OHDSI Home | Forums | Wiki | Github

Making Drug Exposure Concept Sets When Drugs are Frequently Coded at the Ingredient Level

Hi There,

I am an engineer supporting the National Covid Cohort Collaborative (N3C) and we are trying to provide some documentation on building medication concept sets for OMOP data. We are particularly concerned with the case when drugs are frequently coded at the ingredient level, as happens in N3C. I was wondering if I could get feedback on the attached decision tree for medication concept sets.

Thank you very much!
Andrew Girvin

Hi @Andrew_Girvin,

Here are couple of my comments:

And then you say to use Clinical drug component.
We also can use Ingredient but additionaly look at the Quantity field. When there is an Ingredient in DRUG_EXPOSURE.drug_concept_id, then the DRUG_EXPOSURE.Quantity stands for a total amount of an Ingredient given.


we can also use Ingredients, if we specify the route.

And for cases when dosage and route needed and the data is on Ingredient level, you can use ingredients with dosage in Quantity and route specified.

Apart from these comments, I would also think about another parameter - daily dosage.


Not sure what problem you are trying to solve. The data should have as detailed concepts as possible. But sometimes the detail is lacking, and in the extreme case you just get ingredients (in more extreme cases you get only drug classes, such as “chemotherapy”). If you have the detail and you are asking for Clinical Drug Forms (e.g. acetaminophen oral tablet) you will get all such tablets no matter what the strength. Likewise, if you are looking for strength you should as for a Clinical Drug Component and it’s descendants. If you don’t have the detail in the data not much you can do. The Dose Form you can somewhat substitute by the Route of Administration, but that is not a one to one.

Let us know.

Posts crossed.

@Christian_Reich, @Dymshyts, thanks for your comments here. In our data, taken from many OMOP sites, we frequently see missing quantity, dose form, and route, and the drugs are frequently coded at the ingredient level. So we are trying to develop best practices for that scenario.

If you only have ingredient you put the ingredient concept in. That’s the least you need for a DRUG_EXPOSURE record.

It seems that Andrew doesn’t want to use ingredients where ingredients can be used for completely different indications (like corticosteroids for topical and oral use) and exclude patients with such ingredients altogether. That’s a fun task on it’s own albeit hard. I’d think something along you decision tree with the main fork being indication. Doses aren’t usually differential. If trying to spend less time on decisions, something like ATC/NDFRT can be used. If there ATC concepts for the same ingredient with different routes (like here) then route/indication matters. Good thing that 80% of the cases using ingredient will be totally fine.