educational level, current smoking status, rural or urban residence, geographical region, household wealth, drinking water sources, and whether individuals were taking medications to treat diabetes. A subsample of one-third of the households were surveyed, and eligible members were selected to participate in the biomarker testing components, blood glucose testing, height and weight measurements. BMI was calculated from measured weight and height as kg/m2 [15].
We suspected socio-economic status could be an important confounder of the association between drinking water chemistry and diabetes-related traits, so we controlled for wealth and food security as indicators of socio-economic status. The wealth index serves as an indicator of household level wealth that is consistent with expenditure and income measures. The wealth score is constructed in three steps by inventorying household assets and summarizing by principal components analysis [15]. In the first step, principal components are calculated among a subset of household assets commonly seen in households across the country. In the second step, separate principal component scores were calculated for urban or rural households based on assets specific to urban, or to rural settings. In the third step, a wealth index was produced by regressing the area-specific asset scores on the general asset scores [15].
To gauge food security, the BDHS collected data from eligible participants using the Woman’s Questionnaire. These questions were developed from the 2011 Nepal DHS food insecurity module and the Household Food Insecurity Access Scale indicators established by USAID’s Food and Nutrition Technical Assistance project, and were modified to be specific to Bangladesh [15]. The kinds of questions asked included, “How often did you eat three `square meals’ (full stomach meals) a day in the past 12 months (not a festival day)?”, and “In the last 12 months, how often did you or any of your family have to eat wheat (or another grain) although you wanted to eat rice (not including when you were sick)?” Based on the responses to questions, four categories of food security were created, all the responses were then summed to create a food security indicator score. A score of 0 was considered as food secure, all the way to a maximum score of 15, being severe insecurity [15]. For our analyses and using BDHS descriptive scores, we dichotomized this indicator into either “food secure” or “food insecure” and assumed that individuals coming from the same household had similar food availability and security.
2.3. Fasting Blood Glucose Outcomes
Women and men who were age 35 or older in selected households were eligible to have their blood glucose tested [15]. BDHS 2011 indicated that 4,311 women and 4,524 men age 35 and older were eligible for blood glucose testing. Among these individuals, 89% of women and 83% of men participated in the blood glucose measurement [15].
The protocol for measuring fasting blood glucose in the BDHS 2011 survey has been previously described [15]. Briefly, participants in the biomarker sub-study were asked if they had eaten or had anything to drink (except water) before the glucose test. If the participant had not been fasting, an appointment was scheduled for the next morning [15]. Blood was obtained from the middle or ring finger of eligible participants who had fasted overnight. Before being pricked with a non-reusable lancet, the participants’ finger was swabbed with 70% isopropyl alcohol and allowed to dry. The first two drops of blood were wiped away, and the third drop was used to perform the field test. Blood glucose was measured using a HemoCue 201+ blood glucose analyzer (HemoCue America, Brea, California).
Within the subsample, there were three observations with fasting blood glucose measurements greater than 400 mg/dL. This is biologically possible, but unlikely. These observations were included in the main analysis and excluded in sensitivity analyses.
We considered fasting blood glucose to be in the impaired fasting glucose or diabetes ranges using cut-points from the World Health Organization [16]. Fasting blood glucose ≤ 110 mg/dL was considered the reference category, and this category included 499 individuals in our sample with hypoglycemic status as defined by low fasting blood glucose (< 70 mg/dL). We defined fasting glucose in the impaired fasting glucose range as a fasting blood glucose measurement between 110 and 126 mg/dL. We defined fasting blood glucose in the diabetes range as a fasting blood glucose ≥ 126 mg/dL.
2.4. Groundwater Chemistry Data
Chemical concentrations in wells across Bangladesh were measured by the British Geological Survey (BGS), which collects groundwater information within the United Kingdom and internationally [17]. In 1998–99, BGS staff in close collaboration with the Department of Public Health Engineering (DPHE) of Bangladesh carried out a groundwater chemical survey to develop maps showing the regional distribution of elements in Bangladesh groundwater [3]. The survey employed a stratified random sampling in which stratification was by units of area (km2) to ensure a uniform distribution of sites [3, 17]. Water samples were collected from 3,534 well water samples across Bangladesh (excluding the Chittagong Hill Tracts), covering one sample for 37 km2 area [3]. All samples were collected from drinking water wells, which ranged in depth from 7–362 meters deep [18]. The GPS coordinates of each well were recorded. Arsenic was measured using hydride generation-atomic fluorescence spectrometry (HG-AFS) with a detection limit of 0.25 or 0.5 μg L−1 [3]. Other chemicals were measured by inductively-coupled plasma-atomic emission spectrometry (ICP-AES), and in a few cases by inductively-coupled plasma-mass spectrometry (ICP-MS) [3]. All analyses were carried out in BGS laboratories using filtered (0.22 μm) samples; results from both ICP-AES and ICP-MS methods were in good agreement [17].
Arsenic exposure for the main analysis was grouped into three categories: local groundwater arsenic ≤ 10 μg/L; local groundwater arsenic >10 and ≤ 50 μg/L; and local groundwater arsenic > 50 μg/L. For continuous analyses of arsenic, values below the limit of detection (0.5 μg/L) were recoded as equal to 0.4 μg/L. Other groundwater chemicals were dichotomized at their medians to create low and high strata of each chemical, circumventing limit of detection issues for chemicals where at least 50% of the sample was above the limit of detection. Dichotomized chemicals included: aluminum, barium, calcium, iron, potassium, lithium, magnesium, manganese, sodium, phosphorous, silicon, sulfate, strontium, and zinc. We excluded boron, cobalt, chromium, copper, and vanadium from our analysis as > 50% of the samples were below the limit of detection (Table 2).
Table 2.
Geometric mean ratios of fasting blood glucose across three categories of arsenic exposure.
| Model 1 | Model 2 | Model 3 | Model 4 | |
|---|---|---|---|---|
| Arsenic (μg/L) | ||||
| As ≤ 10 | Referent | Referent | Referent | Referent |
| 10 < As ≤ 50 | 1.02 (0.99, 1.05) | 1.01 (0.98, 1.04) | 1.01 (0.98, 1.04) | 1.01 (0.98, 1.04) |
| 50 < As ≤667 | 1.02 (0.99, 1.05) | 1.00 (0.97, 1.03) | 1.00 (0.97, 1.03) | 1.00 (0.97, 1.03) |
| P for Trend | 0.34 | 0.99 | 0.98 | 0.99 |
Model 1. Unadjusted.
