The clinical vocabulary: SNOMED CT, LOINC, ICD and the Italian landscape

AIHealth

On-premise clinical platform with local LLMs, RAG on FHIR/DICOM data, diagnostic support, remote follow-up. Architecture designed for the MDR pathway.

Discover AIHealth →

Digital Health

Medical software development compliant with CE and MDR regulatory standards. Clinical decision support systems, AI integration in clinical workflows.

Discover →

Why terminologies matter

Exchange standards — HL7 v2, CDA R2, FHIR — define the structure of messages and documents. They do not tell you what words to name a diagnosis, a lab test or an active ingredient. The semantic part is handed over to controlled clinical terminologies: structured vocabularies in which a meaning maps to a stable code, with a hierarchy or a graph of relationships, and a governance that keeps stability and maintenance.

Without shared terminologies, a code in a CDA message is valid locally but not comparable across systems, regions or countries. With shared terminologies, aggregating, querying and analysing clinical data produced in heterogeneous contexts becomes feasible — the technical precondition for secondary use of healthcare data, for multi-centre research and for epidemiological planning.

The three dominant families today — with distinct areas of competence — are SNOMED CT, LOINC and the ICD family.

SNOMED CT

SNOMED CT (Systematized Nomenclature of Medicine — Clinical Terms) is the most extensive clinical terminology in existence. It was born from the 2002 merger of SNOMED (College of American Pathologists) with the British NHS’s Clinical Terms Version 3. Since 2007 it has been owned by the International Health Terminology Standards Development Organisation, today SNOMED International, a UK-based non-profit with national members in over thirty countries.

The content of the International Edition exceeds 350,000 active concepts: diagnoses, procedures, observations, infectious agents, anatomical structures, generic drugs, findings. Each concept has a SCTID identifier, a preferred description (Fully Specified Name) and multilingual synonym descriptions. The main relationships are IS-A (subclass), which forms a polyhierarchical taxonomy, and a set of domain relationships — finding site, associated morphology, causative agent, has active ingredient.

Two elements make SNOMED CT particularly expressive:

• Post-coordinated expressions — the ability to compose a concept from others: fracture of shaft of tibia + due to fall encodes a composite diagnosis that does not necessarily exist as a pre-coordinated concept
• Reference Sets (refsets) — official or national subsets (for example a refset for cardiology, for ICU diagnoses) that restrict the allowed concepts in a given context

Formal querying uses ECL (Expression Constraint Language): expressions such as <<64572001 (all subclasses of “disease”) enable semantic searches over a coded record.

Adopting SNOMED CT requires membership of SNOMED International by a State or authorised entities, with fees scaled to GDP. Italy is not currently a member of SNOMED International; use in Italian projects occurs through research or trial licences or in specific authorised contexts. This lag — most visible when compared to the United Kingdom, Scandinavia, Germany and the Netherlands — is a recurring theme in the debate on the Italian FSE’s evolution.

LOINC

LOINC (Logical Observation Identifiers Names and Codes) is the reference terminology for laboratory observations, clinical observations, document sections and documents. It is developed by the Regenstrief Institute in Indianapolis since 1994 and distributed free of charge under the LOINC licence, which permits use in any commercial or open source product.

Each LOINC code — for example 1558-6 Fasting plasma glucose [Mass/volume] in Serum or Plasma — results from a six-axis model:

• Component — what is measured (glucose)
• Property — physical property (mass per volume, molar concentration, …)
• Time — temporal interval (point, interval)
• System — biological material or context (serum, plasma, urine, …)
• Scale — measurement scale (quantitative, ordinal, nominal)
• Method — method (optional)

The LOINC universe has grown well beyond the lab: as of 2017 it includes over 90,000 concepts, including codes for documents and sections — used extensively in CDA R2 to identify document type (code in the header) and section type. Code 34133-9 identifies, for instance, a Summarization of Episode Note; 11502-2 a Laboratory report.

In Italy LOINC is adopted in HL7 Italia’s CDA R2 profiles for coding laboratory observations and document sections. An official Italian translation of LOINC has been published since 2012, managed by an Italian consortium, which localises descriptions while keeping the same numeric codes of the International Edition.

