A boosting inspired personalized threshold method for sepsis screening.

Area of Science:

• Critical Care Medicine
• Biomedical Informatics
• Clinical Decision Support

Background:

• Sepsis poses a significant patient safety risk with high mortality (25-50%) and diagnostic challenges.
• Current diagnostic standards are lacking, and the quick sequential organ failure assessment (qSOFA) shows low specificity in intensive care units (ICUs).

Purpose of the Study:

• To develop and evaluate a personalized threshold method for qSOFA using patient baseline characteristics.
• To compare the performance of this personalized method against standard qSOFA, existing single-biomarker methods, and machine learning algorithms.

Main Methods:

• A personalized threshold approach for qSOFA was developed, incorporating easily measurable patient baseline data.
• Performance was evaluated using data from the MIMIC-III database, comparing against qSOFA, five published methods, and logistic regression/AdaBoosting machine learning models.

Main Results:

• The personalized threshold method demonstrated higher accuracy than standard qSOFA and the five published methods.
• Its performance was comparable to complex machine learning algorithms in sepsis detection.

Conclusions:

• Personalized qSOFA thresholds offer a more accurate and practical approach to sepsis diagnosis compared to existing methods.
• This method is significantly easier to implement and interpret in real-world clinical settings than machine learning, facilitating wider adoption.