OHDSI Home | Forums | Wiki | Github

Concept id 38004515, Hospital, rolls up to both 9201 Inpatient Visit and 9202 Outpatient Visit

Hi @MPhilofsky,
Any update?

Before making the final decision about ‘Hospital’, let me summarize:

  • Initially, the wrong hierarchy were reported. Why do we want to get rid of the concept? It could be hanging in the concept table, but out of the hierarchy. The same way as other 6 top dogs do.
  • I didn’t really meet effective approaches in ETL to distinguish between IN/OUT patient Hospital Visits, only if it’s not specified explicitly in the source. In some data sources, encounters are specified as ‘admission’, ‘hospital’, especially when it goes to the admitted_from_concept_id area. So, all such generic Hospital stuff would be mostly mapped to 0.
  • One of the possible usecases when the differentiation between hospital and ambulatory/home visit is useful: any type of hospital visit (ER/IN/OUT-patient) might be easily associated with healthcare-acquired infection, while ambulatory visit or doctor’s hands during the home visit may cause it just sporadically.

Hi @Alexdavv,

No update. I never brought it up in the WG.

Back to Christian’s question:

The general test to distinguish an outpatient hospital visit from an inpatient hospital visit: does the visit last overnight or > 24 hours? It’s not perfect, but I think almost every visit will be correctly classified. There will be a few outliers, but there always are with observational data.

I still vote to

Let the ETL figure it out, so it’s as unambiguous as possible for the researchers.

@MPhilofsky I would like to revive this topic for the renewed Themis WG.

I recently came across this myself for a data source where the visits were not specified as inpatient or outpatient, only as ‘hospital visit’. So the concept 38004515 was the logical choice. However, I was confused that a visit concept could both be inpatient and outpatient.

My proposal would be to remove both parents, and have ‘Hospital’ as a stand-alone concept (like @Alexdavv proposed before).

@MaximMoinat:

Is there a use case, my friend? If the analyst doesn’t know if it is an inpatient or an outpatient, how does a “I don’t know what the hell it is” Hospital concept help? The analyst uses that to determine the healthcare setting the patient is in, and often uses it as a measure of severity of disease.

YES!

I was just disussing Visits and questions about how to define inpatient versus outpatient visits, how to define visits which belong in Visit Occurrence, visits which belong in Visit Detail, what to do when some source data have datetime and other data only have dates, etc. with @PriyaDesai yesterday. We really need more documented conventions on how to ETL these data.

I also enthusiastically agree concept_id = 38004515 isn’t well defined and quite ambiguous, not only for the ETL team, but even more so for the researcher where IP/OP visit is used as a proxy for severity of a person’s physiological state. (I see @Christian_Reich replied similar sentiment while I was writing my reply)

I have added the Themis tag to this thread. And I will create a Themis GitHub issue for this topic.

@MaximMoinat:

Do you have access to the ADT records that you could cross reference the visit to? If so, you could use that to choose the correct IP/OP type.

ADT -> Admit, Discharge, Transfer

I suppose I am approaching this from an ETL’er standpoint. Maybe I am off base again.

This is the main issue; the ambiguity of parent concepts creates confusion both during ETL and later analysis.

As @Mark hints at, we can solve this during ETL with some additional context. But the naive ETL developer who has a hospital visit sees the concept 38004515 ‘Hospital’ and thinks this maps well. Unknowingly they create ‘hidden’ problems for downstream analysis, especially network analysis.

Right. Because the hospital visit is a logical “or” of IP, OP and ER, it has to be a parent of them, not a child. But it doesn’t work this way because OP visit could be not in the hospital.
So yes, let’s make it a stand-alone concept.

For the use cases: it still could be useful to distinguish from home, ambulatory visits and “no visit”. Why not to use it as a proxy of patient’s state this way?

In the US, a person may have an ER, IP or OP visit at the hospital. A hospital provides all these services. And @MaximMoinat is correct, if we don’t force the ETLer to follow the conventions, then they might chose the easy, generic “hospital visit”.

This is documented in the conventions for outpatient visits:
“For outpatient visits the start date and end date will be the same.”

This describes inpatient visits: " * Inpatient Visit: Person visiting hospital, at a Care Site, in bed, for duration of more than one day, with physicians and other Providers permanently available to deliver service around the clock".

Since we already have documented conventions in place, I vote to make 38004515 non-standard and force the ETL to chose a top level IP, OP, ER, ER to IP, or LTC visit :slight_smile:

Perhaps naively, I have been truncating the encounter end datetime to date, for the end date. We only have OP but there are times that the encounter, in the past, has crossed the midnight boundary. These are extreme corner cases, true, and due to local state laws, none get OMOP’ed.

I dithered on posting this for a while, as this is bit of a tangent, but I think it is relevant to point out that elapsed dates are not the only factor in deciding OP vs IP during ETL.

… or Emergency Room stays for that matter (would cross midnight more often than your typical clinics visit). Do we treat those as OP?

True. Hospitals are institutions, and they can provide outpatient and ER services and get paid for that. But by “hospital” in the OMOP sense we mean what the patients would general assume is a hospital: Lying in bed, nurses and doctors around 24/7 and coming into the room, active treatment, highly unpleasant, particularly when in a shared room. Watch the “Bucket List” with Jack Nicholson and Morgan Freeman. One of my favorite movies.

