EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

1825. Predicting delayed discharge from hospitals: A Decision support system using machine learning

Invited abstract in session MA-15: Machine learning and game theory in healthcare, stream OR in Health Services (ORAHS).

Monday, 8:30-10:00
Room: 18 (building: 116)

Authors (first author is the speaker)

1. Majid Taghavi
Sobey School of Business, Saint Mary's University
2. Mahsa Pahlevani
Dalhousie University
3. Peter Vanberkel
5269 Morris Street, Room 201, Dalhousie University

Abstract

The increasing demand for healthcare services poses significant challenges in effectively managing patient flow, particularly concerning patients classified as Alternative Level of Care (ALC). These patients, although no longer in need of acute care, often encounter obstacles to discharge and cause several issues such as hospital overcrowding and compromised health outcomes. This study uses administrative health data from Canadian hospitals and proposes using machine learning models to identify potential ALC patients and estimate their hospital length of stay as early as their admission time. The findings show the efficacy of the eXtreme Gradient Boosting algorithm in accurately predicting potential ALC patients, while the Random Forest regression model surpasses others in forecasting the length of stay for ALC patients. To understand how the predictions are made from the features of the dataset, the Shapley values were analyzed and used to identify the most important features of the dataset and their impact on machine learning outcomes. Using the most important features, two sets of user-friendly and easy-to-follow guidelines were developed for the hospital staff to proactively identify the ALC patients and estimate their length of stay, and mitigate the patient flow challenges posed by the ALC patients. Hospitals can use such decision-making tools to optimize resource allocation, enhance operational efficiency, and ultimately improve patient care outcomes.

Keywords

Status: accepted

Back to the list of papers