OHDSI Home | Forums | Wiki | Github

How to add empirical therapy feature to a drug event

Dear experts,
I need to include the “empirical treatment” feature for an antibiotic treatment. That is, there is the DRUG event corresponding to the antibiotic treatment and the PROCEDURE domain event “Empirical antibiotic therapy” (4153354). To add the characteristic of “empirical treatment” to “antibiotic treatment”, both events (DRUG-PROCEDURE) should be linked, but there is no concept of RELATIONSHIP between both domains. I also do not find the possibility of including a modifier within the DRUG event (in the style of “condition_status_concept_id” in the CONDITION domain that makes it possible to qualify aspects of a diagnosis),
Could you suggest me any possibility?
Thank you very much in advance for your help.

A few Ideas:

  1. Put one record each (DRUG/Procedure) and assign the same day to both records, and the logic to build a criteria around it is that you either look for the drug on the same day as the procedure, or you look for the procedure on the same day as the drug.

  2. Create a visit out of the encounter, and link the procedure and drug to the same visit. You would build a criteria looking for a drug with a nested procedure ‘at same visit’ or a procedure with a nested drug ‘at the same visit’

I don’t recommend using Fact Relationshp. There’s too many ways to build those and there’s no real standard around it.

Thank you for your advices.
As a complement to the second possibility that you offer me (a visit that includes both DRUG-PROCEDURE events), would it be possible to use the EPISODE/EPISODE_EVENT mechanism offered by CDM5.4 in the same way? I haven’t decided yet which scheme to use (CDM5.3.1 or CDM5.4).


All your ideas are legitimate, except you will probably fail. Because the data isn’t there. Who is recording “Empirical antibiotic therapy”? Even if they did, by OMOP standards this isn’t a therapeutic procedure. It is also not a drug, because it doesn’t say which drug. The fact that a physician tries some antibiotic without having concrete knowledge of the sensitivity of the causing agent is actually what is happening in 99% of antibiotic therapy.

The Episodes are also not your friend. They were introduced to characterize the dynamics of a disease, mostly cancers. They grow, go into remission, and relapse. And spread to different parts of the body. Again, nothing to do with antibiotic therapy.

What are you trying to do? Why don’t you just use the antibiotics that the patients actually got?

I agree with your comments.
I am going to discuss it with my colleagues, but I think that your proposal is the most viable: exclusively, record the antibiotic treatment.
However, I would like to know agreed mechanisms in OMOP to link events that do require linking (a need in general caused by lack of pre-coordination of concepts). CDMv5.4 includes a proposal for measurement and observation events; and, on the other hand, I constantly read that fact_relationship is not advised.
Other mechanisms with using “visits” (as a colleague proposes), although may be acceptable for the data scheme, on the other hand I suspect that it does not contribute to exploitation; it’s not a semantic mechanism, but a data schema based mechanism. I think that’s what you mean when you say that ideas could be legitimate but they will fail (in exploitation phases)
In any case, thanks for the help from all of you.

@mpascualc, see this proposal by @rtmill that adds semantics to the FACT_RELATIONSHIP table via standard relationship concepts. I am helping formalize it for proposal to the CDM WG. The challenge will be to avoid explosion of the set of relationship concepts for miniscule cases. Please comment in that thread if you have thoughts to share.

@jmethot , thank you for your suggestion and the reference; I will do so.