Explainable AI for Hospitalization Duration Predictions

Effective bed management is essential for reducing hospital expenses, improving operational efficiency, and enhancing patient care. This study introduces a predictive framework for ICU length of stay (LOS) at the time of admission, utilizing electronic health records (EHR). Our research applies supervised machine learning classification models to estimate ICU patients’ LOS within hospital clinical information systems (CIS). Notably, this work represents the first known application of explainable artificial intelligence (xAI) to real-world hospital stay data for interpretable machine learning predictions. We assessed predictive classification models using various performance metrics, including Accuracy, AUC, Sensitivity, Specificity, F1-score, Precision, Recall, and others, to classify ICU stays as short or long upon admission. XGBoost demonstrated a 98% AUC in predicting LOS categories. This study highlights how hospitals and ICUs can integrate machine learning to forecast patient stays at admission. Additionally, our findings enhance clinical information systems by incorporating xAI to ensure robust and interpretable LOS prediction models