01Establish that the comparison is meaningful

Start with the physical samples. Were laboratories sent portions of one homogeneous preparation, separate vials from one lot or material from different lots? The design determines whether sample variation is mixed with analytical variation.

Next align the reported quantity. A percentage based on chromatographic peak area should not be compared numerically with a mass fraction as though they were two estimates of one measurand. The same warning applies to results reported on different water or counterion bases.

If these details are missing, the responsible conclusion may be that the reports are not yet directly comparable. Converting units can solve a unit mismatch; it cannot solve a mismatch in what was measured.

02Separate instrument comparison from the whole workflow

Abbatiello and colleagues’ 2015 study involved 11 laboratories and 14 LC–MS systems measuring peptides related to proteins in plasma. Its phased design distinguished peptide-level measurements from additional preparation of protein-containing samples.Abbatiello and colleagues — Large-scale interlaboratory quantitative peptide study (opens in a new tab)

The study incorporated system-suitability monitoring and evaluated digestion recovery, including the use of isotope-labelled protein standards. These measures helped investigate where performance was gained or lost across the workflow.Abbatiello and colleagues — Large-scale interlaboratory quantitative peptide study (opens in a new tab)

The general lesson is about study design. Distributing ready-to-inject material tests a narrower chain than asking every laboratory to prepare the material independently. Neither design is automatically superior; they answer different questions.

A comparison should say which chain was tested. Agreement for a prepared solution does not establish agreement in extraction, digestion or other stages that the participating laboratories did not perform.

03Inspect differences, not just whether values move together

Consider an original example with three matched samples. Laboratory A reports 9.5, 19.0 and 28.5 units. Laboratory B reports 10.0, 20.0 and 30.0. The values rise together perfectly, but A is consistently 5% below B.

A correlation calculation would not remove that proportional difference. Conversely, a small numerical difference close to a decision boundary can matter even when the overall relationship between laboratories is strong.

Examine differences across the concentration range and against the uncertainty and purpose of the measurement. A difference that grows with concentration suggests a different question from a nearly constant offset.

Repeats within each laboratory help distinguish a persistent between-laboratory difference from noisy individual results. Repeated injections alone still leave preparation variation incompletely assessed.

The example does not identify which laboratory is closer to a valid reference. Agreement with another laboratory and agreement with a justified reference value are separate comparisons.

04Use discrepancies to narrow the next question

A focused investigation might compare standard assignments, integration rules or sample-basis calculations before requesting entirely new testing. The order should follow the observed discrepancy, rather than assume the instrument is at fault.

If laboratories share a reference preparation, they may also share an error in its assigned value. Strong agreement then provides evidence of consistency on that shared scale, but less independent reassurance about the scale itself.

Keep original results, subsequent explanations and any corrected calculations distinguishable. A revised number is easier to assess when the reason for the change is documented.