Reference materials, calibrators, controls, and proficiency-testing samples all stand in for patient samples. You measure them and act on the result as if it told you how the method handles real specimens. That substitution only works if the material behaves like a patient sample across the measurement procedures involved. That property is called commutability.
A non-commutable material can actively mislead. The failure is silent, which makes it one of the more easily missed problems in the calibration chain. CLSI EP14-A3 is the protocol for detecting it.
A material is commutable between two measurement procedures when it behaves like a real patient sample on both. Put precisely: the numerical relationship between the two methods’ results for that material is the same as the relationship for genuine patient samples.
Processing is what breaks it. Freezing, lyophilising, spiking with a purified analyte or adding a matrix can all alter how a material interacts with an assay, in ways that native patient samples never do. The material may then read relatively high on one method and low on another, compared with how patient samples behave. A material that sits off the patient-sample relationship in this way is non-commutable.
A non-commutable material gives a method-specific bias that has nothing to do with how the method performs on real patients. Use such a material as a calibrator and you transfer that artefact into every patient result. Use it in an external-quality scheme and a method may fail, or pass, for reasons that would never show up on patient samples. Good methods get penalised. Poor ones can look good. Everything looks fine on the surface: the material gives a number, the number is off, and nothing reveals that the material, not the method, is the problem.
The assessment is a comparison. Measure a panel of individual patient samples on the two (or more) measurement procedures and establish the relationship between them: the line that genuine samples follow, with a prediction interval around it. Then measure the candidate material on the same procedures and see where it falls.
A material landing inside the prediction interval of the patient-sample relationship behaves like a patient sample, and is commutable. A material falling outside it does not, and should not be used to calibrate those methods or to judge them against one another. In effect, commutability is a method-comparison question asked of one special sample: does it lie on the line the real samples define?
Commutability is central wherever a value is transferred between methods through a material. That includes metrological traceability chains that carry a reference value down to routine assays, the validation of calibrators and controls, and the design of proficiency-testing schemes that compare methods. In each case, a non-commutable material breaks the assumption that the material stands in for patient samples. Establishing commutability is a prerequisite, not an afterthought, for any material meant to bridge methods.
Assuming a processed material behaves like a patient sample. Freezing, lyophilising, or spiking can change how a material interacts with an assay. Commutability has to be demonstrated, not assumed.
Blaming the method for a material’s bias. A method-specific offset on a non-commutable material is an artefact of the material, not a fault in the method. Check commutability before concluding a method is biased.
Calibrating with a non-commutable material. Its artefact transfers into every patient result. Establish commutability before a material is used to calibrate.
Assuming it only matters for reference materials. The same trap catches the material used for recurring comparability checks between instruments and for AMR verification. A non-commutable material measures itself rather than the method.
Judging methods in a scheme built on non-commutable samples. Pass or fail becomes about the sample, not the method. Commutable materials are a prerequisite for a fair comparison.
Commutability is a method-comparison question — whether a material falls inside the prediction interval of the patient-sample relationship. Analyse-it fits that relationship on your own data, inside Excel:
Every feature from all five editions for 15 days. Method comparison is in the Method Validation and Ultimate editions, from US$ 475 a year. Validated against NIST and CLSI reference datasets.