OHDSI Home | Forums | Wiki | Github

Any thoughts on how we should generally handle drugs that are procedures with non-specific titles?

Vocabulary Users
#1

Interesting example: http://www.ohdsi.org/web/atlas/#/concept/2213405

This is CPT-4 code for RSV antibody, which gets classified as domain ‘Drug’, but doesn’t map into RxNorm or any subsequent drug classes. That could make sense since the CPT4 concept string is: ‘Respiratory syncytial virus antibody for injection into tissue or muscle’. However, from everything I can find, the only product that can use this code is palivizumab (brand name: synagis), and this is consistent with CMS billing guidelines.

The question to the community: how would we want to handle this situation, either in the specific or more general case?

In my analytical use case, where I want to study palivizumab, I would like for this code to be mapped into RxNorm (the NDC and HCPCS associated with it are much less commonly used in the claims databases I have access to). However, I don’t know how, as a systematic process, we would have any insight into the fact that this code, while non-specifically described, is used for a specific (unlisted) ingredient. Should it be the case that, when these situations arise, we maintain an external reference list to clean the vocabulary as much as possible? Or do we tell researchers that they can’t be assured that the procedural administration codes fully capture the drugs?

#2

@Patrick_Ryan, this is a great example (one of many) of the vexing challenge of using procedure codes to capture drug exposure. This case is a bit unusual, in that the CPT code referenced reflects only a single possible drug (as opposed to say a pneumococcal vaccine). In multi-drug cases, it is not possible to know which drug was actually received from the procedure and thus I would recommend that people develop cohort definitions that include those procedures but not make them equivalent to a specific drug exposure event. It worries me a bit that this marked as being in the Drug domain, since such specificity cannot be relied upon for parallel CPT concepts.

That said, I like your notion that it would be handy to maintain some external relationships connecting CPT codes like RSV Ab, HPV vaccine, etc with their RxNorm associated drug concepts. Even if 1-to-many, it will help with building cohorts more consistently and comprehensively.

So to your questions:

Yes.

Yes. :slight_smile:

#3

Friends:

Two things:

  1. For the vaccines, we are entering the CVX vocabulary, which caracterizes the vaccines properly by its mechanism of action as well as its pathogenetic mechanism. CPT and RxNorm will be mapped to them. It’s half done, I just don’t get the time to finish it.
  2. The problem of mapping when we know the destination space: This is one of three algorithms currently in development:
  • Creation of a new vocabulary with the assumption that something that doesn’t map will be created.
  • Mapping of codes to an existing vocabulary that is construed to be complete (à la RxNorm): Here, we will only map “uphill”, meaning that if something doesn’t map we go up the hierarchy till we find something that maps unequivocally. This is what we do for HCPCS and those “dirty” drug vocabularies where we don’t assume all information is available
  • There is a third version, where the mapping goes uphill and downhill, and if something is hit unequivocally we take it, starting from the most detailed one. This is the case in CPT (and could be also used in HCPCS), where we assume that the codes have to explicitly avoid making direct product referrals for reasons of commercial neutrality, even though there is only one product on the market and we can be certain that that product was used.

These algorithm exist in beta stage, but the general principle of automated mapping of drugs is described here (really only the first use case is mentioned in the wiki right now).

#4

Hi Christian,
I’m new to OMOP (I guess I’ll have to put a short introduction to myself into the forum shortly). I’ve recently installed a local copy of OMOP including vocabulary content.

Anyway, as I’m getting familiar with the terminologies, I read your CVX comments. I’ll probably be ingesting CVX data into OMOP. I noticed GitHub has CVX scripts and thought it might be possible to run these scripts to add CVX crosswalks into the OMOP vocabulary tables.

Can you comment on this? Will these scripts add the CVX terminology and applicable crosswalks? Or are they still under development?

Any advice will be appreciated.

Regards,

George McCullen

t