Early Length of Stay Prediction at Admission in Short-Stay Hospitals
Résumé
Predicting hospital length of stay (LOS) is crucial for optimizing resource allocation and improving patient care. Accurate estimation of a patient’s discharge date enhances operational efficiency and reduces stress on hospital staff. While many studies have addressed this challenge using classification and regression models that consider data from admission to discharge, this research focuses on predicting LOS using only the features available at admission, specifically demographic details and primary diagnosis data, in a case study of patients in surgical wards dealing with musculoskeletal and connective tissue conditions. Advanced machine learning techniques, including boosting algorithms (LightGBM, XGBoost, CatBoost), Random Forest, and Feedforward Neural Networks (FNN), were employed to achieve this. LightGBM, demonstrated the best performance in this context, achieving a precision R score of 0.77 and a mean squared error (MSE) of 1.07. These measures indicate the model’s ability to predict LOS with a high degree of accuracy while minimizing prediction errors. These results highlight the potential of machine learning, particularly LightGBM, to optimize hospital operations and enhance patient outcomes. Ultimately, the findings underscore the value of leveraging admission-based data for early and accurate LOS predictions, providing a foundation for more efficient resource planning in healthcare settings.
Domaines
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |