LOINC

In one line

LOINC (Logical Observation Identifiers Names and Codes) gives every laboratory test and clinical observation a universal code — so "serum glucose, fasting" is 1558-6 in every hospital on Earth, whatever the local lab calls it.

The problem it solves

Walk into three hospitals and ask for a potassium result. One calls it "K+", one "Potassium, serum", one "Serum K (lytes panel)". To a human, obviously the same test. To software, three unrelated strings — so trends don't line up, decision rules don't fire, and a national lab network can't aggregate anything. LOINC fixes this by giving the test itself a permanent, universal identity, independent of whatever each lab prints on its report.

The six axes

A LOINC term isn't a label — it's a fully specified name built from six parts, and that precision is the whole point. "Glucose" alone is ambiguous; the axes pin down exactly which question is being asked:

Axis	Question it answers	Example (2339-0)
Component	What is measured?	Glucose
Property	What kind of quantity?	Mass concentration
Time	Over what interval?	Point in time (not a 24-h collection)
System	In what specimen?	Blood (vs serum, urine, CSF)
Scale	What kind of result?	Quantitative (vs ordinal, nominal, narrative)
Method	By what technique?	(any — often left open)

Change any axis and you have a different test: fasting vs random glucose, serum vs whole blood, quantitative vs a dipstick ordinal. Conflating them is how a chart ends up plotting incomparable numbers on one line.

Beyond single tests: parts, panels and answer lists

LOINC is more than lab chemistry. It also codes:

• Panels and groupings — a CBC, a basic metabolic panel, or a standardised assessment instrument (PHQ-9, Braden scale) is a LOINC panel that bundles its member codes.
• Answer lists — for ordinal and nominal results, LOINC supplies the valid answers too (e.g. "Negative / Trace / 1+ / 2+"), so the result is coded, not free text.
• Document and survey codes — clinical document types and survey questions also get LOINC identifiers, which is why it reaches well past the laboratory.

It is maintained free of charge by the Regenstrief Institute, released twice a year, with tooling (RELMA) to map local codes.

Where it shows up in digital health

• FHIR — Observation.code binds to LOINC almost everywhere; US Core mandates it for lab results, and India's NRCeS profiles use it for diagnostics.
• Lab interfaces — connecting an analyser or a reference lab to an EHR is largely a local-code → LOINC mapping exercise (see the Integration Engine lab).
• Analytics & quality measures — population queries (every HbA1c in the last year) only work if every site coded that test the same way.

LOINC vs SNOMED CT — question vs answer

The single most useful thing to memorise: LOINC codes the question, SNOMED CT codes the answer. The test "Glucose [Mass/vol] in Blood" is LOINC 2339-0; the finding "Diabetes mellitus" is SNOMED 73211009. They are partners, not rivals — LOINC names what was measured, SNOMED names what it means. ICD then groups that finding for billing and statistics.

Common pitfalls

• Mapping to the wrong axis — picking a serum code for a whole-blood result, or a quantitative code for an ordinal one. The value then looks valid but isn't comparable.
• Over- or under-specifying method — binding to a method-specific code when the lab can change analysers (or vice versa) breaks trends silently.
• Treating the long common name as the identity — the code is the identity; names are for humans.

Key takeaways

• LOINC is the universal identity for what was measured — six axes, one code, globally.
• It covers panels, answer lists and documents, not just single lab tests.
• Pair it with SNOMED (answer) and UCUM (units) for a complete, computable observation.
• Most real LOINC work is careful local-code mapping — and the axes are where mistakes hide.