An interpretable machine learning model for real-time sepsis prediction based on basic physiological indicators.

Area of Science:

• Critical Care Medicine
• Biomedical Informatics
• Machine Learning in Healthcare

Background:

• Sepsis prediction models are crucial for clinical diagnosis and treatment.
• Existing models often lack timeliness and interpretability.
• There is a need for real-time, clinically interpretable sepsis prediction tools.

Purpose of the Study:

• To develop a real-time sepsis prediction model with high timeliness and clinical interpretability.
• To address the limitations of current sepsis prediction methodologies.
• To enhance early warning systems for critically ill patients.

Main Methods:

• Utilized eight real-time physiological monitoring indicators (heart rate, respiratory rate, SpO2, MAP, SBP, DBP, temperature, blood glucose).
• Extracted three-hour dynamic feature sequences and calculated linear parameters (mean, standard deviation, endpoint value).
• Constructed a 24-dimensional feature vector and a real-time sepsis prediction model using Local Interpretable Model-Agnostic Explanation (LIME).

Main Results:

• Extremely Randomized Trees model achieved an AUROC above 0.76, outperforming other models.
• Imbalance XGBoost demonstrated high specificity (0.86) in sepsis prediction.
• LIME provided detailed prediction probabilities and feature influence, aiding clinical decision-making.

Conclusions:

• The developed model offers real-time dynamic early warnings for critically ill patients.
• It serves as a valuable reference for clinical decision support systems.
• Interpretive analysis enhances the credibility and clinical utility of sepsis prediction models.