Medical Ontologies Explained: SNOMED CT, ICD-10, RxNorm & LOINC

A patient visit produces a story — a complaint, a diagnosis, a lab draw, a prescription. To make that story usable by software, statisticians, or a machine-learning model, every piece of it has to be turned into a code that means exactly one thing. That is the job of medical ontologies. This guide explains the four that matter most — SNOMED CT, ICD-10, RxNorm, and LOINC — what each one codes, who runs it, and how they all point back to the same underlying clinical reality.

What a Medical Ontology Actually Is

In everyday use, "terminology," "vocabulary," "classification," and "ontology" get blurred together. The distinctions matter.

A terminology is a controlled list of terms with stable identifiers, so that "myocardial infarction" and "heart attack" resolve to one code. A classification groups things into mutually exclusive buckets, usually for counting and reporting. An ontology goes further: it encodes formal, machine-readable relationships between concepts — "is a," "finding site," "causative agent" — so that software can reason about meaning, not just match strings.

The reason healthcare needs all of this is simple: free text does not aggregate. You cannot reliably count, compare, or train on "pt c/o SOB" until it becomes a discrete, defined concept. Codes are how clinical reality becomes data.

Importantly, no single system covers everything. Diagnoses, procedures, drugs, and lab results each have their own best-fit standard — which is exactly why interoperability depends on using several together.

SNOMED CT: The Clinical Meaning Layer

SNOMED CT (Systematized Nomenclature of Medicine — Clinical Terms) is the most comprehensive clinical terminology in use. It is designed to capture what clinicians actually document: findings, disorders, procedures, body structures, organisms, substances, and more.

• What it codes: clinical concepts across the whole patient encounter — symptoms, diagnoses, procedures, observations, and the relationships among them.
• Format: each concept has a numeric SNOMED CT identifier, one or more human-readable descriptions, and logic-based relationships that arrange concepts into poly-hierarchies. A concept can have multiple parents, and definitions are built from defining relationships rather than a single tree position.
• Who maintains it: SNOMED International (formerly the IHTSDO), with national release centers managing country-specific extensions.
• Where it's used: electronic health records for structured clinical documentation, decision support, and as the semantic backbone for exchanging meaning between systems.

The defining feature is scale and structure: SNOMED CT contains hundreds of thousands of active concepts with over a million semantic relationships, which is what lets it function as a true ontology rather than a flat code list.

ICD-10: The Classification for Counting and Billing

ICD-10 — the International Statistical Classification of Diseases and Related Health Problems, 10th Revision — answers a different question. SNOMED CT asks "what is true about this patient?" ICD-10 asks "which standardized bucket does this case belong to, for statistics and reimbursement?"

• What it codes: diagnoses and, in some variants, procedures, organized into a classification optimized for reporting and analysis.
• Format: alphanumeric codes with a hierarchical, category-then-subcategory structure (e.g., a three-character category refined by additional characters for detail).
• Who maintains it: the World Health Organization maintains the base ICD. Countries adopt national modifications — in the United States, the CDC's National Center for Health Statistics maintains ICD-10-CM (Clinical Modification) for diagnoses, with a separate procedure system for inpatient billing.
• Where it's used: mortality and morbidity statistics, epidemiology, claims, and reimbursement.

Because ICD is built for clean, mutually exclusive grouping, it is intentionally less granular than SNOMED CT. That is a feature for counting and a limitation for capturing nuance — which is why the two coexist rather than compete.

RxNorm: Normalizing the World of Drugs

Medications are uniquely messy. The same active ingredient appears under dozens of brand names, strengths, dose forms, and packaging. RxNorm exists to normalize all of that into stable concepts.

• What it codes: clinical drugs at multiple levels — ingredient, strength, dose form, and combinations — plus brand names and packs.
• Format: every drug concept gets an RxCUI (RxNorm Concept Unique Identifier). Concept types include ingredient, clinical drug component, semantic clinical drug (SCD), branded variants, and packs, all linked by defined relationships.
• Who maintains it: the U.S. National Library of Medicine.
• Where it's used: e-prescribing, medication reconciliation, drug interaction checking, and any pipeline that needs to know two records describe the same medication.

RxNorm's superpower is that it sits between proprietary and regulatory drug vocabularies, providing a normalized hub so a prescription written one way can be matched to a dispensing record written another.

LOINC: A Universal Language for Observations

If RxNorm tames drugs, LOINC (Logical Observation Identifiers, Names, and Codes) tames tests and observations. Without it, "glucose" from one lab and "blood sugar" from another never line up.

• What it codes: laboratory tests, clinical measurements, and other observations — and, increasingly, survey instruments and document types.
• Format: each LOINC code carries a fully specified name assembled from defined "parts" (axes) such as the component/analyte, the property measured, timing, the specimen/system, the scale, and method where relevant. This part-based model is what makes two differently named tests comparable.
• Who maintains it: the Regenstrief Institute.
• Where it's used: lab result exchange, public health reporting, and harmonizing observations across institutions and instruments.

LOINC's value is precision about what was measured and how, which is essential before any result can be trusted for analysis or aggregation.

How They Cross-Walk to One Clinical Reality

Here is the key insight: these systems are complementary lenses on a single encounter, not rivals.

Picture one diabetic patient with chest pain. The problem ("type 2 diabetes," "chest pain") is captured natively in SNOMED CT and rolled up to an ICD-10-CM code for the claim. The ordered lab ("HbA1c") is identified by a LOINC code, and its result flows back under that same code. The prescription ("metformin 500 mg tablet") resolves to an RxNorm RxCUI. Four ontologies, one reality.

Crosswalks make the lenses interoperable. Curated maps — such as the SNOMED CT to ICD-10-CM map maintained through the National Library of Medicine — let a clinically rich SNOMED concept translate into the right classification code for reporting. Umbrella resources like the UMLS Metathesaurus link concepts across vocabularies so that synonymous codes can be reconciled.

This is also why multi-ontology coding matters for grounding. A record coded in only one system is ambiguous or lossy: ICD alone misses clinical nuance; SNOMED alone is hard to count; neither describes the drug or the lab properly. Coding the same fact across the appropriate standards — and preserving the maps between them — is what turns scattered records into longitudinal, queryable, model-ready ground truth.

> Note: this article is an educational explainer, not medical, legal, or regulatory advice. Coding for billing, compliance (including HIPAA de-identification), or regulatory real-world evidence submissions should follow current official guidance and qualified review.

Key Takeaways

• SNOMED CT captures clinical meaning (findings, disorders, procedures) as a true ontology with logic-based relationships; maintained by SNOMED International.
• ICD-10 / ICD-10-CM is a classification for statistics and billing — coarser by design; the WHO maintains the base, national bodies (CDC/NCHS in the U.S.) maintain modifications.
• RxNorm normalizes medications into RxCUIs across ingredient, strength, and dose form; maintained by the NLM.
• LOINC standardizes observations and lab tests via a part-based naming model; maintained by Regenstrief.
• They are complementary lenses on one encounter — and multi-ontology coding plus curated crosswalks is what makes clinical data interoperable and trustworthy at scale.

Why This Matters for Grounding AI and Evidence

Models and analyses are only as reliable as the structure beneath them. Data that is coded once, narrowly, or inconsistently inherits ambiguity that no algorithm can fully recover. Data that is coded across the right standards — and kept consistent over time — becomes a dependable reference for both AI training and clinical evidence. At Prometheus Bio, multi-ontology coding is treated as part of what "ground truth" means: not just records, but records grounded in the standards the field already trusts.