OHDSI Home | Forums | Wiki | Github

NOTE_NLP table question


(Yoonhee Park) #1

There are some questions about the Note_NLP table field.

  1. section_concept_id : what is the “section of the extracted term”?
  2. note_nlp_source_concept_id : what is the “code in the source vocabulary used by the NLP system”?

(Christian Reich) #2

@Yoonhee_Park:

  1. I think we have yet to put that in. I just realize we don’t have it. Nobody has pointed it out so far. We do have documentation concepts for NOTE.note_class_concept_id, but nothing here. Thanks for pointing it out.
  2. If the NLP doesn’t use standard concepts (SNOMED for Conditions for example) but some other codes (like ICD10 or so), this is where it would go. Hopefully we won’t need this too often. Use the standards. But since you need to do it in Korean the standard SNOMED and RxNorm may not have Korean lexical terms, so you may have to go to some other sources.

Makes sense?


(Yoonhee Park) #3

Thanks for your reply.
Your answer is helpful.


(George Hripcsak) #4

Oops. I see that my response from 11/8 never made it. Here is what I thought:

Good to clear this up. For section concept id, I see that there are type concepts that have a concept class id of Record Artifact, and they include things like history of present illness section and physical exam section. But the 2016 note_nlp document I have says you should use LOINC codes for the section, which also includes history of present illness. So I assume the LOINC code is the correct one. Plus the LOINC codes are standard but the Record Artifacts are not.

And for note_nlp_source_concept_id, put in whatever the NLP system outputs. E.g., it might put out a UMLS code and that would go in that field. Then map that to SNOMED (most likely) to put into the note_nlp_concept_ID. If the system puts out SNOMED codes, then they go in both fields.


(Evan Sholle) #5

To follow up regarding note_nlp_source_concept_id with a concrete example for @Yoonhee_Park or anyone else looking at this thread in future: we have an NLP pipeline that runs against surgical pathology reports and extracts several things, including TNM staging and the ICD-9/ICD-10 codes the pathologist entered. We map the TNM staging concepts manually to SNOMED, but for the ICD-9/ICD-10 codes, we insert the corresponding concept_ids as the note_nlp_source_concept_ids, then map to SNOMED using concept_relationship and insert the SNOMED code for the condition as the note_nlp_concept_id.


t