A cross-sectional study of well water arsenic and child IQ in Maine schoolchildren

Recruitment and sample

We initially planned to recruit children in grades 3–5 in school districts where prior investigation suggested that the As content of the water supply for some households would be elevated above the EPA Guideline of 10 μg/L [10], and where families used household wells. Recruitment began in 2006–2007, in two New Hampshire (NH) school districts. Examination of well water characteristics for the first 53 recruited NH children revealed very low levels of WAs [on average 2.76 μg/L, with only 5.7% (3 wells) exceeding the US standard of 10 μg/L]. For this reason, in 03/2008, we shifted focus to school districts surrounding Augusta Maine (ME), where colleagues had identified higher exposure [11, 12] . This report considers children attending 11 participating elementary schools in three Regional School Units (RSUs) in ME (Districts A, B, and C).

Recruitment was open to those in grades 3–5 in participating schools. Parents received information distributed by the school, informing them of the study and seeking participation, with notices and response cards sent home in children’s backpacks. Families returning a card indicating interest were given more information by study staff by telephone, and then an appointment was scheduled. We excluded children with conditions with known adverse impact on intellectual functioning (e.g., multiple births, neurodevelopmental disorders) and children who were receiving special education services. These restrictions excluded two children with neurodevelopmental conditions (one a “processing deficit”, and the other “delayed cerebral development impacting motor skills and speech”, both receiving support services at school) that would have made our standard assessment difficult to conduct. Parent report of these conditions was confirmed with school personnel. Exclusion likely biased our findings toward the null. We further limited enrollment to those residing at the present address for three or more years.

From the 1595 children who attended grades 3–5 during the assessment period, 581 families (36%) agreed to participate. One child from each family, selected at random, was eligible for inclusion in analyses. Of the 581, appointments were scheduled with the families of 377 children; appointments were not scheduled with the remainder because of scheduling conflicts, inability to contact the family or change in the family’s interest. Of the children with assessments scheduled, 36 were excluded because they were additional children in the same home; two because they were receiving special education services, and 67 because they had not resided in the home for at least three years. We selected this cutoff to balance minimizing the imprecision of defining exposure using home water As level while maximizing sample size, and because few children were expected to have resided in this home since infancy; indeed, only 61 eligible children had resided in this home since their first year of life. This resulted in a final sample of 272 children in three ME school districts and their families, residing in the present home for an average 7.3 years, which represents more than three-fourths of the children’s lifetimes.

Procedures

Procedures were approved by Columbia University Medical Center and University of New Hampshire Institutional Review Boards. Families received reports summarizing results of well water and developmental testing, information on mitigation procedures (if appropriate) and $25; children received t-shirts.

Measures

Data analysis

We used linear regression analysis to estimate associations between WAs exposure (from the kitchen tap) and child IQ, with and without adjustment for sociodemographic characteristics (i.e., maternal education and intelligence, number of children in the household, qualities of the HOME environment, school district). Preliminary analyses examined bivariate associations between WAs and covariates, including maternal education, maternal intelligence, HOME scores, well characteristics, arsenic mitigation behavior and school district.

Our analysis plan first considered covariate-unadjusted associations between exposure and child intelligence (Model 1). Next, we determined a set of features to be examined in a covariate-adjusted model (Model 2). From the pool of potential covariates, we considered the contributions of those investigated in our prior work on child intelligence and lead [25, 26] and arsenic [1, 2] exposure, including maternal education and intelligence, number of children in the household, and qualities of the HOME environment. Although maternal education and intelligence were expectably correlated [Spearman r₍₂₇₀₎ = 0.53, p < 0.0001], both were included in Model 2. Because, as we discuss below, school district was related both to exposure and to other contributors to child intelligence, it was considered in our adjusted models, as well.

In order to describe non-linear patterns, we examined associations between child IQ, with WAs presented both as a continuous (log-transformed) measure, as well as stratified into four categories [WAs < 5 μg/L (n = 141); ≥5 μg/L to <10 (n = 46); ≥10 μg/L to <20 (n = 52); and ≥20 μg/L (n = 33)].

Covariate-unadjusted associations between well water As and child IQ (Model 1)

Covariate-adjusted associations between WAs and child intelligence (Model 2)

As Table 2 shows, both maternal intelligence and education (categorically defined) contributed significantly to most measures of child intelligence; associations were least strong for Processing Speed. Number of additional children in the home was not significantly associated with child intelligence. Children from families with missing data on the HOME obtained lower scores on some measures of child intelligence. School district was unrelated to child outcomes. Adjusting for the features in Model 2, results for WAs persisted, although effect sizes were expectably lower. WAs between 5 and 10 μg/L remained significantly associated with FSIQ and with all Index scores (except for Processing Speed). Just as in Model 1, significant associations for other levels of WAs concentration were relatively infrequent: WAs between 10 and 20 μg/L remained significantly associated with Perceptual Reasoning scores, and exposure to levels ≥ 20 μg/L remained negatively associated with Working Memory scores, although no longer statistically significant. On average, in adjusted analyses WAs explained 2-3% of the variance in child outcomes, except for Processing Speed, where its contribution was considerably lower. As in Models 1 and 2, there were no significant differences between estimated points lost across the three exposure groups, though power was limited, particularly in the two highest groups.

To examine differential associations at extremely low exposures, we reproduced these analyses, further dividing the < 5 μg/L group into two groups: WAs < 1 μg/L (n = 83) and ≤1 μg/L < WAs <5 μg/L (n = 58: data available upon request). There were no statistically significant differences in estimated IQ points lost for any subscale between these two groups. Results were unchanged when we removed school district from Model 2. Results were essentially unchanged when we repeated the analyses among children who lived at their present address for 5 or more years, although reduced power expectably lowered statistical significance (n = 215: data available upon request). Ln(WAs) was significantly associated with decrements in IQ in simple models only (data available upon request).

Estimating IQ decrements attributable to WAs exposure

Based on Model 2, compared to those with WAs < 5 μg/L, children with WAs ≥ 5 μg/L showed reductions of 6.09 FSIQ points (95% CI = −9.99, −2.19; p < 0.01), 4.97 points in Perceptual Reasoning scores (95% CI = −9.18, −0.76; p < 0.05), 4.88 points in Working Memory scores (95% CI = −9.30, −0.46; p < 0.05) and 6.22 points in Verbal Comprehension scores (95% CI: −11.12, −1.32; p < 0.02). There were no statistically significant differences in estimated points lost among the exposure groups.

Nail As and intellectual function

NAs and WAs were significantly correlated [Spearman r ₍₂₄₆₎ = 0.21, p < 0.001)]. NAs, however, was not significantly associated with changes in IQ scores, in either continuous or stratified models (data available on request).

Comparisons with findings from Bangladesh

Cross-cultural accommodations in the measurement of intelligence in our earlier work prompted the present US replication. Our findings of adverse impact in a US sample, particularly in Performance-related functioning, gives confidence to the generalizability of findings from our work in Bangladesh, where we also observed a steep drop in intelligence scores in the very low range of WAs concentrations (see Figure one in our earlier work [2]).

Because there is no currently standardized test of child intelligence for use in Bangladesh, in earlier work we adapted widely-used instruments. Most particularly, we used weighted sums of items, rather than more commonly used IQ scores. Moreover, we dropped certain items from our battery because of differences in experience that were practically universal for our study children; as examples, rural Bangladeshi children are unfamiliar with Christopher Columbus (Information subscale). Most subtests were unchanged (e.g., Digit Span, Block Design); alterations affected three or fewer items or else the entire subtest was eliminated. While these changes were minimal, they added some measure of uncertainty.

Children in Bangladesh and in the US also differ in regularity of school attendance, health and nutritional status. Stunting has consequences for later intellectual function and school progress [35]. Almost 30% of children included in our Bangladesh studies would have been characterized as stunted [36]. The negative association between WAs and Performance intelligence in the Bangladesh sample persisted, even with adjustment for stunting [2].