ICD

The International Classification of Diseases is the classification maintained by the World Health Organization. The relevant versions:

• ICD-9 (1975) — used globally until the 1990s. In the United States the clinical adaptation ICD-9-CM remained in use until 2015
• ICD-10 (1990, with subsequent revisions) — the current WHO version for coding causes of death in mortality statistics, adopted as the standard for WHO reporting
• ICD-10-CM — US clinical adaptation, adopted for morbidity reporting from 1 October 2015, replacing ICD-9-CM
• ICD-11 — the revision in final stages; WHO is completing the ballot for adoption of the full edition in May 2018

In Italy the situation is peculiar: for the Hospital Discharge Record (SDO) and for much of the health statistics, the code still in use is ICD-9-CM, in the official Italian version published by the Ministry of Health (the 2007 edition is the one mostly in production). For causes of death, ISTAT uses ICD-10. The transition of SDO to ICD-10 has been announced for years but has not yet been completed nationally. The definitive switch is also tied to the upcoming move to ICD-11, which will again shift the landscape.

Other relevant terminologies

Around the SNOMED CT / LOINC / ICD triad orbit specialised vocabularies addressing specific use cases:

• ATC (Anatomical Therapeutic Chemical) — codes drugs by active ingredient and therapeutic class, maintained by the WHO Collaborating Centre for Drug Statistics Methodology in Oslo. In Italy it is the reference terminology for drug coding in health flows
• AIC — identifier of the marketed medicinal product, assigned by AIFA; not a semantic code but a product code
• UCUM (Unified Code for Units of Measure) — standard coding for units of measure in observations (LOINC uses UCUM)
• RxNorm — US drug reference terminology, maintained by the National Library of Medicine
• MedDRA — Medical Dictionary for Regulatory Activities, for coding adverse reactions in clinical trials and pharmacovigilance
• ICD-O — International Classification of Diseases for Oncology, codes tumour topography and morphology
• TNM — cancer staging system, maintained by the Union for International Cancer Control
• HGNC — human gene nomenclature
• ICPC-2 — International Classification of Primary Care, for general practice

Terminologies and exchange standards

Terminologies plug into exchange models through the concepts of:

• CodeSystem — the vocabulary itself (SNOMED CT, LOINC, ICD-10-CM), identified by OID or canonical URI
• ValueSet — a selected subset of codes from one or more CodeSystems, suited to a specific context (e.g. discharge diagnoses, infectious agents)
• ConceptMap — formal mapping between codes from different CodeSystems (ICD-9-CM → ICD-10, LOINC → SNOMED CT observation)

In FHIR these three concepts are first-class Resources (CodeSystem, ValueSet, ConceptMap); in CDA R2 they are expressed as OID references with terminologies coded in <code> and <value>; in observational settings like OMOP CDM CodeSystems are harmonised into the Standardized Vocabularies, the subject of a separate piece.

Tooling

As of 2017 the tooling ecosystem around terminologies is varied:

• Snow Owl (B2i Healthcare, Hungary) — open source terminology server with SNOMED CT, LOINC, ICD, ValueSet and ConceptMap support
• Ontoserver (CSIRO, Australia) — FHIR terminology server, used in the Australian HealthTerminologies service
• BioPortal (NCBO, Stanford) — web portal for biomedical ontologies, with mapping and navigation tools
• OHDSI Athena — browser of the OMOP Standardized Vocabularies, unifying SNOMED CT, LOINC, RxNorm, ICD, ATC, HCPCS, CPT and other terminologies into a single view
• HAPI FHIR Terminology — module of the HAPI FHIR Java server for exposing CodeSystem, ValueSet, ConceptMap via FHIR REST API

The unresolved node for Italy remains formal membership of SNOMED International, without which SNOMED CT uptake in national systems stays partial and tied to specific projects. The growing demand for semantics — driven by clinical AI, observational research and European secondary use programmes — will make this theme increasingly central.

References: SNOMED International — SNOMED CT International Release. Regenstrief Institute — LOINC. WHO — International Classification of Diseases (ICD-9, ICD-10, ICD-11). WHO Collaborating Centre for Drug Statistics Methodology — ATC. AIFA — drug database. Snow Owl, Ontoserver, BioPortal, OHDSI Athena.