Modelling the effect of sampling methods on detection tests for powdered products

Abstract Grab sampling is often a convenient and cost effective way to sample bulk food materials, such as milk powder. On the other hand, modern auto-samplers can sample very small increments directly from the production process and they can be set to collect primary increments systematically. While the quantity of sampled bulk material is important, it is also necessary to consider the impact of sampling on quantitative risk assessment. When grab samples are drawn, the principle of randomisation is only partially met because of the inability to draw small primary increments at random. Food contamination (microbiological or otherwise) does not occur uniformly, and is often patchy or heterogeneous within a batch. Hence, even random sampling of primary increments does not amount to random sampling of pathogens or contaminants. As a consequence, the consumer's risk is underestimated under the holistic assumption of complete randomisation. In this theoretical study, a correlation parameter is introduced to allow for lack of independence in the presence and absence of contamination, and then the effect of the various sampling methods on the consumer's risk is examined. The main conclusion from this study is that grab sampling can increase the consumer's risk by as much as 50 % and hence additional sampling is necessary when grab samples are used for lot disposition when compared to directly sampling the product from the process, which can be done using auto-samplers.