OHDSI Home | Forums | Wiki | Github

How to store the body site of a performed procedure in OMOP CDM?


I would like to store the procedure ‘MRI’ on ‘Left knee’ in the OMOP CDM.
Any idea how I store the body site ‘Left knee’ in the OMOP CDM and link it to the procedure ‘MRI’ in the procedure_occurrence table?

Hi @CDV:

Do you have an analytical use case having that information? Do you want to compare left knee MRI with right knee MRI as a description of the procedure? The result of the MRI is a different matter.

Yes we would like have two patient populations, left-knee and right-knee!

And then what? What’s the scientific question?

It’s a generic question, we don’t have a specific scientific question yet. I was wondering if there is any convention to store bodysites in the OMOP model or if not, what the different options are depending on the use case.

There rarely is. Unless there is an explicit reason the laterality matters, e.g. blood pressure measurement on the right or left arm if you have an aneurysm or obstruction in the aortic arch, it is irrelevant. For treating the patient it is very relevant. But we are not doing that, we are studying populations, and use procedures like an MRI on the knee as a criterion for cohort definition. So, I would drop it for now.

@Christian_Reich I agree that their is rarely a need for laterality. I will offer though that if I break my Left arm today it becomes phenotypically impossible for me to break my Right arm for some arbitrary time period as we are not able to disentangle follow-up care from another arm breaking. Given that I have two arm it’s still physiologically possible for me to break my contralateral arm. Of course I would seek care at two different health care systems with OMOP CDMs to create two events in EHR data systems across the street (or river) from one another!

Isn’t it the case, tho, from a ‘phenotyopical perspective’, that you can’t actually break your left arm? It’s not possible to represent your left arm as broken (is what you’re saying, correct?), so instead we can just phenotype an arm break. An arm break wouldn’t prevent a subsequent arm break unless you defined an arm break to persist until the end of observation. But there is probably the notion of a ‘clean window’ between breaks so you can distinguish one break from another. Am I understanding correctly?

How does this play out for things like fingers? I have 10 fingers, do I need to represent a phenotype for each specific finger breaking? And if I’m doing a study on finger breaks, which finger is the primary outcome? Index the primary event on my index finger? Sorry for the pointed questions, but I’m curious where we draw the line, and if it’s really the case that we are unable to study effects on lateral appendages because we don’t distinguish them in the vocabulary.

Just going to leave this standard, valid, SNOMED concept here: MRI of left knee, which ‘is a’ MRI of knee so also useable for a generic MRI of knee phenotype.