Survival in the Intensive Care Unit: A prognosis model based on Bayesian classifiers

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Decision Support Systems

Background:

• Intensive Care Units (ICUs) require accurate prognostic models for patient management.
• Early mortality prediction and risk stratification are critical for efficient clinical decision-making.
• Existing models may lack the precision or confidence assessment needed for complex ICU environments.

Purpose of the Study:

• To develop and evaluate a novel machine learning model for enhanced early mortality prediction in ICUs.
• To improve clinical decision-making for high-risk patients and treatment effectiveness evaluation.
• To provide a prognostic tool that offers confidence levels for its predictions.

Main Methods:

• Development of a hierarchical machine learning model named Ensemble Weighted Average (EWA).
• EWA is an ensemble of five Bayesian classifiers using a weighted average criterion.
• The model was trained and validated on a real-world ICU patient cohort.

Main Results:

• The EWA model demonstrated superior performance compared to state-of-the-art machine learning predictive models.
• EWA provides an advantage over majority vote ensembles by offering associated confidence levels.
• Local recalibration of the APACHE II score model was beneficial but yielded a weaker predictive model than EWA.

Conclusions:

• The Ensemble Weighted Average (EWA) model offers a significant advancement in early mortality prediction within ICUs.
• EWA's ability to provide confidence levels enhances its utility for clinical decision support.
• The study highlights the potential of advanced machine learning techniques to improve critical care outcomes.