Model 2. Adjusted for age, sex, BMI, and region.
Model 3. Further adjusted for smoking status, educational attainment, urban vs rural residence, household wealth, and diabetes medication use.
Model 4. Further adjusted for food security.
2.5. Exposure Assignment
BDHS participants were assigned arsenic and other groundwater chemical exposures based on data from their cluster’s nearest BGS well. Only one well (the nearest) was assigned to each cluster. The mean distance between the randomly selected point of the cluster and the nearest well was 2.3 (95% CI: 1.3 — 3.6) kilometers. Specific information on the distance between British Geological Survey monitoring wells and each household’s water sources was not available for privacy reasons. To protect household identity, one randomly selected GPS location was taken per cluster.
To do the spatial data merge and exposure assignment, we used administrative shapefiles for Bangladesh from DIVA-GIS [19]. We extracted the GPS locations of the BDHS clusters and BGS-DPHE wells, then imported and projected these in ArcGIS 10.4.1 using the UTM 1984 45 N projection system. We determined the nearest BGS-DPHE wells for each BDHS 2011 clusters using spatial joining in ArcGIS to calculate the distances in kilometers from clusters to wells.
2.6. Statistical Analysis
Our analysis is focused on men and women in Bangladesh who were at least 35 years and who indicated using groundwater as their primary drinking water source. Therefore, we used survey estimation methods [20] to draw inferences for that subpopulation. Supplemental Figure 1 illustrates the process used for their selection.
Missing data were handled using multiple imputation by chained equations [21]. Four variables (current smoker, BMI, taking diabetes medications, and food security) were imputed. Smoking status had 7,527 complete observations (38 imputed), BMI had 7,329 complete observations (236 imputed), diabetes medication use had 7,018 complete observations (547 imputed), and food security had 5,191 complete observations (2,374 imputed). The analytic sample size, after weighting, was 6,587.
We estimated the population proportions for categorical variables (current smoker, taking diabetes medications, urban residence, household wealth, and regional distributions), the population arithmetic means of approximately normally distributed continuous variables (age, BMI, and years of education) and the population geometric means of skewed continuous variables (arsenic, other groundwater chemicals, and fasting blood glucose). The fasting blood glucose distribution was approximately log-normal.
Survey estimation linear regression methods [20] were used to assess the association of groundwater arsenic with natural log-transformed fasting blood glucose. Models were sequentially adjusted: Model 1 was the unadjusted association; Model 2 further adjusted for age, sex and BMI; Model 3 further adjusted for current smoking status, education, household wealth, and whether an individual was currently taking medications for diabetes, and Model 4 further adjusted for food security. Survey estimation multinomial logistic regression methods were used to assess the relative odds of prevalent high fasting blood glucose (i.e., odds ratio of having fasting blood glucose in the range of impaired fasting glucose vs. range of normal glycemia or hypoglycemia, and having fasting blood glucose in the diabetes range vs. having fasting blood glucose in the range of normal glycemia or hypoglycemia) with increasing local well water arsenic. For the analyses treating arsenic exposure as categorical, we calculated a P for trend by generating a “trend score” variable with values = 0.4 for persons in the lowest arsenic exposure category, = 26.5 for persons in the middle arsenic exposure category, and = 143 for persons in the highest arsenic exposure category, and used this trend score variable as a continuous predictor in lieu of the arsenic category indicators.
Because sex [22–27] and BMI [27–30] have been associated in several studies with arsenic toxicokinetics and the related one-carbon metabolism pathway, we considered possible effect modification of the hypothesized arsenic toxicity, by sex and by BMI categories, using models with interaction terms between BMI category and arsenic category, or sex and arsenic category, and adjustment for the same suite of confounders as in our fully adjusted main analyses (Model 4).
To examine potential effect modification by other water well chemicals, we fit separate models with interaction terms between arsenic categories and each of the 14 other groundwater chemicals dichotomized at their medians. We used Bonferroni correction to account for the 14 well water chemicals considered in F test hypothesis tests (α = 0.05/14) [31]. We applied this same standard to the unadjusted and the adjusted models separately, under the assumption that analyses of the same chemical were dependent hypotheses.
To assess whether associations were dominated by urban populations, in a sensitivity analyses we excluded urban and Dhaka residents. In an additional sensitivity analysis, we assigned participants who reported diabetes medication use to the FBG ≥126 mg/dL category to reflect the glycemia that likely would have been observed had these persons not been treated.
All statistical analyses were performed in Stata/SE, version 15.1.
2.7. Ethics Approval
This secondary data analysis protocol was approved by Emory University IRB (IRB00088075). The DHS Program provided the survey and GPS data after examining the project goal and all DHS survey participants provided informed consent. Permission was attained from the copyright section of the British Geological Survey Environmental Science Centre to use the publicly available BGS-DPHE dataset.
3. Results
The survey-estimated characteristics of the adult population age ≥ 35 drinking groundwater in Bangladesh in 2011 are presented in Table 1. In our sample, there were 6,281 participants with fasting glucose < 126 mg/dL: 3,090 men and 3,191 women. There were 306 participants with fasting blood glucose ≥ 126 mg/dL: 146 men and 160 women. The higher and lower blood glucose populations had similar mean ages, but the population with fasting blood glucose ≥ 126 mg/dL had a mean BMI of 22.6 (95% CI: 22.1, 23.2) kg/m2 compared to a mean BMI of 20.5 (95% CI: 20.3, 20.6) kg/m2 among persons with fasting blood glucose < 126 mg/dL. The wealth distributions were dissimilar between these populations, with a greater proportion of persons with fasting blood glucose ≥ 126 mg/dL belonging to the highest two household wealth quintiles. There were more years of education in the population of men with fasting blood glucose < 126 mg/dL [mean 3.9 (95 CI: 3.7, 4.2) years] than among women with fasting blood glucose < 126 mg/dL [mean 1.9 (1.7, 2.1) years], but these sex differences in educational attainment were eclipsed by the differences in blood sugar categories: the mean education attained by men with fasting blood glucose ≥ 126 mg/dL was 6.1 (95% CI: 5.1, 7.1) years, while among women with fasting blood glucose ≥ 126 mg/dL the mean educational attainment was 4.8 (95% CI: 4.1, 5.4) years. Smoking was more common among the population with fasting blood glucose < 126 mg/dL than among persons with fasting blood glucose ≥ 126 mg/dL. Among the population with fasting glucose < 126 mg/dL, 1.6% (95% CI: 1.3, 2.0) take diabetes-related medications, if self-report data are accurate. Among persons with fasting glucose ≥ 126 mg/dL, the proportion using diabetes-related medication was 35.3% (95% CI: 28.6, 42.0). More of the population with fasting blood glucose ≥ 126 mg/dL lived in urban environments, compared to the population with lower fasting blood glucose.
