OHDSI community needs a reliable and usable drug classification system. Currently ATC is used for this purpose, which was largely improved with tremendous effort of the OHDSI vocabulary team, kudos to @Polina_Talapova and @aostropolets. Despite this effort, ATC classification is far from perfect,
- with terms such as ‘J01GB | Other aminoglycosides’ when you can’t know what those “other” mean
- duplicate or ambigous terms
see, S03C has only one child S03CA, so they have the same number of descendants,
but S02CA and S01CA have different set of descendants (as there are different DRC counts) - it’s monoaxial, so you can’t classify the drug by it’s chemical structure, mechanism of action or indiction at the same time
- original intent of ATC is to classify clinical drugs, not individual ingredients, (it might imply the dosage, route or combination with the other ingredients, while our most often need is to classify ingredients.
The drug classification is needed in 2 cases mostly:
- building features
- building phenotypes
Currently ATC concepts as features are rather problematic, because ATC that have information additional are linked to ingredients, for example
in this case all 4 concepts are linked to the same ingredients. And we built a wrong feature like if the patient was on 3 different forms of sulfamides at the same time.
To solve this, we are going to exclude specific ATC concepts from the feature extractor. Thus we’ll end up only with drugEraGroup: Sulfonamides.
To improve the Phenotype creation process we need to hear your needs, do you mostly need the indication or chemical class of drug or mechanism of action or something else?
OR just share your use cases for drug groups usage.
Attaching the multiaxial classification of antibiotics for the reference
antibiotics_classification.xlsx (12.1 KB)
When better knowing the needs we might either
- continue improving ATC
- use another ontologies instead or together with ATC, these ontologies might be specified on different ways of classification. DrugBank and SNOMED are the good candidates.
tagging @Patrick_Ryan as we discussed this worked this through together