Description Logic Enhancements - Community of Practice Consultation
SNOMED International is proposing enhancements to SNOMED CT's logic features, to enable improvements to the quality and analytics capabilities of SNOMED CT. These proposed logic enhancements will allow clinical concepts to be defined more accurately using an extended set of Description Logic features. The enhanced clinical concept definitions will be expressed using the Web Ontology Language (OWL), and distributed using a standard RF2 reference set format. While consumption of the new OWL file will be optional for most end users1, all users can benefit from the new features as a result of improvements in the content of the automatically-generated (inferred) Relationships file.
While this enhancement will initially target improvements in the Pharmaceutical / biologic product and Substance hierarchies, it also opens the door to modeling improvements in other hierarchies - notably Body Structure and Observable entity.
Since its inception, SNOMED International has sought to improve the quality and usefulness of SNOMED CT, while at the same time reduce the maintenance burden of an ever increasing body of content. A primary way of achieving these aims is to create content that is sufficiently well defined to allow a logic reasoner (also called a classifier) to compute the appropriate hierarchical relationships. This reduces the likelihood of human error and allows for an automatic construction of hierarchies as concepts are added, inactivated or modified, using the logical definition of the concepts, rather than relying on a fixed structure chosen and maintained manually by content authors.
During the classification process, relationships stated by authors are converted into OWL for the classifier to reason over. Currently these stated relationships used by the classifier are represented in the RF2 Stated Relationship file. However, there are inherent restrictions in the structure of this file which limit the information that can be made available to the classifier. SNOMED International proposes to enhance this input to the classifier by specifying the meaning of clinical concepts directly in OWL. The distributed output from the classifier (used by the majority of implementers) will continue to be represented in the standard RF2 Relationships file format, as a list of inferred relationships. While the structure of the RF2 Relationships file will remain unchanged, its content will benefit from the improved classification results made possible by the enhanced input file.
The cost of not making the recommended changes would be to severely curtail or even block planned advancement in concept modelling that has, to date, been severely limited to a number of workarounds.
SNOMED International proposes to replace the existing Stated Relationship file with an RF2 reference set containing OWL statements that represent the meaning of clinical concepts. The Stated Relationship file is not normally used by consumers of SNOMED CT, unless they have a use case that requires classification. The format of the (inferred) Relationship file - which is the one normally used by consumers of SNOMED CT - is not expected to change. It is therefore anticipated that this enhancement will not require any action to be taken by the majority of SNOMED CT users. However, organisations should carefully review the checklists in the briefing papers listed below to determine if any action is required on their part. Organisations who wish to classify Stated Relationships may choose to obtain and use the open source classifier supplied by SNOMED International2, which works with the new RF2 OWL reference set file.
The enhanced information that will be captured in the new OWL reference sets will include:
- Further details about the logical behaviour of specific relationship types. For example, the only relationship type that is currently treated as transitive is the |is a| relationship. This means that if we were to state that |Golden Delicious| |is a| |Apple| and |Apple| |is a| |Fruit|, then is can be inferred that a |Golden Delicious| |is a| |Fruit|. It is desirable to also be able to specify this transitive behaviour for other types of relationships; for example |part of|. So if |Finger| is |part of| |Hand| and |Hand| is |part of| an |Upper limb|, then we want the classifier to be able to infer that a |Finger| is |part of| an |Upper Limb|. Other types of relationship behaviour that will be possible to specify include Reflexive Properties and Role Chains. These logical features, called "Property Axioms", are explained further in the briefing papers attached.
- Additional statements about the meaning of clinical concepts. This gives authors the opportunity to state, for example, multiple true statements about concepts and also to indicate hierarchical relationships between concepts which meet specified criteria. The enhancements also make it possible to state sufficient but not necessary conditions using a logic feature known as a "General Concept Inclusion". This provides the ability to state alternative definitions for concepts.
Proposed Schedule for Change
The table below shows the features that will be introduced over the next few releases of the SNOMED CT International Edition. This is an optimistic timeline that has been driven by the desire to fulfil obligations for successful delivery of work being done in the Pharmaceutical / biologic product and Substance hierarchies.
Stated Relationship File
Additional Logic Features in International Release
Production OWL Refset
Technical preview OWL refsets
Snapshot of stated relationships plus:
Expected publication prior to March 2018.
Will not include additional logic features.
Insufficient on its own for correct classification.
First official release of OWL refset.
Full, Snapshot and Delta of all content.
Support all features in SNOMED CT logic profile.
Content changes authored as normal.
SNOMED International request that users of SNOMED CT review the briefing paper listed below and provide feedback via this form. SNOMED International will begin to respond publicly to feedback received by 11 December 2017 on this page on our confluence site and continue until the end of this consultation exercise on 28 February 2018.
1. Except for use cases where classification is required.
2 Currently completing testing. Link will be supplied here when available.