How to verify a new POCT analyser

Verification, validation and what ISO 15189 asks

The manufacturer validated the analyser: they established that the method works and published claims for precision, bias, linearity and detection limits. Verification is your job: showing that the method meets those claims in your setting, with your operators, your samples and your consumables, before it is used on patients. ISO 15189:2022 keeps this distinction in clause 7.3.2 (verification of examination methods) and 7.3.3 (validation), and the point-of-care annex applies the same expectation to POCT devices under the laboratory's responsibility.

A verification is not a research project. For a CE-marked quantitative POCT analyser used as intended, the core of it is two studies, precision and comparison, planned in advance, run over a few days and written up as one file.

Plan before you touch the analyser

• Which analytes and levels. Every analyte you will report, at least two concentrations spanning the clinically important range, and the decision limits your clinicians use.
• What you are comparing against. The main laboratory method, or an existing analyser of the same model already in use. Agree how samples will be split and how quickly they will be run on both.
• Acceptance criteria, written down first. Take them from your own quality specification, or from a published total allowable error (TEa): CLIA proficiency-testing limits, RCPA analytical performance specifications or biological-variation targets from the EFLM database. Deciding after the data are in is how verifications quietly become opinions.
• Who does what. The operators who will run the device day to day should run the studies. Their precision is the precision you will live with.
• Materials. Fresh control lots, calibrators as per the instructions for use, and patient samples with consent handled under your normal policy.

Step 1: precision

Precision tells you how much the analyser scatters around its own mean. CLSI EP15-A3 (user verification of precision and estimation of bias) is the practical protocol for a verification: five days, five replicates per day, at each level, which gives 25 results per level. From those you estimate repeatability (within-run) and within-laboratory precision (including day-to-day variation), and compare each against the manufacturer's claim using the verification limit in the standard. If you need a fuller characterisation, for example for a laboratory-grade analyser, CLSI EP05-A3 uses 20 days with two runs a day and two replicates per run.

Practical notes: use the same control lot throughout; do not exclude a day because it looks odd without a documented reason; and if your estimate is worse than the claim but within your own allowable error, say so explicitly in the write-up rather than leaving it to the assessor to spot.

Step 2: method comparison and bias

Comparison tells you whether the new analyser agrees with the method you already trust. CLSI EP09c is the reference protocol; for a verification, at least 40 patient samples spread across the measuring range, run on both methods within the sample stability window, ideally in duplicate and over at least five days, is the accepted minimum. Plot the difference (or percentage difference) against the mean, fit a regression that allows error in both methods (Deming or Passing-Bablok rather than ordinary least squares), and read the bias at each medical decision level from the fitted line with its confidence interval.

Two mistakes are common. The first is choosing samples that cluster in the normal range, which leaves the decision limits unverified. The second is reporting the correlation coefficient as evidence of agreement; correlation measures whether two methods rank samples the same way, not whether they give the same number.

If a comparison method is not available, EP15-A3 allows bias to be estimated from a certified reference material or an EQA sample with an assigned value.

Step 3: linearity and the measuring range

Linearity is often assumed from the manufacturer's claim for a verification and only checked when the comparison shows curvature at the ends of the range, or when you intend to report values near the limits. When you do check it, CLSI EP06 describes the approach: a series of at least five levels prepared by dilution or mixing across the claimed range, run in duplicate or more, and evaluated by fitting first-, second- and third-order polynomials to find whether non-linear terms are significant and whether the deviation from linear matters clinically.

Step 4: detection limits, when they matter

For most POCT chemistry and HbA1c this is not part of verification. For assays reported near zero, such as high-sensitivity troponin, CRP at the low end, or qualitative molecular targets, it can be. CLSI EP17 defines the limit of blank, limit of detection and limit of quantitation, and describes how to verify a manufacturer's claimed LoD with a modest number of low-level samples. Confirm the claim rather than re-establishing it.

Step 5: reference intervals

You will usually adopt the manufacturer's or the laboratory's reference interval rather than establish your own. CLSI EP28 describes a transference check: run 20 samples from apparently healthy people and confirm that no more than two fall outside the adopted interval; if three or more do, run another 20 and reconsider. Establishing an interval from scratch needs 120 reference individuals per partition and is rarely justified for a POCT device that agrees with the laboratory.

Acceptance criteria and sigma

Precision and bias are judged against total allowable error. A useful single figure is the sigma metric: sigma = (TEa minus absolute bias) divided by CV, all in the same units, typically percent. A method at six sigma or above rarely produces a clinically significant error and can be controlled with simple QC rules; a method under three sigma will fail QC often and may not be fit for the intended use. Sigma also tells you which Westgard rules and how many controls the analyte needs, which is the subject of the companion article.

The verification file and sign-off

An assessor wants to see one document, or one folder, that contains: the plan with the criteria; the raw data for every study; the calculations and charts; the comparison of each result against its criterion; any deviations, exclusions and their reasons; the conclusion (fit for purpose, or fit with limitations, or not fit); and the signature of the person authorised to release the method, with the date. Under ISO 15189:2022 that authorisation needs to be from someone the laboratory has designated as competent to make it. Store the file with the equipment record so that it is found when the device is.

• Plan and criteria dated before the data
• Precision: 5 x 5 at two or more levels, repeatability and within-lab against the claim
• Comparison: 40 or more patient samples across the range, difference plot, Deming or Passing-Bablok, bias at decision levels with confidence intervals
• Linearity and detection limits only where relevant, and stated as such
• Reference interval transference with 20 samples
• Sigma per analyte with the TEa source named
• Deviations and exclusions explained
• Conclusion and authorised signature, filed with the equipment record

Where software helps

The hard part of a verification is rarely the statistics; it is assembling the data and the write-up. If the new analyser and the comparison method are both connected to the same platform, the paired patient results can be drawn automatically, the precision grid can fill from the control runs, and the pack can be signed and locked as one document. The Catenix Statistics Workbench runs the scenario described here (EP15 style precision, EP09 style comparison with Deming regression and jackknife confidence intervals, EP06 linearity, EP17 limits, EP28 transference and sigma from a TEa library) and produces a signed verification pack. The judgement about fitness for purpose remains yours.

Questions people ask

This article is general guidance for laboratory and point-of-care professionals. It is not a substitute for the standards themselves, your accreditation body's requirements or the analyser manufacturer's instructions for use. Catenix does not interpret clinical results and provides no clinical decision support.