ICDO2 to ICDO3 map

[gkennos]

As part of our MOSAIQ ETL, we primarily have ICDO3 codes, but there are some residual ICDO2 codes included in our data.

As a consequence, I extracted conversion maps from the appendices of this document: https://apps.who.int/iris/bitstream/handle/10665/96612/9789241548496_eng.pdf

We would like to make these mappings available for the oncology working group, and therefore are requesting the addition of the ICDO2 vocabulary, so that we can then add these mappings to the concept_relationship table. Note that there are some 1-n mappings in this conversion, although at least in our data, we observe only 1-1 and n-1 mappings.

If this request is accepted, I will compile the necessary ICDO2 files from the ICDO3 concepts, merging in the deletions / additions / mappings mentioned above.

[dimshitc]

ICDO2 was not used since year 2000. Do you want to analyze such a historical data?

[gkennos]

These data were collected since 2016, so I can't answer why these codes were used, but regardless they are present 🤷🏻‍♀️

The vast majority are icdo3, but there's a significant minority of icdo2 (and a non-significant minority of icdo1)

[cgreich]

Can you send the ones you have? Should be a finite list that is no longer growing. So we know is it a problem of dozens, hundreds and thousands.

[gkennos]

It's dozens. There are 74 distinct codes that exist in our system's reference table that belong to ICDO2 rather than ICDO3 (actually, 2 of them are ICDO1). This means that they are available in the user interface of the system and can still be added by users right now for a reason that is not clear to me. In the data that I'm currently looking at, I have examples of 15 of these codes, but at the moment I am looking at the newest slice of the data, so I would feel most comfortable accounting for all 74, especially as there's no way to stop clinicians using these codes going forward.

There are 142 codes that I scraped from that PDF file for conversion, but seems like the rest of them aren't available in the system so aren't expected to be entered in either historical or future data.

I will follow up with another group that I know uses the same system to see if their reference table has a similar mix of codes.

[dimshitc]

Then you can attach the link to the google document, containing these codes. We need code and description, but if you have any additional information about these concepts, please include

[mgurley]

@dimshitc @cgreich @gkennos

Can we just do an analysis of how much it would take to fully support the full ICDO2 to ICDO3 mapping? I feel it is likely that means-testing codes to only put ones found in real data causes more work than just supporting the full ICDO2 to ICDO3 mapping. We are only talking about one axis, morphology, so it should not be overwhelming. More ICDO2 to ICDO3 mapping work can be found here (done by SEER):

https://seer.cancer.gov/tools/conversion/

I believe this must be based on the appendix pointed to by @gkennos, but is in an Excel spreadsheet.

[gkennos]

Ha, yes @mgurley I spotted that spreadsheet only yesterday - after extracting from the pdf 🤦‍♀️🤦‍♀️🤦‍♀️

I think the thing we do still need to confirm is that the ICDO2 codes still being available to enter in the UI is not a single-institution issue. If it's just our configuration, then fair play we will just maintain a custom mapping for it in our ETL, but I suspect that it is actually an issue for other groups using the same OIMS, which would make the case for supporting the mapping centrally I think?

[cgreich]

We could support it in three ways:

• Centrally, with the entire ICDO2 officially as a source vocabulary
• Centrally, but only the ICDO2s you have in your data
• Locally (and maybe shared amongst a few institutions)

It's no difference to ICD9CM and ICD10CM. Except there is (was) widespread adoption of that one. Let's make a pragmatic decision.