Is there an optimal number of urinary samples to assess exposure to fast elimination endocrine disruptors?

Abstract

In epidemiological studies investigating associations between exposure and health outcomes, exposure is usually assessed by the concentration of biomarkers (pollutants or metabolites) in one urine or blood sample. The biomarkers concentration is used as a surrogate of the internal dose suspected to be responsible for the observed adverse biological effect. Contaminants such as phthalates or bisphenols, which are suspected endocrine disruptors, are rapidly metabolized and eliminated in urine, resulting in a highly variable concentration over a short time. This high variability represents the main limitation to the classical approach of using only one sample to assess exposure and may lead to erroneous conclusions. Although increasing the samples may provide more reliable information, no rational is available yet to determine the optimal number of samples needed. Sixteen volunteers (8 men, 8 women) provided between 1 and 3 urine samples per week over 6 months leading to a total number of 42–57 per subject. Samples were analyzed for a series of endocrine disruptors including 16 phtalates metabolites, 9 pesticides metabolites and 4 bisphenols. In order to determine the minimum number of urine samples necessary to reliably classify the volunteers according to their level of exposure (based on the average concentration), 2 different algorithms based on bootstrap-like simulations were tested. In the first one, two types of ranking have been implemented: individual-based and quartile-based. On this algorithm, several parameters have been tested: numbers of bootstrap iterations, re-sampling samples after selection or not, log-transformation of the data. The second algorithm aimed at determining the number of samples needed to reach a stable mean value over time. Most biomarkers presented high detection rates, with 21 metabolites detected in at least 80% of samples. The first algorithm, based on the average of 40 samples per subject, demonstrated that individual-based ranking provided poor results, with a maximum of 3% of correct ranking whatever the biomarker, whereas quartile-based ranking allowed reaching up to 46% of correct ranking. Log-transformation of the data improved the rate of correct ranking up to 65%. The results were not influenced by the number of bootstrap iterations but the rate of correct ranking was significantly lower when the re-sampling was allowed. The second algorithm demonstrated that between 15 and 20 samples per subject were needed to reach a stable mean concentration. The high detection rate of the urinary biomarkers tested here highlights the frequency of human exposure to endocrine disruptors and their ubiquity in human surroundings. The statistical analyses applied here demonstrated the complexity to classify individuals according to their exposure level and underlines the irrelevance of classification based on a single sample. The two approaches (individual-based and quartile-based) used in the first algorithm demonstrated that for fast elimination pollutants, increasing the number of samples only slightly improves the reliability of the classification. The present study highlights the limitation of the classical approaches generally used in epidemiological studies, where exposure assessment is based on a single urine sample. Although increasing the number of samples provides and average value, which is more reliable than a spot sample, the variability in biomarkers concentration may still limit the benefit of multiple sampling. These results strongly incite to take into account the variability of urinary biomarkers concentration in the design of epidemiological studies or to consider other biological matrices (e.g. hair) more adapted to assess exposure to fast elimination pollutants.