Post-coordination & expressions

The answer in one paragraph

Most of the time you record a pre-coordinated concept — a single existing code like Fracture of femur. When the precise meaning doesn't exist as one code (left side? caused by what? severity?), SNOMED CT lets you post-coordinate: compose an expression from a base concept plus attribute–value refinements, written in Compositional Grammar (SCG). The expression means something the terminology can still classify and query — you're not inventing a local code, you're speaking the same language in longer sentences.

The grammar in one example

71620000 |Fracture of femur| :
  { 363698007 |Finding site| = ( 71341001 |Bone structure of femur| :
                                 272741003 |Laterality| = 7771000 |Left| ) }

Read it as: the concept Fracture of femur, refined so that its finding site is the femur — and that femur is the left one. The colon introduces refinements; each refinement is attribute = value; both sides are themselves concepts. Multiple refinements separate with commas; grouped attributes wrap in { } when several belong together.

Note where the laterality sits. It is nested inside the body structure, in parentheses — not attached to the disorder. The tempting shorter version:

71620000 |Fracture of femur| :
  272741003 |Laterality| = 7771000 |Left|      ← invalid

...is not legal, because Laterality may only be applied to a body structure, and Fracture of femur is a clinical finding. A fracture is not left; the femur is left. Getting this wrong is the single most common post-coordination error, and the concept model below is what rules it out.

Pre vs. post: how to choose

	Pre-coordinated	Post-coordinated
What	One existing concept ID	Expression composed at recording time
Pros	Simple to store, query, exchange	Says exactly what happened
Cons	May not exist for your exact meaning	Systems must store/classify expressions
Rule of thumb	Always prefer it when it exists	Use when precision genuinely matters

In the real world, adoption of post-coordination is uneven — many EHRs store only pre-coordinated codes, and interfaces (including FHIR's CodeableConcept) carry expressions less commonly. Knowing the grammar still pays: it's how the terminology itself defines concepts, and it's the foundation ECL queries build on.

The attributes you're allowed to use

You can't attach any attribute to any concept: the concept model (Machine Readable Concept Model, MRCM) defines which attributes apply to which domains — Laterality applies to body structures; Causative agent to clinical findings; Method to procedures. Authoring tools (and Shabda Studio, later) validate against these rules so expressions stay classifiable.

Why a practitioner should care

• Reading definitions: open any concept and its defining relationships are exactly this grammar — you can now read them.
• Edge-case capture: registries and research often need the precision single codes lack.
• Mapping work: understanding composition explains why some local terms map to expressions rather than single SNOMED codes.

Where to go next

ECL — querying SNOMED turns the same grammar into a one-line query language over the whole graph.