Development and validation of QuadraSynth ML: a novel classifier for sepsis prediction

Area of Science:

• Biomarkers and diagnostics
• Machine learning in medicine
• Trauma and critical care

Background:

• Sepsis is a life-threatening complication following trauma, necessitating early detection.
• The role of RalA in inflammatory diseases is known, but its specific link to sepsis requires elucidation.
• Developing effective biomarkers and predictive models is vital for managing traumatic sepsis.

Purpose of the Study:

• To identify novel biomarkers for early post-traumatic sepsis detection.
• To develop and validate a composite early-warning model for traumatic sepsis.
• To assess the potential of RalA as a biomarker for early sepsis diagnosis.

Main Methods:

• Collected laboratory variables and plasma RalA expression from trauma patients within 24 hours of admission.
• Utilized Boruta, LASSO, and logistic regression to identify key features for sepsis prediction.
• Developed a predictive model (QuadraSynth ML) using machine learning classifiers and validated it across multiple cohorts.

Main Results:

• The QuadraSynth ML model achieved an AUC of 0.902 in the primary cohort.
• Internal and external validation cohorts showed favorable AUCs of 0.836 and 0.840, respectively.
• Admission RalA levels showed promise as an early sepsis detection biomarker.

Conclusions:

• Admission RalA levels show potential for early sepsis detection in trauma patients.
• The QuadraSynth ML model demonstrates significant ability in early identification of sepsis.
• Further prospective validation is recommended for RalA and the predictive model.