Stream: terminology
Topic: SNOMEDCT expressions
Bret H (Nov 05 2021 at 16:05):
I am curious about the real-world experience of using SNOMEDCT expressions. This stream was recommended as a good place to ask. Could anyone share if you know of systems/groups that are routinely using expressions, especially in transactions across systems? I'm curious to see how their exchange works and how well/if expressions are being used in real-time. Much appreciated!
Robert McClure (Nov 05 2021 at 18:00):
@Michael Lawley @Rob Hausam @Peter Jordan @Suzy Roy
Rob Hausam (Nov 05 2021 at 19:55):
@Bret H I'll take an initial pass at it and will say that I think you will find that the answer is largely no - that there is not much use (especially routine use) of SNOMED CT expressions for real-time data exchange, especially across systems. My understanding is that University of Nebraska (Jim Campbell and others) does use them, but I think that's internal use in essentially an "expression library" approach (if I'm recalling the details correctly). I'm sure that @Michael Lawley will have some thoughts and likely many more details.
Grahame Grieve (Nov 05 2021 at 21:13):
it's possible to use them in FHIR, transparently - if you're careful to use a terminology server consistently, the a snomed code is just a list of characters, and only the terminology server needs to know anything about it. But terminology servers don't fully support expressions yet (I believe from discussing with Michael, who's furthest ahead on this), so that this possiblity is theory not practice at this point
Lloyd McKenzie (Nov 05 2021 at 21:20):
The other possibility is to pre-define enumerated expressions that convey your concepts and have everyone just use those as if they were stand-alone codes. (Much as may implementations do with post-coordinated UCUM expressions like "cm" and "mL".)
Rob Hausam (Nov 05 2021 at 21:27):
Yes, @Lloyd McKenzie - that's similar to the "expression library" approach. One thing to keep in mind is that determining true equivalence now with SNOMED CT will generally require description logic classification - rather than relying on string matching or simpler "normal form" approaches.
Peter Jordan (Nov 05 2021 at 22:22):
@Rob Hausam makes a key point about the classification requirement. IMO, the distinction between post-coordinated and pre-coordinated concept creation has narrowed to the extent that a common set of tooling is required and the logical progression (in my eyes, anyway :)) is for that tooling to assign SCTIDs to post coordinated concepts and reduce the massive adoption barrier that's created by the use of expressions as codes.
Bret H (Nov 06 2021 at 16:09):
Thanks all. I appreciate your insight and thoughts. It will be interesting to hear what @Michael Lawley has to say.
You all rock! :tada:
Bret H (Nov 06 2021 at 16:18):
@Peter Jordan but does that mean we could have messages where pre or post-coordination is used - requiring local logic when processing incoming transactions to handle both?
Personally, I like the specificity of a post-coordinated model but that's mainly because that means as a developer I feel like I should be able to rely (more or less) on the message to tell me what the code means rather than use a terminology server to calculate the meaning (but a terminology server would be needed for other purposes - no getting away from it : ^ ) - such as understanding the codeded meaning of the data elements). Both approaches have their benefits and drawbacks. What interests me in the SNOMEDCT expression model is the ability to send specific meanings through coding the data elements as well as the value.
Michael Lawley (Nov 06 2021 at 21:32):
Hi Brett, I think the others have covered most points. I would concur with @Rob Hausam s answer regarding exchange of expressions. From what I've seen, the majority of the use of them is either in theoretical exercises, academic/research settings, or closed environments, that is, where the producers and consumers of the expressions are the same group. A common case there is secondary data use where expressions are constructed from free text using NLP techniques.
Michael Lawley (Nov 06 2021 at 21:46):
If you're looking for advice on using them, my first response would be "don't", followed by "what problem are you trying to solve?"
In many many cases, it is far simpler and more robust to deal with multiple codes / composition at the information model level. This is because the reasoning is simpler (classification not needed) and the use-cases are often ones of relatively simple refinement.
Ultimately, the real value of post coordination comes from the subsumption relationships that can be inferred by the classifier. If you're
not looking to get that kind of value then it's likely that "post coordination" is really just being used as a convenient syntax for aggregating multiple codes together in a single field.
Bret H (Nov 12 2021 at 17:26):
Thanks @Michael Lawley et al. Appreciate you taking time out for this brief foray. :+1:
Last updated: Apr 12 2022 at 19:14 UTC