Why reference ranges differ between laboratories

A reference interval is a sample, not a law

The standard way to establish a reference interval is to measure the analyte in a group of apparently healthy individuals who meet stated inclusion criteria, then report the central 95% of the resulting distribution - the 2.5th to 97.5th percentiles. The laboratory standards that govern this work call for a minimum of around 120 qualifying individuals per partition, and a partition is needed for each subgroup that behaves differently, such as each sex or each paediatric age band.

Two consequences follow immediately. First, the interval inherits the characteristics of whoever was recruited. Second, because a 95% band is used, one in twenty healthy reference individuals sits outside the interval by construction - the tails were trimmed to make a usable band, not because those people were unwell.

Recruiting 120 healthy volunteers per subgroup is expensive, so many laboratories do not derive intervals from scratch. They verify a manufacturer's or a published interval against a smaller local sample and adopt it if it holds. That is legitimate practice, and it is another reason two laboratories can land on different numbers: they may have started from different sources.

The analyzer and the assay

Different instruments do not measure the same molecule the same way. Enzyme activity assays depend on reaction temperature, substrate and buffer; immunoassays depend on which antibodies the manufacturer chose and what exactly those antibodies bind. Change the platform and the numbers shift, sometimes substantially, even though the patient has not.

This is why a reference interval belongs to a method as much as to a population, and why results from different platforms are not always directly comparable. When a hospital replaces an analyzer, published intervals for affected analytes are re-verified and may be reissued, and a patient's series of results can step up or down at the changeover for purely technical reasons.

Standardisation and harmonisation programmes exist to shrink these gaps - traceability to reference materials and reference measurement procedures has genuinely narrowed the spread for several common analytes. But the work is analyte by analyte and far from complete, and where an assay is not standardised, the manufacturer's own interval remains the honest one to use.

• Enzymes (ALT, AST, ALP, CK) are reported as activity and are sensitive to assay conditions.
• Immunoassays (hormones, tumour markers, troponins) vary with antibody design and calibration.
• Calculated values (eGFR, anion gap, transferrin saturation, LDL by calculation) inherit the variability of everything they are calculated from - and change outright when the equation is revised.
• Point-of-care devices frequently carry their own intervals, separate from the main laboratory's.

Who was in the reference group

A reference population is a real group of people in a real place. Its age distribution, sex balance, ancestry, diet and even altitude are baked into the resulting band. Haemoglobin intervals derived at high altitude are genuinely higher than those derived at sea level, because the physiology genuinely differs - neither band is an error.

Sex is the partition most often visible on a report. Haemoglobin, haematocrit, creatinine and ferritin all carry published sex-specific intervals, and where this site's dataset holds a subgroup band it is shown separately rather than blended into an average that describes neither group well.

Age is the partition most often missed. Alkaline phosphatase is expected to be high in a growing child; several analytes shift across the neonatal period, through puberty and into older age. Reading a paediatric result against an adult band is one of the most common and most avoidable misreadings in the whole subject.

Some numbers are not reference intervals at all

A significant group of tests is reported against decision limits rather than population percentiles. These are thresholds set by guideline committees to separate risk categories or to define a diagnosis, and they are chosen for clinical outcome rather than derived from a healthy distribution.

Lipids are the clearest example: the familiar cholesterol targets are risk thresholds, and it would make little sense to call the upper 2.5% of a population's cholesterol "abnormal" and the rest normal when the whole distribution has shifted. Glycated haemoglobin cut-points for diagnosing diabetes work the same way. Cardiac troponin is different again - it is interpreted against an assay-specific upper reference limit taken from the 99th percentile of a healthy population, so the threshold moves when the assay does.

Therapeutic drug levels invert the logic entirely. A digoxin or vancomycin range is a target window for treatment, not a description of health, and it depends on the indication and on when the sample was drawn relative to the dose. A trough and a peak are not interchangeable, and neither is meaningful without the timing.

The specimen matters too

Before any of the above applies, the sample has to arrive intact. Serum and plasma are not identical matrices, and potassium in particular reads higher in serum because platelets release it as the sample clots. Haemolysis raises potassium and several enzymes by spilling cell contents into the fluid measured. A tube filled below its draw volume changes the ratio of anticoagulant to blood and skews coagulation results.

Timing and preparation shape results as much as physiology does. Fasting state, posture in the minutes before a draw, tourniquet time, recent exercise and time of day all move specific analytes measurably. None of that is laboratory error, and none of it is disease - it is why collection instructions exist and why an unexpected result is often repeated before anyone acts on it.

For a student, the useful mental model is a chain: the physiology, the collection, the transport, the assay and the interval it is compared against. A surprising number sends you back along that chain rather than straight to a diagnosis.

So which range should you use?

For a real result: the interval printed on that report, from that laboratory, on that analyzer, for that patient's sex and age partition. Nothing else has standing.

For studying: a consistent set of teaching values is genuinely useful, because exams test the shape of the data - which direction a value moves in a given condition, which analytes travel together, and which numbers demand immediate attention. That is what the ranges on this site are for, and why every page repeats that they vary by laboratory.

For reading the literature: check which units and which assay a paper used before comparing its numbers to anything you know. A value that looks wildly wrong is more often a unit system or a different platform than a real disagreement.

Frequently asked questions