Paoletti & esbensen: Journal of aoaC international Vol. 98, no. 2, 2015 295 Distributional Assumptions in Food and Feed Commodities— Development of Fit-For-Purpose Sampling Protocols Claudia Paoletti 1 European Food Safety Authority (EFSA), Via Carlo Magno 1/A, 43100 Parma, Italy Kim H. esbensen Geological Survey of Denmark and Greenland, Copenhagen, Denmark; ACABS Research Group, Department of Chemistry and Bioscience, Aalborg University, Campus Esbjerg, Denmark Guest edited as a special report on “Representative Sampling of Food and Feed Materials” by Kim Esbensen, Claudia Paoletti, and Nancy Thiex. 1 Claudia Paoletti is employed by the European Food Safety Authority (EFSA). The positions and opinions presented in this article are those of the author and do not necessarily represent the views or scientifc works of EFSA. Corresponding author’s e-mail: claudia.paoletti@efsa.europa.eu DOI: 10.5740/jaoacint.14-250 SPECIAL GUEST EDITOR SECTION Material heterogeneity infuences the effectiveness of sampling procedures. Most sampling guidelines used for assessment of food and/or feed commodities are based on classical statistical distribution requirements, the normal, binomial, and Poisson distributions—and almost universally rely on the assumption of randomness. However, this is unrealistic. The scientifc food and feed community recognizes a strong preponderance of non random distribution within commodity lots, which should be a more realistic prerequisite for defnition of effective sampling protocols. Nevertheless, these heterogeneity issues are overlooked as the prime focus is often placed only on fnancial, time, equipment, and personnel constraints instead of mandating acquisition of documented representative samples under realistic heterogeneity conditions. This study shows how the principles promulgated in the Theory of Sampling (TOS) and practically tested over 60 years provide an effective framework for dealing with the complete set of adverse aspects of both compositional and distributional heterogeneity (material sampling errors), as well as with the errors incurred by the sampling process itself. The results of an empirical European Union study on genetically modifed soybean heterogeneity, Kernel Lot Distribution Assessment are summarized, as they have a strong bearing on the issue of proper sampling protocol development. TOS principles apply universally in the food and feed realm and must therefore be considered the only basis for development of valid sampling protocols free from distributional constraints. A fundamental part of quality control protocols is the defnition of appropriate frameworks for optimal sampling procedures to ensure accuracy and precision of surveys. Here we focus on sampling protocols for the detection and quantifcation of possible contaminants in food and/or feed commodities, but the same principles can be applied to many other food and/or feed products as well without loss of generality, regardless of the concentration level of the analyte(s) of interest: high, intermediate, or trace. Bulk food and feed sampling is a multistep procedure in which typically a composite sample is frst produced by pooling primary increments, and then mass-reduced (possibly in several steps) to obtain an ultimate laboratory sample of suitable size for analysis, i.e., the test portion, or the analytical aliquot. Among all sampling steps involved, application of composite sampling is the most critical. If the primary sample cannot be proven to be representative, all ensuing steps of mass-reduction, sample preparation, and analysis are in vain for reasons recently explained in full in the horizontal standard DS 3077 (1), where the specifc requirements for ensuring representativeness a , are addressed. Several guidelines defning sampling strategies specifc for, or routinely applied to, food and feed products are available. However, most of them are based on stringent distributional assumptions, seldom justifed or discussed in suffcient detail. Such sampling plans are based on the assumption that the analyte of interest is distributed at random. Under this assumption the mean and the SD of the analyte concentration in the lot as well as both producer’s and consumer’s risks can be effectively estimated, based on classical statistical distributions, e.g., normal, binomial, and Poisson (2–4). Yet, the critical randomness assumption is most often undocumented and therefore unjustifed and dangerous. The high likelihood of unrecognized irregular spatial heterogeneity, sometimes due to the nature of the analyte itself (e.g., mycotoxins, bacterial infections) or to agitation and transportation of lots and/or the spatial and time order intrinsic to industrial manufacturing, processing and production processes (5), will inevitably promote signifcant segregation and/or clumpiness. This type of heterogeneity has to be counteracted by the specifc sampling process used, which has to secure unbiased sampling of the entire lot. As a consequence, whenever the distribution of the material is not random, but is characterized by irregular spatial heterogeneity, a The Theory of Sampling (TOS) is the only scientifc framework that offers a valid defnition of the term “representativeness,” which is used throughout this Special Section. Representative means an analytical result, obtained from a documented unbiased sampling process with a precision variance below an a priori set threshold. The strict mathematical defnition is presented in DS 3077 and in the dedicated TOS literature. Downloaded from https://academic.oup.com/jaoac/article/98/2/295/5654444 by guest on 02 June 2023