Wow! We are no inscribed into state law? Where? :slight_smile:

Yeah. Visits are not clean things. Again, If in doubt, take the perspective of the patient. If the patient believes to be in the hospital as above it is an inpatient visit. If it was an emergency and after seen the doctor the patient goes home it is a emergency room visit, even if the poor thing had to wait forever to be seen. Usually, it is pretty obvious which one.

The idea of outpatient hospital vs inpatient hospital in US claims is justification for using UB04 Claims forms. From care delivery it is possible that in outpatient the patient maybe sleeping in a hospital bed. Rarely overnight, mostly day bed. It is different from the clinic visit where the billing is mostly cms 1500 form with rare exceptions.

The challenge is our attempts to standardize EHR and claims from across the world into one data model.

Historically we have thought of visit domain as patients perspective. Is the patient sleeping in their own bed = inpatient. Else outpatient. It was clean and simple to implement. It was patients perspective. Ie Visit domain was patients perspective.

Then we muddied that by adding clinic, emergency room, which are partly from patients perspective. We made these part of visit domain - continuing patient perspective.

Then we added more to the visit domain to capture more granular details into the visit domain. Some were patients perspective, others were details the patient wouldn’t care (eg HIPAA place of service codes). Those details were less patient perspective, and more administrative/billing. Eg. Tribal clinic vs dialysis clinic.

Now we are talking about more and more administrative codes. Eg. An oncologist may not see a patient while admitted to an orthopedic procedure, because they don’t have privileges in the hospital. So the hospital wheel chairs the patient to the oncology clinic, from the inpatient ward.

Standardization is messy and not clean. But maybe we should think about who’s perspective are we standardizing in what field?

We are using OMOP for AOU, which means the records leave our place of business; Tennessee has stricter laws about this than standard U.S. HIPPA regulations.

Whist I agree with why Tennessee does this, it does make my life lots more fun :roll_eyes:

How is this different from concept_id = 9201, IP?

The problems with 38004515, ‘hospital’:

  1. We already have concept_id = 9201 with the same exact meaning
  2. ‘hospital’ is ambiguous, generic, and not defined anywhere except in this thread

Looks like we are in agreement :slight_smile:

Emergency Room is concept_id = 9203 is on the same level as the IP, OP, LTC, ER to IP concepts

Correct, you know your data the best. So, if you have another way to distinguish IP from OP, please use it. This is just a general guideline for folks.

Yes, as expected.

That is true, but it doesn’t cancel the fact that “OP” concept could be more general than “hospital” concept in a situation where the OP services are provided outside the hospital. That’s why there’s no good way to link them all in a single hierarchy.

Why is it bad if it’s the only information they have in the source? They don’t know whether it’s IP, ER or OP visit within the hospital, but they know that it’s one of 3 (logical OR). If they keep it with the concept_id 38004515 ‘Hospital’, they can distinguish it from home, ambulatory visits and “no visit” situations in studies.

It says it’s only an ambulatory institution (meaning not a hospital), which is probably wrong. There are some discrepancies between the conventions (ideal world) and RWD where hospital could handle the OP visits.

Ok, it makes sense if each ETL could scrape IP and OP from “hospital” visits. But how would you map a “hospital” visit of less than a day duration?

True, but we don’t map from OMOP. We map to OMOP from the various sources where a “hospital” visit could be OP and ER also. Because if they don’t have any precise information, they can capture the institution’s piece rather than the patient’s perspective “watched the movie”.

It’s different and we (and you :slightly_smiling_face:) pretty much defined it. It’s generic, it’s a bit more institutional perspective rather than a patient’s perspective. But it’s ambiguous only unless we found an agreement.

Maybe, but how would we map the generic thing on the vocabulary level? Throw it over the fence to the ETLer without mapping and without a clear solution is not a solution.

Love the debate :slight_smile:

At its core, “hospital” is a care site, NOT a visit. It does not belong in the Visit domain. Neither do the other care sites that got lumped into the Visit domain with the changes years ago.

It is bad because researchers rely on IP vs OP as a proxy for severity of disease. If a person has an inpatient visit for covid, they are much more ill than the person being seen as an outpatient.

So, in following with the OHDSI mantra, the ETL’er needs to analyze their source data and make the determination if the visit at a “hospital” was for an outpatient visit, inpatient visit or ER visit. Analysis MUST occur before/during ETL, we can’t hand over ambiguous data in CDM format to researchers and tell them to figure it out.

This code comes from Medicare Specialty vocabulary. Nobody is making visits out of the Medicare Specialty vocabulary because at the source these do not represent visits. This is a healthcare provider taxonomy.
ETL’ers are chosing “hospital” because the visit happened at the hospital, which is a care site. Not because they have a code = ‘A0’ with vocabulary_id = ‘Medicare Specialty’ with data and attributes representing a visit. But some folks might have gone ahead and mapped a visit to ‘hospital’ because it was a good semantic match.
I’m not sure how the Vocab team should roll back this change. In the past, some other concepts which were hardcoded in the ETL (type_concept_ids) were deprecated without a replacement or a mapping. And sites were told to figure it out themselves. I’m not particularly a fan of this process, but we can’t accurately map the hospital to a true IP/OP/ER/ER to IP/LTC visit without additional context about the visit. The ETLer has to analyze the data, make the decision, document the business rules and rationales, then hardcode the visit_concept_id.

@Gowtham_Rao and @MPhilofsky are right. Should we clean up and make some explicit Themis convention out of them?

t