Developing a Machine Learning Model to Shorten Emergency Department Length of Stay: Model Testing and Nurses Acceptance

Abstract

This study aimed to develop and evaluate a portfolio of ML models to predict ED LoS and examine nurses’ acceptance of AI-based clinical decision support. Secondary data from two public hospitals in Jakarta were analysed using three ML algorithms Random Forest, Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) to classify ED LoS into short, medium, and prolonged categories. Predictor variables included triage level, arrival time, referral source, disposition, number of diagnostic tests, and consultations. Model performance was assessed using precision, recall, and F1-scores across training, testing, and blind validation datasets. Additionally, nurses’ readiness to adopt ML tools was evaluated using a survey. Across 687 ED cases, XGBoost achieved the best overall performance (precision, recall, and F1-score = 1.00), indicating excellent discrimination and balance between sensitivity and specificity. SVM also demonstrated strong external validation (blind-test F1 = 1.00), confirming robust generalisation across hospital sites. High-performance metrics across all models indicate consistent accuracy and calibration. Most nurses (89.3%) expressed high performance expectancy, and 95.7% high effort expectancy toward technology adoption. The developed ML framework accurately predicts ED LoS in Jakarta’s hospital settings, providing a foundation for data-driven resource management.

Ähnliche Arbeiten