Semantic Forms

Similar to the stated and inferred view of precoordinated concepts, as described in Precoordination and Postcoordination, the meaning represented by postcoordinated expressions can also be represented in different ways. When enabling postcoordinated expressions within a system, it is important to be support these different views and understand why they are needed.

Close to User Form

The Close to User Form (CTU) Expression is the faithful representation of the clinical meaning as it was entered by the user. It is the{ primary stored and communicated view of the clinical information that’s encoded using SNOMED CT. The CTU Expression includes any refinement that was applied by the system, based on the selections made in a data entry form, or those made explicitly.

And, importantly, it does not include any additional relationships that are added, based on classifier rules to make the expression complete or to normalize it. It only includes the parts of the clinical meaning that were specifically intended by the user.

Classifiable Form

The Classifiable Form (CF) expression is a syntactically valid and concept model compliant representation of the CTU expression. It serves as the input to the classifier, enabling expressions to be classified together with other SNOMED CT content.

And in the classifiable form, any ungrouped attributes and role groups have been validated and appropriately applied to the definition of the focus concept.

The Necessary Normal Form (NNF) expression is the inferred view of the expression, and it includes all the relationships that are necessarily true, with redundancy removed. This is the output of the process that took the classifiable form expression and classified it with a given SNOMED edition. The NNF represents the necessary relationships used for querying, so this becomes part of your substrate when you’re running an Expression Constraint query.

The NNF includes refinements that represent inferred relationships, without any redundant refinements or redundant relationship groups.

Within the released SNOMED CT content, the stated definitions of SNOMED CT concepts are represented as Description Logic OWL axioms, and these are the clinical definition that SNOMED authors state when they define the meaning of a concept. For postcoordinated expressions, the CTU expression can be compared to the stated concept definition, as it represents the expression as it was created.

When querying SNOMED CT concepts , the inferred definitions are used. The inferred definition of a concept is derived (or inferred) by classifying the stated definitions using a Description Logic reasoner. The output of this classification process is represented in the release using the relationship table and includes the necessary normal form for each concept. The necessary normal form represents the set of ‘necessary’ inferred relationships. For more information on concept definitions, please refer to Appendix D of the release file specification: . In the same way as for precoordinated concepts, the inferred view, or the NNF expressions, needs to be generated to support querying over postcoordinated expressions, see the image below.

To enable the classification of the CTU expression, a transformation is required to provide a form that can serve as input to a Description Logic reasoner. Such reasoner requires a form which is syntactically valid and complies to the concept model rules. This is why all support for postcoordination also requires the ability to generate the CF of each expression.

Although the Compositional Grammar syntax supports a standard way of representing SNOMED CT expressions, expressions that contain exactly the same concept identifiers and refinements, may still differ from one another in the following ways:

• Inclusion of whitespace between elements

• Inclusion of specific terms associated with identified concepts

• The order in which focus concepts, refinements, attributes, and attribute groups appear

The canonical form of an expression is a serialized representation produced by applying a set of rules that ensure a single unique representation for any expression.

Precoordination and Postcoordination

SNOMED CT can be considered as a knowledge graph (direct acyclic graph), where each concept is represented as node in the graph structure, and the position of each node corresponds to its meaning and relationship to other concepts in the graph. These positions can be thought of as coordinates. Precoordinated concepts are predefined and accurately placed in the graph, while postcoordinated expressions require coordination into the graph by a description logic classifier. This is necessary to ensure that the combined concepts are positioned correctly in the graph to represent their relationships, and to allow for effective search and retrieval of the resulting concepts. In this context, the use of coordinates refers to the position of nodes in the SNOMED graph, and precoordination and postcoordination refer to the placement of concepts in the graph at a certain moment in time.

Concepts and Expressions

A SNOMED CT concept represents a clinical idea with a unique SNOMED CT identifier that is distributed as part of a SNOMED CT release. Concepts are linked to terms by descriptions, which provide a human-readable representation of the meaning of the concept. Additionally, they are related to each other by relationships, where each relationship represents a defining property of the concept being the source of the relationship. These defining relationships provide a machine-processable representation of the concept’s meaning.