Differences in the distributions of WAs between the present US sample and our earlier work with Bangladeshi children likely explain some points of contrast between present and earlier findings. In much of Asia, WAs levels range far higher than in the US. The differences across samples in exposure characteristics may have affected our ability to detect effects at all points across the exposure/outcome curve. We lacked high-end exposures in the Maine sample, and we had far fewer low-end exposures in Bangladesh. For example in our initial Bangladesh studies of 6- and 10-year old children, mean WAs levels were, respectively, 120.1 and 116.6 μg/L. To illustrate further, approximately 70% of the present sample were exposed to WAs < 10 μg/L, compared to 30% among our Bangladeshi 10-year-olds [2]. Differences are also substantial when we consider WAs < 5 μg/L (52% in Maine, 25% in Bangladesh), or WAs >20 μg/L (12% in Maine, 67% in Bangladesh). At the far lower levels reported here (mean WAs = 9.9 μg/L), we still detected an adverse association attributable to WAs for Index and FSIQ scores. Conceivably, a US sample with children exposed at the higher levels seen in Bangladesh might detect continued adverse associations at those levels.

Magnitude of associations

With adjustment for other contributors, WAs ≥ 5 μg/L was associated with reductions of 4.5-6.5 points in FSIQ and in most Index scores. The magnitudes of these associations are similar to those observed with modest increases in blood lead, an established risk factor for diminished IQ. An increase in blood lead from 2.4 to 10 μg/dl or from 10 to 20 μg/dL is associated with estimated decrements of 3.9 and 1.9 IQ points, respectively [37]. Household use of the pesticide chlorpyrifos has been associated with a decline in FSIQ and Working Memory of 1.4 and 2.8 points, respectively, for each standard deviation of cord blood levels [38]. Among 7-year olds, prenatal (but not postnatal) exposure to organophosphate pesticides was associated with a loss of 7 FSIQ points, comparing highest to lowest quartile, as well as with all Index Scores [39].

Overall, the variance explained in predicting component indices of the WISC-IV was slightly smaller, relative to what we have noted in our work in Bangladesh (where R² is generally in the range of 20%-30%), most likely a consequence of the lower and narrow range of exposures examined here.

NAs as a biomarker of exposure

The significant correlation between NAs and home WAs levels suggests that household well WAs contributes to the body burden of As. On the other hand, NAs was not significantly associated with changes in IQ scores. This lack of a significant association might reflect that only 248 participating children provided nails, and that the range of NAs concentrations was narrow, so that we had less power to detect a small effect size. Alternatively, NAs may reflect exposure less accurately in children than in adults, due to the rapid growth of other organ systems in children that may alter the relative distribution of As across tissues.

Limitations

Although we were able to characterize levels of As exposure in home drinking water, we did not collect information on quantity of water consumed. At the time of our study, public water sources, such as schools, were required to evaluate drinking water and to maintain WAs levels < 10 μg/L. Compared to circumstances in many other parts of the world, where children’s consumption can be directly linked to use of wells at home and/or at school, US children access water through a wide range of processed sources (bottled water, soft drinks), complicating consumption estimates. Scheduling home visits across our largely rural communities at times convenient for staff and families proved challenging, and resulted in children occasionally not being present during the home visit; defining these children as “Missing” on HOME allowed their retention in analyses, but we were unable to characterize the quality of the home environment for 29 families. Finally, in this cross-sectional study, we were unable to characterize WAs exposure retrospectively across the lifespan. We measured current well characteristics, and length of residence, but few children had resided in the present home for their entire lives. And while using “entire life residence” as a more stringent exclusion criterion would have better characterized exposure during period(s) of peak brain development, lack of residential mobility may be a marker for other social features not considered here. As we note, when we restricted analysis to the 215 children with 5 years of continuous residence, results were essentially unchanged.

Do new analytic approaches need to be considered?

Since the 1980’s [40], the field of developmental neurotoxicology has relied upon descriptive models that adjust for other possible contributors to child intelligence, such as maternal intelligence and education. The underlying assumption in that approach is that these known predictors of child intelligence are not associated with exposure. In older studies, and in developing nations, this assumption likely remains reasonable. However, with growing public awareness of the hazards associated with certain environmental exposures, contemporary studies in developed nations need to be aware that maternal characteristics may influence exposure levels; for example, more educated mothers may be more likely to avoid exposure. Our observations have implications for the design of future studies of highly publicized potentially toxic exposures. We suggest that future studies collect data related to parental knowledge of exposure patterns and health risks of these agents. Such data can be used in deciding which parental factors to include in statistical models.