692 Enhancing Burn Patient Discharge Predictions: Random Forest and Logistic Regression with Six-Clicks Oversampling

TL;DR

This study shows that adding the Six-Clicks tool to patient data improves predictions of burn patients' discharge outcomes, helping with better care planning.

Contribution

The study demonstrates that incorporating the Six-Clicks score into predictive models enhances discharge disposition accuracy for burn patients.

Findings

Including Six-Clicks scores in the model significantly improved classification across four discharge categories.

Group 4, which used Six-Clicks, outperformed Group 8 in predicting outcomes, especially for patients with higher mobility impairments.

Enhanced prediction accuracy can improve multidisciplinary care coordination and healthcare resource management.

Abstract

Predicting discharge disposition for burn patients with large total body surface area (TBSA) is essential to optimizing care and improving long-term outcomes. Patient recovery, expectations, and goal setting require early physician planning, but recovery is rarely based solely on TBSA. A multidisciplinary approach, paired with simple yet comprehensive assessment tools, can enhance patient care discussions and family preparedness. The Activity Measure for Post-Acute Care Six-Clicks, a functional screening tool used by physical therapists in hospital settings, has shown promise in predicting discharge disposition and can significantly improve treatment planning, resource allocation, and the overall patient experience in burn care.

Patient data collected to predict discharge disposition included demographics, TBSA, and Six-Clicks scores. The original dataset had 238 patients, which was expanded to 580 total samples using SMOTE to address class imbalance. Oversampling ensured uniform distribution across the four discharge disposition categories: home discharge, skilled nursing facilities/long-term care, transfers to another institution, and expired/hospice. Logistic regression and Random Forest models were trained using feature sets. Group 4 and Group 8 included predictors such as age, TBSA, intubation status, and inpatient surgeries, with Group 4 additionally incorporating the Six-Clicks scores to assess its predictive value. Models were evaluated using accuracy, ROC AUC, and confusion matrices.

Six-Clicks scores as a feature in Group 4 significantly improved the prediction of discharge disposition shown in the confusion matrix. It exhibited better classification across the four discharge categories, with more correct predictions than Group 8, particularly in categories representing higher mobility impairments.

Group 4 significantly improved model predictions, emphasizing that adding the Six-Clicks score to existing assessment criteria such as TBSA can better evaluate discharge disposition and improve treatment plan effectiveness.

Enhancing prediction accuracy has far-reaching impacts beyond the hospital. It improves the coordination and effectiveness of multidisciplinary teams by ensuring the appropriate level of care is provided. Including a tool such as AM-PAC Six-Clicks improves discharge planning accuracy, leading to more efficient management of healthcare resources and better patient outcomes.

No external funding was provided for this study.

Figures