OHDSI Home | Forums | Wiki | Github

SNOMED to ICD-10-CM: Context-dependent mappings

(Emily) #1

Question: How does the OHDSI vocab team handle context-specific mappings between SNOMED and ICD-10-CM?

Background info: The mappings published by the NLM between those two vocabs include lots of “context-dependent mappings,” like SNOMED code A maps to ICD-10-CM code B, but only if the patient is female and has conditions C, D, or E. In those cases, you can’t really map it without knowing the specific use case.

Even if the “context-dependent” label is stored as a Class in the concept table, how does that shake out in Atlas or Athena, unless you’re paying really close attention? For example–do these context-specific mappings end up buried deep in lists of descendants, where they may (or may not) be inappropriate based on context?

I feel like a user who is really attuned to vocab issues might be able to sort this stuff out, but I’m thinking about the users who just click the “Descendant” box and go on their merry way. Although, perhaps trying to fix user behavior is not a concern for the vocab team. :slight_smile:

Really appreciate any insight on this topic!

(Kristin Kostka, MPH) #2

Amplifying this post. @Dymshyts @Christian_Reich @aostropolets, I’m curious to know the answer to this as I’ve seen the Class but am not sure how to explain how this all shakes out. I just assumed the Vocabularies solved all my problems. :wink:

(Dmytry Dymshyts) #3

Hi @epfaff
We build the mappings from ICD10CM to SNOMED (opposite to UMLS).
In most of cases, we have equivalent mappings, but if not, additional context is represented by multiple mappings, for example
here you can see the mapping to two concepts via Maps to (Non-standard to Standard map (OMOP)) relationship:
together they make a semantic equivalent of the source “Drug or chemical induced diabetes mellitus with hypoglycemia with coma” concept.

Or we can use additional Maps to Value (Non-standard to value_as_concept map (OMOP)) relationship,
which is the attribute of a ‘Maps to’ related concept.
Maps to value indicates exact procedure the subject has a history of.

Please let me know if it expains enough.

Hope they really do:)

(Kristin Kostka, MPH) #4

This is super helpful @Dymshyts!

I think @epfaff’s question is upstream. So the mappings exist and preserve this detail… but if they’re context dependent, how does one know which map to use? E.g. Wouldn’t the context-dependency be kind of a priori to the ETL specification? (Maybe I am missing somthing)

(Dmytry Dymshyts) #5

We just don’t have context dependant mapping as UMLS has,
one of the reason we can avoid it, is the opposite direction of mapping.
So if a concept is mapped to something, it’s always like that.