Statistical considerations for the safety evaluation of GMOs
This opinion proposes: 1) updated statistical guidelines and possible approaches for the analysis of compositional, agronomic and phenotypic data from field trials carried out for the risk assessment of GM plants and derived foods/feeds; 2) minimum requirements that should be met in the experimental design of field trials, such as the inclusion of commercial varieties, in order to ensure sufficient statistical power and reliable estimation of natural variability. A graphical representation is proposed to allow the comparison of the GMO, its conventional counterpart and the commercial varieties with respect to many variables, taking into account natural variability. It is recommended to quantify natural variability from data on non-GM commercial varieties treated in the same way and in the same experiments as the GM and the conventional counterpart test materials. Only when such estimates are unavailable may they be estimated from databases or literature. Estimated natural variability should be used to specify equivalence limits to test the difference between the GMO and the commercial varieties. Adjustments to these equivalence limits allow a simple graphical representation so that a single pair of confidence limits may be used to display statistically significant differences and to visually assess equivalence. The possible types of outcome of this graphical representation are described and a proposal is made when further evaluation should be performed. In addition to providing specific recommendations for the interpretation of compositional analysis, this opinion highlights some statistical issues of a more challenging nature, such as the simultaneous assessment of many characteristics (i.e. multivariate analysis), which will require further research. The principles proposed in this opinion may be used, in certain cases, for the evaluation of GMOs other than plants.