An alternative statistical approach for method validation under EU Regulation 2021/808

Commission Regulation 2021/808 sets performance criteria for methods for analysing pharmacologically active substances in food-producing animals. Due to the vast numbers of veterinary drugs and sample matrices, conventional re-validation of existing methods to this regulation is costly and disruptive to the delivery of routine analyses. This paper presents an alternative approach to method validation based on adding fortified samples to routine screening batches that is cost-effective and fits seamlessly within routine analytical workflows. The relationships between added concentration and each of three quantities: (1) mean measurement result, (2) the uncertainty of the mean result, and (3) the within laboratory reproducibility standard deviation are directly estimated by a maximum likelihood fit of a statistical model to measurement results. Hence, the model provides estimates of trueness, relative standard deviation and measurement uncertainty at concentrations of interest that are within the range of concentrations used in the model fit. The CCα (decision limit for confirmation) and CCβ (detection capability for screening) are calculated from estimates of measurement uncertainty. Because this approach uses data collected during routine analyses across a large number of analytical batches, rather than relying on specifically constructed validation studies, the resulting estimates of trueness and within-laboratory reproducibility better reflect real-world method performance. Estimates of performance do not rely on patterns of replication at specific concentrations or the availability of a particular number of results. Instead, the sufficiency of validation data is judged by the relative standard errors of estimates of within lab-reproducibility at concentrations of interest gained by a parametric bootstrap of the fitted statistical model. R script (a programming language used for statistical analysis) which demonstrates model fitting, performance parameter estimation and assessment of validation quality via a parametric bootstrap is provided here along with further scripts that demonstrate the method in two examples.