Table 1:
Estimated population proportions and 95% confidence intervals for characteristics of the population of Bangladesh in 2011, age ≥ 35 and using groundwater as primary drinking water source.
| Participant Characteristics | Persons with fasting blood glucose < 126 mg/dL (n=6,281) | Persons with fasting blood glucose ≥ 126 mg/dL (n=306) |
|---|---|---|
| Age (in years) | 51.8 (51.4, 52.2) | 52.9 (51.2, 54.6) |
| Sex (male) | 49.2% (48.0, 50.4) | 47.7% (42.1%, 53.3%) |
| BMI (kg/m2) | 20.5 (20.3, 20.6) | 22.6 (22.1, 23.2) |
| Years of education | 2.9 (2.7, 3.1) | 4.8 (4.1, 5.4) |
| Fasting blood glucose (mg/dL) | 91.6 (90.8, 92.5) | 165.1 (158.9, 171.4) |
| Current Smoker | 14.0% (12.4, 15.6) | 9.0% (5.5, 12.6) |
| Taking diabetes medication | 1.6% (1.3, 2.0) | 35.3% (28.6, 42.0) |
| Household characteristics | ||
| Urban residence | 15.5% (13.8, 17.1) | 23.0% (18.4, 27.6) |
| Household wealth index | ||
| Quintile 1 | 21.7% (19.6, 23.8) | 11.8% (6.9, 16.6) |
| Quintile 2 | 21.7% (20.0, 23.4) | 12.4% (7.4, 17.3) |
| Quintile 3 | 21.9% (20.3, 23.6) | 19.6% (14.4, 24.8) |
| Quintile 4 | 21.6% (19.8, 23.3) | 24.5% (18.7, 30.2) |
| Quintile 5 | 13.1% (11.4, 14.7) | 31.8% (25.8, 37.8) |
| Regional distribution | ||
| Barisal | 6.1% (5.5, 6.6) | 8.0% (5.4, 10.7) |
| Chittagong | 16.8% (15.7, 17.9) | 23.7% (18.3, 29.1) |
| Dhaka | 29.4% (27.7, 31.2) | 19.0% (13.2, 24.9) |
| Khulna | 13.4% (12.3, 14.5) | 12.2% (8.5, 15.9) |
| Rajshahi | 15.2% (14.1, 16.3) | 20.6% (15.5, 25.8) |
| Rangpur | 13.4% (12.6, 14.2) | 11.4% (7.7, 15.1) |
| Sylhet | 5.7% (5.1, 6.2) | 5.0% (2.9, 7.1) |
There were three observations that had fasting blood glucose measurements above 400 mg/dL that slightly skew the mean fasting glucose
Water chemistry measurements are described in Supplemental Table 1. The most common chemicals were sodium with a geometric mean of 40.22 mg/L, calcium with a geometric mean of 26.21 mg/L, and silicon with a geometric mean of 18.82 mg/L. The geometric mean of arsenic was 4.14 μg/L. Groundwater arsenic levels varied geographically across regions (Supplemental Table 2). The region with the highest geometric mean arsenic concentration was Chittagong (16.7 μg/L), followed by Khulna (8.4 μg/L), and Sylhet (5.8 μg/L). The lowest geometric mean arsenic was in the Rangpur region (1.1 μg/L).
Table 2 presents the association of arsenic with the geometric means of fasting blood glucose. In unadjusted models, persons with groundwater arsenic in the 10 to 50 μg/L range had a geometric mean ratio of 1.02 (0.99, 1.05) compared to individuals in the ≤10 μg/L range, and persons with groundwater arsenic >50 μg/L had a geometric mean ratio of 1.02 (95% CI: 0.99, 1.05) compared to persons with groundwater arsenic ≤10 μg/L. However, these associations were not statistically significant. The ratio of geometric means of fasting blood glucose adjusting for age, sex, BMI, and region was 1.01 (95% CI: 0.98, 1.04) among the population in the 10 to 50 μg/L arsenic exposure category compared to the ≤ 10 μg/L category. The fully adjusted models had similar associations, with a geometric mean ratio of 1.01 (95% CI: 0.98, 1.04) for fasting glucose comparing persons with arsenic in the 10 – 50 μg/L range compared to persons in the ≤ 10 μg/L category, and 1.00 (95% CI: 0.97, 1.03) increase for > 50 μg/L arsenic concentration category compared to the ≤ 10 μg/L category. None of these associations of fasting blood glucose with groundwater arsenic category were statistically significant. Restriction to persons outside Dhaka and urban areas did not change these from null associations (Supplemental Table 3). Considering a possible linear relationship of the geometric mean of fasting blood glucose with arsenic as a continuous predictor, the association remained null (Supplemental Table 4). Recoding the participants reporting diabetes medication use as if they had been observed as having FBG ≥126 mg/dL (Supplemental Table 5) attenuated associations compared to the estimates obtained from analysis categorizing participants by their observed FBG values (Table 3); however, neither set of associations was statistically significant.
Table 3.