Aside from the human-readable representation, all released SNOMED CT concepts can be represented in various ways, three of which are:

• their concept identifier, which uniquely identifies the concept

• their stated view, represented through the explicit representation of their defining properties as stated by the author of the concept

• their inferred view, which represents the formal definition of concepts as logically derived by applying a description logic classifier to the stated view

Please explore the example below which provides an example of three different views that can be used to refer to the clinical meaning 'appendicitis'

To support a standardized and computer-processable way of representing and referring to clinical meanings in SNOMED CT, irrespective of whether this involves a single concept identifier, or a combination of concept identifiers, SNOMED CT supports expressions.

A SNOMED CT expression is a structured combination of one or more that represents a single clinical idea. SNOMED International distinguish between two main types of expressions, i.e. precoordinated expressions and postcoordinated expressions.

A precoordinated expression is the simplest form of an expression. It contains a single concept identifier, and optionally one of the terms associated with the concept to make the expression human readable. For example, a precoordinated expression, which means “fracture of tibia” can be as represented as “31978002”, optionally followed by the term ‘fracture of tibia’ placed in vertical bars known as ‘pipes’. This means that all available concepts in SNOMED CT can be represented as precoordinated expressions, by using the assigned concept identifier.

Examples of precoordinated expressions are:

Postcoordinated expressions contain two or more concept identifiers and are used to specify a clinical meaning by referencing existing SNOMED CT concepts.

For example, the clinical meaning "Pain in left arm" can be represented by the following expression including three SNOMED CT concepts:

Another example is the clinical meaning of "Edema of right lung", which can be represented as:

The SNOMED CT Compositional Grammar is used as the syntax for expressions, and the SNOMED CT Concept Model rules determine how concepts can be combined. These topics are elaborated in the subsequent pages of this guide.

Postcoordinatied expressions are useful to support the authoring of new concepts in SNOMED CT, but they may also be created to support specific implementation use cases, e.g. when no precoordinated concept is available for a specific clinical meaning, or when an information model or implementation strategy requires a certain representation of the clinical meanings.

The ability to create postcoordinated expressions greatly increases the depth of detail that SNOMED CT can represent without having to include every possible specific site for every possible disorder via a released concept.

This chapter provides an introduction to SNOMED CT expressions and covers several aspects related to their representation and use.

It begins by outlining the key differences between pre- and postcoordination, including their relative benefits and limitations. It covers the syntax used to represent expressions, providing a detailed overview of the various expression parts and operators that are used. The chapter introduces the different forms that expressions can take, and it discusses the role of the concept model and editorial guidance required to ensure the accuracy and consistency of expressions.

There is a range of use cases in which it may be helpful to use postcoordination, including:

Capture new clinical meanings in a health record, message structure, or query specification

• To support situations where clinicians need to record and share a clinical meaning, which has not been defined in any release of SNOMED CT. This applies to direct data entry, coding values in UIs, or other clinical situations. SNOMED CT postcoordinated expressions can be used to represent a new clinical meaning as an alternative to concept authoring.

Enable interoperability between different structural representations of the same clinical meaning

• To support transformation between different representations within a clinical system or while exchanging data between systems. The postcoordinated expression is a common representation of clinical meaning. It can be exchanged and compared between models and shared as part of interoperability standards like HL7 FHIR.

Represent maps from other code systems or from interface terms to an equivalent representation of the meaning in SNOMED CT

• To map other code systems to SNOMED CT may require the representation of new meanings or combinations of meanings already represented in SNOMED CT. In these cases, when a suitable precoordinated concept is not available, postcoordination offers an alternative to concept authoring.

Support Natural Language Processing

• To transform the meaning of free text into structured SNOMED CT expressions. Postcoordination allows the representation of specific context or a greater amount of detail in the free text than you can capture with a single precoordinated code

Before you launch into a postcoordinated solution it’s important to understand that in some cases there are alternatives to postcoordination that may be easier to implement.

The two main alternatives to postcoordination are:

Create extension concepts:

• Firstly, and perhaps the most obvious alternative to postcoordination is to add the new clinical meanings to an extension, as precoordinated content. This approach enables the use of standard SNOMED authoring tools to classify the concepts and distribute them in a way that supports queries. However, this alternative is not an option for everyone, and in those cases, postcoordination may be considered.

