OHDSI Home | Forums | Wiki | Github

concept_id maps to more than one concept_ids on the concept_relationship table of the OMOP CMD standardized vocabulary

Hello!

I am writing to ask why do some concept_ids on the concept_relationship table map to more than one valid value. Please find an example in the screenshot below.

I thought each concept_id maps to exactly one standard and valid concept_id. Please correct me if I am wrong. For the record, last time I updated my concept_relationship table from Athena was the 2nd of July 2018.

I would deeply appreciate your assistance!

1 Like

You could have 1 non-standard map to multiple standard concepts, that is fair game. This is usually found in cases where one code is associated to multiple conditions, so the concept_relationship will map the single code to each specific condition that is packaged in it.

I think the rule you are thinking is that there’s exactly one standard concept for a ‘medical idea’, meaning you won’t find multiple concepts for Still’s disease.

In the case you brought up, it looks like the ICD10 code M06.14 is mapping over to 2 concepts, one for Still’s Disease, the other for RA of hand joint. From what I read, Still’s shares many characteristics with RA so perhaps that is why there is the dual mapping. But, I checked the vocab and it doesn’t look like Still’s rolls up to anything under RA, so the mapping isn’t redundant.

I can’t give you the actual reason why this particular concept got mapped to two conditions. Most likely the vocabulary team got that mapping from some source, or you did find a mapping error. But, to your original question, it is valid to have a non-standard concept map to multiple standard concepts.

1 Like

In ontology, it is possible to desribe info like that if the relation between two concepts is one-to-many relationship.

Hello!
Unfortunately, you have found real bug.

As you see, concept ‘42616600 M06.14 Adult-onset Still disease, hand’ from ICD10 is non-standard. And it simultaneously MAPS TO standard concepts: ‘4344166 239920006 Adult onset Still’s disease’ and ‘4115161 287007001 Rheumatoid arthritis - hand joint’. And those are the main conclusions from vocabulary team:

firstly, this case contradicts our rules of data representation. Concepts can have more than one ‘Maps to’ relationship_id only when they are pre-coordinated concepts listing two or more semantic components through AND or OR (if there is no an equivalent combination concept that is also hierarchically well connected).
Evidently, your case is not about it. “Read more” link: http://www.ohdsi.org/web/wiki/doku.php?id=documentation:vocabulary:mapping

Secondly, mapping to the second SNOMED concept in this list (4115161) is incorrect, because Still’s disease and rheumatoid arthritis are different diseases from the big category of autoimmune pathology.

Review and bug fix of ‘ICD10’ vocabulary is already in our backlog. We apologize for inconvenience caused.

1 Like

@Chris_Knoll thank you for your reply. I see that it is valid for a non-standard concept to map to multiple standard concepts. However, how do I treat these double mappings in my ETL? 0.3% of my conditions (nearly 1000) map to two concept_ids. Is there an efficient way to map each of them to the most correct concept_id? Do I have to go over these conditions one by one and pick the most correct concept_id?

You’ll insert 2 records in your target CDM table, one for each mapping.

You don’t need to pick, you map them both. The topic of this thread was pointing out an error in a map (where Still’s disease ICD10 was mapping to 2 standard concepts: Still’s disease and RA of joint). That wasn’t a correct mapping, but there are legitimate cases where a source code can map to multiple concepts (which could possibly be to different domains, ie drug_exposure and procedure_occurrence). For example, it might be reasonable if there’s a specific source code for “chemotherapy using Paclitaxel”, you might get a procedure record for ‘chemotherapy’ and a drug exposure of ‘Paclitaxel’.

1000 sounds like a lot, but perhaps you could go to your source system, group by…count() those concepts and let us know the top 10 conditions (by occurrence count) in your data which map to multiple concepts? Perhaps you’ve uncovered additional errors in the vocabulary mapping.

I see! It is all clear now. Thank you for the example

Here are the first few concept_ids that map to more than one concept_id in the concept_relationship table.

If you could (next time) add the concept name so that we can see quickly what the concept in question is, and also it would be helpful to understand the frequency of these multi-mapping concepts that appear in your data (ie: it wouldn’t make sense to tackle a concept with 3 mapping_destination_counts when it appears once in your data).

Just for kicks, the first concept in there with the multi map is:
Closed fracture of base of skull with subarachnoid, subdural, and extradural hemorrhage, with prolonged [more than 24 hours] loss of consciousness, without return to pre-existing conscious level

Fun! This does look like a combo of things happening together (facture, hemorrhage, loss of consciousness) so this is an interesting case. It’s very specific tho, I wonder how many times this appears in data.

1 Like
t