OHDSI Home | Forums | Wiki | Github



(Erica Voss) #1

I got a great question from @Eldar today . . .

In some databases we can have information about patient readmission to hospital after surgery admission
that we would like to store in CDM.
I think of using visit_occurrence.preceding_visit_occurrence_id , that can be filled with visit_occurrence_id of
corresponding surgery admission visit. This way seems elegant but I’m not sure
that it’s correct as due to CDM documentation it can be used to link a visit IMMEDIATELY preceding.
So, lets say we have sequence of visits 1,2,3,4, where 1 is a surgery admission visit, 4 readmission visit,
2 and 3 are some other visits happened between 1 and 4 and are not related to surgery admission visit.
Is it legit for visit 4 to use preceding_visit_occurrence_id = 1 to link primary admission with readmission?

I would have thought it was okay to use VISIT_OCCURRENCE.PRECEDING_VISIT_OCCURRENCE_ID in the way Eldar explains - surgery happens (VISIT 1) and they are readmitted later (VISIT 2 - linked to VISIT 1). However based on the definition of this column this would not be correct.

Could we use the column in the way Eldar explains or would this introduce inconsistencies in how this column is used?

Tagging @clairblacketer, @Christian_Reich, @mvanzandt

(Clair Blacketer) #2

@ericaVoss You are correct in that the original stated intent of visit_occurrence.preceding_visit_occurrence_id was to capture the visit immediately preceding the current visit record. In reality I take this to mean that the field is attempting to capture readmissions. Readmissions are notoriously tricky to logically define because I think of them as strictly inpatient artifacts so if a person was in the hospital, then had an outpatient follow-up, and then had another inpatient stay a few days later, you could make the argument that preceding_visit_occurrence_id should be the inpatient stay and not the outpatient inbetween. With that being said, I don’t believe this field is being used currently in standard analytics so this is our chance to refine the definition a bit and I think the one @Eldar laid out is great. Essentially this field can hold the visit_occurrence_id of a known associated prior visit, if available. We may want to even limit to only inpatient visits as I don’t see this being useful in the outpatient setting.


(Christian Reich) #3

Not sure what this would do. The Preceding Visit field is used to chain together a sequence of Visits that from the patient perspective would be considered one long treatment. Readmission is the opposite: It says the patient was sent home, but then it turned out that was too early. So, readmission is admission after a non-adjacent Visit situation. If you smooshed it together you would not be able to find these.

Just find all Visits with surgeries, and then look for new Visits within a short period of time. Maybe with some limited Diagnosis set (e.g. wound infection).

(Eldar Allakhverdiiev) #4

Hi @Christian_Reich, in case of data I’m working with there can be several visits within a short period of time after surgery and only one of them is actually readmission related to surgery ( let’s say it was cancer surgery and there were several visits for chemotherapy etc. and one - readmission). So it’s not enough just to look for new visits. From source data I do know for sure what of the following visits was readmission related to the surgery and I need to have corresponding ‘flag’ in CDM.

(Chris Knoll) #5

Any reason why this wouldn’t simply be an observation of a readmission (where you can put additional information in value_as_concept to indicate the nature of the complication?)

I’m only suggesting because my understanding of the prior_visit field was to handle the case where you have multiple visits in the same day (or even multiple visits in the same hour) and this field was allowing you to order them so you know if a complication arises between one visit and the next, you know which visit to attribute it to.

what is being described here (visits using the prior_visit field are used to ‘chain’ associated visits together) I thought was covered using the episode extension of the CDM?

(Christian Reich) #6

Exactly. Don’t use the ETL to build a specific research question.

Definitely there, but also to generally chain things together that are part of an overall healthcare experience.

That is a higher level umbrella for more than just Visits. It is to put together everything belonging to a Disease or Treatment. Like everything including concrete histologies, affected anatomical sites, lymph nodes, etc. belonging to Breast Cancer.

(Eldar Allakhverdiiev) #7

Thank you @Chris_Knoll,
Now it became clear for me purpose and logic of preceding_visit_occurrence_id.

The way you suggest to link visits by an observation looks fine and elegant. But could you please clarify what information you’d suggest to store in value_as_concept_id ?

(Chris Knoll) #8

I don’t have a specific recommendation for that (it’s something that a THEMIS convention would specify) I was just musing that just like we have other ‘higher-order’ concepts like ‘personal history’ that become observations, or ‘drug order vs. drug administration’, I feel we’re introducing another meta-concept of ‘readmissions’ where visits or observations have a additional context that they are related to some prior healthcare interaction. While I do see the value of this (I can easily imagine a type of study/analysis where the readmission for X is the outcome of interest), my worry is that if we bury a clinical observation under a re-admission observation, we now have to look in 2 places to find the event (the literal condition_occurrence vs. readmission observation + value_as_concept ID).

I recognize that I suggested the use of value_as concept in my previous post, but now I am thinking that I wouldn’t want to suggest a course where we now have 2 locations of clinical observations. I think if there were concepts for ‘readmission’ observations, and I was doing a study on readmissions with outcome, then the logic for defining the outcome would be 'condition occurrence of {outcome) with observation of ‘readmittance’ at same visit (or within some days, etc).That wouldn’t exactly say which prior event the re-admittance was for, but it does give a little context about ‘this person is being readmitted and isn’t just some random encounter of this person to the healthcare system’.

Disclosure: I’m not a health care provider and am only speaking from a context of data capture (also something I’m not an expert). My comments stem from how this data might be leveraged in tools. While I don’t think our clinical/analytical thinking should be limited by what the tools can perform, thinking about tools gives us a concrete examples on how we’d leverage the data. It’s not my intent to cast any doubt on any ideas expressed herein.