Arsenic is a ubiquitous element in the environment, where it occurs in both organic and inorganic forms. Inorganic forms are primarily present in water, rendering drinking water the primary route of human exposure to these species. In contrast, exposure to organic arsenicals primarily occurs through food ingestion [1
]. Although the primary route of total arsenic exposure globally is food ingestion, some areas (e.g., Bangladesh, Taiwan, and Chile) have elevated inorganic arsenic levels in the drinking water, rendering this the primary regional exposure pathway [2
]. Finally, some occupations, including nonferrous smelting, arsenic production, wood preservation, glass manufacturing, production and application of arsenic-based pesticides, and electronics, elevate the risk of arsenic inhalation [2
], which also entail higher urinary arsenic (UAs) concentrations and adversely affect health, similarly to arsenic ingestion [3
Concern about the role of arsenic exposure in relation to global public health has risen as some forms of organoarsenic compounds and inorganic arsenic have been classified as “carcinogenic to humans” by the International Agency for Research on Cancer and have been shown to induce oxidative stress and DNA damage [2
]. Furthermore, inorganic arsenic exposure has also been found to be a risk factor for hypertension, cardiovascular disease, and diabetes, even in populations with low levels of exposure [5
The European Food Safety Authority has detected low levels of arsenic in almost all food items (typical concentrations less than 0.25 mg/kg). Factors influencing total arsenic in the diet include food type, growing conditions, food-processing techniques, and the arsenic concentration in cooking water [1
]. Many previous studies on dietary predictors of UAs in populations with low exposure through drinking water have focused on fish and shellfish consumption [9
], as they are generally considered the foods with the highest total arsenic levels [1
]. However, several other dietary factors have been proposed to be predictors, including rice and poultry [1
], as have also demographic and behavioral factors, such as age, gender, and body mass index [11
]. However, only a few studies have explored predictors of UAs concentrations in low-level exposure populations, employing both a broad number of potential determinants and using well-validated information on UAs concentrations. From a public health perspective, a deeper understanding of the predictors of UAs concentrations will facilitate the prevention of chronic, noncommunicable diseases induced by arsenic exposure.
The purpose of the present study was to identify the demographic, dietary, and lifestyle factors that are predictors of urinary arsenic levels among postmenopausal women in the Danish Diet, Cancer and Health (DCH) cohort.
2. Materials and Methods
From 1 December 1993 through 31 May 1997, a total of 57,053 individuals (29,875 women and 27,178 men), aged 50–64 years, born in Denmark, and with no previous cancer diagnosis in the Danish Cancer Registry, were enrolled in the prospective DCH cohort [19
]. At enrolment, each participant gave a urine sample and completed a self-administered, interviewer-checked, 192-item semiquantitative food frequency questionnaire, as well as a comprehensive lifestyle questionnaire including information on smoking, occupational history, and health status. Anthropometric measures were collected by trained personnel. Among the 29,875 women, 338 had a cancer diagnosis before baseline, one had an unknown date of cancer diagnosis, and 157 did not provide a urinary sample, and were thus excluded. Among the remaining 29,379 women, 900 out of 1121 eligible postmenopausal breast cancer cases and a comparison group of 898 postmenopausal women were selected for a case-cohort study on cadmium and breast cancer, as described in detail previously [20
]. For 745 randomly selected women among these 1798 women, we measured not only cadmium but also UAs levels, and these women were included in the present study. Of the 745 women, 388 developed breast cancer between 4 years after the baseline visit and the end of 2012, and 357 women did not develop breast cancer between baseline and 2012. In order to account for this, case status is included as a covariate in the analyses.
The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the regional research ethic committee for Copenhagen and Frederiksberg (Approval no.: (KF) 01-345/93). Written informed consent was obtained from all study participants upon enrolment into the cohort. The present analysis was carried out without contact to the cohort members or their families.
Spot urine samples from each of the 745 women were shipped on dry ice and analyzed at Research Triangle Institute’s Analytical Science Laboratory (Research Triangle Park, NC, USA). Arsenic was measured in the urine samples, which were collected in transparent polypropylene cups (USON Plast, Svendborg, Denmark) and stored in transparent 1 ml polypropylene cryotubes (NUNC, Roskilde, Denmark). UAs concentrations were determined using a Thermo X-Series 2 (Bremen, Germany) inductively coupled plasma mass spectrometer (ICP-MS) following digestion in the presence of high-purity acids and oxidants in a Class 100 clean hood to prevent contamination from environmental sources. A detailed description of the ICP-MS procedure is described elsewhere [20
]. A detailed description of sample preparation procedures, analytical method, and quality control measures is provided in the supplementary methods section. Urinary creatinine concentration was quantified using a Cayman Chemicals Creatinine Assay Kit No. 500701 (Cayman Chemicals, Ann Arbor, MI, USA) with UV-VIS measurement at 500 nm employing a Beckman Coulter DU800 UV/VIS Spectrometer (Beckman Instruments Inc., Brea, CA, USA).
The distribution of the measured values of UAs concentration was right-skewed, and we therefore log-transformed the concentration prior to statistical analyses in order to obtain a normal distribution.
Association between log-transformed UAs levels and baseline study characteristics were analyzed in linear regression analyses, including univariate regression, multivariate regression including all potential predictors, and finally using an unsupervised forward regression model to determine a reduced model. The model optimization procedure added variables at each step that provided the greatest value of the adjusted R2 statistic, stopping at the step where the significance level corresponding to the addition of a predictor was greater than 0.2.
