SNOMED CT

SNOMED CT statistics (as of November 2016). SD stands for sufficiently defined.

SNOMED CT can be characterized as a multilingual thesaurus with an ontological foundation. Thesaurus-like features are concept–term relations such as the synonymous descriptions "Acute coryza", "Acute nasal catarrh", "Acute rhinitis", "Common cold" (as well as Spanish "resfrío común" and "rinitis infecciosa") for the concept 82272006.

Under ontological scrutiny, SNOMED-CT is a class hierarchy (with extensive overlap of classes in contrast to typical statistical classifications like ICD). This means that the SNOMED CT concept 82272006 defines the class of all the individual disease instances that match the criteria for "common cold" (e.g., one patient may have "head cold" noted in their record, and another may have "Acute coryza"; both can be found as instances of "common cold"). The superclass (Is-A) Relation relates classes in terms of inclusion of their members. That is, all individual "cold-processes" are also included in all superclasses of the class Common Cold, such as Viral upper respiratory tract infection (Figure).

Common cold as a primitive concept in SNOMED CT

SNOMED CT's relational statements are basically triplets of the form Concept₁ – Relation_x – Concept₂, with Relation_x being from a small number of relation types (called linkage concepts), e.g. finding site, due to, etc. The interpretation of these triplets is (implicitly) based on the semantics of a simple Description logic (DL). E.g., the triplet Common Cold – causative agent – Virus, corresponds to the first-order expression

forall x: instance-of (x, Common cold) -> exists y: instance-of (y, Virus) and causative-agent (y, x)

or the more intuitive DL expression

Common cold subClassOf causative-agent some Virus

In the Common cold example the concept description is "primitive", which means that necessary criteria are given that must be met for each instance, without being sufficient for classifying a disorder as an instance of Common Cold . In contrast, the example Viral upper respiratory tract infection depicts a fully described concept, which is represented in description logic as follows:

Viral upper respiratory tract infection as a defined concept in SNOMED CT

 Viral upper respiratory tract infection equivalentTo
 	Upper respiratory infection and Viral respiratory infection and
 		Causative-agent some Virus and
 		Finding-site some Upper respiratory tract structure and
 		Pathological-process some Infectious process

This means that each and every individual disorder for which all definitional criteria are met can be classified as an instance of Viral upper respiratory tract infection.

As of 2011, SNOMED CT content limits itself to a subset of the EL++ formalism, restricting itself to the following operators:

• Top, bottom
• Primitive roles and concepts with asserted parent(s) for each
• Concept definition and conjunction but NOT disjunction or negation
• Role hierarchy but not role composition
• Domain and range constraints
• Existential but not universal restriction
• A restricted form of role inclusion axiom (xRy ^ ySz => xRz)
• The logic will be extended in the near future to include General Concept Inclusion Axioms.

For understanding the modelling, it is also important to look at the stated view of a concept versus the inferred view of the concept. In further considering the state view, SNOMED CT used in the past an modelling approach referred to as 'proximal parent' approach. After 2015, a superior approach called "proximal primitive parent" has been adopted.

SNOMED CT provides a compositional syntax[19] that can be used to create expressions that represent clinical ideas which are not explicitly represented by SNOMED CT concepts.

For example, there is no explicit concept for a "third degree burn of left index finger caused by hot water". However, using the compositional syntax it can be represented as

284196006 | burn of skin | :
   116676008 | associated morphology | = 80247002 | third degree burn injury |
 , 272741003 | laterality | = 7771000 | left |
 , 246075003 | causative agent | = 47448006 | hot water |
 , 363698007 | finding site | = 83738005 | index finger structure

Such expressions are said to have been 'postcoordinated'. Post-coordination avoids the need to create large numbers of defined Concepts within SNOMED CT. However, many systems only allow for precoordinated representations. Reliable analysis and comparison of post-coordinated expressions is possible using appropriate algorithms machinery to efficiently process the expression taking account of the underlying description logic.

Major Electronic Health Record Systems (EHRS) have repeatedly complained to IHTSDO and other standards organizations about the "complexity" of post-coordinated expressions.

