Machine Learning Prediction of Discharge Destination in Patients with Parkinson's Disease; A Nationwide Cohort Study.

Background: Non-home discharge after hospitalization is common among patients with Parkinson's disease (PD) and is associated with adverse outcomes. Early identification of patients likely to require post-acute facility care may improve discharge planning. Methods: We conducted a retrospective cohort study using a nationwide healthcare database, including adults aged ≥50 years hospitalized with PD between November 2017 and June 2023. The first hospitalization of each patient was defined as the index admission. Discharge destination was categorized as home, facility, or in-hospital death. Data were split into training (80%) and testing (20%) sets. Random forest models were developed to predict discharge destination, and performance was evaluated using area under the receiver operating characteristic curve (AUC). An elastic net logistic regression model was additionally developed for facility discharge. Results: Among 281,664 index hospitalizations, 48.0% were discharged home, 44.8% to a facility, and 7.2% died in-hospital. In the test set, random forest models achieved AUCs of 0.775 for home discharge, 0.774 for facility discharge, and 0.832 for in-hospital death. An elastic net model achieved an AUC of 0.752, and a seven-item risk score (fracture history, dementia, transfer admission, fall history, marital status, insurance type, hospital region) identified a high-risk group with a 73.8% facility discharge rate, compared with 40.6% in the low-risk group. Conclusions: Using nationwide claims data, this disease-specific prediction model identified discharge destination in hospitalized patients with PD. A simplified, interpretable risk score enables early risk stratification at admission and may facilitate multidisciplinary discharge planning and post-acute care allocation.