OHDSI Home | Forums | Wiki | Github

Tumor Registry ETL using LOINC?

Vocabulary Users
#1

Greetings,

We have tumor registry data in NAACCR format that we want to put into the CDM (along with our EHR data). @Mark_Danese mentions that “virtually all cancer codes are in LOINC” in a relevant post found here. Some questions regarding this:

(1) Would mapping to LOINC make the most sense for tumor registry data? According to the OHDSI LOINC specifications found here, it would appear that most of the data would be going into the Measurement domain.

(2) Would using LOINC codes as standard concepts allow full functionality of the OHDSI platform? In other words, would mapping to the LOINC vocabulary instead of SNOMED cause any queries or tools or break?

This topic seems to have been visited several times in these forums but never resolved. One suggestion was that we may need to include an oncology vocabulary? Any other insights into the problem would be greatly appreciated.

1 Like
#2

Loinc is a standard vocabulary for measurements, so that would make good
sense if you have labs or other quantified observations in your registry.
If you have qualitative observations, they can also be stored in the
observation table and can use any standard concept, snomed or loinc.

All ohdsi tools will work so long as you follow the cdm conventions, which
means you apply standard concepts from the appropriate domain within each
cdm table.

If you find you can’t map your data or there are other specific
vocabularies that are useful for a particular analytic use case, please let
us know on the forums.

#3

Friends:

Probably a good time to start this conversation. Tumor information is popping up, but it hasn’t really made it through the the kind of urgency that we would move. Should we?

The idea is that you have pathology/histology, stage and metastatic information. Is there anybody who has strong opinions how to deal with that?

#4

We have been working on mapping SEER data. It is a bit challenging for many reasons. I will post what we have done with oncology data to date in a tab delimited file on dropbox for now: https://www.dropbox.com/s/2gl5az5duq7z72q/Final%20SEER%20variable%20with%20codes.txt?dl=0

The main challenge is that the oncology vocabularies combine concepts together. For example histology codes (e.g., the codes for small cell lung cancer) also include behavior (e.g., “malignant”). You can separate them, but you then can’t use the text descriptors from ICD-O which are specific to clinically relevant combinations of histology and behavior. Another issue is that there are “recoded” values from SEER that are useful for grouping things into commonly used cancers. Generally this is by site, but non-solid tumors need to be handled differently.

LOINC is very good for most oncology vocabulary work. It needs to be updated in a few places but it works. I would encourage us to use that for everything rather than mixing vocabularies. At least, it would be good to do it in LOINC as much as possible.

In the end, I think we are going to need a separate “oncology” vocabulary.

NAACCR to ICD-10-CM to SNOMED?
t