OHDSI Home | Forums | Wiki | Github

In hospital death


I am new to the OMOP CDM, and I try to figure out how in hospital deaths should be indicated in contrast to deaths at home.
I could not find a location field or something similar in the DEATH table, there is the death_type_concept_id, but this field should be used for the source of information if I understood it correctly. Also, the cause of death has nothing to do with the location of the death.

Could you give me some directions on how this is meant to be mapped to the CDM v5.4?

Thank you very much!

Hi @HeideNei:

Welcome to the family.

Death, like any other event, is not linked to a location. But it is linked directly or indirectly through timing to the Visit. Granted, you could be discharged from the hospital and die at home in the evening. But it’s probably not that relevant, unless you want to study things like the effectiveness of code blue responses of a hospital. Do you?

It is relevant in the surveillance of healthcare associated infections.

I found the concept 4061268 death in hospital. So I could create a record in the OBSERVATION table with this concept without any value additionally to the record in the DEATH table. Or does this approach violate any guidelines?

Well, you only have an issue if the death is the same day as discharge. Any day after that - you know it is at home and not in the hospital. So, that’s good.

No, this doesn’t work. You can do that, but nobody is looking for such concept (except you who put it in). Death is in the DEATH table and only there.

But then what is this concept “death in hospital” for and how/when can it be used?

Yeah. 4061268 is a standard SNOMED concept, which means, you are allowed to use it. We should de-standardize it (standard_concept=null). Problem: There are 300 thousand of them, and there is no chance that we go through them all and set some of them to non-standard. The way this should be handled is through the THEMIS Working Group that sets the convention standards we want to use. But I agree, it is an issue we need to keep figuring out.

But death goes into the DEATH table only.

Hello @HeideNei and welcome to OHDSI and the OMOP CDM!

As Christian states, if a Death occurs during a hospitalization, you will have a Death record and Visit Occurrence record on the same day. If the Death occurs at home, the Death date and Visit dates won’t overlap. This is the cleanest and easiest way for researchers to determine if the death happens at the hospital or at home. Sure, you can take it a step further and add in an additional record in the Observation table, but when it comes to clinical event data, most folks take their source values and map them to standard concept_ids before inserting in the clinical event tables. And most source systems do not have SNOMED code = 16983000 and most sites do not add in this additional step to create these records during the ETL process. ETLing data to the OMOP CDM is quite the large undertaking. Making it more complex by trying to represent every fact in obscure ways isn’t advisable. Especially for your first ETL. Take the pragmatic approach and ETL the basic data. If your researchers (concept_id = 16983000 won’t be use in a network study) still have an unmet need, then add in this additional row of data.

Per the conventions in CDM v5.4 “The death domain contains the clinical event for how and when a Person dies”. So, if you have a death record it must go in the death table. If you have additional attributes or details about a death, such as died in hospital, you can use another standard concept_id to represent these details.

@Christian_Reich I don’t think we should de-standardize concept_id = 4061268. We could add stronger language to the Death table. Something along the lines, “if you have a death record it must go in the death table. If you have additional attributes or details about a death, you can use another standard concept_id in its given domain to represent these details”.

Let’s THEMIS it, @MPhilofsky.

1 Like

If you do not want it in the observation table, it does need to be de-standardized. If there is a standard code that I can map to (this has been done under advisement), then I will map to it, no exceptions. If one is ok with it being in both tables, then it does not matter.
We are out patient only, so in this specific case, it is highly unlikely (unless we receive a C-CDA to update our records from an in-patient facility) that we would have this code, but this could happen with other records.

There is no standard code, @Mark. It’s a record in its own table, called DEATH. Death, like birth, is an event, but we don’t treat it like the other events, which don’t have an effect on our being on this earth. So, no mapping record. The ETL has to do some extra gymnastics.

Yes, I am putting it in the Death table; I am just informing that it will be duplicated in the observation table as well if, for some unknown reason, 4061268 should show up in our EHR. I was agreeing with your recommendation below:

Oh sure. But it won’t matter. You are thinking from the perspective of “where should I put something?”. You should be thinking in terms of “where is the analyst going to look for something?”. And nobody is looking for an “observation” when there is an official DEATH table. So, this is the typical “attic” use case.

But on the other hand it wouldn’t hurt or do any harm.