Odds ratios of fasting blood glucose in the ranges of impaired fasting glucose (fasting blood glucose ≥ 110 mg/dL and <126 mg/dL) or diabetes (fasting blood glucose ≥126 mg/dL) relative to a reference category of fasting blood glucose < 110 mg/dL across three categories of arsenic exposure.
| Fasting blood glucose in impaired fasting glucose range | Fasting blood glucose in diabetes range | |
|---|---|---|
| Model 1 | ||
| Arsenic (μg/L) | ||
| As ≤ 10 | Referent | Referent |
| 10 < As ≤ 50 | 0.89 (0.62, 1.28) | 1.15 (0.78, 1.69) |
| 50 < As ≤667 | 1.21 (0.84, 1.74) | 1.33 (0.97, 1.82) |
| P for trend | 0.33 | 0.07 |
| Model 2 | ||
| Arsenic (μg/L) | ||
| As ≤ 10 | Referent | Referent |
| 10 < As ≤ 50 | 0.87 (0.61, 1.25) | 1.14 (0.80, 1.64) |
| 50 < As ≤667 | 1.11 (0.76, 1.62) | 1.35 (0.97, 1.88) |
| P for trend | 0.59 | 0.09 |
| Model 3 | ||
| Arsenic (μg/L) | ||
| As ≤ 10 | Referent | Referent |
| 10 < As ≤ 50 | 0.89 (0.62, 1.27) | 1.19 (0.80, 1.78) |
| 50 < As ≤667 | 1.14 (0.79, 1.67) | 1.40 (0.97, 2.04) |
| P for trend | 0.58 | 0.08 |
| Model 4 | ||
| Arsenic (μg/L) | ||
| As ≤ 10 | Referent | Referent |
| 10 < As ≤ 50 | 0.88 (0.62, 1.27) | 1.19 (0.80, 1.78) |
| 50 < As ≤667 | 1.14 (0.79, 1.66) | 1.40 (0.96, 2.04) |
| P for trend | 0.59 | 0.08 |
Model 1. Unadjusted.
Model 2. Adjusted for age, sex, BMI, and region.
Model 3. Further adjusted for smoking status, educational attainment, urban vs rural residence, household wealth, and diabetes medication use.
Model 4. Further adjusted for food security.
Table 3 reports the odds ratios from multinomial logistic regressions of a person having fasting glucose in the impaired fasting glucose (fasting blood glucose ≥ 110 mg/dL and <126 mg/dL) or diabetes (≥126 mg/dL) ranges, relative to fasting blood glucose < 110 mg/dL, as groundwater arsenic increases. The unadjusted odds ratio comparing persons in the 10 – 50 μg/L arsenic range to those in the ≤ 10 μg/L arsenic for impaired fasting glucose range fasting glucose was 0.89 (95% CI: 0.62, 1.28), and among persons with arsenic > 50 μg/L vs. arsenic ≤ 10 μg/L, the odds ratio of impaired fasting glucose range glucose was 1.21 (95% CI: 0.84, 1.74). The fully adjusted odds ratio for impaired fasting glucose-range fasting glucose among the population in the > 50 μg/L vs. ≤ 10 μg/L arsenic category was 0.88 (95% CI: 0.62, 1.27), and for persons with arsenic > 50 μg/L vs. ≤10 μg/L arsenic category was 1.14 (95% CI: 0.79, 1.66). The unadjusted odds ratios for having fasting blood glucose in the diabetes range for persons in the 10 – 50 μg/L arsenic category vs. ≤ 10 μg/L arsenic category was 1.15 (95% CI: 0.78, 1.69), and the odds ratio for having fasting glucose in the diabetes range among persons in the > 50 μg/L arsenic category vs. ≤ 10 μg/L arsenic category was 1.33 (95% CI: 0.97, 1.82). The fully adjusted odds ratio for diabetes-range fasting glucose among persons in the > 50 μg/L compared with the ≤ 10 μg/L arsenic category was 1.19 (95% CI: 0.80, 1.78), and the adjusted odds ratio for persons in the > 50 μg/L arsenic category vs. ≤10 μg/L arsenic category was 1.40 (95% CI: 0.96, 2.04). In sensitivity analysis excluding urban populations and the Dhaka region (Supplemental Table 6), there were no significant odds ratios, but these estimates were highly imprecise likely reflecting the small number of persons in the joint strata.
We considered the possible role of BMI or sex as possible effect modifiers in the fully adjusted arsenic - fasting blood glucose associations, by examining the ratios of geometric mean ratios (GMRR) to contrast the magnitude of associations of arsenic with fasting blood glucose, comparing persons in the normal weight range (18.5 ≤ BMI < 25), in overweight persons (25 ≤ BMI < 30), or in obese persons (BMI ≥ 30) compared to the reference associations for arsenic with fasting blood glucose among underweight (BMI < 18.5) persons. The normal weight vs. underweight GMRR for the fasting blood glucose contrast between persons with arsenic concentrations 10 – 50 μg/L relative to the ≤ 10 μg/L arsenic category was 1.03 (95% CI: 0.99, 1.07), and the normal weight vs. underweight GMRR for the contrast between persons in the > 50 μg/L vs. ≤ 10 μg/L arsenic category was 1.06 (95% CI: 1.02, 1.10). The GMRR for overweight persons vs. underweight persons in the population for the contrast in blood glucose between persons in the 10 – 50 μg/L relative to the ≤ 10 μg/L arsenic category was 1.02 (95% CI: 0.96, 1.07), and the GMRR for the contrast between the > 50 μg/L vs. ≤ 10 μg/L arsenic category was 1.04 (95% CI: 0.99, 1.10). The GMRR for obese persons versus underweight persons in the population for the contrast between persons with groundwater arsenic in the 10 – 50 μg/L range vs. ≤ 10 μg/L arsenic category was 1.09 (95% CI: 0.98, 1.22), and the GMRR for the contrast between persons in the > 50 μg/L category vs. the ≤ 10 μg/L arsenic category was 1.10 (95% CI: 0.99, 1.22). There was no significant effect modification by sex. The adjusted GMRR for men vs. women in the contrast in fasting blood glucose between persons with arsenic in the 10 – 50 μg/L category vs. the ≤ 10 μg/L arsenic category was 1.00 (0.98, 1.03), and the GMRR for the contrast in fasting blood glucose between the > 50 μg/L arsenic category vs. the ≤ 10 μg/L arsenic category was 1.02 (0.99, 1.04).
