Two-Step Approach for Occupancy Estimation in Intensive Care Units Based on Bayesian Optimization Techniques.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Analysis

Background:

• Intensive care units (ICUs) face high occupational pressure, necessitating accurate patient length of stay (LoS) predictions.
• Effective LoS estimation aids in resource management, treatment planning, and patient outcome prediction.
• Biomedical sensor data from ICUs offers a rich source for developing predictive models.

Purpose of the Study:

• To develop a novel methodology for estimating patient LoS in ICUs using data from the initial 24 hours.
• To optimize the XGBoost algorithm for improved precision and computational efficiency in LoS prediction.
• To validate the proposed methodology against existing state-of-the-art approaches.

Main Methods:

• Utilized the XGBoost algorithm as the core estimator for LoS prediction.
• Implemented a novel two-step Bayesian optimization approach to fine-tune XGBoost hyperparameters.
• Designed the methodology for high-performance computing using graphics processing units (GPUs) to reduce execution time.
• Analyzed the scalability of the developed algorithm.

Main Results:

• Identified an optimal set of XGBoost hyperparameters for LoS estimation.
• Achieved a Mean Absolute Error (MAE) of 2.529 days for LoS prediction.
• Demonstrated improved performance compared to current state-of-the-art methods.
• Validated the utility and accuracy of the proposed approach.

Conclusions:

• The proposed AI-driven methodology effectively estimates patient LoS in ICUs using early clinical data.
• Bayesian optimization significantly enhances the performance of XGBoost for this critical healthcare application.
• The approach offers a valuable tool for improving ICU management and patient care.