Exactitude et intervalles statistiques en validation de methode

Abstract

Accuracy is an important concept in method validation because it represents the global performance of the method. In general, accuracy is quantified by combining trueness and measurement precision. From the metrological point of view, it is not correct because accuracy is a qualitative concept. In method validation, practitioners use measurement error quantity instead of accuracy. They evaluate an interval IM = Xaverage± k.SFIwhere Xaverageis the average of the measurement and SFI the intermediate precision standard deviation. This interval is compared with an acceptation interval to validate the method. To choose the k value, two approaches are commonly used. The first approach takes k = 2. The second named "accuracy profile" evaluates a ß-Expectation tolerance interval which contains in average a proportion ß of measurement values. We compare these intervals to different statistical intervals (confidence, tolerance and prediction interval). Our goal is to give explanations on the interpretation of the ß-Expectation tolerance interval in the context of validation and a more user-friendly expression.