An approach to rapidly assess sepsis through multi-biomarker host response using machine learning algorithm.

Area of Science:

• Biomedical Science
• Computational Biology
• Immunology

Background:

• Sepsis is a life-threatening condition requiring accurate patient stratification.
• Current sepsis endotyping lacks accuracy, hindering timely clinical decisions and management.
• Understanding host immune response is critical for effective sepsis patient stratification.

Purpose of the Study:

• To demonstrate the feasibility of a data-driven model for predicting sepsis host immune response.
• To support clinical decision-making through AI-driven patient stratification.
• To validate the predictive potential of combining multiple biomarker measurements.

Main Methods:

• Utilized a machine learning approach to analyze host immune response biomarkers.
• Combined measurements of cytokines and chemokines: IL-6, IL-8, IL-10, IP-10, and TRAIL.
• Employed supervised learning algorithms (naïve Bayes, decision tree) on a 70% test dataset.

Main Results:

• Achieved high accuracy in predicting sepsis host immune response.
• Naïve Bayes algorithm demonstrated 96.64% accuracy.
• Decision tree algorithm achieved 94.64% accuracy.

Conclusions:

• The proposed AI approach shows significant potential for clinical decision support in sepsis.
• This method offers a valuable testing resource for improving sepsis patient stratification.
• Data-driven validation models can enhance the management of sepsis through accurate endotyping.