Sepsis mortality prediction with the Quotient Basis Kernel.

Area of Science:

• Critical Care Medicine
• Biostatistics
• Machine Learning in Healthcare

Background:

• Sepsis is a life-threatening condition requiring accurate mortality prediction in intensive care units (ICUs).
• Existing scoring systems like SAPS and SOFA have limitations in predicting sepsis-related mortality.
• Improved prediction accuracy can enhance clinical decision-making and patient outcomes.

Purpose of the Study:

• To develop and evaluate a novel algorithm for assessing the risk of death in sepsis patients.
• To improve the accuracy, sensitivity, and specificity of mortality prediction compared to current methods.
• To integrate a new algorithm into clinical decision support systems for ICU use.

Main Methods:

• Utilized Simplified Acute Physiology Score (SAPS) and Sequential Organ Failure Assessment (SOFA) data for algorithm development.
• Developed novel kernels, including the Quotient Basis Kernel (QBK), based on linear algebra, geometry, and statistics.
• Employed soft-margin support vector machines for mortality prediction and compared kernels using Jensen-Shannon metric.

Main Results:

• The QBK achieved 80.18% accuracy, outperforming standard methods (71.32%-71.55%) and other kernels.
• QBK demonstrated superior sensitivity (79.34%) and specificity (83.24%) in mortality prediction.
• Cross-validation on 400 patients confirmed the favorable performance of the proposed methods.

Conclusions:

• The developed algorithm, particularly the QBK, offers a more effective approach to sepsis mortality risk assessment.
• The algorithm enhances the sensitivity and specificity of prediction, surpassing existing scoring systems.
• This method holds potential for improving clinical decision support in managing sepsis patients.