IFCC Working Group Recommendations for Assessing Commutability Part 1: General Experimental Design

Abstract

Commutability is a property of a reference material (RM) that relates to the closeness of agreement between results for an RM and results for clinical samples (CSs) when measured by ≥2 measurement procedures (MPs). Commutability of RMs used in a calibration traceability scheme is an essential property for them to be fit for purpose. Similarly, commutability of trueness controls or external quality assessment samples is essential when those materials are used to assess trueness of results for CSs. This report is part 1 of a 3-part series describing how to assess commutability of RMs. Part 1 defines commutability and addresses critical components of the experimental design for commutability assessment, including selection of individual CSs, use of pooled CSs, qualification of MPs for inclusion, establishing criteria for the determination that an RM is commutable, generalization of commutability conclusions to future measurements made with the MPs included in the assessment, and information regarding commutability to be included in the certificate for an RM. Parts 2 and 3 in the series present 2 different statistical approaches to commutability assessment that use fixed criteria related to the medical decisions that will be made using the laboratory test results.

Fig. 1. The bias for logarithmic transformation of concentration [ln(concentration)] between 2 measuring systems is shown for a panel of CSs (black diamonds) and for 5 candidate RMs (red squares) vs the mean concentration measured with each MP

The error bars for each RM indicate the uncertainty of the difference in bias between each RM and the average bias for the CSs. The bias values for in-transformed results can be multiplied by 100 to give an approximate percent bias. The black solid line is the mean bias (18%) for all the CSs. The blue dashed lines indicate the maximum allowable commutability-related bias for the RM to be considered commutable with the CSs. (A) shows results for 2 MPs (x and y) for which the random errors of the MPs and the sample-specific influences are reasonably small. (B) shows results for 2 MPs (x and z) with more scatter for the CS results, suggesting that MP z has poorer precision and/or larger sample-specific influences in relation to MP x.