OHDSI Home | Forums | Wiki | Github

Mapping vague terms

Vocabulary Users
#1

Hi everyone,
We are working on mapping our pathology data into CDM however we have come across few terms which are vague and don’t have any particular answer to them. for example Number involved, Number examined, etc. We are facing difficulty in mapping terms like this Please let me know if you have any suggestion.

Additionally I also want to ask, can we use two codes for a single term for example I have a term “involvement status of an organ” so can I use two different codes for it like “involvement status” and a second code for “the organ”. How can I do this using Usagi.

Question about non-standard to standard code mapping... is it always one-to-one?
#2

@Anisa:

You are running into a very typical problem, so don’t feel bad. There are a few guidelines you can apply:

  • The OMOP CDM creates facts about the patient. “Number examined” is something relevant to the institution. It has no effect on what happens to the patient. Be generous with dropping those.
  • “involvement status of an organ” - we have concepts in the “Cancer Modifier” vocabulary that might fit the bill: Extension/Invasion and Margin concepts. The rule is here that we need pre-coordinated concepts with the full meaning. The OMOP CDM does not allow post-coordination, which means cobbling things together from many different concepts. So, splitting things into “involvement” and “organ” won’t work.
  • If nothing helps consider the use case. Is there a scientific question concerning a population of patients that this particular piece of data will support? If not kick it out. For example, for most organs that exist twice the laterality is very important for the patient but not for the population. I know dropping data creates anxiety, but keeping them in some awkward way won’t help create any evidence.
#3

I think our issue is that we are trying to use the OMOP CDM to support research projects that need access to detailed observations such as # LNs, gross tumour size etc. The parameters are important when trying to train predictive models using imaging and pathology data. We wanted to use OMOP so that the same central platform could be used to support all researchers across our institution. It also means that we are trying to support future as well as current use cases, always a tricky proposition (although these current issues are related to current, well defined use cases)! I guess we need t o work closely with the oncology and imaging working groups to move this forward.

#4

Wonderful. Please come and bring it on. Also, feel free to drop an issue into the Oncology WG Github.

t