Use information model structure:

• The second alternative is to define the postcoordination structure using the information model, using separate coded data elements for the focus concept, and each of the refinement values in the expression. For example, rather than using an expression for ‘fracture of the foot’, two separate data elements could be applied, one for the SNOMED CT concept for 'fracture' and the other for the SNOMED CT concept for 'foot'. This approach has good accuracy for data capture but introduces significant limitations for the execution of semantic queries over the clinical data, as the focus concept is separated from the refinements and is not combined into a single SNOMED expression. In some situations where data is captured separately and follows the SNOMED CT concept model, post-processing can be applied to convert information models into postcoordinated expressions, after data capture.

Compositional Grammar

The syntax used to represent SNOMED CT expressions is called the Compositional Grammar. This is a formal grammar describing the symbols you can use and how the different parts of an expression can be constructed. Please refer to Compositional Grammar - Specification and Guide for the full detail on this.

While alternative syntaxes are possible, the compositional grammar syntax is considered to be the normative standard for interoperability purposes. Therefore, the recommended practice for anyone supporting postcoordination within their clinical information system, is to represent expressions using this syntax.

Structure

The diagram below illustrates the key parts of a SNOMED CT expression through an example. The expression used in this example represents the stated view of the concept 16554361000119106 | PET CT of brain| .

A SNOMED CT Compositional Grammar expression begins with an optional definition status, contains one or more focus concepts (represented by a concept identifier) and optionally has a refinement. Each refinement may contain grouped or ungrouped attributes (or both). An attribute consists of the attribute name (represented by a concept identifier) together with the value of the attribute. The attribute value is either an expression or a concrete value (i.e. string, integer, decimal or boolean). Please note that, in the context of an expression, the word 'Attribute' is used to refer to the name/value pair within a refinement. However, in the context of the SNOMED CT concept model, the word 'Attribute' is often used to refer specifically to the 'attribute name' (i.e. the concept used as the relationship type).

Overall, postcoordination may be performed to either refine or qualify the meaning of a SNOMED CT concept. The refinement part of a postcoordinated expression may thus represent either a 'refinement' or a 'qualification' of the stated focus concept.

Refining the meaning of a focus concept is done by applying more specific values to particular attributes of the concept. The definition of the focus concept, therefore, determines the refinements that are possible for the concept.

Consider the following expression representing the meaning 'Spiral fracture of tibia':

The focus concept of this expression has the following stated definition:

The attribute value of used in the expression (|Fracture, spiral (morphologic abnormality)|) is a subtype of the attribute-value present in the definition of the concept (|Fracture (morphologic abnormality)|), as shown below.

Qualifying the meaning of a concept can be done by adding a qualifying characteristic to a concept, e.g. by applying and attribute-value pair to the focus concept, where that attribute-value pair does not refine an existing attribute-value pair present in the definition of the focus concept. If a particular qualifying characteristic is applied to a concept, the resulting expression represents a more tightly defined subtype of that concept. All attribute-value pairs stated to qualify a focus concept should fall within the acceptable domain and range determined by the SNOMED CT concept model.

Examples of qualifications include:

The concept may be qualified according to its clinical course (<< 288524001 |Courses (qualifier value)| ) or severity ( << 272141005 |Severities (qualifier value)| )

The concept may be qualified according to its priority (<< 272125009 |Priorities (qualifier value)|) or intent (<< 363675004 |Intents (nature of procedure values) (qualifier value)|)

As SNOMED CT evolves with each new release, maintaining the relevance and accuracy of postcoordinated expressions becomes imperative. Postcoordinated expressions allow for detailed and nuanced clinical concepts by combining multiple SNOMED CT codes. However, without a robust maintenance process, these expressions can become outdated, leading to failures in subsumption by any concept in SNOMED CT and exclusion from Expression Constraint Language (ECL) results.

Importance of Maintenance After Each Update

Alignment with Latest SNOMED CT Versions

• Concept Evolution : SNOMED CT regularly updates its content, which may include the addition of new concepts, modification of existing ones, or retirement of outdated concepts.

