Decision Tree in hospital discharge delays

[chuie]

Hi There,

Does anyone has developed a decision tree on a hospital discharge delays. The goal is to discharge a patient within two hours of discharge order written date/time. I need to develop a decision tree on factor affecting a prolonged discharge delay ( > 2h vs <2 hours).

Is there anyone who can navigate me thru  their experience/result that could be shared?

Really appreciate  your valuable comments

Thanks

C

[Reeza]

Do you have data that you can use to determine this? Usually delays relate to possibly waiting for a ride or discharge to family member, or a resident/nurse is late that day, or if there's an emergency in another area of the hospital, or if the transfer is late. If you don't have data on those type of events not sure how you'd model it. You do need to talk to the business areas to see what you should include in your model and what they think is the biggest issue.

[chuie]

Hi Reeza
Yes I do have data with following variables.
Basically we are trying to find what are the factors influencing discharge
delays and which node of DT has most significance impact on discharge
delays.
Will you please guide if the DT is appropriate method to figure out where
the bottle necks are for the discharge delays?

Facility
Account Number
Patient Type - Patient Indicator
Patient Type
Admit Date
Admit Time
Disc Date
Disc Time
Nursing Station
Patient Encounter - Medical Surgical Flag
Attending Physician
Primary Insurance group
Primary Insurance
Financial Class
Reporting DRG - Code
Reporting DRG - Description
Age
Discharge Disposition
Group Name
Room Number
Patient Name
DOB
Disposition
LOS
GMLOS
Reporting DRG Weight
Opportunity Days
Month
Week
DT
OPDay
OPdayRange
Opp.Index
ObsHrs
48HRSFlag
DOWAdmit
DishTime
DischOrdDT

[Reeza]

I don’t know what all those variables mean, but I suspect you do not have the data needed. If you want to use the data you have you can use a decision tree and treat outcome as either a binary outcome or as a categorical variable if you bin the delay into groups. 

Otherwise, you can also treat it as a continuous variable to look at length of delays or even a two stage model - probability of a delay, then if there is a delay, how long can it be. 

Note that that your data should have a portion that has delays and  no delays to ensure you’re getting actionable results. You may want to do a case control matching to ensure you a good ratio of events to non-events. 

---

@chuie wrote:
Hi Reeza
Yes I do have data with following variables.
Basically we are trying to find what are the factors influencing discharge
delays and which node of DT has most significance impact on discharge
delays.
Will you please guide if the DT is appropriate method to figure out where
the bottle necks are for the discharge delays?

Facility
Account Number
Patient Type - Patient Indicator
Patient Type
Admit Date
Admit Time
Disc Date
Disc Time
Nursing Station
Patient Encounter - Medical Surgical Flag
Attending Physician
Primary Insurance group
Primary Insurance
Financial Class
Reporting DRG - Code
Reporting DRG - Description
Age
Discharge Disposition
Group Name
Room Number
Patient Name
DOB
Disposition
LOS
GMLOS
Reporting DRG Weight
Opportunity Days
Month
Week
DT
OPDay
OPdayRange
Opp.Index
ObsHrs
48HRSFlag
DOWAdmit
DishTime
DischOrdDT

---

[chuie]

Thank you Reeza

My outcome variable is the time between physician's discharge Order to Patient discharge time. The goal is have it less than 2 hours. So I will create a new binary variable where if the time between  physician's discharge Order time  to Patient discharge time is less than 2 hours vs more than 2 hours.

My independent variable  would be nursing unit, age , DRG, insurance etc.

I have some understanding but not as expert as you guys are. I am looking  for some sample, articles on how it might looks like  and showcase this to my director so that they might buy me a sas visual statistics and miner.

Please share if you have any thing related to these.

Thanks

c

[Reeza]

I don't actually use Visual Statistics, though I do have SAS VA. So not sure how that would work. I'm fairly certain that SAS has some healthcare demos in SAS VA somewhere though....