Evaluation of sampling plans for measurement of gluten in oat groats

The presence of gluten-containing grains in oat groats is not uncommon. Many countries with regulations on “gluten-free” labeling of foods, such as the US, EU and Canada, have a 20 mg/kg (ppm) limit and allow “gluten-free” claims on oat products, provided that the limit is not exceeded. The non-uniform spatial distribution of gluten-containing grains in a bulk lot poses a challenge when selecting a representative sample of oat groats from a lot to determine if the lot is within regulatory compliance limits. A probability-based method for evaluating sampling plan designs was developed. A balanced nested experimental design was used to estimate gluten concentration in 16 laboratory samples from each of 10 mini-lots of oat groats spiked with varying amounts of wheat kernels used as the gluten source. The total variance of the gluten test procedure was partitioned into the variances between laboratory samples (Vs), test portions (Vtp) and aliquots tested (Va). From regression analysis, each variance was found to be a function of gluten concentration (G): Vs=(100/Ns)35.0880G, Vtp=(1/Ntp)20.0078G and Va=((1/Na)0.0264)G1.5167, where Ns is laboratory sample size in g, Ntp is test portion size in g, and Na is number of aliquots analyzed for gluten. The observed gluten distribution among samples of oat groats tended to follow gamma and negative binomial distributions compared to normal and lognormal distributions, especially at low gluten concentrations. An R program was developed using the relation of variance with gluten concentration and negative binomial distribution to compute and plot an operating characteristic (OC) curve for various gluten sampling plan designs. The OC curve was used to predict the acceptance (or rejection) probability of a bulk lot at given gluten concentration by a specific sampling plan design. The effect of change in laboratory sample size, test portion size, number of laboratory samples and manufacturer's accept/reject limit (generally lower than the regulatory limit) on the OC curve was determined.