Class-Modeling of Septic Shock With Hyperdimensional Computing

Area of Science:

• Medical Informatics
• Computational Biology
• Critical Care Medicine

Background:

• Sepsis is a life-threatening condition where the immune system overreacts to infection, potentially leading to septic shock and organ failure.
• Early detection of septic shock is vital for timely treatment but remains challenging due to symptom variability.
• Modeling patient stability changes offers a promising approach for identifying the acute transition to septic shock.

Purpose of the Study:

• To develop and evaluate a novel one-class classification model for early septic shock detection.
• To utilize hyperdimensional computing for robust patient state modeling in critical care.
• To create adaptable models that can prioritize sensitivity or specificity based on clinical needs.

Main Methods:

• Implementation of a one-class classification framework using hyperdimensional computing.
• Development of multiple models considering diverse patient data contexts.
• Adaptation of models to dynamically adjust predictions and prioritize detection outcomes.

Main Results:

• The developed models achieved 90% sensitivity in detecting septic shock among septic patients.
• Accurate predictions were made up to three hours prior to the onset of septic shock in 60% of cases.
• Models demonstrated flexibility in adjusting predictions and prioritizing sensitivity or specificity.

Conclusions:

• Hyperdimensional computing provides an effective approach for early septic shock detection.
• The proposed models offer accurate and adaptable tools for critical care clinicians.
• Early prediction of septic shock can significantly improve patient management and outcomes.