Mapping procedure codes from CCI/CIHI: ICD10PCS heirarchical or SNOMED?

1

We have procedure codes in CCI format and these are structured in a hierarchical manner:

  • The first character (field 1) of each code represents a CCI Section which identifies broad types of interventions contained in CCI (e.g. diagnostic, therapeutic, obstetrical, etc.)
  • The next two characters (field 2) represent CCI Groups which identify logical groupings of related terms (Note: in the case of at least 3 sections, these groups are anatomy-driven, while others are section-dependent)
  • The fourth and fifth characters (field 3) represent CCI Interventions which identify generic procedures/interventions contained in a given section (Note: interventions are section dependent).
  • The sixth to tenth characters represent the CCI Qualifier which, may be broken into three separate fields each of which has a discrete meaning. In the case of therapeutic interventions, the first two characters relate to the approach/technique used, the next two characters relate to devices used/implanted, and the last character relates to any tissue used (to perform the intervention).

We need to map 100% of the source records to something in omop, even if it is a generic code. Ideally, we want standard codes so that ATLAS is useful, but if necessary, the source_value field can be queried by end users. We started by trying to map these using usagi but the error rate was too high.
To make the manual task manageable, one option is to choose SNOMED codes for the procedures and then add the anatomical site as a modifier. The other option is to try to map to ICD10PCS hierarchical codes.
As a concrete example:
CCI code: 1.YM.52.HA = Drainage, breast using percutaneous (needle) approach (L side is given in another field)
could goto

  • ICD10PCS:
    2863326: Medical and Surgical @ Skin and Breast @ Drainage @ Breast, Left @ Percutaneous
    or
  • SNOMED
    4046266 Drainage procedure
    with SNOMED modifier_concept
    4298444 Breast Structure

Has anyone else looked at this? Are there advantages when trying to generate cohorts with atlas of one approach over the other? I can cope with some loss of granularity since the source codes will be preserved but I would like to sync with what others are doing with administrative procedure codes from different sources.
Thanks!

2

Hi @amartel

From what you described, CCI looks like the ideally structured vocabulary from the mapping feasibility standpoint.

You don’t want that! It may seem you’re closer to the Standard representation, but you’re not. You’d come up with another locally defined approach which doesn’t allow you to practically leverage any of the systems - neither OHDSI vocabularies nor the source hierarchy.
You would rather want to make the decision: you’d stay with the source representation using source entities as the source values, source concepts, or even fake standard concepts, or you’d map to standard, but in this case, it’s highly recommended to stick to the existing approaches and conventions as much as possible.
If you’re mostly thinking about the local application/use cases, you may adopt CDM as the data model, and then you can stay with the source representation of the content. And it may work pretty well.
If you’re aiming for collaboration with the network, distributed studies, or full-fledged tools integration, I would recommend the latter option.

And you definitely don’t need to do tons of manual work. The easiest way is to utilize the source structure and attributes because OMOP targets have them too. Usually, plain SQL and good knowledge of vocabularies are your best friends.

3

Thanks @Alexdavv - I was leaning towards the SNOMED as I wanted to limit the number of vocabularies and we have other source systems from radiology where SNOMED has been used already. Your argument against modifies does make sense though and we will loose too much information without them.
I had thought of keeping it local and using the source representation. If we were only going to collaborate with other Canadians, that would work well, but tools like ATLAS are very valuable, and I want to make sure we can still use them.
Do you know if many groups use the ICD10 hierarchical codes? Has anyone in OHDSI examined what vocabularies are most commonly used to represent procedures?