Table 4 considers potential interactions of categorical arsenic exposure with 14 other groundwater chemicals dichotomized at their median concentrations; the presence of a significant effect of any arsenic category in any chemical stratum was assessed by a separate F test per chemical. There were no Bonferroni-significant interactions, after accounting for the multiple F tests. In fully adjusted models, among persons with low drinking-water calcium (≤ 25.5 mg/L), arsenic exposure in the 10 to 50 μg/L range had a fasting blood glucose geometric mean ratio of 1.03 (95% CI: 1.00, 1.06) compared to the referent group of ≤ 10 μg/L water arsenic. Among persons with low drinking-water iron (≤ 0.69 mg/L), the geometric mean ratio of fasting blood glucose for persons in the > 10 and ≤ 50 μg/L arsenic range vs. the ≤ 10 μg/L arsenic range was 1.03 (95% CI: 0.98, 1.08), and for persons with arsenic in the > 50 μg/L range vs. arsenic ≤ 10 μg/L was 1.06 (95% CI: 1.01, 1.11). Low potassium (≤ 3.0 mg/L) may have potentiated arsenic toxicity, as in that stratum the geometric mean ratio of fasting blood glucose contrasting persons in the > 10 and ≤ 50 μg/L arsenic range vs. ≤10 μg/L arsenic range was 1.03 (95% CI: 1.00, 1.07), and contrasting persons in the > 50 μg/L arsenic range vs. ≤ 10 μg/L arsenic range was 1.06 (95% CI: 1.02, 1.11). Among persons in the population with low water magnesium (≤ 12.1 mg/L), persons with arsenic concentrations in the 10 to 50 μg/L range compared to those with ≤ 10 μg/L arsenic had a fasting blood glucose geometric mean ratio of 1.05 (95% CI: 1.02, 1.07). Among persons with high-zinc water (> 0.014 mg/L), persons in the 10 – 50 μg/L arsenic concentration group, compared to persons with arsenic < 10 μg/L, had a fasting blood glucose geometric mean ratio of 1.06 (95% CI: 1.01, 1.11), but among persons exposed to high zinc the contrast in fasting blood glucose between the >50 μg/L and ≤ 10 μg/L was a null association of 0.98 (95% CI: 0.93, 1.03). There were no significant associations detected in other groups. A sensitivity analysis using stratified analyses rather than interaction-term models did not have Bonferroni-significant adjusted associations in any water chemistry stratum after accounting for the number of stratified models (Supplemental Table 7).
Table 4.
Associations of arsenic with fasting blood glucose (geometric mean ratios, GMR) assessing for interaction by other groundwater chemicals dichotomized at their median concentration, considered separately. P values are from F test per chemical, for a significant association of any higher than lowest category arsenic with fasting blood glucose, in either the high chemical or low chemical subgroup. The Bonferroni-adjusted significance threshold considering 14 chemicals’ hypothesis tests is α=0.004. Adjusted models controlled for age, sex, BMI, region, smoking status, educational attainment, urban vs rural residence, household wealth, diabetes medication use, and food security.
| Unadjusted GMR | F-test P Value | Adjusted GMR | F-test P Value | |
|---|---|---|---|---|
| Aluminum: Low (≤ 0.04 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.01 (0.98, 1.04) | 1.00 (0.97, 1.03) | ||
| 50 < As ≤667 | 1.01 (0.98, 1.05) | 1.00 (0.97, 1.03) | ||
| Aluminum: High (> 0.04 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.03 (0.98, 1.09) | 0.535 | 1.02 (0.96, 1.08) | 0.765 |
| 50 < As ≤667 | 0.99 (0.93, 1.05) | 0.99 (0.93, 1.05) | ||
| Barium: Low (≤ 0.05 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.03 (0.99, 1.08) | 1.02 (0.98, 1.06) | ||
| 50 < As ≤667 | 1.03 (0.98, 1.08) | 0.98 (0.93, 1.03) | ||
| Barium: High (> 0.05 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.98 (0.93, 1.03) | 0.687 | 0.98 (0.94, 1.03) | 0.492 |
| 50 < As ≤667 | 0.99 (0.94, 1.05) | 1.03 (0.97, 1.10) | ||
| Calcium: Low (≤ 25.5 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.03 (1.00, 1.07) | 1.03 (1.00, 1.06) | ||
| 50 < As ≤667 | 1.02 (0.97, 1.06) | 0.99 (0.94, 1.04) | ||
| Calcium: High (> 25.5 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.95 (0.91, 0.99) | 0.064 | 0.94 (0.90. 0.98) | 0.019 |
| 50 < As ≤667 | 0.98 (0.93, 1.04) | 0.99 (0.93, 1.04) | ||
| Iron: Low (≤ 0.69 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.04 (1.00, 1.08) | 1.03 (0.98, 1.08) | ||
| 50 < As ≤667 | 1.09 (1.05, 1.14) | 1.06 (1.01, 1.11) | ||
| Iron: High (> 0.69 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.98 (0.93, 1.03) | 0.002 | 0.98 (0.93, 1.03) | 0.054 |
| 50 < As ≤667 | 0.92 (0.88, 0.97) | 0.94 (0.90, 0.99) | ||
| Potassium: Low (≤ 3.0 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.04 (1.00, 1.08) | 1.03 (1.00, 1.07) | ||
| 50 < As ≤667 | 1.06 (1.02, 1.11) | 1.06 (1.02, 1.11) | ||
| Potassium: High (> 3.0 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.98 (0.93, 1.03) | 0.137 | 0.97 (0.93, 1.02) | 0.018 |
| 50 < As ≤667 | 0.96 (0.91, 1.00) | 0.94 (0.89, 0.98) | ||
| Lithium: Low (≤ 0.004 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.01 (0.98, 1.05) | 1.00 (0.97, 1.04) | ||
| 50 < As ≤667 | 1.00 (0.96, 1.04) | 0.99 (0.95, 1.02) | ||
| Lithium: High (> 0.004 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.04 (0.99, 1.09) | 0.008 | 1.04 (0.99, 1.09) | 0.030 |
| 50 < As ≤667 | 1.07 (1.02, 1.12) | 1.05 (1.01, 1.10) | ||
| Magnesium: Low (≤ 12.1 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.05 (1.02, 1.08) | 1.05 (1.02, 1.07) | ||
| 50 < As ≤667 | 1.04 (0.99, 1.10) | 1.04 (0.99, 1.09) | ||
| Magnesium: High (> 12.1 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.94 (0.90, 0.99) | 0.031 | 0.93 (0.89, 0.98) | 0.005 |
| 50 < As ≤667 | 0.97 (0.92, 1.03) | 0.96 (0.90, 1.01) | ||
| Manganese: Low (≤ 0.26 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.02 (0.99, 1.06) | 1.01 (0.98, 1.05) | ||
| 50 < As ≤667 | 1.05 (1.01, 1.09) | 1.03 (0.99, 1.07) | ||
| Manganese: High (> 0.26 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.99 (0.94, 1.03) | 0.055 | 1.00 (0.95, 1.04) | 0.090 |
