Are your analysers interchangeable? Comparability of results across instruments and sites

What ISO 15189:2022 asks

ISO 15189:2022 sets the requirements for quality and competence in medical laboratories. Its clause on the comparability of examination results covers examinations done on different methods or equipment, or at different sites. As a published summary puts it, the laboratory needs a process to assess that comparability, and where differences are found, their impact on biological reference intervals and clinical decision limits should be evaluated and acted on, with users told of any clinically significant difference [Marrington et al., 2026].

Two things in that requirement are easy to miss. It is not limited to two analysers of the same model: a point-of-care device and the main laboratory analyser measuring the same analyte are in scope. And the test of success is clinical, not statistical: whether a difference changes what a result means at a reference interval or decision limit.

The EP31 approach

CLSI EP31 gives recommendations for verifying the comparability of quantitative results for individual patients between two or more instruments within one health care system, whether the instruments differ in method, in model, or are separate instruments of one type [CLSI EP31, 2025]. Its protocol compares patient samples, chosen at the concentrations of interest, across all the instruments, using a range test for up to ten instruments and modified strategies above that.

In routine use it works, with a practical caveat. A laboratory that applied it to two dozen chemistry tests across several analysers found the guideline useful for periodic verification, but noted that it may need acceptance criteria adjusted to the analytical performance of the technology available, and a number of measurements matched to the laboratory’s resources [Guiñón et al., 2024].

Patient samples, not only control material

Control and EQA material is designed to be stable, not to behave exactly like a patient sample. Two methods can agree on control material and still disagree on patient blood, and the difference can depend on the sample type as much as the instrument. That is why EP31 is framed around patient results, and why a periodic comparison with real samples, near the limits the laboratory reports against, is worth more than a control chart for this particular question.

Setting matters as well as sample. The same device used by a different operator, on capillary rather than venous blood, away from the laboratory, is in practice a different measuring system. The worked example below shows how much that can change.

Comparability at the decision limits

A comparability check usually ends in one number per analyte: a mean difference, or a maximum difference against an allowable limit. For the clinical question the standard asks, three further numbers are more useful, each one per decision limit.

• Bias at the limit. Read from the comparison line at the limit itself, not averaged over the range. A proportional difference can be small at one limit and large at another.
• Results that change category. Of the paired results, how many fall on one side of the limit on one instrument and on the other side on the second.
• The repeat-testing baseline. How many of those changes one instrument measuring the same samples twice would produce anyway, from its imprecision. Only the excess is due to the instruments differing.

The third number needs the imprecision of each instrument. Where replicate measurements exist, use them; where a default is assumed, say so. The guide to bias at clinical decision limits works through the method in detail.

One analyser, three meters, two settings

Public data

A public study compared three point-of-care haemoglobin meters with one laboratory haematology analyser as the reference, first in a central laboratory and then in a community clinic, where the meters were used on finger-prick blood and the reference on a venous sample drawn at the same visit [Jaggernath et al., 2016]. Its data are deposited under CC BY 4.0 [Gelderblom, 2016]. The study also repeat-tested a subset of samples, so imprecision can be measured rather than assumed, which makes these the only comparisons in our public set where it is. We ran all six comparisons through our program with limits at 8, 12 and 13 g/dL; the study itself used 12 and 13 g/dL as its anaemia thresholds for women and men.