For example, the postcoordinated expression above can be transformed using a set of standard rules to the following "normal form expression" which enables comparison with similar concepts.

64572001 | disease | :
   246075003 | causative agent | = 47448006 | hot water |
 , 363698007 | finding site | = ( 83738005 | index finger structure | :
          272741003 | laterality | = 7771000 | left | )
 , { 116676008 | associated morphology | = 80247002 | third degree burn injury |
 , 363698007 | finding site | = 39937001 | skin structure | }

The International Edition of SNOMED CT only includes human terms. In 2014, clearly veterinary concepts were moved into a SNOMED CT veterinary extension. This extension is managed by the Veterinary Terminology Services Lab[20] at the Va-Md College of Veterinary Medicine at Virginia Tech.

Earlier SNOMED versions had faceted structure ordered by semantic axes, requiring that more complex situations required to be coded by a coordination of different codes. This had two major shortcomings. On the one hand, the necessity of post-coordination was perceived as a user-unfriendly obstacle, which has certainly contributed to the rather low adoption of early SNOMED versions. On the other hand, uniform coding was difficult to obtain. E.g.,Acute appendicitis could be post-coordinated in three different ways[21] with no means to compute semantic equivalences. SNOMED RT had addressed this problem by introducing description logic formula. With the addition of CTV3 a large number of concepts were redefined using formal expressions. However, the fusion with CTV3, as a historically grown terminology with many close-to user descriptions, introduced some problems which still affect SNOMED CT. In addition to a confusing taxonomic web of many hierarchical levels with massive multiple inheritance (e.g. there are 36 taxonomic ancestors for Acute appendicitis), many ambiguous, context-dependent concepts have found their way into SNOMED CT. Pre-coordination was sometimes pushed to extremes, so there are, for example, 350 different concepts for burns found on the head.

A further phenomenon which characterizes parts of SNOMED CT is the so-called epistemic intrusion.[22] In principle, the task of terminology (and even an ontology) should be limited to providing context-free term or class meanings. The contextualization of these representational units should be ideally the task of an information model.[23] Human language is misleading here, as we use syntactically similar expression to represent categorically distinct entities, e.g. Ectopic pregnancy vs. Suspected pregnancy. The first one refers to a real pregnancy, the second one to a piece of (uncertain) information. In SNOMED CT most (but not all) of these context-dependent concepts are concentrated in the subhierachy Situation with explicit context. A major reason for why such concepts cannot be dispensed with is that SNOMED CT takes on, in many cases, the functionality of information models, as the latter do not exist in a given implementation.

With the establishment of IHTSDO, SNOMED CT became more accessible to a wider audience. Criticism of the state of the terminology was sparked by numerous substantive weaknesses as well as on the lack of quality assurance measures.[24] From the beginning IHTSDO was open regarding such (also academic) criticism. In the last few years considerable progress has been made regarding quality assurance and tooling.

The need for a more principled ontological foundation was gradually accepted, as well as a better understanding of description logic semantics. Redesign priorities were formulated regarding observables,[25] disorders, findings,[26] substances, organisms etc. Translation guidelines[27] were elaborated as well as guidelines for content submission requests and a strategy for the inclusion of pre-coordinated content. There are still known deficiencies regarding the "ontological commitment" of SNOMED CT,[28] e.g., the clarification of which kind of entity is an instance of a given SNOMED CT concept. The same term can be interpreted as a disorder or a patient with a disorder, for example Tumour might denote a process or a piece of tissue; Allergy may denote an allergic reaction or just an allergic disposition. A more recent strategy is the use of rigorously typed upper-level ontologies to disambiguate SNOMED CT content.

The increased take-up of SNOMED CT for research into applications in daily use across the world to support patient care is leading to a larger engaged community. This has led to an increase in the resource allocated to authoring SNOMED CT terms as well as to an increase in collaboration to take SNOMED CT into a robust industry used standard. This is leading to an increase in the number of software tools and development of materials that contribute to knowledge base to support implementation. A number of on-line communities that focus on particular aspects of SNOMED CT and its implementation are also developing.

