Predicting Discharge Destination of Critically Ill Patients Using Machine Learning.

Area of Science:

• Critical Care Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Discharge destination decisions for critically ill patients are complex, involving multiple factors and stakeholders.
• Accurate prediction of discharge disposition is vital for effective care planning and assessing functional outcomes.
• Existing methods for predicting discharge destination in intensive care units (ICUs) have limitations in scope and generalizability.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in predicting ICU patient discharge destination within 24 hours of admission.
• To compare the predictive power of the Acute Physiology and Chronic Health Evaluation (APACHE) IV score versus its constituent patient characteristics.

Main Methods:

• A retrospective study utilizing the eICU Collaborative Research Database (eICU-CRD) from 2014-2015.
• Development and testing of machine learning models on 115,248 adult ICU admissions to predict four discharge categories: death, home, nursing facility, and rehabilitation.
• Application of synthetic minority over-sampling techniques to address class imbalance and use of hierarchical and ensemble classifiers.

Main Results:

• The XGBoost model demonstrated superior discrimination, achieving an area under the receiver operating characteristic curve of 90% (recall: 71%, F1: 70%).
• Individual patient characteristics within the APACHE IV model proved more effective predictors of discharge destination than the overall APACHE IV score.
• The findings highlight the potential of machine learning for early and accurate discharge disposition prediction.

Conclusions:

• Machine learning models incorporating detailed patient characteristics outperform the APACHE IV score alone in predicting ICU discharge destinations.
• Integrating these predictive models into clinical decision support systems can facilitate timely disposition planning for patients, caregivers, and ICU teams.
• Early identification of discharge disposition can significantly improve care coordination and resource management in critical care settings.