OHDSI Home | Forums | Wiki | Github

How to pick a standard concept when a non standard concept maps to multiple standard concepts?

Implementers
#1

It looks like the concept with concept_id = 44830486 is a non-standard concept that maps to several standard concepts that seem to be somewhat not the same:

https://athena.ohdsi.org/search-terms/terms/44830486

image
image1021×201 6.92 KB

It looks like this list can not be narrowed down to a single standard concept using the concept_relationship valid_start_date and valid_end_date, the concept valid_start_date and valid_end_date, domain_id, and invalid_reason.

How do I pick which standard concept to use?

Here is a query that provides some useful information:

select 
  rel.concept_id_1,
  rel.concept_id_2,
  rel.valid_start_date,
  rel.valid_end_date,
  con.*
from 
  omop_concepts.concept_relationship rel
  join omop_concepts.concept con on rel.concept_id_2 = con.concept_id
where 1=1
  and rel.concept_id_1 = 44830486
  and rel.relationship_id = 'Maps to'
order by 
  rel.valid_start_date,
  con.valid_start_date
;

image
image1694×218 12.7 KB

image
image1244×193 8.32 KB

#2

Are you talking ETL of data to OMOP, @greshje.gmail john? In that case you pick all 4 and write 4 records for each one record coming in.

Or what is your use case?

#3

Thanks for the quick reply @Christian_Reich ! Yes, we are doing ETL, but writing multiple records for each instance where there is a mapping to multiple concepts is not an option as one of the metrics we are most interested in is how often each DX occurs in our data set.

#4

Understand, but you may have a very technical sense of what constitutes a diagnosis. Each diagnosis has a varying set of attributes, which go from generic (prostate cancer) to very detailed (stage 4 metastatic castration-resistant prostate cancer). Same disease. So, you can easily split things up. The total number of diagnoses is not a fixed thing.

The reason codes get split can be one of two:

  • There is no good equivalent standard concept in SNOMED (which we use for diagnoses),
  • The ICD code actually contains two things (viral hepatitis with hepatic coma) and should be split up.

So, in sum: Have no fear, split them up and write the records. It’s the OMOP law. The denominator of all diagnoses is still good (and doesn’t change much from the little bit of splitting).

#5

Yes, looking at it more carefully, a patient with the non-standard code did have all four of the conditions for the mapped to concepts.

Thanks for the quick responses and the help!

t