The identification of the major associations of pesticides to which the population is exposed is the first step for the risk assessment of mixtures. Moreover, the interpretation of the mixtures through the individuals' diet and the characterization of potentially high-risk populations constitute a useful tool for risk management. This paper proposes a method based on Non-Negative Matrix Factorization which allows the identification of the major mixtures to which the French population is exposed and the connection between this exposure and the diet. Exposure data of the French population are provided by the Second French Total Diet Study. The NMF is implemented on consumption data to extract consumption systems which are combined with the residue levels to link dietary behavior with exposure to mixtures of pesticides. A clustering of the individuals is achieved in order to highlight clusters of individuals with similar exposure to pesticides/consumption habits. The model provides 6 main consumption systems, 6 associated mixtures of pesticides and the description of the population which is most exposed to each mixture. Two different ways to estimate the matrix providing the mixtures of pesticides to which the population is exposed are suggested. Their advantages in different contexts of risk assessment are discussed.
Keywords: BIC; BMI; Bayesian information criterion; Body Mass Index; CS; Combined exposure; Dietary patterns; Food risk assessment; LB; LOD; NMF; Non-Negative Matrix Factorization; Pesticides mixtures; TDS; Total Diet Study; UB; consumption system; limit of detection; lower-bound; upper-bound.
Copyright © 2013 Elsevier Ltd. All rights reserved.