Random Survival Forests for Predicting the Bed Occupancy in the Intensive Care Unit

Area of Science:

• Medical Informatics
• Critical Care Medicine
• Machine Learning

Background:

• Intensive care unit (ICU) bed occupancy prediction is crucial for resource management.
• Uncertainty in patient prognosis and random admissions complicate capacity planning.
• Reactive measures are often necessary due to unpredictable ICU bed demand.

Purpose of the Study:

• To develop a predictive model for ICU bed occupancy using Random Survival Forests.
• To assess the model's performance against baseline and standard Random Forest regression.
• To evaluate the efficacy of group-based versus individual patient length-of-stay assessments.

Main Methods:

• Utilized Sequential Organ Failure Assessment (SOFA) scores from 4098 ICU patients over four years.
• Developed a predictive model employing Random Survival Forests.
• Compared Random Survival Forest performance with baseline and Random Forest regression models.

Main Results:

• Random Survival Forests demonstrated effectiveness in predicting ICU bed occupancy.
• The group-based approach of Random Survival Forests outperformed individual patient length-of-stay assessments.
• The developed model shows promise in assisting physicians with occupancy estimations.

Conclusions:

• Random Survival Forests offer a robust solution for ICU bed occupancy prediction.
• Group-based predictive modeling is superior to individual assessments for ICU capacity planning.
• This approach can aid in optimizing ICU resource allocation and management.