Transferring and verifying a reference interval Establishing a reference interval needs 120 individuals per partition. Transferring one you can verify with about 20, when the transfer is valid, and how to check it holds.

Establishing a reference interval from scratch is expensive. The non-parametric method wants at least 120 reference individuals per partition. A partitioned analyte multiplies that. Much of the time you do not have to.

CLSI EP28-A3C lets you transfer a well-characterised interval and confirm it holds in your population with a far smaller study. The interval can come from the manufacturer, from the literature, or from another method. Knowing when that shortcut is legitimate, and how to check it, is the difference between weeks of recruitment and an afternoon.

When a transfer is valid

A reference interval is a property of a population measured by a method. A transfer is only sound when both are comparable. The reference population should be similar in the things that move the analyte: age and sex distribution, and any relevant demographic or geographic factors. The measurement procedure and its pre-analytical handling should be comparable, or the two methods related by a known comparison so the interval can be converted across.

Where the populations or methods differ materially, the borrowed interval may simply be wrong for you. No amount of verification rescues a transfer that should not have been attempted.

Verifying with a small study

Verification asks a modest question: does the transferred interval fit my population? Answering it takes only a modest sample. EP28-A3C describes verifying with around 20 reference individuals per partition. Measure them, then count how many fall outside the transferred interval. If no more than a small proportion do, the interval is acceptable. If too many fall outside, the interval does not fit your population and you cannot simply adopt it.

The threshold is not arbitrary. A 95% reference interval should, by definition, exclude about 5% of a matching reference population. In 20 individuals you expect roughly one outside. A binomial test formalises this. It tells you whether the number falling outside is consistent with the interval being correct, or too many to be chance. The worked example below verifies a transferred mercury interval this way.

A small verification sample plotted against a transferred reference interval, showing how many individuals fall outside the limits.
Verifying a transferred reference interval against a small sample (CLSI EP28-A3C, Example 3 — mercury). The count falling outside the limits is tested against what the interval predicts.

If verification fails

Too many individuals outside the transferred interval is a useful result, not a wasted study. First check the obvious: a genuine population difference, a pre-analytical or methodological mismatch, or an outlier or two in your small sample distorting the count.

If the interval genuinely does not fit, your population really is different, then you have your answer. The right response is to establish an interval of your own rather than force a borrowed one that does not describe your patients. A larger verification sample can also help distinguish a real misfit from the noise inherent in judging 20 individuals.

A decision flowchart. Start: does a valid interval exist and are the population and method comparable? No, establish from scratch. Yes, transfer it, measure about 20 individuals and count how many fall outside; then decide whether more fall outside than the binomial test allows. No, accept the interval. Yes, investigate and establish your own.
The transfer-and-verify decision in full. If a comparable interval exists, transfer it and check it on about 20 individuals with a binomial test. Accept it when the count outside is consistent with chance. Establish your own only when no valid interval exists or the check fails.

Downloads

Download the CLSI EP28-A3C transfer example workbook (.xlsx) — a mercury reference interval transferred and verified against a small sample with a binomial test, ready to open in the Analyse-it trial.

Common mistakes

Transferring across mismatched populations or methods. Verification cannot fix a transfer that was never valid. Confirm the population and method are comparable first.

Reading a small verification sample too strictly. With 20 individuals, one or two outside is expected. Use the binomial test rather than reacting to a single exceedance.

Adopting a failed interval anyway. If too many fall outside, the interval does not fit your population. Establish your own rather than report one you have shown to be wrong.

Establishing when you could transfer. If a valid interval exists and your population matches, a 20-sample verification saves recruiting 120. Transfer first, establish only if it fails.

Verify a reference interval with Analyse-it

Analyse-it transfers and verifies intervals per EP28-A3C, on your own verification sample, inside Excel:

  • The binomial proportion test on the verification sample, which is what EP28 asks for
  • The transferred interval and the verification result reported together
  • So a small verification study is not mistaken for an establishment

Every feature from all five editions for 15 days. Reference intervals are in the Medical, Method Validation and Ultimate editions, from US$ 340 a year. Validated against NIST and CLSI reference datasets. See choosing a reference interval method, or the reference interval reference guide for the technical detail.