OHDSI Home | Forums | Wiki | Github

Concept ID hierarchy

Vocabulary Users
#1

Hi all,
I am Subhajit and very new to this forum. I’m also not sure if this is the right sub-group I should post my question. I am really looking forward to learn from you all. I’m a primarily interested in person level predictive model development.

Ok, here is my question. I am trying to consolidate the conditions (condition_concept_id) into larger categories which would be easy to use for our work. I understand that the hierarchy in the concept_id is taken from SNOMED CT hierarchy (or relationship model).
So one way to do this by using the SNOMED hierarchy tree we could count. But I’m wondering if there exists such a tree based on the concept_id for the OMOP or any code that can achieve this grouping directly from the OMOP tables.

I appreciate your help. Please let me know if anything that might help me figuring out this consolidation process.

Thanks much,
–Subhajit

#2

Sounds like perfectly good to me. Plus, not a big deal if you didn’t pick it.

Yes, that is a good one. Because you don’t know how high to go when you roll up. If you go too high you end up with the top dog “Clinical Finding”, which is correct but useless. We need to figure out some cross-section through the hierarchy with a good level of summary. Any good idea?

#3

I believe @ericaVoss has developed a solution to look up the hierarchy and settle in on a certain level based on numbers of levels above and numbers below.

But, she could share the algorithm/query and describe it.

#4

Thank you Christian.
Yes, we definitely don’t want to end up with “Clinical Finding” :slight_smile:
Right, finding the correct level of that hierarchy tree is a key thing.

One I was thinking could be helpful for us something like (broader categories)
https://icd.who.int/browse10/Help/Get/hierarchy/en

Another could be little more granular. If I can first have a target set of categories, then using a tree search algorithm I could roll the descendants up to their parents.

#5

Thank you for the information Chris.

t