OHDSI Home | Forums | Wiki | Github

Mapping a local procedure vocabulary to OMOP - questions about best practice

Hello everyone,

I am part of a project group from Hungary, from two medical universities, joining EHDEN soon. For that purpose, we need to map our local procedure vocabulary to OMOP, using Usagi, and we have a few issues to clarify, about what is the best practice for mapping.

  1. When automapping with Usagi, the “concept class” should be set for “procedure” or the “domain”, or both? Should we always only map to SNOMED?

  2. Can we map one local concept to more than one OMOP concepts (“one to many mapping”)? If yes, what is the maximum number for that?

  3. When there is no equal match in SNOMED procedures, can we map to other sets, for example ICD10 for clinical findings, SNOMED for devices or RxNorm for drugs? If yes, is it needed to always map “one to many” and map at least one SNOMED procedure?

  4. Is it better to map only one concept (one to one mapping), even if it is wider, or is it better to map more concepts (one to mapy mapping), if that way we can find an equal match?

Can you please offer some advice on this? I read through a lot of topics but I couldn’t find answers to my questions yet. @Christian_Reich, maybe?
Thank you in advance!

Some examples of your source values will be helpful.

  1. I am not sure about the best setting for Usagi because I have not used it for a while. ‘Should we always only map to SNOMED?’. No, concepts in the Procedure domain belong to a number of different vocabularies.

  2. Yes you can map to multiple concepts, but prior to giving guidance it will help if you provide some examples of procedures were you think this is required.

  3. Yes, there is nothing magical about SNOMED. What is important is to understand the difference between a standard and non-standard concept. Procedure_concept_id, drug_concept_id, condition_concept_id, etc. should only be filled by ‘Standard’ concepts. Use columns that end with ‘_source_concept_id’ if you cannot find an appropriate ‘Standard’ concept but there is a non-standard that matches your source value. There is nothing special about SNOMED, it is NOT necessary to map to a SNOMED concept.

  4. It is better to map to one concept. It is important to understand that a row with a concept stands alone. Your question sounds like you think that you can get a better description by using multiple concepts, but any analysis will not know about your ‘group of concepts’. Again it will help if you can provide examples.

Thank you for your answer! I will give some examples to clarify.
We are mapping our procedures, but some of the concepts are drugs or devices, or drugs with a diagnosis, or device with a procedure, or some different combinations.

Example A. procedure → measurement + device
Our local procedure code: 89445 Continuous invasive pressure measurement in the pulmonary artery with a Swan-Ganz catheter
We would map it into:

  • Pulmonary artery catheter (SNOMED, Device domain, Phisical object concept class)
    AND
  • Pulmonary artery pressure monitoring (SNOMED, Measurement domain, Procedure concept cass)
    Is it okay, that we map it into two concepts, both with “maps to” realtionship? Or do we need to choose one target concept, if this problem comes up?

Example B. procedure → procedure + procedure
Our local procedure code: 92530 Induction of labor with intramuscular drug administration
We would map it into:

  • Medication administration: intramuscular (SNOMED, Procedure domain, Procedure concept class)
    AND
  • Induction of labor (SNOMED, Procedure domain, Procedure concept class)
    Same question, is it okay like this, can we map two procedures, or should we choose just the second one (Induction of labor)?

Example C. procedure → drug
Our local procedure code: 98105 Levosimendan for the treatment of heart failure
We would map it into:

  • Levosimendan (RxNorm, Drug domain, Ingredient concept class)
    The reason behind this, that the clinical finding, the heart failure will be mapped from our diagnosis codes to ICD10 anyways. Is it okay that we map our procedure to a Drug domain?

Example D. procedure → device
Our local procedure code: Artificial lens
We would map it into:

  • Intraocular lens implant (SNOMED, Device domain, Phisical object concept class)
    Is it okay that we map a procedure code to a device domain?

Hi @Agota_Meszaros

maybe instead of mapping to

it would be better to map to
4279768 65677008 Pulmonary catheterization with Swan-Ganz catheter

I would just map to 4150970 31208007 Medical induction of labor
the “Medication administration: intramuscular” unfortunately doesn’t help as it’s impossible to track which drug administration is related to it.

that’s ok

Probably the actual procedure means 4288992 69724002 Implantation of prosthetic intraocular lens
and you need to map to that.

Usually if we are not sure in the meaning of the concept, we look at the context, i.e. look at a couple of patient stories and look at other events assosiated with the given concept.

Thank you @Dymshyts!

I understand, so it is better to map to only one target concept. But is it okay, that our source concept means someting else? In this case our sorce concept means the procedure of contuinuous invasive measurement of blood pressure using Swan-Ganz catheter, but the target concept you suggested means the procedure of catheterisation.

I can happily accept if that is the case and these little differences don’t matter at all, but I want to be sure we are on the same page regarding this. It is just one example, but we have a lot of similar concepts.

Okay, we will try this. And I agree, in most our cases we can assume that the device itself means implanting the device.
I have an other question regerding this. Since we don’t code devices in our electronic health record, the device itself won’t appear on patient’s history as an OMOP CDM concept. Will it be a problem for later research, or it doesn’t matter?

I meant the mapping to 2 concepts:
4279768 65677008 Pulmonary catheterization with Swan-Ganz catheter
and
Pulmonary artery pressure monitoring.

I don’t see a lot of studies around devices, that’s why if we need to chose between the mapping to Procedure and Device, I would chose a Procedure.

t