Evaluation of a Food Frequency Questionnaire-Food Composition Approach for Estimating Dietary Intake of Inorganic Arsenic and Methylmercury1

Inorganic arsenic intake in 969 men and women and methylmercury intake in 785 men and women from across the United States were assessed by a semiquantitative food frequency questionnaire, in combination with a database for the content of those elements in foods, and by toenail concentrations of arsenic and mercury. In addition, empirical weights for foods on the dietary questionnaire were derived from multivariate regression models to estimate associations between diet and toenail arsenic and mercury levels, independent of the assumptions about inorganic arsenic and methylmercury in foods, which are based upon limited residue measurements. The use of empirical weights significantly improved the correlation of arsenic consumption with toenail arsenic levels (r 033, P 0.0001), compared with the weak correlation obtained using the food residue method to calculate intake (r = 0.15, P = 0.0001). Mercury consumption computed using empirical weights yielded a significant correlation with toenail arsenic (r = 0.42, P = 0.001), similar to the correlation using energy-adjusted intake calculated from food residue tables (r 0.35, P 0.001). These results illustrate the potential use of empirically derived weights for foods in estimating toenail levels of selected heavy metals and support the validity of published food residue data that are used to estimate mercury consumption. Introduction Dietary consumption is recognized to be a primary pathway of exposure to heavy metals and other compounds that are potential toxicants (1), and yet the relationship between ingestion of specific food items and individual exposure to such metals has not been elucidated. This information would be valuable for evaluating the relationship between personal exposure and possible health effects. For example, both arsenic and mercury are known to be associated with serious adverse health effects, including cancer, but methods for estimating exposure to these elements from food intake have not been validated. Epidemiobogical studies of populations that are exposed to arsenic have indicated that oral ingestion of arsenic is a factor in the induction of nonmebanoma skin cancer in humans (2) and may also be a cause of some internal cancers, such as bladder and kidney cancers (3-5). Associations between arsenic exposure and noncancer health outcomes, e.g. , diabetes meblitus, have been observed as well (6, 7). However, there is also some evidence for an arsenic requirement in mammals (8). Associations between mercury exposure and selected cancer outcomes in humans have been observed in occupational settings (9, 10). In addition, mercury exposure has been associated with various noncancer health outcomes in humans, including myocardial infarction, cardiovascular disease, neurodevelopmental effects in children, and nephrotoxicity (11-13). Exposures to inorganic arsenic and methylated mercury are of principal interest because of their relative uptake, accumulation, and toxicity in humans (14, 15). The objective of our research was to develop a simple reliable method for estimating chronic dietary consumption of arsenic and mercury. We evaluated the ability of dietary inorganic arsenic and methylmercury scores, derived from a semiquantitative food frequency questionnaire and food residue tables, to predict bevels of arsenic and mercury in toenails collected from human volunteers. To investigate alternative methods for assessing dietary exposure, we compared the results of the food composition approach to the accuracy of toenail arsenic and mercury as dietary predictors, estimated from regressions of the toenail levels and food consumption data alone. Materials and Methods Toenail mercury, demographic, diet, and lifestyle (e.g., smoking) data were obtained from 384 women, ages 30-55 years, in 1976 and 585 men, ages 40-75 years, in 1986 at time of enrollment in the Nurses’ Health Study and the Health Professionabs Follow-Up Study, respectively. Sixty % of the men were dentists and, thus, may have been occupationally exposed to mercury. Women were controls in a nested case-control study of toenail trace element levels and breast cancer (16), and the data from men are currently being used in a nested casecontrol study of trace elements and heart disease. Toenail on August 27, 2017. © 1997 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from /044 Toenail As and Hg as Biomarkers of Dietary Intake arsenic data were available for 785 members of this population, 210 women and 575 men. Among the women, a set of nail clippings from all 10 toes were obtained between 1982 and 1983, whereas those from the men were obtained in 1987. Concentrations of I 6 elements, including arsenic and mercury, were determined by instrumental neutron activation analysis at the University of Missouri Research Reactor; details of this study may be found in Garland et a!. (17). Reproducibility of arsenic and mercury in toenails, collected 6 years apart, was shown to be high (Spearman correlation coefficients of 0.54 for arsenic and 0.56 for mercury), indicating that toenail levels of these elements are reasonably stable indicators of exposure (17). Every 2 years, participants in these prospective cohort studies completed follow-up questionnaires that updated information on potential risk factors and disease, including a semiquantitative food frequency questionnaire of the previous year’s diet (18, 19). We used food frequency questionnaires and demographic and lifestyle data that were collected in 1984 for the women and in I 986 for the men in our analyses, except for state of residence for the women, which was collected at the time of enrollment. Average daily dietary intakes of inorganic arsenic and methylmercury were estimated for each participant using methods described elsewhere (20). Food composition data for total arsenic and mercury were adjusted to reflect the best