Predicting Discharge Destination From Inpatient Rehabilitation Using Machine Learning.

Area of Science:

• Rehabilitation Medicine
• Health Informatics
• Artificial Intelligence in Healthcare

Background:

• Predicting patient discharge destination from inpatient rehabilitation is crucial for care coordination and efficient healthcare resource use.
• Previous studies have explored factors influencing discharge disposition, but advanced predictive modeling remains an area for development.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting discharge disposition (home vs. nonhome) for patients in inpatient rehabilitation.
• To identify key patient and encounter-specific factors associated with home versus nonhome discharges.

Main Methods:

• Utilized a dataset of 4922 patient encounters from 4401 patients at a Midwestern academic inpatient rehabilitation facility.
• Tested fifteen distinct machine learning models, including a bagging classifier with a decision tree base, employing random undersampling.
• Input variables comprised demographic and hospital encounter-specific data.

Main Results:

• The best-performing model, a bagging classifier, achieved an area under the receiver operating characteristic curve of 0.722.
• Shapley value analysis identified length of stay, intravenous medication, urinary dysfunction, age, white blood cell count abnormalities, plasma sodium levels, and fatigue as significant predictors.
• The dataset included 3687 home discharges and 1235 nonhome discharges.

Conclusions:

• Machine learning models demonstrate efficacy in predicting inpatient rehabilitation discharge disposition.
• The study successfully identified critical factors influencing whether patients are discharged to home or nonhome settings.
• Findings can inform clinical decision-making and resource allocation to optimize patient transitions from rehabilitation.