| 50 < As ≤667 | 0.94 (0.90, 0.99) | 0.95 (0.91, 0.99) | ||
| Sodium: Low (≤ 34.6 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.01 (0.97, 1.05) | 1.01 (0.97, 1.04) | ||
| 50 < As ≤667 | 0.99 (0.95, 1.04) | 0.98 (0.93, 1.03) | ||
| Sodium: High (> 34.6 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.04 (0.99, 1.09) | 0.065 | 1.01 (0.96, 1.06) | 0.360 |
| 50 < As ≤667 | 1.05 (0.99, 1.10) | 1.04 (0.98, 1.09) | ||
| Phosphorous: Low (≤ 0.3 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.00 (0.96, 1.05) | 1.00 (0.96, 1.05) | ||
| 50 < As ≤667 | 0.92 (0.81, 1.04) | 0.91 (0.81, 1.03) | ||
| Phosphorous: High (> 0.3 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.03 (0.98, 1.08) | 0.104 | 1.01 (0.97, 1.07) | 0.237 |
| 50 < As ≤667 | 1.12 (0.99, 1.27) | 1.11 (0.98, 1.25) | ||
| Silicon: Low (≤ 19.6 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.01 (0.98, 1.05) | 1.00 (0.97, 1.04) | ||
| 50 < As ≤667 | 1.02 (0.99, 1.06) | 1.01 (0.97, 1.04) | ||
| Silicon: High (> 19.6 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.02 (0.97, 1.07) | 0.734 | 1.02 (0.97, 1.07) | 0.796 |
| 50 < As ≤667 | 0.99 (0.94, 1.04) | 1.00 (0.95, 1.05) | ||
| Sulfate: Low (≤ 0.8 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.01 (0.97, 1.05) | 1.00 (0.97, 1.04) | ||
| 50 < As ≤667 | 1.02 (0.99, 1.05) | 1.01 (0.99, 1.05) | ||
| Sulfate: High (> 0.8 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.00 (0.95, 1.05) | 0.527 | 1.01 (0.96, 1.06) | 0.220 |
| 50 < As ≤667 | 0.97 (0.91, 1.03) | 0.95 (0.90, 1.01) | ||
| Strontium: Low (≤ 0.16 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.02 (0.99, 1.06) | 1.02 (0.99, 1.05) | ||
| 50 < As ≤667 | 1.02 (0.97, 1.07) | 1.01 (0.96, 1.06) | ||
| Strontium: High (> 0.16 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.97 (0.92, 1.02) | 0.468 | 0.95 (0.91, 1.00) | 0.135 |
| 50 < As ≤667 | 0.98 (0.93, 1.04) | 0.97 (0.92, 1.03) | ||
| Zinc: Low (≤ 0.014 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 0.99 (0.95, 1.03) | 0.98 (0.94, 1.02) | ||
| 50 < As ≤667 | 1.03 (0.99, 1.06) | 1.02 (0.99, 1.05) | ||
| Zinc: High (> 0.014 mg/L) | ||||
| As ≤ 10 | referent | referent | ||
| 10 < As ≤ 50 | 1.06 (1.01, 1.11) | 0.039 | 1.06 (1.01, 1.11) | 0.024 |
| 50 < As ≤667 | 0.98 (0.93, 1.03) | 0.98 (0.93, 1.03) | ||
Fully adjusted for age, sex, BMI, region, smoking status, education, urban vs rural residence, wealth, if individuals were taking diabetes medications, and food security.
4. Discussion
We did not detect an overall association between arsenic, measured a decade prior at proxy locations, with fasting blood glucose, or with the prevalence of having elevated fasting blood glucose in the range of impaired fasting blood glucose or diabetes, among adults age ≥ 35 in Bangladesh in 2011. Results were similar when treating arsenic as a categorical or continuous exposure (Table 3, 4). Arsenic levels in urban areas and in Dhaka were low as measured in the British Geologic Survey, but there are other risk factors in these communities that could influence the development of diabetes. However, the association remained similar in sensitivity analyses excluding urban and Dhaka participants (Supplemental Table 2). In another sensitivity analysis, we assigned participants reporting diabetes medication use to the FBG ≥126 mg/dL category to account for what their underlying fasting blood glucose level would have been without medication (Supplemental Table 5). The results were consistent with the original analysis and suggest a null overall association between local historical groundwater arsenic and fasting blood glucose.
There was a small number of persons with fasting blood glucose in the diabetes range in our sample. This could be important if the effects of arsenic on fasting blood glucose differ across quantiles of the glycemia distribution - in particular, if arsenic has impacts primarily at high levels of fasting blood glucose (e.g., among persons with uncontrolled diabetes). In a few previous studies of arsenic and glycemia, the association of arsenic with glycemic biomarkers was primarily among persons with very poorly controlled diabetes; this pattern was reported both in an American Indian sample [32] and in a nationally representative sample of Canadians [8], although in both of those North American cross-sectional studies the exposure metric was a urine arsenic biomarker rather than groundwater arsenic. In this study, although the associations had wide confidence intervals and were not significant, there was a monotonic increase in the odds ratio of having fasting blood glucose ≥ 126 mg/dL versus fasting glucose < 110 mg/dL with increasing arsenic categorical exposure (Table 3), and these odds ratios were robust to adjustment for confounders. Whether our finding is another replication of an arsenic association peculiar to the tail of the glycemia distribution, but not significant in our analysis because of the small number of persons with poorly controlled diabetes in this study, is ambiguous. Our estimate for the prevalence of having fasting blood glucose ≥126 mg/dL was 4.6%, which is lower than the estimated prevalence of diabetes in a more recent 2016 review that suggested a diabetes prevalence in Bangladesh of 7.4% (95% CI 7.2–7.7%) [31].
Arsenic toxicity in the context of other water chemicals has only recently begun to be studied, but mixtures including arsenic are common in areas such as the Bengal delta. Hoque et al. investigated the status of several groundwater toxicants and nutrients relevant to human health in Asian deltas [33], compiling data on 5,256 tube-wells from published literature for Bengal, Mekong, and Red River deltas. Most of the high calcium and magnesium wells are located in the south-west and southern parts of the Bengal delta, and contained between 25–50% of the recommended daily intake for magnesium. Northeastern regions of Bangladesh had much lower magnesium concentrations, with wells containing < 10% of the recommended daily intake.
