First Steps toward Harmonized Human Biomonitoring in Europe: Demonstration Project to Perform Human Biomonitoring on a European Scale

Background For Europe as a whole, data on internal exposure to environmental chemicals do not yet exist. Characterization of the internal individual chemical environment is expected to enhance understanding of the environmental threats to health. Objectives We developed and applied a harmonized protocol to collect comparable human biomonitoring data all over Europe. Methods In 17 European countries, we measured mercury in hair and cotinine, phthalate metabolites, and cadmium in urine of 1,844 children (5–11 years of age) and their mothers. Specimens were collected over a 5-month period in 2011–2012. We obtained information on personal characteristics, environment, and lifestyle. We used the resulting database to compare concentrations of exposure biomarkers within Europe, to identify determinants of exposure, and to compare exposure biomarkers with health-based guidelines. Results Biomarker concentrations showed a wide variability in the European population. However, levels in children and mothers were highly correlated. Most biomarker concentrations were below the health-based guidance values. Conclusions We have taken the first steps to assess personal chemical exposures in Europe as a whole. Key success factors were the harmonized protocol development, intensive training and capacity building for field work, chemical analysis and communication, as well as stringent quality control programs for chemical and data analysis. Our project demonstrates the feasibility of a Europe-wide human biomonitoring framework to support the decision-making process of environmental measures to protect public health. Citation Den Hond E, Govarts E, Willems H, Smolders R, Casteleyn L, Kolossa-Gehring M, Schwedler G, Seiwert M, Fiddicke U, Castaño A, Esteban M, Angerer J, Koch HM, Schindler BK, Sepai O, Exley K, Bloemen L, Horvat M, Knudsen LE, Joas A, Joas R, Biot P, Aerts D, Koppen G, Katsonouri A, Hadjipanayis A, Krskova A, Maly M, Mørck TA, Rudnai P, Kozepesy S, Mulcahy M, Mannion R, Gutleb AC, Fischer ME, Ligocka D, Jakubowski M, Reis MF, Namorado S, Gurzau AE, Lupsa IR, Halzlova K, Jajcaj M, Mazej D, Snoj Tratnik J, López A, Lopez E, Berglund M, Larsson K, Lehmann A, Crettaz P, Schoeters G. 2015. First steps toward harmonized human biomonitoring in Europe: demonstration project to perform human biomonitoring on a European scale. Environ Health Perspect 123:255–263; http://dx.doi.org/10.1289/ehp.1408616


Identification of determinants of exposure
For each biomarker of exposure the relation with possible determining factors was studied by univariate and multiple regression techniques. All analyses were done separately for the mothers and the children.
Confounders and covariates are listed in Supplemental Material, Table S1. Confounders are a priori defined variables that are known to be related to the biomarker. Covariates are possible determinants; its relationship with the biomarker is tested within the study group. All confounders and covariates were put in the models as categorical variables.
Both for univariate and for mutiple models, linear mixed models were used instead of the ordinary linear regression models. Hence, the clustered design was taken into account in mixed effect analysis. Within a country, participants were recruited in a similar way, and thus the mothers (or children) within one country may be considered as dependent measures. Mothers (or children) from one country may have 'more identical' biomarker values than mothers (or children) from another country. This dependency between the biomarker values could be introduced into a model by a random effect. The correlation between the biomarker values within a country was estimated by the model. The intraclass correlation coefficient gives the proportion of variability in the biomarker values due to the variability between countries. The introduction of the random effects into the model will change the confidence intervals of the estimates.
Explanatory variables of interest were included in the model as fixed effects.
First, univariate models were developed for all covariates. In a second step, multiple regression models were built including those determining factors which are significant at the 0.25 3 significance level in the univariate analyses. The confounders are fixed into the model. Important determining factors were identified by stepwise selection procedures (this is a combination of forward and backward selection procedures) in which we set p<0.05 to stay in the model. As such a final linear mixed model is obtained.
Quantitative relationships between the covariates and the biomarkers were calculated from the estimates of the linear mixed model, assuming that, when quantifying the relation of one covariate with the biomarker, all other covariates in the model are fixed at the population mean.
The problem of multicollinearity, that is the existence of a high degree of linear correlation amongst two or more explanatory variables in a regression model (Neter et al. 1996) was examined. Multicollinearity makes it difficult to separate the effects of the explanatory variables on the dependent variable. In the presence of multicollinearity, the estimate of one variable's impact on Y while controlling for the others tend to be less precise than if the predictors were uncorrelated with one another. Spearman correlation coefficients between the different explanatory variables were calculated; highly correlated variables were not included in the same model. The effects of multicollinearity were analyzed using variance inflation factors. If the variance inflation factor was larger than 10 then multicollinearity was concluded (Fox 1991). Table S2. List of confounders (Conf) and covariates (Cov) to be examined in relation with biomarkers of exposure.

Descriptive statistics
The descriptive statistics for all biomarkers are presented in Supplemental Material, Table S3.
The data are given in µg/g hair for mercury and both in µg/L and µg/g creatinine for urinary markers, separately for children and mothers.
Geometric means and percentiles are calculated on the basis of the 'raw' data, i.e. without weighing or correction for clustering, and without adjustment for confounders or covariates.

Identification of determinants of exposure
The results of the multiple regression models are given in Supplemental Material, Table S4 to   Table S19. • not significant (p=0.22) after weighing for unequal numbers per country.

Comparison of results between countries
The results of the comparison between countries are given in Supplemental Material, Table S20 to Table S35. LOQ: limit of quantification; GM: geometric mean; CI: confidence interval. a p-values for comparison between countries are calculated as follow: in a first step, overall significance of country is tested by a linear regression model (see p-value ALL in first row); in case of an overall significant difference between the countries (p<0.05), a post hoc 27 28 analysis is done and the mean exposure level in each country is compared with the European exposure value (see p-value per country in the following rows).        MEP: mono-ethyl phthalate; LOQ: limit of quantification; GM: geometric mean; CI: confidence interval.
a Adjusted for age, gender and creatinine. b Adjusted for age and gender  MBzP: mono-benzyl phthalate; LOQ: limit of quantification; GM: geometric mean; CI:confidence interval.
a Adjusted for age, gender and creatinine. b Adjusted for age and gender. MBzP: mono-benzyl phthalate; LOQ: limit of quantification; GM: geometric mean; CI: confidence interval.
a Adjusted for age and creatinine. b Adjusted for age. MnBP: mono-n-butyl phthalate; LOQ: limit of quantification; GM: geometric mean; CI:confidence interval. a Adjusted for age, gender and creatinine. b Adjusted for age and gender. MnBP: mono-n-butyl phthalate; LOQ: limit of quantification; GM: geometric mean; CI: confidence interval.
a Adjusted for age and creatinine. b Adjusted for age.  MiBP: mono-iso-butyl phthalate; LOQ: limit of quantification; GM: geometric mean; CI: confidence interval.
a Adjusted for age and creatinine. b Adjusted for age.