6.3 Clinical Terminologies: SNOMED CT, LOINC, RxNorm, and ICD
Key Takeaways
- SNOMED CT codes clinical concepts—problems, procedures, organisms, findings—with relationships. It is a clinical reference terminology, not a claim form.
- LOINC identifies observations and documents: laboratory tests, vital signs, surveys, and some report titles. It names the question, not the problem list.
- RxNorm normalizes medication concepts (ingredient, strength, dose form) for medication lists, e-prescribing, and drug decision support.
- ICD (ICD-10-CM/PCS in the United States) is a classification for billing, statistics, and many quality value sets. It is not a clinical interface terminology.
- Prefer native capture in the terminology built for the meaning, then map outbound. Do not make clinicians shop ICD codes as the only way to record a problem.
6.3 Clinical Terminologies: SNOMED CT, LOINC, RxNorm, and ICD
Quick Answer: SNOMED CT = clinical concepts (problems, findings, procedures). LOINC = observations (labs, vitals, many documents). RxNorm = medications. ICD = administrative classification for claims, statistics, and many reporting value sets. ICD is not a clinical interface terminology. Capture natively, then map.
Tasks A.1 and A.2 both collapse if the coded layer is wrong. A perfectly designed allergy module still fails if the only code you can store is a billing ICD string. A lab interface that files results to a homemade mnemonic will not exchange, trend, or decision-support cleanly. This section is the semantic backbone of Domain 2: what each family is for, what it is not for, and when you map instead of forcing clinicians to speak administrative code.
SNOMED CT — concepts, not claims
SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms) is a comprehensive clinical reference terminology. It represents concepts—clinical findings, procedures, body structures, organisms, substances, events—and the relationships among them (for example, bacterial pneumonia is a pneumonia and has causative agent a bacterium). Concepts have identifiers; humans see descriptions (fully specified names and synonyms). That design is why SNOMED can power problem lists, procedure documentation, allergy substances, infectious-disease organisms, and decision-support logic that walks a hierarchy (“if any beta-lactam allergy…”).
SNOMED is not a claim. Payers do not adjudicate a CMS-1500 from a raw SNOMED identifier. You may map a SNOMED problem to an ICD-10-CM code for billing or for a quality eCQM value set. That map is a second artifact. If you skip SNOMED and let the clinician pick only ICD on the problem list, you lose laterality nuance, clinical hierarchy, and the ability to group related problems without maintaining a private synonym table.
Interface versus reference: clinicians search the words they say (“heart failure,” “HFrEF,” “CHF”). An interface terminology is that searchable, friendly layer. SNOMED often sits behind it as the reference concept. Some products expose SNOMED descriptions directly. Either pattern is still “clinical terminology.” ICD is not that layer. ICD is a classification optimized for statistics and payment, with residual “other” and “unspecified” bins that are useful for coding and harmful as the only problem-list language.
LOINC — observations, not problems
LOINC (Logical Observation Identifiers Names and Codes), maintained by the Regenstrief Institute, identifies questions: laboratory tests, vital signs, survey instruments, some clinical documents and report titles. A LOINC term is built from parts such as component, property, timing, system (specimen), scale, and method. “Glucose mass concentration in serum” is a different LOINC than “glucose mass concentration in CSF” or a point-of-care strip method.
LOINC tells you what was measured or asked. It does not, by itself, tell you the patient’s problem. A sodium result of 118 mmol/L is an observation (LOINC). Hyponatremia on the problem list is a clinical concept (SNOMED, then possibly ICD for the claim). Filing “hyponatremia” as a LOINC, or using LOINC as the problem-list code system, is a category error.
Answers to LOINC questions may themselves be coded. A microbiology organism or a nominal survey answer often uses SNOMED (or another answer list) while the test name stays LOINC. Informatics builds should keep that split: LOINC for the observation identity, SNOMED (or a declared answer list) for many coded values.
RxNorm — medications
RxNorm, from the U.S. National Library of Medicine, is the normalized terminology for clinical drugs and related medication concepts: ingredients, precise ingredients, dose forms, strengths, branded products, and relationships among them. It is the lingua franca for medication lists, e-prescribing, formulary alignment, and much drug-allergy or duplicate-therapy CDS.
RxNorm is not the bottle on the shelf. Dispense-level identifiers such as NDC describe packaged products. Inpatient pharmacies still need NDC, inventory, and barcode product identifiers. The informatics pattern is: clinicians think in RxNorm-like clinical drugs (“lisinopril 10 mg oral tablet”); supply chain and BCMA think in the package that was scanned; maps connect them. Do not use ICD to mean a drug, and do not use LOINC to mean a medication order.
