Total error vs. measurement uncertainty: the match continues.

Although 5 years have passed since the publication of a previous editorial in this journal that tried to reconcile the concepts [1], laboratory professionals are still playing the total error (TE) vs. uncertainty game. Two papers published in this issue of Clinical Chemistry and Laboratory Medicine further discuss the subject [2, 3], which was revitalized by the organization of the 1st EFLM Strategic Conference in Milan on the analytical performance specifications at the end of 2014 [4]. The organizers considered it timely to address the topic of performance specifications both because it was 15 years since it was previously addressed and because performance specifications are more and more central for the clinical application of test measurements [5, 6]. Being members of the scientific program committee of the Milan conference, we could be biased in commenting on papers dealing with the conference topics. However, it appears useful to add some thoughts here. We should first underline that the consensus statement from this conference in itself does not address the problem of TE vs. uncertainty [7]. This is indeed outside the scope of the statement as it concentrates on models to set performance specifications. The TE and measurement uncertainty concepts were discussed by some of the speakers at the conference and taken forward by one of the Task and Finish Groups (TFG) established during the conference [4]. Westgard’s opinion paper contains some criticisms that deserve clarifications [2]. Similarly to a previously published paper on the same topic [8], the paper is mainly a defense of the TE concept the same author promoted 40 years ago more than an objective comment on the need to improve the definition and the quality of performance specifications, which was the main topic of the Milan conference. We note some misconceptions in the interpretation of the Milan consensus [7], for example, when the author considers the “Milan simplification from 5 to 3 levels of models a modest adjustment of the Stockholm guidance”. The three models proposed in the consensus document from Milan are not necessarily a hierarchy. The recommendation is to use model 1 or 2 depending on the nature and characteristics of the measurands and depending on their clinical application (i.e. diagnosis, monitoring, etc.). Accordingly, Westgard is ignoring the need, which explains the creation of a TFG by EFLM [4], to allocate different tests or clinical applications of the same test to different models for performance specifications recognized in the consensus statement of the Milan conference, preferentially 1 or 2. Another ignored aspect is the quality of the information, which in our view is one of the novelties of the Milan conference. In Westgard’s paper, biological variability is strongly promoted as a source of analytical performance specifications. This can be a good and usable model for many measurands, but it must be underlined that there is a clear need for evidence-based data and that studies and papers with information about biological variation have to be critically appraised [9, 10]. The utility to elaborate specifications at different levels of quality (i.e. minimum, desirable and optimum) to move, in case, from desirable to minimum quality goals and, in the meantime, ask IVD companies to work for improving the quality of assay performance is also an important aspect [11]. Finally, a confounding approach in the paper of Westgard [2] is the tendency to mix statements contained in the individual papers published in the conference proceedings with agreed concepts that became part of the consensus statement. An example is represented by the content of Petersen’s paper [12], which is considered “tout court” the “preference” of the Milan audience. It should be underlined that each of the published papers from the Milan conference represents the opinion of their authors and that there is only one “consensus” paper from the conference [7]. Referring to the TE utility, in his paper Westgard tries to create a sort of “black and white” situation between TE and measurement uncertainty. If we agree that TE may still be useful, for example, in evaluating single EQAS results, there is also a need that all stakeholders working in the field of Laboratory Medicine look carefully to the measurement uncertainty as this help very much in the identification of important sources of