How the analyser-change report works

The report is something we offer: this page describes how it is computed, not a record of work for a customer. The method has been run end to end on public data, shown at the foot of this page.

What goes in

• The paired results from your method comparison or parallel run: the same samples measured on the old and the new analyser.
• Your decision limits for each analyte, in the units you report.
• Your repeat-test imprecision for both analysers if you have it, from QC data or a set of repeated samples. Without it we use a stated default and label the figures as assumed.
• For a network, one file per site. For an EQA or proficiency-testing scheme, the scheme’s own history.

What comes out

1. The comparison line with its interval. A Passing-Bablok line of new results on old, so you see whether the difference is constant, proportional or both.
2. Bias with limits of agreement. The Bland-Altman average difference and the range most differences fall in, for the familiar overall picture.
3. Bias at each of your decision limits. Because the difference that matters is the one at the concentration where a clinical decision changes.
4. Patients crossing each limit, in each direction. Set against what repeat testing alone would move, so the count reflects the change and not ordinary noise.
5. A re-test band for the transition. A narrow range around each limit where a new result is repeated, sized to a budget you choose.
6. A one-page note for requesting clinicians. What changed, at which limits, and what the laboratory is doing about it.
7. For networks, whether sites report interchangeably. The same analysis across sites, at your limits, showing which site moves which patients.

What this adds to your method comparison

Your validation software tells you the bias. It does not tell you which patients now sit on the other side of a decision limit, how many of those moves repeat testing would have caused anyway, or what to re-test while both analysers are in use. Those three answers are what the report adds, and it can run on the same paired results you already collected.

Bias at the decision limit, not on average

An average bias of a few percent can hide a larger shift at the one concentration where treatment changes. We estimate bias at each of your limits from the comparison line, with an interval from bootstrap resampling of your pairs.

Patients across the line, net of noise

Repeat testing on one unchanged analyser already moves some patients across a line. We count the crossings after the change and set them against that expectation, so you see what the change itself did. In some public series the change moved fewer results than repeat testing would have; the report says so rather than implying a problem.

A re-test band for the transition

A narrow range around each limit, sized to a re-test budget you choose, inside which a new result is repeated. We report how much of the change it catches on data it was not fitted to, by fitting the band on part of the pairs and scoring it on the rest. Studies with too few pairs for a full band get an indicative band or none, and the report says which.

Where it runs

• On your own computer. The program needs no network connection and opens none.
• It prints checksums (sha256) of its configuration and of your input file, so anyone can confirm what was run on what.
• Only aggregate files leave: the aggregate results, the report and a statement listing each released file with its checksum, which you sign.
• Counts below the minimum cell size you set, five by default, are shown as “<5”.
• The list of which of your samples crossed a line stays on your computer, keyed by your own sample reference.

Fixed before we look

The configuration, including decision limits, acceptance limits and the re-test budget, is agreed and hashed before the data are opened. Changes after that are amendments, dated and sealed before they are run. Each one is listed with its date, version and checksum on the changelog.

What happens when a test fails

It stays on the record. In September 2026 our first test of a change detector on raw instrument files failed: on public flow-cytometer files it flagged streams in which nothing had changed, because its threshold assumed the noise between samples was better behaved than real samples are. We kept that result, fixed the method under a sealed amendment, and re-ran it once on files it had never seen. The re-test passed. The detector is a separate offer for EQA schemes; it is not part of the analyser-change report, and it has not yet been run on any scheme’s data. Both results are on the changelog.

What it is not

• Not an ISO 15189 verification. It supports your verification work and does not replace it.
• Not a device approval or regulatory clearance of any kind.
• Not advice about an individual patient.
• Repeat-test imprecision is assumed unless you supply it, and every figure that depends on it says which.

Tested on public data

Public data

We ran the program end to end on 10,579 paired results in 34 method comparisons at 14 stand-in sites, from 12 open-access studies; each data set plays the part of one laboratory. Of the 34 method comparisons, 21 get a full re-test band and 5 an indicative one, because they have fewer pairs than the minimum we set; the other 8 are shown as a comparison only or are too small for a band. All five checks we fixed before the run passed. The checks are ours, not an outside review.