Journal of chromatography. A

Derivation of uncertainty functions from validation studies in biological fluids: application to the analysis of caffeine and its major metabolites in human plasma samples.

PMID 24925447


Procedures for estimating the measurement uncertainty (MU) of the concentration of a given analyte in a sample are of major concern for analytical chemists. Unfortunately, it is still unclear how and why MU should be assessed. While several possibilities exist, an appropriate approach consists in using method validation data for the evaluation of MU. This was demonstrated by a validation study achieved in the framework of a clinical study related to caffeine in sports medicine, where the results were used for the evaluation of MU. After validation of the method developed using ultra-high pressure liquid chromatography-mass spectrometry for caffeine and its three main metabolites, accuracy profiles were built for each analyte. The first important conclusion is that the developed method was valid for all compounds and met the given specifications for the application (fit for purpose). Relevant estimates of combined standard uncertainty were computed to obtain uncertainty functions, which allow obtaining values of MU as a function of the concentration of the analyte. The great advantage of both uncertainty function and uncertainty profile is the development of a continuous model that enables easy calculation of the standard, expanded and relative expanded uncertainty at any concentration within the validation domain. In fact, the expanded uncertainty interval is assumed to contain 95% of all possible measurements, regardless of the concentration. Finally, the uncertainty function enables the determination of the lowest limit of quantification by selecting adequate acceptance limits, with the limit of quantification being defined as the point where the relative uncertainty equals the acceptance limit threshold. It has to be noted that further discussions remain mandatory to establish which criteria should be applied to define an adequate decision threshold, and the proposal afforded in this work may open new avenues in this direction.