OHDSI Home | Forums | Wiki | Github

Drug concepts with lower granularity (medicine groups)

Vocabulary Users
#1

Hi OHDSI community,

We are a Danish research group currently working on vocabulary mapping of several major Danish Health Data databases to the OMOP CDM, and have a question concerning drug and therapy mappings.

We are mapping drugs directly to RxNorm, wherever possible, and thankfully, the majority of the database can follow OHDSI mapping guidelines nicely. However, our database also has concepts that are higher up in the hierarchy, e.g. not a specific drug ingredient, but a group of ingredients. While this is a problem in itself for the medical interpretation, I am wondering how this will affect analysis, e.g. in prediction models.

Example: In treatments with biological medications, a patient can get the drug ‘Infliximab’, but it can also be registered as ‘Biological therapy’. A biological therapy can be found only as a procedure code. I am wondering how this will affect analysis at a later stage, as procedure codes obviously will not show up in the drug era table.

The problem is similar for concepts like ‘TNF-alpha inhibitors’, ‘Hormone treatment’, ‘Steroid treatment’, ‘Antacids’, ‘Treatment with VEG-F antibody’… (the list goes on) that are only available as procedure codes (or classification, which we obviously cannot map to).

I was wondering if anyone has thoughts on this, how to remedy this in mapping the vocabulary or analysis, or whether there are any plans to create higher-up concepts that can be used in drug studies.

I looked through the forums, and I don’t think this question has been posted before, but if I overlooked it, my sincere apologies and please send me a link to the thread.

Best regards,

Viviane

@cssdenmark

#2

As for now, there’s no good option. The only work-around is certain procedural codes.

And we’re thinking of a couple of solutions. One of them is to allow to map to Classifications (ATC) and use them as event_concept_id.

But “biological therapy” won’t be useful anyway - too unspecific. Unless you have a good way to map to a drug group, it’s not helpful in studies.

#3

Thank you!

Yes, we agree it is very unspecific. On the other hand we also think leaving this information out will be a problem when doing analysis. If we leave it out, it looks as if patients did not receive any type of biological therapy at all, that way they could potentially end up in a ‘control group’.

Thank you for the link to the previous discussion. I can see there are others that are wishing for a classification-type concept to become standard :slight_smile:

How would using a mix of procedure codes and drug codes affect analysis? Would the result be equal when doing prediction models, or does the difference in domain and the use of different tables have an effect? Or should I be asking this question somewhere else in the OHDSI forum?

Best regards,

Viviane

@cssdenmark

#4

Isn’t this where concept set expressions can help you? You can group all these unspecific drug codes together, and create cohort eras out of them using the standard tools?

#5

Hi Chris,

Thanks for your input!
So I read that it was generally not advised to mix concept domains within a concept set, but can be done. So that is definitely a good idea. As we are still in the preparatory (mapping) phase, we were trying to think ahead how this would affect not only the practical running of an analysis, but whether getting drug information from different domains would influence results. But I am guesing that due to the unspecific low granular nature of some of these concepts, we would mostly be using it to exclude certain patients, where drug information is not specific enough.

#6

@Viviane:

What @Chris_Knoll is saying that there is a proper solution. You can combine all ingredients in a drug class into a concept set and use that one. E.g. for TNF-alpha inhibitors you create a concept set containing the RxNorm ingredients adalimumab, certolizumab, etanercept, golimumab, infliximab.

However, that does not help you if the source data doesn’t give you the exact ingredient, but only the wishy washy class. The problem with these classes is that they are nowhere defined. For example, is naloxone an opioid or not? Depends. If you are looking from a chemical class perspective it is. If you are looking from the mechanism of action perspective it is not. That is the reason we have shirked away from allowing drug classes into DRUG_EXPOSURE.

#7

Thank you @Christian_Reich, that makes total sense to decrease the risk of allowing non-specific and possibly differently defined drug concepts into the drug_exposure table that can distort results. We will definitely consider this in our analysis, and potentially exclude data where the granularity is this low.

t