Readmission Times in eICU

[supatuffpinkpuff]

Our initial issue, e-mailed to Alistair Johnson and Tom Pollard:

We are currently working on a project studying ICU readmissions, using the eICU database. Through our exploratory analysis, we found that a large number of our ICU readmissions seem to have 0 minutes between their ICU stays. More specifically, out of the entire ~200,000 ICU admissions in eICU dataset, we found 28,242 ICU readmissions within 48 hours of the original admission in the eICU dataset, of which 23,473 (83.1%) that have 0 minutes between the discharge time of the first ICU stay, and the admission time of the next ICU stay (which we’ll call the “readmission time”).

Among these 23,473 stays that have 0 minutes between their ICU stays, their original stay discharged to: • “Step-Down Unit (SDU)” - 17,820 (76%) • “Other ICU”or ”ICU” - 5,482 (23%) • All other locations - 171 (<1%)

We further examined these 0 readmission times, and found that hospitals had varying proportions of 0 readmission times. In 10 hospitals, all ICU readmissions had 0 readmission times, and in 50 hospitals, no ICU readmissions had 0 readmission times. A full list of all hospital IDs and the proportion of readmissions that had 0 readmission times is attached (Hospital_0_Admission_Times_Proportions.xlsx).

We were wondering what these 0 readmission times mean, given their high prevalence. We calculated readmission times from the columns unitDischargeOffset and hospitalAdmitOffset, which were provided in the Patient table. Exact methodology can be found in the attached document (Calculating Time Between ICU Stays.docx)

From a clinical standpoint, 0 minutes between subsequent ICU stays is difficult to interpret literally. We are wondering what could have led to these values. With that in mind, how were the Patient columns “unitDischargeOffset” and “hospitalAdmitOffset” originally obtained in the data set?

First response from Alistair:

I highly recommend posting this question on the eICU GitHub so we may discuss it publicly. In short, eICU contains non-ICU stays (step down wards), which look to be the majority of your "readmissions".

Response:

To clarify, does that mean some patientunitstayids actually correspond to step-down unit stays, as opposed to ICU unit stays?

And how can we identify the step-down unit stays, vs. ICU unit stays? We did check unitType, but for these stays we’ve been looking at, they don’t appear to list “SDU/Step down” as the next unit type. unitStayType looks to be another alternative, although admittedly I haven’t checked that yet.

[alistairewj]

@jraffa maybe you have some insight into how to classify an ICU stay vs. a non-ICU stay

[obadawi]

This is a good analysis of the data and shows the importance of really understanding the data source. I'll try to explain why you're seeing these patterns...

First, it's important to remember that this database records patient stays in a transactional manner so every unique stay in a different ward or unit generates a new patient unit stay ID. The other thing that is unique is that a patient can be in a specific ICU bed but have their status changed from an ICU patient to a "step down" patient if their care is downgraded. When this happens, the system effectively ends the first unit stay and starts a new unit stay, leading to two stays with zero minutes between them. In the versions of software used for this database, you will find these subsequent stays are associated with an admission source of "ICU to SDU". So when you see this pattern, you can interpret it as a patient remained in the same ICU bed but had their care downgraded to step down status.

Hope that helps.

On Fri, Oct 23, 2020 at 10:37 AM supatuffpinkpuff [email protected] wrote:

Our initial issue, e-mailed to Alistair Johnson and Tom Pollard:

We are currently working on a project studying ICU readmissions, using the eICU database. Through our exploratory analysis, we found that a large number of our ICU readmissions seem to have 0 minutes between their ICU stays. More specifically, out of the entire ~200,000 ICU admissions in eICU dataset, we found 28,242 ICU readmissions within 48 hours of the original admission in the eICU dataset, of which 23,473 (83.1%) that have 0 minutes between the discharge time of the first ICU stay, and the admission time of the next ICU stay (which we’ll call the “readmission time”).

Among these 23,473 stays that have 0 minutes between their ICU stays, their original stay discharged to: • “Step-Down Unit (SDU)” - 17,820 (76%) • “Other ICU”or ”ICU” - 5,482 (23%) • All other locations - 171 (<1%)

We further examined these 0 readmission times, and found that hospitals had varying proportions of 0 readmission times. In 10 hospitals, all ICU readmissions had 0 readmission times, and in 50 hospitals, no ICU readmissions had 0 readmission times. A full list of all hospital IDs and the proportion of readmissions that had 0 readmission times is attached (Hospital_0_Admission_Times_Proportions.xlsx).

We were wondering what these 0 readmission times mean, given their high prevalence. We calculated readmission times from the columns unitDischargeOffset and hospitalAdmitOffset, which were provided in the Patient table. Exact methodology can be found in the attached document (Calculating Time Between ICU Stays.docx)

From a clinical standpoint, 0 minutes between subsequent ICU stays is difficult to interpret literally. We are wondering what could have led to these values. With that in mind, how were the Patient columns “unitDischargeOffset” and “hospitalAdmitOffset” originally obtained in the data set?

First response from Alistair: I highly recommend posting this question on the eICU GitHub so we may discuss it publicly. In short, eICU contains non-ICU stays (step down wards), which look to be the majority of your "readmissions".

Response: To clarify, does that mean some patientunitstayids actually correspond to step-down unit stays, as opposed to ICU unit stays?

And how can we identify the step-down unit stays, vs. ICU unit stays? We did check unitType, but for these stays we’ve been looking at, they don’t appear to list “SDU/Step down” as the next unit type. unitStayType looks to be another alternative, although admittedly I haven’t checked that yet.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/MIT-LCP/eicu-code/issues/149, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD6USVC64C36YRM3IMJLO4DSMGIKJANCNFSM4S4USJHQ .

[jraffa]

I was in the middle of writing my e-mail and Omar said most of what I was going to say, so listen to him. Below might help too.

I don't think there's a super easy answer for this, but yes, some patientunitstayids will be SDU stays. I have a couple of other suggestions:

1. I would say the majority of stays with no discernible time gap between the discharge of the first stay and the admission of the next stay are more likely to be 'transfers'. In many cases the first ICU stay may be very short (e.g., 0 minutes) and is purely there for administrative reasons (i.e., patient must have been admitted to ICU before SDU).

2. The variation in the % of hospitals with what you call 0/non-0 readmissions is likely representing those hospitals with and without step-down units (or alternatively those with the above 'ICU before SDU' policy).

[supatuffpinkpuff]

Thank you all for your help. I'll try to investigate ways to tell apart SDU stays from ICU stays.

[supatuffpinkpuff]

We ended up using the "UnitStayType" column from the Patient table, which seems to identify which stays are actually SDU stays, and most cleanly resolved the issues we were seeing. After no longer counting patientunitstayids that had "stepdown/other" as their UnitStayType, the number of readmissions that that had a 0-minute readmission time was reduced to about 3%.