Machine Learning-Augmented Triage for Sepsis: Real-Time ICU Mortality Prediction Using SHAP-Explained Meta-Ensemble...

Area of Science:

• Artificial Intelligence in Medicine
• Machine Learning for Healthcare
• Computational Biology and Bioinformatics

Background:

• Sepsis poses a significant mortality risk, especially in intensive care units.
• Traditional scoring systems (qSOFA, SIRS, NEWS) lack precision for timely sepsis management.
• Optimization algorithms and metaheuristics offer advanced solutions for complex problems in healthcare.

Purpose of the Study:

• To develop a novel, interpretable machine learning framework for predicting in-hospital mortality in sepsis patients.
• To improve upon the diagnostic accuracy of existing clinical scoring systems for sepsis.
• To leverage bio-inspired optimization algorithms for enhanced predictive modeling in critical care.

Main Methods:

• A retrospective dataset of sepsis patients was analyzed, incorporating clinical and laboratory features.
• Synthetic Minority Oversampling Technique and imputation methods addressed data imbalance and missing values.
• A hybrid model combining ensemble ML and deep learning, optimized by the Red Piranha Optimization algorithm, was developed and validated.

Main Results:

• The proposed AI model achieved a high predictive performance with an area under the receiver operating characteristic curve of 0.96.
• The model demonstrated a Brier score of 0.118 and a recall of 81%, indicating superior accuracy.
• The AI framework significantly outperformed conventional scoring systems in predicting sepsis mortality.

Conclusions:

• AI-driven tools show significant potential for enhancing clinical decision-making in sepsis management.
• Early and accurate prediction of sepsis mortality can facilitate timely interventions.
• This research highlights the value of advanced machine learning and optimization techniques in critical care settings.