That we did not see strong evidence for effect modification of the hypothesized arsenic toxicity by strong correlates of arsenic toxicokinetic differences between people (i.e., no significant interactions by sex, and no monotonic trend of interactions with increasing categories of BMI) could suggest that our local, historical groundwater arsenic exposure measure may have limited construct validity as a surrogate for people’s drinking water arsenic exposures at the etiologically relevant time for impacting fasting blood glucose measures [34]. However, it is also possible that there exists an arsenic toxicity for glucose regulation that is not modified by sex nor by BMI. A United States-based cohort study that found a positive association between drinking water supply arsenic and incident diabetes did not show significant interactions by sex, nor by waist circumference [35].
Our study has several key limitations. We think the most important limitation is that groundwater arsenic and other chemical concentrations were measured in the BGS survey conducted in 1998/99, not concurrently with the glucose measurements, and in nearby areas, not in household water supplies. The appropriateness of these measures as a proxy for drinking water chemicals encountered by groundwater-consuming BDHS participants (and the population to whom inference is extended) in the etiologically relevant time for glycemic outcomes in 2011 depends on the extent of water chemistry differences between the measured BGS survey wells and the unmeasured BDHS household water supply wells. Deeper aquifers (> 30m) likely have arsenic concentrations that have less temporal measurement error and are lower overall than in the shallow aquifers [36, 37]. In addition to random error differing by well depth, in an Araihazar, Bangladesh temporal variability study there were systematic changes in arsenic concentration for 11 of their 26 monitoring wells observed over a period of 2–3 years, with trends that differed in direction across wells [37]. The water samples from the BGS-DPHE were collected from drinking water wells that ranged from seven to 362 meters deep, the median depth was 35 m (25%ile: 22 m, 75%ile: 56 m) [3]. In addition to these hydrological sources of temporal error in the same well over time, since the time of the BGS-DPHE survey, millions of pumps have been implemented to improve access to safe drinking water, potentially leading to major and systematic changes in exposures if the BGS well and BDHS water supply wells are at systematically different depths [38]. We lack data on the well depth distribution for the BDHS household water supplies, but if the discrepancy in well depths between the two populations is spatially patterned similarly to the distribution of diabetes across our study participants, this could introduce differential measurement error, dependent measurement error, or both, of exposure by outcome, contributing to information bias that could be in any direction. A rigorous study using household water supply tube well measures did detect a significant association with diabetes prevalence in strata of Bangladesh [13], and the difference between our findings may be due to their study having more relevant exposure measures.
Another concern for potential information bias is that fasting blood glucose has temporal variability as well as other sources of measurement error, and this study is based on a single glucose measurement. Past research by Selvin et al, suggests that studies using only a single measurement could arrive at considerably different prevalence estimates for diabetes compared to using two glucose measurements [39].
Lastly, our modeling approach has a possible limitation: the distributional assumption of fasting blood glucose being lognormal is a strong assumption, and our P values may be sensitive to departures of the empirical distribution from a lognormal distribution. This could bias standard errors of our estimates and therefore our claims of significance or lack thereof; but the point estimates of associations are likely reasonable. This may be important for interpreting the odds ratios of high blood sugar with low arsenic, as these point estimates were substantial even if not statistically significant, and showed a monotonic dose-response.
To better understand the relationship between arsenic and fasting blood glucose in Bangladesh, prospective epidemiological studies with more participant-relevant, longitudinal arsenic exposure measures could reduce exposure measurement error and allow assessment of the potential timing of an arsenic exposure-glycemic trait association. We did not identify groundwater chemicals that were Bonferroni-significant modifiers of groundwater arsenic’s association with fasting blood glucose outcomes; but these putative modifiers may have shared many of the same sources of measurement error as our arsenic exposure measure. More precise measurements of drinking water chemistry joint exposures could improve ability to detect interactions where present.
5. Conclusions
We did not detect a strong relationship between our surrogate measure of arsenic exposure and fasting blood glucose in Bangladesh in 2011. This could be due to measurement error but also due to a limited number of persons with diabetes, if the association of arsenic with higher blood sugar is specific to very poorly controlled diabetes. We were unable to detect Bonferroni-significant interactions of arsenic with other groundwater chemicals, but this does not preclude the existence of interactions that could be observed in studies with less measurement error, a greater number of persons affected by diabetes, or both.
Supplementary Material
Acknowledgements
Our analysis was supported by a grant from Unilever Ltd., and supported in part by funding from the National Institute of Environmental Health Sciences (P30 ES019776). We thank the DHS, BGS, our collaborators, and study participants for sharing their data for this analysis.
Funding sources: This work was supported by a grant from Unilever Ltd., and supported in part by funding from the National Institute of Environmental Health Sciences (P30 ES019776). Unilever Ltd. supported the retrieval of secondary data for this manuscript but had no involvement in the decision to publish or preparation of this manuscript.
Footnotes
Declarations of interest: None.
