Biological Reference Intervals: How to Establish and Verify

A biological reference interval is the range of values expected in a defined healthy reference population, usually the central 95% between the 2.5th and 97.5th percentiles. Laboratories either establish an interval from their own reference individuals or verify that an interval from a manufacturer, the literature or another laboratory fits their method and population.

Reference intervals shape how every result is read. An interval that does not suit your population or analyser flags healthy people as abnormal, or hides real disease, and the error repeats on every report.

Key definitions

• Reference individual: a person selected using defined health criteria.
• Reference population: all individuals who would meet those criteria.
• Reference limits: the lower and upper values, usually the 2.5th and 97.5th percentiles.
• Partition: a subgroup, such as adult women, that needs its own interval.
• Clinical decision limit: a threshold tied to a clinical decision, not to the healthy distribution.

By definition, about 5% of healthy people fall outside a 95% reference interval. A result just outside the interval is not automatically abnormal, which is why the older term "normal range" is discouraged.

The CLSI EP28 approach

The most widely used guidance is CLSI EP28, Defining, Establishing, and Verifying Reference Intervals in the Clinical Laboratory, developed with the IFCC. Check that you are using the current edition. Establishing an interval broadly involves:

1. Define the reference population and exclusion criteria, such as illness, medication, pregnancy or recent alcohol.
2. Select reference individuals using a questionnaire and, where needed, screening tests.
3. Match pre-analytical conditions (fasting, posture, sample type, processing) to routine practice.
4. Analyse samples on the routine method.
5. Inspect the data and remove outliers using a defined method.
6. Decide on partitions.
7. Calculate reference limits and, ideally, their confidence intervals.
8. Document, review and approve.

For the non-parametric method, EP28 is commonly cited as recommending at least 120 reference individuals per partition. This allows the 2.5th and 97.5th percentiles to be estimated with a 90% confidence interval around each limit.

Non-parametric percentiles

The non-parametric method does not assume a normal (Gaussian) distribution, which suits the skewed data seen for many analytes. Rank the values from lowest to highest. The rank of the lower limit is 0.025 × (n + 1) and of the upper limit 0.975 × (n + 1), so with 120 values the limits fall at about the 3rd and 118th values.

Parametric methods (mean ± 1.96 SD, after transformation if needed) can be used only when the data are, or can be made, Gaussian, and that must be checked rather than assumed.

Partitioning by age, sex and other factors

Partitions are needed when subgroups differ enough to matter. Haemoglobin differs between adult men and women, alkaline phosphatase is much higher in growing children, and many analytes change in pregnancy.

Partition only when the difference is both statistically and clinically meaningful. Statistical approaches such as the Harris and Boyd method help, but clinical judgement still matters. Every partition multiplies the number of reference individuals needed, which is why paediatric intervals are so hard to establish locally.

Establish, verify or derive indirectly

Indirect methods apply specialised statistics to routine patient results to separate out the healthy distribution. They need careful data selection and expert review.

Verifying a transferred interval with 20 individuals

Under the commonly cited EP28 approach, the laboratory tests samples from 20 reference individuals from its own population, selected using criteria comparable to the original study. If no more than 2 of the 20 results (10%) fall outside the proposed limits, the interval can be accepted.

If 3 or more fall outside, test a further 20. If 3 or more of the second group also fall outside, the interval is unlikely to suit your method or population, and you should consider establishing your own. Check current CLSI guidance for the detail.

Before verifying, confirm the source used a comparable method, calibration traceability, sample type and pre-analytical conditions. Verification tests transferability; it cannot rescue an interval from an incompatible method.

Reference intervals vs clinical decision limits

A clinical decision limit is based on outcome studies or guidelines, not the healthy distribution. An HbA1c of 48 mmol/mol (6.5%) for diagnosing diabetes, as recommended by WHO, is a decision limit; lipid treatment targets are another example.

For analytes like these, printing a population reference interval can mislead clinicians. Report what fits the test's purpose and label it clearly. Critical limits are different again, as our guide to critical results reporting explains.

What ISO 15189:2022 expects

ISO 15189:2022 expects laboratories to define biological reference intervals and clinical decision limits suited to their patients and examinations, record the basis for them, communicate them to users and review them when pre-examination or examination procedures change. Assessors will ask for the source of each interval, verification data and records of review.

Practical triggers for review include a new analyser or reagent, a change in calibration traceability, a new sample type, a change in the population served and clinician feedback that too many results are flagged.

Worked example: verifying a haemoglobin interval in Abu Dhabi

A hospital laboratory in Abu Dhabi replaces its haematology analysers. The manufacturer offers adult haemoglobin intervals of 130 to 170 g/L for men and 120 to 155 g/L for women, derived from a population abroad. The laboratory serves Emirati patients and a large expatriate population, and haemoglobinopathy traits and iron deficiency are common in the region.

With the haematologist, the team sets exclusion criteria: known anaemia, recent illness or blood donation, pregnancy, known haemoglobinopathy trait and ferritin below the laboratory's cut-off. Healthy adult volunteers are recruited to reflect the patient mix, and samples are collected and processed as in routine work.

For men, 1 of 20 results falls outside the interval, so it is accepted. For women, 3 of 20 fall below 120 g/L, so a second group of 20 women is tested. Only 1 result falls outside, and the interval is accepted.

The haematologist remains cautious about the female lower limit. The laboratory documents the verification with raw data and approval, adds the interval to its annual review list and plans an indirect analysis of routine outpatient results.

Common mistakes

• Copying intervals from a textbook or insert without verification
• Verifying with unscreened volunteers or patient samples
• Ignoring pre-analytical differences such as fasting or sample type
• Assuming data are Gaussian and using mean ± 2 SD
• Reporting a population interval where a decision limit applies
• Not reviewing intervals after an analyser or method change

Key takeaway: Establish intervals with at least 120 well-selected individuals per partition or verify transferred ones with about 20, document the basis, and review them whenever your method or population changes.

Frequently Asked Questions

What is a biological reference interval?

It is the range of results expected in a defined healthy reference population, usually the central 95% between the 2.5th and 97.5th percentiles. It helps clinicians interpret a result, but because 5% of healthy people fall outside it by definition, a result outside the interval is not necessarily a sign of disease.

How many samples are needed to establish a reference interval?

CLSI EP28 is commonly cited as recommending at least 120 reference individuals per partition for the non-parametric method, which allows confidence intervals around each limit. Each partition, such as adult women or a paediatric age band, needs its own group. Smaller samples can be used with robust methods, but uncertainty is greater.

How do you verify a reference interval?

Under the commonly cited CLSI EP28 approach, test 20 healthy individuals from your own population. If no more than 2 fall outside the limits, accept the interval. If 3 or more do, test another 20; if that group also has 3 or more outside, the interval probably does not fit and needs review.

What is the difference between a reference interval and a clinical decision limit?

A reference interval describes results in healthy people. A clinical decision limit is a threshold linked to a specific action, such as diagnosis or treatment, based on outcome studies or guidelines. HbA1c of 48 mmol/mol for diagnosing diabetes is a decision limit, not a reference limit.

Next steps

To manage reference intervals within a complete quality system, our ISO 15189 Advanced: Medical Laboratory Quality & Competence course covers examination processes and reporting, and Laboratory Quality Management System builds the foundations. For related reading, see method validation vs verification, the ISO 15189:2022 checklist for medical labs and pre-analytical errors. For help choosing, speak to a counsellor.