Uncertainty across instruments and sites

MEASUREMENT UNCERTAINTY

Uncertainty across instruments and sites

Match the estimate to the result’s origin and the intended comparison.

Revised 30 September 2026. The original publication date and address are retained.

Define the reporting situation

A result produced on a known analyser can have a result-specific uncertainty estimate for that procedure and concentration. A service-wide estimate can be useful when it represents the actual reporting process, but pooling every analyser is not automatically the correct answer.

SituationWhat the model needs
Known procedure and analyserEvidence representative of that result
Interchangeable routine analysersWithin-analyser variation and relevant between-analyser effects
Comparison across sitesProcedure comparability, differences, shared effects and intended use
Serial patient resultsAnalytical comparability plus appropriate biological and clinical context

Do not erase mean differences by centring

Pooling within-group variances describes within-group dispersion. It does not include persistent differences in group means. If analyser or site differences matter to a result drawn from the service, investigate and account for them with a suitable design and model.

Illustrative case

Two analysers each have SD 1 unit, but their means differ by 4 units for comparable specimens. A pooled within-analyser SD can remain 1. That number alone does not describe the variability of an unidentified result drawn from both analysers. Conversely, adding every cross-site difference to every known-analyser result can answer the wrong question.

Use a study that separates the effects

Paired patient-specimen comparisons, appropriately designed precision studies and suitable EQA evidence can answer different parts of the problem. Shared reagent or calibration sources may create correlation. A patient median can change with case mix, so it is not by itself proof of an analyser shift.

Avoid league tables based on one MU number

A larger uncertainty estimate can reflect a broader assessment scope or more complete evidence. Compare the same measurand, concentration, coverage basis and included contributions before interpreting differences between services. Do not infer better clinical performance from the smaller number alone.

Take the next step

Keep the next question tied to the procedure and result being compared.

Free Word and PDF pack with fictional examples. No payment or account required. Newsletter and course sign-ups are separate.


Discover more from Pathology Uncertainty

Subscribe to get the latest posts sent to your email.

Leave a Reply