References:
- 1.Alam MG, et al. , Arsenic contamination in Bangladesh groundwater: a major environmental and social disaster. Int J Environ Health Res, 2002. 12(3): p. 235–53. [DOI] [PubMed] [Google Scholar]
- 2.Smith AH, Lingas EO, and Rahman M, Contamination of drinking-water by arsenic in Bangladesh: a public health emergency. Bull World Health Organ, 2000. 78(9): p. 1093–103. [PMC free article] [PubMed] [Google Scholar]
- 3.Kinniburgh D and Smedley P, Arsenic contamination of groundwater in Bangladesh. British Geological Survey, 2001. [Google Scholar]
- 4.Maull EA, et al. , Evaluation of the association between arsenic and diabetes: a National Toxicology Program workshop review. Environ Health Perspect, 2012. 120(12): p. 1658–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Chen Y, et al. , No association between arsenic exposure from drinking water and diabetes mellitus: a cross-sectional study in Bangladesh. Environ Health Perspect, 2010. 118(9): p. 1299–305. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Navas-Acien A, et al. , Arsenic exposure and prevalence of type 2 diabetes in US adults. JAMA, 2008. 300(7): p. 814–22. [DOI] [PubMed] [Google Scholar]
- 7.Peng Q, Harlow SD, and Park SK, Urinary arsenic and insulin resistance in US adolescents. Int J Hyg Environ Health, 2015. 218(4): p. 407–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Feseke SK, et al. , Arsenic exposure and type 2 diabetes: results from the 2007–2009 Canadian Health Measures Survey. Health Promot Chronic Dis Prev Can, 2015. 35(4): p. 63–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Baek K, Lee N, and Chung I, Association of arsenobetaine with beta-cell function assessed by homeostasis model assessment (HOMA) in nondiabetic Koreans: data from the fourth Korea National Health and Nutrition Examination Survey (KNHANES) 2008–2009. Ann Occup Environ Med, 2017. 29: p. 31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Rahman M, et al. , Diabetes mellitus associated with arsenic exposure in Bangladesh. Am J Epidemiol, 1998. 148(2): p. 198–203. [DOI] [PubMed] [Google Scholar]
- 11.Rahman M, et al. , Relations between exposure to arsenic, skin lesions, and glucosuria. Occup Environ Med, 1999. 56(4): p. 277–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Nabi AH, Rahman MM, and Islam LN, Evaluation of biochemical changes in chronic arsenic poisoning among Bangladeshi patients. Int J Environ Res Public Health, 2005. 2(3–4): p. 385–93. [DOI] [PubMed] [Google Scholar]
- 13.Islam R, et al. , Association between type 2 diabetes and chronic arsenic exposure in drinking water: a cross sectional study in Bangladesh. Environmental health : a global access science source, 2012. 11: p. 38–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Tseng CH, A review on environmental factors regulating arsenic methylation in humans. Toxicol Appl Pharmacol, 2009. 235(3): p. 338–50. [DOI] [PubMed] [Google Scholar]
- 15.NIPORT, Bangladesh Demographic and Health Survey. 2011: Dhaka. [Google Scholar]
- 16.WHO, World Health Organization: Definition and diagnosis of diabetes mellitus and intermediate hyperglycemia. 2006.
- 17.BGS. Bangladesh: DPHE/BGS National Hydrochemical Survey. [cited 2017 August 18]; Available at: < http://www.bgs.ac.uk/research/groundwater/health/arsenic/Bangladesh/mapsnhs.html >.
- 18.Kinniburgh D and Smedley P, Arsenic contamination of groundwater in Bangladesh. 2001. 3: hydrological Atlas. [Google Scholar]
- 19.Hijmans RJ, et al. DIVA-GIS. Vsn. 5.0. A geographic information system for the analysis of species distribution data. Manual available at < http://www.diva-gis.org > .
- 20.STATA. svy estimation — Estimation commands for survey data. [cited 2017 August 19]; Available at: < https://www.stata.com/manuals13/svysvyestimation.pdf >.
- 21.Azur MJ, et al. , Multiple imputation by chained equations: what is it and how does it work? Int J Methods Psychiatr Res, 2011. 20(1): p. 40–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Fukagawa NK, et al. , Sex-related differences in methionine metabolism and plasma homocysteine concentrations. Am J Clin Nutr, 2000. 72(1): p. 22–9. [DOI] [PubMed] [Google Scholar]
- 23.Mudd SH and Poole JR, Labile methyl balances for normal humans on various dietary regimens. Metabolism, 1975. 24(6): p. 721–35. [DOI] [PubMed] [Google Scholar]
- 24.Gamble MV, et al. , Folate and cobalamin deficiencies and hyperhomocysteinemia in Bangladesh. Am J Clin Nutr, 2005. 81(6): p. 1372–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Konstantinova SV, et al. , Divergent associations of plasma choline and betaine with components of metabolic syndrome in middle age and elderly men and women. J Nutr, 2008. 138(5): p. 914–20. [DOI] [PubMed] [Google Scholar]
- 26.Shen H, et al. , Factors Affecting Arsenic Methylation in Arsenic-Exposed Humans: A Systematic Review and Meta-Analysis. Int J Environ Res Public Health, 2016. 13(2): p. 205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Hudgens EE, et al. , Biological and behavioral factors modify urinary arsenic metabolic profiles in a U.S. population. Environ Health, 2016. 15(1): p. 62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Gribble MO, et al. , Body composition and arsenic metabolism: a cross-sectional analysis in the Strong Heart Study. Environ Health, 2013. 12: p. 107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Gomez-Rubio P, et al. , Association between body mass index and arsenic methylation efficiency in adult women from southwest U.S. and northwest Mexico. Toxicol Appl Pharmacol, 2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Gomez-Rubio P, et al. , Indigenous American ancestry is associated with arsenic methylation efficiency in an admixed population of northwest Mexico. Journal of toxicology and environmental health. Part A, 2012. 75(1): p. 36–49. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Hochberg Y and Tamhane AC, Multiple comparison procedures. 1987: John Wiley & Sons, Inc; 450. [Google Scholar]
- 32.Gribble MO, et al. , Arsenic Exposure, Diabetes Prevalence, and Diabetes Control in the Strong Heart Study. American Journal of Epidemiology, 2012. 176(10): p. 865–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Hoque MA and Butler AP, Medical Hydrogeology of Asian Deltas: Status of Groundwater Toxicants and Nutrients, and Implications for Human Health. International Journal of Environmental Research and Public Health, 2016. 13(1): p. 81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Cronbach LJ and Meehl PE, Construct validity in psychological tests. Psychol Bull, 1955. 52(4): p. 281–302. [DOI] [PubMed] [Google Scholar]
- 35.Brauner EV, et al. , Long-term exposure to low-level arsenic in drinking water and diabetes incidence: a prospective study of the diet, cancer and health cohort. Environ Health Perspect, 2014. 122(10): p. 1059–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.van Geen A, et al. , Spatial variability of arsenic in 6000 tube wells in a 25 km2 area of Bangladesh. Water Resources Research, 2003. 39(5). [Google Scholar]
- 37.Dhar RK, et al. , Temporal variability of groundwater chemistry in shallow and deep aquifers of Araihazar, Bangladesh. J Contam Hydrol, 2008. 99(1–4): p. 97–111. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.UNICEF B. Bangladesh National Drinking Water Quality Survey of 2009. Bangladesh Bureau of Statistics 2009 [cited 2017 May 11]; Available at: < https://www.unicef.org/bangladesh/BNDWQS_2009_web.pdf >.
- 39.Selvin E, et al. , Short-term variability in measures of glycemia and implications for the classification of diabetes. Archives of Internal Medicine, 2007. 167(14): p. 1545–1551. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.