We are currently mapping clinician specified disease terminology to SNOMED-CT. A frequent issue is the request for disease-specification with values which are difficult or impossible to integrate with postcoordination. A major problem is that quite often qualifier values exist for a requested term (mostly in the Descriptor/Time patterns hierarchy) which semantically match a requested specification, however none of these is valid as a Range in any domain according to the MRCM.
Could somebody elaborate, why the qualifier value concepts exist at all, if they can’t be used as focus concepts or as attribute values? Wouldn’t it be more helpful to deactivate such concepts until they might be integrated into the concept model at a later time to reduce load on authors/implementers?
Hi Gorjan, at least some of those concepts (for example <307141008 |Time descriptors of days, weeks and years (qualifier value)|) exist not as targets for modelling but for use as concepts themselves.
Instead they are used for implementation in an information model. For example, 307145004 |Monday (qualifier value)| could be used to code the day in a prescription where the dose is ‘take 2 tablets once daily on Mondays’.
The MRCM specifies what can be modelled when authoring a concept, and can also be used to guide (but only guide) sensible post-coordination. Though the MRCM cannot prevent all unsuitable post-coordination.
If the creation of a concept is not permitted by the SNOMED Editorial guide, then the MRCM is unlikely to permit it either.
This is not the same as what a sensible post-coordination might be.
Good answer Stuart. We are often asked the question about the plethora of qualifiers that cannot be used as values within the MRCM. These often do act as values within information model structures, but are not available as values for property values in a concept model.
Gorjan, if you could give us some examples of where you are having issues, it might help us to identify some alternatives to represent the clinical concepts you are trying to support.
On the other part of the question though, I think there are some concepts we probably don’t need in SNOMED CT.
For example, the small hierarchy of random looking years that got into SNOMED CT via the UK’s Clinical Terms V3 (CTV3) under 277267003 |Calendar year (qualifier value)|.
Those years are versions of UICC / AJCC TNM Cancer staging that supported CTV3 templates.
I’ll raise a CRS to make them Inactive as no longer needed.
Thank you for your elaboration. Since we currently only work with clinical disease concepts postcoordination, I have not yet had any experience with the use of information models and SNOMED. Knowing this use-case however facilitates my discussions with colleagues. @jcase Most concrete problems are solvable, the issue is mostly related to differing training-depth.
Examples would be the use of e.g. 272118002 Acuteness instead of 385315009 Sudden onset or 1255665007 Moderate instead of 6736007 Moderate severity for severities grading of a disease. The second example can also be confusing, because 255604002 Mild does not specify “severity” in its FSN or PS, while moderate and severe do.
Other times there would be qualifier values available which would semantically fit, e.g. infectiologists often request symptomatic/asymptomatic for various infectious diseases (CMV, Gonorrhea etc.), however 67335000 Asymptomatic and 255297006 Symptomatic, are not allowed by the MRCM. I have some potential alternatives in mind for this, like using severity instead, however mild and moderate might have too much drift from the actual meaning. Or maybe something with “Has interpretation” and “Finding value”. If you have come across this constellation before, any input would be appreciated!
Of the 12,191 (qualifiers) that are still active concepts in today’s International SNOMED, 6540 of them were already present as far back as 2002. 5425 (ie 82% of them at the time) had been inherited into SNOMED from the UK’s Clinical Terms Version 3 terminology (CTV3).
Although some of CTV3’s “qualifiers” might have been always primarily conceived of as values to be placed on their own into slots in an information model … I think it would be reinventing the truth to claim that most of them were. The information models of that era really weren’t that sophisticated!
Rather, the huge majority of (qualifiers) were in fact always intended to be used in crudely postcoordinated expressions built entirely within the CTV3 terminology, but within the constraints of a “template” file that was originally part of CTV3 but whose semantics lived on in SNOMED, at least for a few years, as the optional qualifier flavour of published sct1_relationships and the “refinability” flag on modelled (inferred) relationships.
The Optional Qualifier rows were poorly maintained until they were ultimately entirely expunged from SNOMED as a deliberate part of the Release Format 2 (RF2) design from around 2014, after which most (qualifiers) immediately became without any technically or semantically clear purpose in SNOMED, barring the very small number that became legitimate targets of the MRCM when that was invented in 2009.
Thank you Jeremy, this pretty much clarifies everything. I guess the only question remaining would be why those concepts haven’t at least been deactivated? Is the main problem that qualifier values originally meant for RF1 postcoordination can’t be reliably distinguished from values still used today in information models? Or is it more of a political issue?
The reason that these qualifier values have not been inactivated is related to member priorities for content development and resource limitations. There is quite a bit of content in the qualifier value hierarchy that is still useful, and evaluating which are or are not of value is a tedious and time-consuming task. We have have been inactivating some qualifiers as identified by members, but there has not been a call to do a comprehensive review.
Whilst the ‘CTV3 template’ origin story explains much of the qualifier value content (and also accounts for the majority of the 1000+ codes beneath 408739003 |Unapproved attribute|- the ‘A’s for each template’s OAV triples), many qualifier value concepts have been added since.
Indeed, of the ~6000 non-MRCM referenced qualifier values in the international release, over 500 have been created since 2020 - and therefore were presumably added with some internationally intended model of use.
A lot of these are types of 224891009 |Healthcare services| (the value of which has been debated elsewhere) and a small number are replacements/reworkings of older qualifier values, but there are many other novel codes representing different ideas (e.g. grading and staging values).
I am particularly interested in the dozen or so concepts beneath 3431000181105 |Gene product function descriptors| and 3441000181103 |Gene product metabolic activity descriptors|) - promoted to the international data in 2022. These hint at providing an alternative (and non-comparable) way of recording functional and activity assertions supported by findings beneath 106221001 |Genetic finding|. It would be good to understand how these recently added qualifier value codes are intended to be used.