OHDSI Home | Forums | Wiki | Github

How to represent a patient admitted from their own home

(Edward Palmer) #1

In the VISIT OCCURRENCE table the admitted_from_concept_id asks for something from the “VISIT” domain. If the patient is admitted from their own home (which is the vast majority of cases for an acute hospital) how is this represented? Or Have I misunderstood the function of admitted_from_concept_id?

I can find “Home Visit” (581476) is that correct? It seems strange to be semantically linking where a patient was admitted from, to a notion of a visit.

Many thanks for clarification.

(Qi Yang) #2

admitted_from_concept_id is supposed to take code from UB-04 form (Vocabulary_id = UB04 Point of Origin). Among the UB-04 Point of Origin code, the first one is “Home/non health care facility point of origin”(http://athena.ohdsi.org/search-terms/terms/32193). But now all UB-04 codes are non-standard without mapping to standard concept. They used to be standard just a few months ago. Why are they all non-standard now? @Dymshyts, @Gowtham_Rao. Is it because this part of the model is subjecting to changes?

(Gowtham Rao) #3

You are correct, they were standard till recent version of vocabulary. I don’t think it the reason was model changes, but was due to a project to standardize vocabulary. @Christian_Reich led this effort and could point to additional context.

(Christian Reich) #4

Sure they have, @QI_omop. Do you have examples where there are not?

Only concepts we explicitly did not map were those about birth and death. This is not a Visit problem. You are not coming from a Visit “Heaven”, and go to a Visit “Heaven” or “Hell”. At least not according to the OMOP CDM.

(Christian Reich) #5

The idea of a Visit is the setting in which healthcare happens. If the patient enters healthcare from outside healthcare (i.e. “home”, that’s where people live, but could be “hotel”, “street”, “bar”, “all-inclusive vacation resort”) we don’t capture that. We generally only capture the healthcare experience of patients. Which means, in all these cases the admitted_from_concept_id concept is NULL.

Home Visits are different. They are healthcare encounters, where a provider visits the patient outside a healthcare institution. It has nothing to do with healthy people living in homes.

Makes sense?

(Qi Yang) #6


Sure they have, @QI_omop. Do you have examples where there are not?

Here are the examples:

32582 3 HMO Referral
32197 6 Transfer from another Healthcare facility
32199 9 Information not available
32587 N Normal Delivery
32588 O Premature Delivery
32589 P Sick Baby
32590 Q Extramural Birth
32591 R Not Available
32592 S Born inside this hospital
32593 T Born outside this hospital

I agree with you that there is no need to map some of them to standard.

(Christian Reich) #7

This is the typical CMS mess, where different categories are mixed up together:

Source Destination Note
HMO Referral NULL HMO is not a Visit, but a Payer
Transfer from another Healthcare facility NULL We may have to add some Über-Visit here. Till then Flavor of Null
Information not available NULL Flavor of Null
Normal Delivery NULL Babies are from where? “Heaven”?
Premature Delivery NULL This is a Condition, we might add a Maps to to the UB04
Sick Baby NULL A Condition, but useless
Extramural Birth NULL Flavor of Null
Not Available NULL Flavor of Null
Born inside this hospital NULL Which means the baby wasn’t admitted
Born outside this hospital NULL But where? Flavor of Null

(Edward Palmer) #8

Thank you all for the responses. There is definitely conflicting information circulating, so I think some clarification is needed.

The GitHub repo for CDM6 for [admitted_from_concept_id] states “A foreign key to the predefined concept in the Visit Domain reflecting where the patient was admitted from.”

If this is true, then the instructions on GitHub need to be updated.

Could I politely challenge this from an inferential perspective. All my patients (critical care) are at greatly increased risk of leaving the hospital in a diminished physiological state. This means we really care about if they came from their own home, and were discharged to a nursing facility. If we don’t capture the prior residence, we cannot control for this transition in state properly. Where they came from is as much a part of their healthcare journey as any other.

(Christian Reich) #9

That is correct. We are revising the entire documentation. The principle is what happened to the patient. So, something like “HMO referral” does not mean anything from a perspective of what the care settings were prior to this Visit.

Not sure I understand, @Doc_Ed. Nursing Facility is definitely something you’d put into ADMITTED_FROM_CONCEPT_ID and DISCHARGED_TO_CONCEPT_ID. If we don’t know where they came from, or they came from no healthcare setting (my home is not a healthcare setting, I know that much), we put NULL in.