The predictor variables used in the model selection procedure were age (years), BMI (kg/m2
), alcohol intake (g/day), abstainer (yes/no), smoking status (never, former, current), years of school attendance (<8, 8–10, >10 years), area of residence (Copenhagen, Aarhus), total fish intake (g/d), total prawn intake (g/day), red meat (g/day), poultry (g/day), tap water (L/d), all vegetables (g/day), cruciferous vegetables (g/day), leafy vegetables (g/day), all fruits (g/day), all potatoes (g/day), rice (g/day), cereals (g/day), dairy products (g/day), ever worked for >1 year in the glass industry (yes/no), ever worked for >1 year in the wood industry (yes/no), estimated arsenic in drinking water (µg/L) (calculated as described in [21
]), later development of breast cancer (yes/no), and urinary creatinine concentration (g/L). Potential predictors included in the selection procedure were selected based on a priori knowledge about determinants for arsenic concentrations.
In order to exclude the possibility that included variables would be selected because of association with creatinine rather than with UAs, we constructed models using UAs as the dependent variable and added urinary creatinine as a predictor instead of using it to normalize the UAs concentration, as recommended by Barr et al. [22
]. This was done because it allows appropriate adjustment for urinary creatinine while permitting the statistical effects of the remaining covariates in the model to be independent of the effects of the creatinine concentration [22
]. As a sensitivity analysis, we also conducted the analyses with creatinine-normalized arsenic concentrations, instead of including creatinine as a covariate.
As further sensitivity analyses, we calculated all models for women in the lowest quartile of fish consumption and for women who did not develop breast cancer in order to investigate the robustness of our findings.
The procedure PROC GLM was used for regression analyses. The procedure PROC GLMSELECT was used for the model optimization procedure, specifying the option “selection = forward (select = ADJRSQ stop = SL SLE = 0.2)”. All analyses were performed in SAS version 9.4 (SAS Institute, Cary, NC, USA).
Data from the Diet, Cancer and Health cohort is not publicly available due to Danish legislation concerning protection of personal data. Admission to accessing data is based on application to the principal investigators Anne Tjønneland and Kim Overvad, through whom the application will be distributed to the steering committee of the Diet, Cancer and Health cohort, who will then process the application and return to the applicant with a final decision regarding access to data. Also, acquisition of data is only allowed after permission to handle data has been obtained from the Danish Data Protection Agency: http://www.datatilsynet.dk/english
. The e-mail address for Dr. Anne Tjønneland is: [email protected]
. The e-mail address for Dr. Kim Overvad is: [email protected]
Among the 745 women initially included in the study, one woman lacked information on covariates and was excluded from the analytical study population, which then consisted of 744 postmenopausal women.
The distribution of lifestyle factors, daily intake of major dietary groups, occupational history, arsenic consumption through drinking water, and urinary creatinine concentrations in the analyzed samples are presented in Table 1
. The study participants had a median age of 57 years and a median BMI of 25 kg/m2
. The majority of the participants lived in the Copenhagen area, and the dominant dietary groups were dairy products, fruits, vegetables, and cereal. Very few participants had worked in industries with potential arsenic exposure.
presents the distribution of UAs concentrations in the study population, both with and without adjustment for creatinine. There was wide variation in urinary arsenic concentrations among participants. The arithmetic and geometric mean UAs concentrations were 37.87 and 16.34 µg/L, respectively. The corresponding values after creatinine normalization were 63.77 and 33.77 µg/g, respectively. The correlation (RSpearman
) between UAs and urinary creatinine was 0.59.
Results of the univariate and multivariate regression model are shown in Table 3
. In the univariate regression models, alcohol, fish, prawn, and poultry intake as well as creatinine were significantly positively associated with log-transformed UAs concentrations, and tap water and dairy products were negatively associated with UAs concentrations. In the multivariate model, including all variables, alcohol, fish, and poultry intake as well as creatinine remained significantly positively associated with UAs concentration, and tap water and dairy remained negatively associated, but fruits and potatoes also showed significant inverse associations with UAs.
The unsupervised forward model selection procedure selected 11 out of 22 possible predictor variables (Table 4
). The final model explained 34.75% of the variation in UAs concentration (adjusted R2
value). This model included (in order of decreasing adjusted R2
value magnitude): Urinary creatinine levels, fish consumption, dairy product consumption, fruit consumption, alcohol consumption, potato consumption, tap water consumption, poultry consumption, working in the glass industry, vegetable consumption, BMI, and leafy vegetable consumption.
As sensitivity analyses, we repeated the above analyses for log-transformed, creatinine-normalized arsenic concentrations. The results of this were highly similar to the non-normalized analyses, with alcohol and fish being the primary predictors of higher UAs levels and fruits, potatoes, and dairy being the primary predictors of lower UAs levels. In contrast to the non-normalized model, however, poultry intake was only nonsignificantly associated with higher UAs concentrations (p = 0.12), whereas estimated arsenic in drinking water was now significantly directly associated with UAs concentration (p = 0.04). We also repeated the analyses among women who did not develop breast cancer only. The results of this were highly similar to the main analysis, with the only difference in the multivariate analysis that alcohol and poultry consumption were now only borderline significantly associated with higher UAs levels (p = 0.07), and tap water was no longer a predictor (p = 0.38).
As a final sensitivity analysis, we repeated the regression analyses for women with the lowest quartile of fish consumption (≤23.9 g/day), in order to investigate whether the predictors of UAs differed in this group (Table S1
). The non-normalized median UAs concentration in this group was 13.3 and the arithmetic mean was 30.4 µg/L. However, in the multivariate models, fish consumption, now along with prawn intake, were still the primary predictors of higher UAs concentrations after creatinine concentration, and alcohol and case status were borderline significantly associated with higher concentrations. In the forward regression model, fruit, dairy, and potato consumption were also included as predictors of lower UAs concentrations (Table S2