Improving sepsis mortality prediction with machine learning: A comparative study of advanced classifiers and...

Area of Science:

• Computational medicine
• Artificial intelligence in healthcare
• Critical care research

Background:

• Sepsis presents a significant global health challenge with high mortality rates.
• Machine learning (ML) offers a promising avenue for enhancing the accuracy and timeliness of sepsis mortality prediction.
• Early prediction is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop a machine learning model for rapid and accurate sepsis mortality prediction.
• To focus on data obtainable within an ambulance setting for practical pre-hospital application.
• To compare ML model performance against the established quick Sequential Organ Failure Assessment (qSOFA) score.

Main Methods:

• Utilized the Medical Information Mart for Intensive Care-IV (MIMIC-IV) dataset.
• Compared 11 ML algorithms, including LightGBM, against the qSOFA score.
• Employed dynamic updating models, SHapley Additive exPlanations (SHAP) for feature importance, and AUC/PRAUC for performance evaluation.

Main Results:

• The LightGBM model achieved the highest performance metrics: AUC of 0.79 and PRAUC of 0.44.
• LightGBM outperformed the qSOFA score (AUC = 0.76, PRAUC = 0.40).
• Dynamically updated and tuned models further enhanced predictive performance, demonstrating practical utility.

Conclusions:

• A LightGBM-based ML model shows superior efficacy in predicting sepsis mortality in an ambulance setting.
• The study highlights the practical applicability of ML for pre-hospital sepsis management.
• Real-time updates and hyperparameter tuning are critical for optimizing ML model performance in clinical settings.