• Accuracy in Clinical Representation : Expressions that reference inactive concepts will no longer effectively represent valid clinical meanings, leading to misinterpretation of patient data.

• Interoperability : Ensuring that the substrate and the expressions are up-to-date maintains interoperability between systems using the latest versions of SNOMED CT.

• ECL Queries : Outdated expressions won't appear in ECL query results, hindering data analytics and reporting.

• Clinical Decision Support : Inaccurate expressions can affect decision-making processes that rely on current clinical terminologies.

• Non-Semantic Identifiers : Assign a unique, non-semantic identifier to each expression, such as:

  • A "16" partition SNOMED Identifier

  • A sequential number

• Post-Upgrade Audit : After upgrading to a new SNOMED CT edition, generate a comprehensive report listing expressions that require updates.

• Identification Criteria : Expressions may need updating if they:

  • Contain references to retired concepts

• Leverage SNOMED CT Historical Data : Utilize historical associations provided by SNOMED CT to map outdated concepts to current ones.

• Configuration of Confidence Levels :

  • High Confidence : Automatic replacements using "Same As" associations.

• Complex Cases : Not all expressions can be updated automatically due to:

  • Ambiguities in historical associations

  • Changes in clinical practice or terminology

• Time and Effort : Manual reviews are time-consuming and require significant human resources.

• Specialized Knowledge : Requires expertise in both clinical domains and SNOMED CT's structure and updates.

• Evolving Concepts : Understanding the nuances of how concepts have changed or been restructured.

• Interdependencies : Changes in one area may affect multiple expressions across different clinical domains.

• Automation Boundaries : Current tools may not fully automate the update process, especially for complex expressions.

• Reporting Accuracy : Generating accurate reports of outdated expressions can be challenging without sophisticated tooling or querying capabilities.

Close to user form is a simpler form of an expression that is later transformed into a "classifiable form". Close-to-user form expressions include only the essential concepts required to convey the desired meaning, and fewer concepts in the expression means that it is less susceptible to be impacted by changes in the dependencies. The steps of automatic updates and manual reviews are much simpler to implement and execute when using close-to-user form expressions.

Best Practices for Effective Maintenance

• Regular Updates : Schedule maintenance activities aligned with SNOMED CT release cycles.

• Store expressions using close-to-user form : the simpler expressions facilitate maintenance.

• Comprehensive Documentation : Keep detailed records of changes made to expressions for future reference and auditing.

SNOMED CT Concept Model

The SNOMED CT Concept Model is a set of rules that govern the ways in which SNOMED CT concepts are permitted to be modelled using relationships to other concepts. These rules are critical to the consistent modelling of SNOMED CT content, which in turn determines the extent to which reproducible logical inferences can be drawn. These logical inferences are the foundation for effective use of SNOMED CT for retrieval and reuse of clinical information.

For each attribute (relationship type) defined in SNOMED CT, the concept model specifies a set of rules to support the proper application of SNOMED CT attributes, including:

• Attribute Domain : The set of concepts to which the attribute may be applied

• Attribute Range : The set of concepts which may be used as the value for the attribute

• Attribute Cardinality : The minimum and maximum number of times that the attribute may appear in a concept definition

• Attribute Grouping : Whether an attribute may or may not belong to a relationship group

• Attribute In-group cardinality : The minimum and maximum number of times that the attribute may appear in each relationship group

The Machine Readable Concept Model (MRCM) represents rules of the SNOMED CT concept model in a form that can be read by a computer and applied to test that concept definitions and expressions comply with the rules.

The diagram below shows a few example of concept model rules, as they can be derived from the MRCM reference sets.

Given the machine processable nature of SNOMED CT, the Compositional Grammar and the MRCM, postcoordinated expressions may be tested against the rules specified in the MRCM, for example to validate the expression, or to support the creation of MRCM-compliant expressions.