In theory, description logic reasoning can be applied to any new candidate post-coordinated expressions in order to assess whether it is a parent or ancestor of, a child or other descendant of, or semantically equivalent to any existing concept from the existing pre-coordinated concepts. However, partly as the continuing fall-out from the merger with CTV3, SNOMED still contains undiscovered semantically duplicate primitive and defined concepts. Additionally, many concepts remain primitive whilst their semantics can also be legitimately defined in terms of other primitives and roles concurrently in the system. Because of these omissions and actual or possible redundancies of semantic content, real-world performance of algorithms to infer subsumption or semantic equivalence will be unpredictably imperfect.

Using consistent rules is important for the quality of SNOMED CT. To that end, in 2009, a prototype Machine Readable Concept Model (MRCM) was created by the SNOMED CT team. In a follow up work, this model is being revised to utilize SNOMED CT expression constraints.

SNOMED CT is a clinical terminology designed to capture and represent patient data for clinical purposes.[29] The International Statistical Classification of Diseases and Related Health Problems (ICD) is an internationally used medical classification system; which is used to assign diagnostic and, in some national modifications, procedural codes in order to produce coded data for statistical analysis, epidemiology, reimbursement and resource allocation.[30] Both systems use standardized definitions and form a common medical language used within electronic health record (EHR) systems.[31] SNOMED CT enables information input into an EHR system during the course of patient care, while ICD facilitates information retrieval, or output, for secondary data purposes.[31][32]

LOINC is a terminology that contains laboratory tests. Since 2017, SNOMED International started creating terms for LOINC components and created a set of SNOMED CT expressions that capture the meaning of many LOINC terms.

More specifically, the following sample computer applications use SNOMED CT:

• Electronic Health Record Systems
• Computerized Provider Order Entry CPOE such as E-Prescribing or Laboratory Order Entry
• Catalogues of clinical services; e.g., for Diagnostic Imaging procedures
• Knowledge databases used in clinical decision support systems (CDSS)
• Remote Intensive Care Unit Monitoring
• Laboratory Reporting
• Emergency Room Charting
• Cancer Reporting
• Genetic Databases

SNOMED CT is maintained and distributed by SNOMED International, an international non-profit standards development organization, located in London, UK..

The use of SNOMED CT in production systems requires a license. There are two types of license:

1. Country/territory membership in SNOMED International (charged according to gross national product).
2. Affiliate license (dependent on the number of end users). LDCs (least developed countries) can use SNOMED CT without charges.

For scientific research in medical informatics, for demonstrations or evaluation purposes SNOMED CT sources can be freely downloaded and used. The original SNOMED CT sources in tabular form are accessible by registered users of the Unified Medical Language System (UMLS) who have signed an agreement. Numerous online and offline browsers are available.

Those wishing to obtain a license for its use and to download SNOMED CT should contact their National Release Centre, links to which are provided on the IHTSDO website.

To facilitate adoption of SNOMED CT and use of SNOMED CT in other standards, there are license free subsets. For example, a set of 7 314 codes and descriptions is free for use by users of DICOM-compliant software (without restriction to IHTSDO member countries).[33]

GPS was released in Sep 2019 and contains 21 782 concepts.[34]

As of 2016, the Event hierarchy does not have a concept model defined. In 2006, some concepts from the 'Clinical Finding' hierarchy were moved to the Event hierarchy. Those concepts retained some of their attributes. (e.g., causative agent)

SNOMED International is working on creating a concept model for observable entities.[35]

Pharmaceutical product example

Pharmaceutical and biologic products are modeled using constructs of active ingredient, presentation strength, and basis of strength. Since 2018, harmonization of SNOMED CT drug content with IDMP standard is an editorial goal. The following types of entities are present:

A higher level term grouping drugs. For example, 398731002 | Product containing sulfamethoxazole and trimethoprim (medicinal product) |

Concept that represents a concrete drug product as used in clinical practice. For example, 317335000 | Product containing precisely esomeprazole 20milligram/1 each conventional release oral tablet (clinical drug)|

Concept representing how the product is delivered. For example, 385219001 | Conventional release solution for injection (dose form) |.