Does that work?

(Gowtham Rao) #10

So, in the absence of information about what other health Care setting the patient came from - we assume they came from some form of non health care setting like home. These are all NULL i.e. no health care information

Is that accurate capture…?

(Christian Reich) #11

OUCH!!! Can you believe it? I fell into the trap.

Both fields are mandatory, so if we have no information about a healthcare setting they came from the content is not NULL, but 0.

But otherwise yes. The fields capture information what other such setting they came from or went to. 0 means we don’t know of any healthcare setting. Whether we because there was no healthcare setting (home, cruise ship etc.) or we just lack the information is indistinguishable. Healthcare data are chronically bad representing facts from outside their realm.

(Edward Palmer) #12

The main problem with this approach is that we now don’t know if absence of information is data missingness, or if the patient came from a non-healthcare facility. In the NHS we have excellent data on where people come to hospital from. From an analytical point of view, those who are homeless, or have a missing point of origin information tend to have worse outcomes than those who come from their own home. So I am keen to record this aspect accurately.

(Gowtham Rao) #13

Why not have many domain neutral concept_id to represent various flavors of NULL (instead of just one 0 - which is essentially a domain neutral concept id to represent NULL)

(Chris Knoll) #14

@Gowtham_Rao, was this topic not settled here?

(Gowtham Rao) #15

Thanks @Chris_Knoll I forgot about it.

(Christian Reich) #16

Two points:

  • That is the reality in most healthcare systems. You have only a partial capture. A missing record has a much lower specificity than an existing record. It’s the plight of RWD.
  • Facilities we have. What we don’t have is “no healthcare”. This is what most people experience. They are just happy, healthy, and roaming around wherever they feel.

Can you give us the list of places people came from?

(Edward Palmer) #17

I wonder if we are talking cross purposes (probably because I am still on an OMOP learning curve). I just mean that we capture patients’ home addresses quite accurately in the NHS, and generally know if they arrived from another healthcare facility, their own home, or if they don’t have a fixed abode. These are the fields that we capture routinely:

  • Own Home
  • Nursing home or equivalent
  • Health-related institution
  • Non-health-related institution
  • Residential place of work/education
  • Hospice or equivalent
  • No fixed address/abode or temporary abode

I would advise against any scenario where you double code something to mean missing data (i.e. NA) and something else (in this case “came from home”) simultaneously (both taking a “0” here). Often there is an informative missing data pattern to capture and this should be represented accurately.

(Christian Reich) #18


We got home addresses. The assumption is that people live at home, unless they are in an healthcare institution. A home for the elderly is considered “home”, but if there are medical services (nurses, doctors routinely visiting) they are healthcare institutions. In other words, as soon as services in those institutions can be reimbursed by the insurance. That’s what it really comes down to. This is different to the NHS.

We could add those you listed very well (hospice is already there). I can see useful use cases supported by them. Will add “Own home”, “Nursing home”, “Residential place”, “Homeless” and make them children under a new Concept “Home” ("Home visit is different - there the provider comes to the home to provide care). “Health-related institution” or “Non-health-related institution” is difficult. They should be on the top of the hierarchy. We could do that, but that would change the way stratification works right now, where we roll up Visits as high as we can, ending up with things like “Outpatient”, “Inpatient” etc.

Do you have the complete controlled vocabulary you are using with codes? Would be better to bring these in as a proper vocabulary.

I know what you are saying. However, right now, “Visit” explicitly refers to a healthcare setting. 0 means no healthcare setting, or no known healthcare setting. That is no different to any other 0 in any other field containing a Concept. If we change that we have to accept that in many countries the data will not support this distinction.

(Edward Palmer) #19

Thanks Christian. As ever this has been an extremely useful conversation.

The vocab we specifically use is ICNARC for critical care in the UK, however they use a closed model and tend not to allow their resources to be used outside their network. I will ask them.

Alternatively, there is the Hospital Episode Statistics data dictionary availible via NHS Digital. They have very similar fields on this sort of thing, and I can’t see any evidence of a licence. I have dropped them an email to ask if they have any licence restrictions.

Both would need to be mapped into OMOP standardised codes. Do you have any recommendations for this? Do you for example do a first pass fuzzy match? Happy to help out here if useful.