OHDSI Home | Forums | Wiki | Github

Translating non-standard Brand Names into standard ones

Hi there,

Could you help me to figure out with an algorithm of converting drugs to standard concepts on the example:
I have RXNorm concept 19080316 Zyrtec https://athena.ohdsi.org/search-terms/terms/19080316 in the data.
What is the right way to translate it into the standard concept id using vocabulary relationships?

Hello @Alexander,

You should read The Book of OHDSI, specifically Chapters 4-6. It will tell you how to map your data using the Concept and Concept Relationship tables.

On top of what Melanie said, Brand Names are always non-standard because they’re drug attributes rather than independent entities. If you want all standard Drug under that brand you may use ‘Brand name of’ relationship (as you can see in Athena)

I checked the book, but I didn’t find how to convert non-standard Drug Brand Names into Standard concepts. They don’t have relationship “maps_to”.

We are parsing medical notes, and sometimes Brand Name is the only information we have there. What is the right way to translate them into standard concepts to keep in the OMOP CDM? In my example there are many other concepts with relationship “Brand name of”

My hypothesis that we should convert it to https://athena.ohdsi.org/search-terms/terms/1149196 to keep in the OMOP CDM, but what is an algorithm for doing it automatically for other brand names using vocabularies?

Yes, you’ll use Brand name of and limit concept_class_id of concept_id_2 to ingredient

1 Like

Thanks, it’s helps me a lot. I had another approach: take “Has precise ingredient” concepts and from them take “Form of” relationship. But it gives the same results.

I learn something new every day :slight_smile: I thought all RxNorm codes/concept_ids would map to a standard.

@aostropolets why don’t they have a ‘maps to’ relationship? I thought (and can definitely be wrong on this) all RXCUIs would map/link to an ingredient

Unfortunately, even inside RxNorm itself brand name may have more than one ingredient combination assigned to it. If you include RxNorm Extension which includes international market drugs, drugs under the same brand name may have different contents in different countries, like different combinations of NSAIDs or OTC cough medicines.

There is a code snippet that we use to find “stable” brand names to ingredient contents relations inside existing CONCEPT and CONCEPT_RELATIONSHIP tables. But it’s sort of last resort and as far as I know it was never tested for ETL. I would only use it if you have high record counts and no way to obtain additional info.

If you are interested, check lines 91 through 222

1 Like

Only those that can be used for drug_concept_id in DRUG_EXPOSURE and DRUG_ERA, and hence are

  • Valid
  • Ingredients, Clinical/Branded Drug Forms/Components, Clinical/Branded Drugs and Clinical/Branded Packs. We added Quantified Clinical/Branded Drugs, which the NLM have de-facto but not explicitly as a TTY. And we have the equivalent Groups, even though I don’t like them.

Brand Names, Dose Forms, Precise Ingredients are not. Multiple Ingredients (MIN) we haven’t even incorporated (to some people’s chagrin).

I’ve linked FDB’s NDDF+ to the UMLS & RxNorm in the past to compensate for limitations in the RxNorm vocabulary.

Which ones, @docBrian?

Thanks Eduard, I’ll check that code.

Sounds like a medication error waiting to happen. Thank you for the explanation, @Eduard_Korchmar!

Primary focus was building an ontology to support drug-related clinical decision support. This was for an EMR that included CUI as well as vocabulary-specific encoding.

For example, the NDDF has drug/disease indications and contraindications, drug/drug contraindications, drug/adverse event maps (e.g., monitor for fever or thrombocytopenia or elevated liver enzymes, etc.), routes of administration, etc. The ontology we built linked drug concepts to the symptom/sign/disease/lab concepts. None of these relationships were in the UMLS (MRREL, etc.) at the time we did the work.

We also did a fair amount of ingredient mapping for the ontology, so that we could easily identify all ACE inhibitors or all ARBs or all cephalosporins etc. for hypothesis testing prior to design of a pragmatic trial using EMR data.

Were you satisfied with the quality of indication relationships? What I saw before is that NDFRT goes way to broad linking too many indications to an ingredient, which is not suitable for the type of research questions we have (e.g. identify drugs that are specific to few conditions). I’d love to hear about your experience (or read about it somewhere)! Drug classes are covered by ATC and they attempt to cover some indications but I’m really looking for a disease-drug links.