To support the creation and validation of SNOMED CT postcoordinated expressions, the following MRCM process can be used:

1. Determine which domains the expression belongs to using the

   • An expression belongs to a given domain if any focus concept is either:

     1. Valid when tested against the associated domainConstraint ; or

Please note that in the above process only rules with a contentType = << should be used. Rules with a ruleStrength of should be enforced when authoring and cause an error during validation, while rules with a ruleStrength of should be used as a recommendation for authoring and result in a warning during validation.

An alternative approach to authoring and validating postcoordinated expressions is to use the domainTemplateForPostcoordination from the to ensure compliance with the full set of attribute rules. The domainTemplateForPostcoordination can also be specialized into an expression authoring template that meets the needs of a particular use case, while still conforming to the overall rules of the domain.

In addition to the concept model, SNOMED International also rely on editorial principles when authoring the content of SNOMED CT. The SNOMED CT Editorial Guide covers a wide range of topics, including the principles of SNOMED CT, and the structure and organization of the terminology. In addition, it documents guidelines for the modelling of content within specific domains and sub-domains that and it defines the role of the attributes available. The SNOMED CT Editorial Guide plays a critical role in ensuring the quality of SNOMED CT content, as it documents the authoring principles that require human analysis and interactions to ensure that concepts are accurate, consistently modelled, and clinically relevant.

Implementing postcoordination with SNOMED CT is similar to authoring SNOMED CT concepts, as it involves creating clinical meanings by combining existing concepts. However, it is crucial to ensure that the implementation of postcoordination does not impact clinical safety.

It is essential to involve relevant stakeholders such as clinicians, informaticians, and quality assurance personnel to ensure that the approach taken is appropriate and safe for use in clinical practice. Thorough testing and evaluation of the postcoordinated expressions can help identify potential safety concerns early in the implementation process, ultimately leading to safer and more effective healthcare.

Postcoordination with SNOMED CT offers significant benefits for users who need to capture and express clinical meanings accurately. Postcoordination enhances the flexibility and expressivity of SNOMED CT, making it more adaptable to current clinical practice. However, postcoordination also presents a variety of challenges, encompassing both human and technical issues that are crucial to handle as part of an implementation, ensuring that postcoordination is used judiciously and consistently to maximize its benefits while minimizing its potential drawbacks.

Benefits

Two overall benefits can be described for postcoordination:

Postcoordination supports the ability to express or capture new meanings or clinical ideas that are not already in SNOMED CT

SNOMED CT implementers without the means to create and manage a proper extension may require clinical meanings to be represented using SNOMED CT without being dependent on specific precoordinated concepts existing. In such cases, postcoordination may be considered as part of a SNOMED CT implementation.

Postcoordination supports the ability to compose clinical meanings from separate input values and then classify to identify existing precoordinated content

Some SNOMED CT implementations are dependent on a fixed information structure, user requirements may enforce a specific way of entering clinical data, or new clinical meanings are created dynamically (for example within a natural language processing tool). These are cases in which implementations may require clinical meanings to be represented as expressions.

To realize the benefits of postcoordination, careful analysis of the following described challenges needs to be undertaken and solutions need to be implemented.

Tooling and guidance in the area of postcoordination are still in a developmental stage

A key challenge for the implementation of postcoordination is the lack of mature or well-developed tools and practical experience within this area.

Creating expressions, in a way that conforms to all the rules and consistent with existing content in order to classify correctly, is not simple!

As described in and , expressions that represent the desired semantics must follow both terminological and editorial principles of SNOMED CT. In addition, expressions should be consistent with the modeling applied for the released SNOMED CT content. Therefore, the creation of expressions should be performed by people with the required knowledge, or assisted by tools that facilitate the proper creation of expressions.

Expressions are tied to a specific SNOMED CT release and upgrading that release requires maintenance

Maintenance processes need to be established to ensure that the expressions used in the system continue to work properly with the updated version of SNOMED CT.

Interpretation by humans is a challenge

Generated expressions may not use natural language

Task of manually assigning terms that accurately represent the meaning of expressions require training and is cognitively demanding

Interpretation by machines is a challenge

Advanced tooling is required to create, maintain, understand and use expressions

• When expressions are communicated to other systems:

  • Interpretation in the receiving system is not guaranteed

  • The burden of advanced tooling is put upon the receiving system