Risk-based sampling design for GMO enforcement in traded plant products

Enforcement of genetically modified organism (GMO) labelling and low-level-presence (LLP) rules depends on how consignments are sampled in the field, yet most sampling standards prescribe a single uniform intensity, typically at least 50 increments per homogeneous lot, regardless of a consignment’s risk profile. We propose a risk-based sampling design that assigns high-, medium- and low-risk consignments to tiers of 50, 35 and 25 increments, with anomaly-triggered escalation and a documented Minimum Viable Protocol (MVP) for constrained field conditions. Binomial and hypergeometric solutions, checked against Monte Carlo simulation, distinguish aggregate containment from end-to-end regulatory detection probability through an explicit sample-reduction stage. Under homogeneous distribution with real kernel counts (2,500–3,300 per 0.5 kg increment), the tiers differ by at most 0.2 containment percentage points and all reach ≥99.8% even at 0.01% GM, the spread being within modelling uncertainty rather than demonstrated equivalence. Under clustered, commingled-pocket distribution, aggregate containment is governed by increment number, with model-conditional 95% boundaries at 4.3–9.9% of lot mass under distinct-segment probing and 5.8–11.3% in the non-distinct case, so suspected heterogeneity must prohibit intensity reduction. Under realistic kernel counts, end-to-end detection at 0.025% LLP remains ≥95.6% for all tiers provided at least one fifth of the aggregate sample is analysed. Quantification precision at the 0.9% labelling threshold is likewise unaffected by tier reduction. Tiering reallocates ≈30% of illustrative field effort towards high-risk cargo; the benefit is conditional on correct classification and bounded residual risk.