Predicting Prolonged Hospital Length of Stay in Trauma Patients Using Machine Learning Techniques: A Cross-Sectional Study.

Background and Aim: Hospitalization due to trauma places a significant financial burden on healthcare systems, patients, and insurers. Predicting the length of stay can support resource management and workflow, ultimately enhancing healthcare interventions. This study uses machine learning techniques to predict trauma patients' hospital length of stay (LOS). Methods: This retrospective study analyzed data from 795 trauma patients registered at Jahrom University of Medical Sciences between March 21, 2021, and December 14, 2022. Data preprocessing, modeling, and evaluation were performed using Python. Seven machine learning algorithms-Support Vector Machine, K-Nearest Neighbors, Random Forest, Adaptive Boosting, Decision Tree, Artificial Neural Network, and Extreme Gradient Boosting-were applied for prediction. The models were compared using evaluation metrics: accuracy, precision, recall (sensitivity), F-measure, and the Area Under the Receiver Operating Characteristic (ROC) curve. Results: The Random Forest and Extreme Gradient Boosting algorithms demonstrated the best performance, with accuracy, precision, and recall of 93%. The Decision Tree algorithm achieved the highest area under the ROC curve (0.74). Key predictive features included oxygen saturation level, Glasgow Coma Scale, ICU stay duration, injury severity score, Abbreviated Injury Scale, and comorbid conditions. Conclusion: Among the tested algorithms, Decision Tree, Extreme Gradient Boosting, and Random Forest exhibited superior predictive performance. These models can support better resource allocation, policy-making, and healthcare planning, ultimately improving hospital efficiency and patient care.