Measurement uncertainty – a revised understanding of its calculation and use

Measurement uncertainty (MU) is an important element of ISO 15189 [1]. Laboratories are required to determine MU for each measurement procedure at the time of validation or verification and to review MU over time. This enables laboratories to assess the measurement procedure as being fit-for-purpose for clinical diagnosis according to a quality standard and in the longer term, whether the procedure maintains its performance for monitoring of disease response. Farrance et al. [2], in their opinion paper on MU, describe how “Westgard style” total error and “guide to the expression of uncertainty in measurement (GUM) style” uncertainty in measurement are not completely opposed methodologies and that the GUM style [3] can be applied where there are not only random errors (imprecision) but also a systematic error (bias) [4]. As the authors explain, although bias can be a problem for many tests for various reasons such as a lack of between-method standardization or harmonization, bias may be minimal and may not need to be incorporated into the MU determination provided the bias is within acceptable limits within the same method group.

They went on to say, “In addition, as result interpretation is largely by comparison (comparison to a population reference interval, comparison to a previous result or comparison to a clinical decision point), bias becomes largely irrelevant provided it remains internally consistent and consistent with the comparator. In this situation, both total error and MU essentially become an assessment of imprecision”.

One concern, however, may be the calculation of MU for clinical decision limits where multiple measurement procedures are used to establish the cutoff value. In the case of glucose, HbA_1c, and cholesterol, for example, all have a complete hierarchy of reference materials and reference measurement procedures with calibration of routine methods for these analytes traceable to a primary reference material. Provided manufacturers’ glucose assays, for example, are traceable to a certified glucose reference material that is listed on the Joint Committee for Traceability in Laboratory Medicine (JCTLM) database [5] and the measurand is the same across the entire traceability chain [6], in theory, the between-method bias is minimal and will not prevent use of the same decision limit for all assays. For “harmonized” measurement procedures where there is a lack of a reference material and a higher order measurement procedure, although bias can be removed by mathematical recalibration using the median or mean (or trimmed mean) value measured by contemporary methods (e.g. troponin I, thyroid-stimulating hormone) [7–10], the comparative method may develop a bias over time. To ensure use of the same harmonized decision limit over time and across methods, MU may require correction for bias as well as imprecision – an example is the harmonization of human growth hormone [11].

Analogously, some decision cutoff values are incorporated in clinical practice guidelines (e.g. LDL-cholesterol, prostate specific antigen) thus requiring clinical laboratories to minimize the bias to enable a satisfactory comparison of results, harmonized clinical interpretation and patient safety [12].

In the editorial by Jones [13], there is a sentence that deserves our attention: “Our goal in laboratory medicine is to produce information which can be used for clinical decisions that improve patient health”. This means that the determination and monitoring of MU should be used for two purposes: first, within the laboratory to provide accurate results; second, to provide the users with an objective tool for interpreting the results. The second issue importantly deserves further work. As described in the opinion paper also by Jones [14], a “one size fits all” calculation of MU is inappropriate; rather MU should be calculated depending on how the “true” value is obtained and applied depending on the type of comparison required for correct result interpretation. Example scenarios of the components of MU calculation are given ranging from the simplest comparison of a previous result from the same patient (e.g. serial troponin measurements within a short time period on the same analyzer within one calibration), to repeat measurements over multiple calibrations, to the interpretation of results against a population reference interval or a clinical decision limit. The challenge here will be the application of this new concept to all analytes.

Therefore, we believe that these papers and editorial should raise the interest and awareness of laboratory professionals about the right approach to MU determination and monitoring over time, particularly in light of its adoption in accreditation programs according to ISO 15189. Currently, the clause on uncertainty in the International Standard (5.5.1.4) underlines the need to determine this essential quality issue with significant flexibility (“as the laboratory shall define”), but there is no prescription of how to present uncertainty data. Therefore, further work should be done to harmonize different approaches to the determination and monitoring of MU.

In addition, the discussion regarding the right approach to laboratory accreditation according to ISO 15189 cannot be focused only on the analytical uncertainty, as current evidence highlights the vulnerability of extra-analytical phases. In particular, a growing body of evidence has been collected to demonstrate how the poor quality of the pre-analytical phase may affect the entire laboratory information [15]. The uniqueness of the pre-analytical phase is that it can influence the subsequent phases (analytical and post-analytical), thus making it a critical step: errors in appropriateness in test request, patient and sample misidentification, and poor quality of specimens strongly affect the ultimate value of laboratory data. Errors related to sample quality and integrity, before and after specimen transportation, may be much higher than analytical errors.

The comprehensive vision of laboratory services provided by ISO 15189 requires that clinical laboratories move to a more “patient-centered” view, addressing the issues of quality and error reduction in the extra-analytical steps. This, in turn, should be managed by adopting a valuable and harmonized list of quality indicators as required by the International Standard itself [16, 17].

It seems quite difficult to incorporate the pre- and post-analytical uncertainty into an MU calculation. The alternative way is to identify and continuously reduce the risk of errors in the extra-analytical phases through a risk management process that, according to ISO 15189, takes into consideration all steps of the cycle, namely the steps that are more vulnerable to error and risk of errors.

The development of an International Standard (ISO 15189) for the accreditation of medical laboratories was inspired by the need to recognize the specificities of this type of laboratory in comparison to testing and calibration laboratories, namely the pre- and post-analytical variables associated with the clinical scenario of test request, sample collection and transportation and result interpretation. It is time to move to an integrated use of the ISO 15189 to promote and assure quality in the total testing cycle as well as patient safety [18].