available information on the percentage present in absorbable chemical forms of these elements, i.e. , inorganic arsenic and methylmercury. Inorganic arsenic was estimated to account for 1 .5% of total arsenic in fish and 20% of total arsenic in shellfish, based upon data reported in the literature (21-23). Because of the absence of arsenic speciation data for nonfish foods, 100% of total arsenic measured in all other foods was assumed to be the inorganic species. Following Bloom (24), 100% of total mercury in fish and 65% in shellfish was assumed to be methylmercury. Similarly, 95% of total mercury in pork and beef and 70% in eggs was assumed to be methylmercury (25). Similar to arsenic, 100% of mercury in all other foods was assumed to be methylmercury due to an absence of food-specific speciation data. For all analyses, the dependent variable (either toenail arsenic or mercury) was first transformed to the natural bogarithm scale and then adjusted for the following variables in a general linear model: age, gender, body mass index (in kg/m2), weight of nail sample, caloric intake, state of residence, smoking status (current, former, or never), and analysis batch; for mercury, the indicator variable “dentist” was also included to account for possible substantial occupational exposures to mercury. The residuals from this model, added to the mean of the logarithmically transformed toenail arsenic or mercury data, then served as the response variable for evaluating food items, allowing an assessment of their predictive value that is apart from demographic and laboratory variability. The validity of the estimated average daily dietary consumption of inorganic arsenic and methylmercury was evaluated by computing the Spearman correlation coefficients between the estimated intakes and the adjusted toenail data. This approach for characterizing intake requires a number of assumptions regarding the food composition data, principally, homogeneity ofelement levels in individual servings of specific food items over time and across geographic regions and food preparation methods. To the degree that such assumptions are not true, the validity of the estimated dietary intakes can be expected to decrease. An alternative procedure that is independent of food composition data exists, in which standard multivariate regression models are fit to evaluate the relationship between consumption of food items and levels of the biomarker of interest (26, 27). The resulting model and associated empirical weights (i.e., regression coefficients) can be used to estimate consumption from diet data alone. The accuracy of the empirical weight models can be evaluated by predicting the biomarker levels for a population that is independent of the one from which the model was developed. Three different approaches toward empirical weight model selection were considered: these will be referred to as the percentage contribution approach, the bivariate regression approach, and the stepwise regression approach. For the percentage contribution approach, only those food items that were estimated to contribute at beast 0.05% of the total daily consumption of the element of interest were considered to be potential predictors of toenail arsenic and mercury and were included in the mubtivariate model. Those food items with associated P-values that were >0.2 were then discarded, and the model was rerun to obtain a final fit for the major predictors. For the bivariate regression approach, each food item was first considered individually as a predictor of the toenail response. All those food items meeting the 0.2 level of significance were then collected as joint potential predictors in a multivariate model. After running this model once, food items were eliminated if their corresponding P-values exceeded 0.2, and the multi-variate model was refit to provide final coefficients for the major predictors. Finally, the stepwise regression model initially included all food items and used a standard forward/ backward approach with entry and exit significance levels of 0.2. The use of these three different approaches for model selection is advantageous, in that it allows an assessment of the consistency of specific food items as major predictors of toenail arsenic and mercury; similar model-checking approaches have been used in other assessments of dietary predictors of biomarkers (28). Because of the barge number of food items and the fact that some food items are likely to be strongly correlated with others, we also undertook a full evaluation of regression diagnostics, such as evaluations of colbinearity between predictors and influential data points. Two methods, halving and cross-validation, were used to evaluate the regression models that were developed from each of the model selection approaches. For the halving method, half of the data were selected at random and used to fit the model, which, in turn, was used to predict adjusted toenail and mercury for the remaining half of the dataset. For the cross-validation method, the model was fit to n 1 observations and was then used to predict for the withheld observation; the process was repeated n times and, thus, afforded a predicted value for every observation. The predictive ability of the regression models was quantified by Spearman correlation coefficients between predicted and observed toenail arsenic and mercury. Results Toenail arsenic was highly variable among individuals, with a mean of0.20 ppm, a SD of 1 .94 ppm, and a range of 0.02-53.98 ppm. Toenail mercury was also quite variable, with mean of 0.74 ppm, a SD of 3.53 ppm, and a range of 0.003-86.37 ppm. The mean estimated average daily dietary consumption of inorganic arsenic was 10.22 pg/day, with a SD of 6.26 p.g/day and a range of 0.93-104.89 pg/day, whereas the mean estimated average daily dietary consumption of methybmercury was 7.76 pg/day, with a SD of 6.54 pg/day and a range of 0.36-123.84 .tg/day. on August 27, 2017. © 1997 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from