We analyzed data from the NHANES survey years 2003–2008. The NHANES is a nationally representative multistage random survey of the noninstitutionalized U.S. population that is conducted by the U.S. National Center for Health Statistics. Information is gathered on health status and health behaviors through in-person interviews and detailed information is collected on diet. NHANES participants also undergo a clinical examination that includes laboratory measures such as blood and urine analyses. For this study, we used data from the NHANES demographics, in-person dietary questionnaire, physical examination, and laboratory and health questionnaire files. Because our study used publicly available and de-identified data, it was determined to be exempt from institutional board review by Dartmouth College’s Committee for the Protection of Human Subjects.
Study population. We analyzed data from all children (< 18 years of age) who participated in the NHANES survey from 2003 through 2008. During this period, 13,208 children participated in the NHANES, and the response rate for the entire survey was 76%. For each NHANES survey, samples from approximately one-third of the participants were randomly selected for urinary arsenic measurements. From 2003 through 2008, 2,477 children (6–17 years of age) had urinary arsenic concentrations measured in the NHANES. Of these, we excluded 154 children because of incomplete dietary information from the 24-hr recall; this yielded a final sample of 2,323 children for our study.
Urinary arsenic assessment.
For NHANES, urine was collected from participants in arsenic-free containers and shipped on dry ice to the Environmental Health Sciences Laboratory at the National Center for Environmental Health (NCEH; Atlanta, GA) (Caldwell et al. 2009
). At NCEH, urine samples were stored frozen (≤ –70°C) and analyzed within 3 weeks of collection following standardized protocols (Aposhian and Aposhian 2006
; NCHS 2004). Total urinary arsenic concentrations were measured using inductively coupled plasma dynamic reaction cell–mass spectrometry on an ELAN DRC II ICPMS or Perkin-Elmer ELAN 6100 DRC plus (PerkinElmer SCIEX, Concord, ON, Canada); arsenic species and metabolites (arsenous acid, arsenic acid, MMA, DMA, arsenobetaine, and arsenocholine) were measured using high performance liquid chromatography (HPLC).
Method detection limits and interassay coefficients of variation (CV) varied among analytes and surveys. For total arsenic, the detection limit was 0.6 μg/L for the 2003–2004 survey and 0.7 μg/L for the 2005–2006 and 2007–2008 surveys. From 2003 to 2008, the detection limit was 1.7 μg/L for DMA, 0.9 μg/L for MMA, 1.2 µg/L for arsenous acid, 1.0 µg/L for arsenic acid, 0.6 µg/L for arsenocholine, and 0.4 μg/L for arsenobetaine. CV across NHANES lots varied from 3.0% to 6.1% for mean total arsenic concentrations, from 3.3% to 6.6% for DMA, and from 5.3% to 7.3% for arsenobetaine.
We focused our analyses on total urinary arsenic and urinary DMA concentrations becasue these were detected in most subjects. Urinary measurements of total arsenic and DMA, samples with levels below the detection limit (0.6% n
= 13 and 1.0% n
= 240, respectively) were assigned the value of the detection limit divided by the square root of 2 (Caldwell et al. 2009
; Jones et al. 2011
; Navas-Acien et al. 2008
; Steinmaus et al. 2009
). Due to uncertain or negligible health impacts of arsenobetaine and arsenocholine concentrations, we subtracted these components from the total urinary arsenic concentrations. For arsenobetaine, 48% of samples (n
= 1,109) fell below the corresponding detection limit and were assigned the detection limit divided by the square root of 2. However, because only 23 participants (1.0%) had arsenocholine measures above the detection limit, we assigned a value of 0 to all measures that fell below the detection limit (Steinmaus et al. 2009
). Thus, our definition of total arsenic included arsenous acid, arsenic acid, MMA, and DMA, consistent with previous studies (e.g., Gilbert-Diamond et al. 2011
). Arsenous acid, arsenic acid, and MMA were not considered separately due to the low levels of detection (only 6.9%, 7.8%, and 40.9% of our study had values above the detection limit, respectively).
24-hr rice consumption.
The in-person dietary questionnaire of NHANES collects detailed information on the study participant’s diet for the 24-hr period preceding the clinical and laboratory examinations (including urinary measurements) and for some measures (such as seafood consumption) up to a 30-day recall period. The NHANES 24-hr recall period is a validated assessment of dietary consumption (Moshfegh et al. 2008
). At the examination, NHANES participants were asked to recall everything they ate and drank in the prior 24 hr, and NHANES staff coded these data and recorded information on the serving size. For children < 12 years of age, the dietary component was conducted with the assistance of a proxy (i.e., a parent or other caregiver), and for children 12–17 years of age the survey was administered without the assistance of a proxy.
We used U.S. Department of Agriculture (USDA) food codes to identify rice consumed during the in-person 24-hr recall period and to classify children as “rice eaters” versus “non-rice eaters.” As in previous studies, all food data from the 24-hr dietary recall period were matched to the Food Commodity Intake Database (FCID) (USDA 2010) to quantify exposure to rice (Batres-Marquez et al. 2009
). The FCID provides conversion data to estimate the total content of food commodities such as rice, tomatoes, beans, and the like in each item with a USDA food code. We estimated the total amount of dry grams of rice consumed by each participant by multiplying the quantity of each food consumed during the 24-hr recall period by the FCID estimate of dry rice content (grams of rice per 100 g of food) for that specific food, and then summing across all foods consumed during the 24-hr recall period. To classify those children who consumed rice (rice eaters) versus those who did not (non-rice eaters), we operationally defined a rice eater as someone who consumed at least 0.25 cup of cooked rice (equivalent to 14.1 g white rice dry weight) in the 24-hr recall period (Batres-Marquez et al. 2009
We also collected data on sociodemographics (age, sex, race/ethnicity, educational status, family income), body mass index (BMI; kilograms per meter squared), exposure to cigarette smoke, drinking-water source, and seafood consumption (obtained from both the 24-hr recall period and 30-day food recall questions). We estimated the percentages of the population that were normal weight, overweight, and obese by converting measured BMI to percentiles based on age and sex (< 85th percentile, normal; 85th to < 95th percentile, overweight; and ≥ 95th percentile, obese) (Centers for Disease Control and Prevention 2010
). We anticipated that race/ethnicity would be related to rice consumption and therefore classified race/ethnicity as non-Hispanic white, non-Hispanic black, Mexican American, and other, multiple races. Due to the small number of individuals, the “other Hispanics” NHANES category was combined with the “other/multiple” race/ethnicity.
