OHDSI Home | Forums | Wiki | Github

Provider table - standardize the way providers are identified

Good evening OHDSI! A thought came to my mind and wanted to get the community’s input…

  1. OHDSI analysis is centered around the construct of a cohort made of a population of subject’s. Subject_id’s are almost always person_id’s. Has anyone performed an analysis where subject_id is a provider_id? If not performed, is there an interest for such use cases @Patrick_Ryan asked this question here
  2. Could you please share if you have access to identifiable provider information such as NPI https://npiregistry.cms.hhs.gov/ in USA. i.e. if you have claims or EHR data, do you have identifiable provider information in the data.
  3. Is there an interest for OHDSI community to maintain provider_id composed of NPI or some international id’s - similar to how we maintain vocabularies. e.g. data from this link may somehow be part of OMOP vocabulary like system, maybe OMOP provider ‘registry’.

We know, and there have been attempts to return it back to person_id. @daniellameeker (or @Daniella_Meeker) initially wanted to keep it open, so we could create cohorts of providers for example. For studying effects that depend on the provider. Of course, you could always use their patients instead…

This question was discussed before, but I forgot in what context. The OMOP CDM doesn’t care what information you have, and if it is identifiable or not. If you ask about what data actually contain - IQVIA’s data have NPI and provider information.

Like create a standard vocabulary with the true providers in there? Sounds interesting, except internationally it will be a challenge. But why not? provider_concept_id=0 if you cannot link or identify.

As a world of caution: Some databases explicitly prevent you from using this information, or use in context with other information. Provider profiling will leave some folks uncomfortable. But again, those data sources wouldn’t use that provider_concept_id.

Most of my studies are intervening on providers (or observing policies intended to influence provider decisions). Since patients can have more than one provider (eg one provider in the intervention group, one in the control group). I can’t simply replace
with patient. however, as provider directories & NPIs are available, it would be nice to leverage standards.

Thank you @daniellameeker.

(Assuming that data being used is in OMOP format.)

For these studies where the subject being analysed is a provider, do you use any ohdsi tools or systems like Atlas, R packages, ohdsi results tables?

Would it be possible to share some use cases?

We used OMOP for all of these multi-site trials, mostly leveraging the vocabulary tables for drug categorizations, which we extend with value sets for quality measures, and we used NPIs for provider IDs:

Anyhoo - I think it would be fine to change the cohort table to be only patients or group providers in a different table. It was just when we were working on OMOP 4 in the age of the dinosaur, this use case was front and center for me.


@Gowtham_Rao or @aostropolets

Have the PLAN_CONCEPT_ID and PAYER_CONCEPT_ID been added to the OMOP Vocabulary yet? For our THEMIS recommendation we want to specifically tell people where to look in the Vocabulary.

There is a github issue open to handle this

1 Like

Discussion continued here. Let’s close this.