Refined methodologies for probabilistic dietary exposure assessment for food contaminants based on the observed...

The Observed Individual Means (OIM) methodology, based on the non-parametric bootstrap, is usually employed to perform basic probabilistic dietary chronic exposure assessment, and assumes independence and identical distribution of occurrence data within food category. However, this assumption may not be valid if several expected distributions of occurrence can be a priori identified within food category. Moreover, OIM assumes each analysed food sample to equally contribute to mean occurrence, as information about relevance of each food item cannot be incorporated into exposure assessment.

In this paper we address the above-mentioned violations and develop two statistical methodologies to accommodate for them into OIM.

The stratified non-parametric bootstrap and weighted mean occurrence are employed to correct for such violations. As a case study, we compare the methodologies by estimating the exposure of the adult Italian population to the process contaminant 3-monochloropropane-1,2-diol.

We propose strategies to interpret their results and show their relevance in conducting exposure assessment.

For the first time in the literature, we critically examine a widely used methodology for Probabilistic Dietary Exposure Assessment from a statistical perspective, focusing on the underlying assumptions and their potential violations in real-world scenarios. We then develop techniques to address these violations, providing a more accurate and robust approach to exposure assessment. This work is particularly relevant for risk assessors and managers, since it offers a refined toolset for more precise exposure assessments.