Because cigarette smoke is a potential source of arsenic exposure (Chen et al. 2004
), we used serum cotinine to estimate passive or active exposure to cigarette smoke. The NHANES measures serum cotinine using an isotope-dilution HPLC/atmospheric pressure chemical ionization mass spectrometry method. For values below the detection limit of 0.015 ng/mL, a value of the detection limit divided by the square root of 2 was assigned (Jones et al. 2011
; Navas-Acien et al. 2008
We used urinary creatinine to account for urinary dilution (Barr et al. 2005
). In the 2003–2004 and 2005–2006 NHANES surveys, urinary creatinine was measured on a Beckman Synchron CX3 (Beckman Coulter Inc., Brea, CA) using a Jaffe reaction, and in the 2007–2008 panel it was measured on a Roche/Hitachi Modular P (Roche Diagnostics Corp, Indianapolis, IN) using an enzymatic method. Therefore, we adjusted 2003–2004 and 2005–2006 urinary creatinine measurements to 2007–2008 equivalents (Gebel 2002
; NCHS 2009).
In the United States, arsenic exposure through drinking water is found primarily in private unregulated water systems (Karagas et al. 2000
; Nuckols et al. 2011
). Although we were unable to obtain measurements of arsenic in drinking water for NHANES participants, we used their self-reported drinking-water source to estimate potential exposure as either public (using a community water source) or private (defined as either a well, spring, or cistern water source) water.
To exclude the possibility that seafood contributed to forms of arsenic exposure other than arsenobetaine or arsenocholine, such as DMA, we used the USDA food codes that correspond with fish, shellfish, mollusks, and/or crustaceans to identify children who consumed any seafood during the 24-hr recall period in our primary analyses [see Supplemental Material, Table S1
)] (Navas-Acien et al. 2011
). Furthermore, because seafood consumption may affect urinary arsenic concentration for up to 3 days, we also performed a secondary analysis in which we restricted our sample to children who reported no seafood consumption in the 30 days before urinary arsenic measurement (Molin et al. 2011
) (see Supplemental Material, Table S2
Statistical analyses. The NHANES uses a stratified sampling methodology that makes it possible to derive national estimates from survey participants’ data. To account for this sampling, we used complex survey design methods in Stata version 12.0 (StataCorp., College Station, TX) for all analyses. These methods account for a respondent’s probability of selection and for the NHANES sampling methodology by calculating weighting factors for each respondent that account for sampling strata, primary sampling units, and person weight variables (NCHS 2005). For all analyses we set the p-value for statistical significance to 0.05 (2-sided).
Because metabolic processes may vary according to a child’s age (Hall et al. 2009
) and NHANES dietary data were collected differently according to age (i.e., with and without a parent or caregiver), we stratified our sample into age groups of 6–11 years and 12–17 years. Analyses were performed on all ages as well as according to these two age categories.
-transformed total urinary arsenic (the original NHANES total arsenic measure minus arsenobetaine and arsenocholine) and urinary DMA concentrations. This transformation produced a linear association with rice consumption for total urinary arsenic (lack-of-fit p
-value = 0.45; Draper and Smith 1981
), and improved the homoskedasticity and normality of model residuals, for both total urinary arsenic and urinary DMA. For this lack-of-fit test, the null hypothesis is that there is no bias—that bias error and pure error are approximately the same; the null hypothesis thus is rejected when an F
-statistic comparing bias error to pure error exceeds a critical value. The exponentiated model coefficients represent the relative (percent) change in the dependent variable from its mean value at the reference level of exposure (Vittinghoff et al. 2005
Urinary creatinine can be a strong predictor of arsenic methylation efficiency; thus, we included it as an independent variable in our multiple regression models (Barr et al. 2005
). However, analyses with and without creatinine yielded similar results. We adjusted for potential confounding using three different models. Our baseline adjustment model (model 1) included age (continuous), sex (boy/girl), race/ethnicity (non-Hispanic white/non-Hispanic black/Mexican American/other, multiple races), and urinary creatinine concentration (continuous). Additionally, we fit a model that further adjusted for BMI (as a continuous variable) and serum cotinine concentration (continuous) (model 2). The final model additionally adjusted for water source (public/private) and was restricted to only those children who reported no seafood consumption during the 24-hr recall period (model 3). As a secondary analysis, we repeated model 3 restricting the model to children who reported no seafood consumption in the 30-day food recall questions [see Supplemental Material, Table S2
We used rice consumption as a predictor variable in two ways. First, we treated rice consumption as a dichotomous variable, evaluating selected population characteristics and urinary arsenic variables (total arsenic and DMA urinary concentrations) according to whether the study participant consumed ≥ 0.25 cup of cooked rice during the 24-hr recall period. To compare the characteristics of study participants we used a chi-square test for categorical variables. We then explored the potential dose–response relationship between 0.25 cup cooked rice consumed during the 24-hr recall period and log10-transformed total arsenic and DMA using multiple linear regression as described.