OHDSI Home | Forums | Wiki | Github

How are OMOP concept names selected from the SNOMED source vocabulary?

Vocabulary Users
#1

I’m in the middle of EHDEN SME Training course with @Rijnbeek and @MaximMoinat and an interesting question came up that I don’t know the answer and couldn’t find the documentation to support:

SNOMED-CT provides multiple descriptions for each concept. One of those descriptions may be a ‘fully specified name’, one or more of those descriptions would have acceptability as ‘preferred’, and the other descriptions would have acceptability as ‘synonyms’. How do we decide which description to put in the OMOP CDM CONCEPT.CONCEPT_NAME, and which descriptions go into the CONCEPT_SYNONYM.CONCEPT_SYNONYM_NAME? In the one example I looked up from this document (https://www.nictiz.nl/wp-content/uploads/2014/06/SNOMED-CT-starters-guide.pdf), SNOMED code 22298006, it appears the OMOP conceptname = ‘Myocardial infarction’ is ‘preferred’ but not ‘fully specified name’, and the ‘fully specified name’ and the ‘synonyms’ all were entered into CONCEPT_SYNONYM.

Vocab team: @Christian_Reich @Dymshyts @aostropolets, can you please advise the convention used to determine which SNOMED descriptions get used where?

ATLAS team @Frank, I know we aren’t using CONCEPT_SYNONYM in our vocab search in ATLAS 2.7.3, but can you discuss plans for using it in future releases?

ATHENA team @pavgra, I don’t know how the ATHENA lookup is using CONCEPT_SYNONYM in search (I can see its displayed once you select a given concept), can you clarify? I know we want to align the vocab search experience between ATHENA and ATLAS, but I am aware that we currently have two different resultsets if we search for the same string so want to make sure we understand why those differences exist.

USAGI team @MaximMoinat, could you reply to this thread to summarize how synonyms are used when mapping source values?

Thanks all for your input.

#2

Hello!

SNOMED CT concepts are different in their nature from OMOP concepts. They have multiple names (descriptions) and rules that recommend in which situation which description is valid to use.

According to SNOMED editorial guide, preferred term is description that is deemed to be the most clinically appropriate way of expressing a concept in a clinical record. So it is the closest match we have to CONCEPT_NAME by OMOP CDM rules.

The most simple answer would be that the preferred term PT is used as concept_name. However, SNOMED version in OMOP currently is a mash-up of international version and UK-specific edition. Since local editions may have their own rules to determine which term is preferred for their cases, PT may differ between editions. In case of UK it is obvious with use of greek-derived diphtongs (esophagus vs. oesophagus) and occasionally locally used eponyms. Most of preferred terms between International edition and UK of course, still do match, and in cases where they don’t we simply pick one assigned most recently.

All other active descriptions, both simple synonyms and fully specified names, from all versions are put indiscriminately in CONCEPT_SYNONYM table.

Considering that preferred terms may differ between versions and editions, conventions for naming synonyms in SNOMED are constantly changed, huge number of descriptions are written following outdated guidelines and technical impossibility to differentiate between kinds of descriptions inside CONCEPT_SYNONYM table, it is necessary to always include CONCEPT_SYNONYM table in querries when searching for SNOMED concept by text.

I am not sure how Athena and Atlas browsers handle this at the moment.

#3

@Patrick_Ryan:

Well, this is a somewhat complex order of preference based on the input tables:

  • Newest in sct2_concept,
  • Followed by sct2_desc,
  • Followed by synonym

The description with class in parenthesis is suppressed.

If you don’t like the description/synonym assignment in a concept you are seeing let us know.

#4

Thanks @Christian_Reich, for me, it isn’t about preference for assignment, its just clarity in the procedure that we are performing. Others that are more expert than I in SNOMED-CT should weigh in if they believe there is an alternative approach that we can consider. Where can I find the documentation that describes how we transformed the SNOMED source files into OMOP vocabulary structure?

t