ICD — classification for administrative use
ICD (International Classification of Diseases) is a classification. In current U.S. practice, ICD-10-CM classifies diagnoses and some external causes; ICD-10-PCS classifies inpatient procedures. WHO publishes the international ICD family; U.S. CM/PCS have their own maintenance processes. Hospitals and clinics use ICD for claims, utilization statistics, public-health reporting, and as the backbone of many quality and risk-adjustment value sets.
What ICD is not:
- It is not a clinical interface terminology. “I50.9 Heart failure, unspecified” is a residual class, not a good way to capture HFrEF versus HFpEF for care.
- It is not an observation identifier. Labs belong in LOINC.
- It is not a medication terminology. Drugs belong in RxNorm (plus NDC where needed).
- It is not proof that a problem list is clinically curated.
Coders, CDI specialists, and billing systems should remain fluent in ICD. Informaticists should keep ICD in the outbound and administrative layer unless a specific reporting workflow truly requires a clinician to choose an ICD code (some problem lists still do; that is a design compromise, not the semantic ideal).
CPT and HCPCS are professional and some facility procedure/service codes for billing. They are not one of the four named families in this section’s title, but the same rule applies: billing procedure codes are not SNOMED procedure concepts and not LOINC observations.
Mapping versus native capture
Native capture means the user records meaning in the code system designed for that meaning, at the time of documentation:
| What the clinician is recording | Native code system | Typical outbound map |
|---|---|---|
| Problem, finding, some procedures, organisms | SNOMED CT | ICD-10-CM/PCS, sometimes CPT, quality value sets |
| Lab test, vital sign, survey, many documents | LOINC | Public-health or HIE observation catalogs; answer values may be SNOMED |
| Medication on a list or prescription | RxNorm | NDC for dispense, formulary codes, some CDS local tables |
| Claim diagnosis or inpatient coded procedure | ICD-10-CM/PCS | Payer edits, groupers, reports |
Mapping is a maintained translation: SNOMED “type 2 diabetes mellitus” to E11.9 or a more specific E11.- code depending on documentation; a local lab mnemonic to a LOINC; an RxNorm clinical drug to the NDC stocked today. Maps are many-to-one, one-to-many, and version-sensitive. They require governance. They are how you get semantic interoperability without making every nurse think in claim codes.
The failure mode CPHIMS loves: a project team “implements SNOMED” by mapping ICD codes to SNOMED after the fact and never changing what the clinician searches. That is a reporting crosswalk, not native capture. The opposite failure is equally real: a lab goes live with local test names and promises to “add LOINC later,” then cannot exchange or trend across sites.
Putting the four families in one encounter
A patient is admitted with shortness of breath. The problem list gets a SNOMED concept for acute on chronic systolic heart failure. The BNP and chest radiograph are LOINC observations (the radiology report title may also be LOINC; findings in the narrative are another layer). Home carvedilol is an RxNorm clinical drug, matched to an NDC for BCMA. At coding, ICD-10-CM heart-failure codes and any procedure codes go on the claim. If you put the ICD code on the problem list and a homemade “BNP” string in the result file, the stay is billable and the record is clinically poorer—and every HIE partner has to guess what you meant.
Scenarios and exam traps
Scenario. A steering committee says “we already have ICD on every account, so we do not need SNOMED.” That sentence treats a classification as a clinical interface terminology. Care management, CDS hierarchies, and problem-oriented charts will stay blunt.
Scenario. An analyst codes “diabetes” as a LOINC because a glucose panel exists. LOINC can identify the glucose observation. It does not belong as the problem.
Scenario. Pharmacy wants all inpatient medication CDS to key off ICD procedure codes. Medications are RxNorm (and products/NDCs). ICD does not identify carvedilol 6.25 mg tablets.
Watch these traps:
- ICD is not a clinical interface terminology. Do not make unspecified billing classes the only words on the problem list.
- LOINC is observations, not problems. Tests and vitals are not diagnoses.
- SNOMED is concepts, not claims. You still need ICD (and often CPT) to bill.
- Mapping after the fact is not the same as native capture.
- RxNorm is not NDC, and neither is ICD.
- Two sites sharing an EHR vendor have not achieved semantic interoperability if each uses private local codes.
When a stem asks “which terminology,” name the meaning first—problem, observation, medication, or claim—then pick the family. That habit is the entire section.
A clinic wants clinicians to maintain the problem list by picking only ICD-10-CM codes because “those are the official codes.” Which statement should the informatics lead defend?
A project files “hyponatremia” as a LOINC code on the problem list because the serum sodium is low. What is the correct placement?
Why is it incorrect to say that documenting a SNOMED CT problem “completes the claim”?