A computational approach to early sepsis detection.

Area of Science:

• Critical Care Medicine
• Biomedical Informatics
• Machine Learning in Healthcare

Background:

• Sepsis is a life-threatening condition requiring early detection for effective treatment.
• Current sepsis prediction methods often lack the sensitivity and specificity for timely intervention.

Purpose of the Study:

• To develop and validate a high-performance early sepsis prediction technology for the general patient population.
• To create an algorithm capable of predicting sepsis onset at least three hours in advance.

Main Methods:

• Retrospective analysis of adult intensive care unit patients from the MIMIC II dataset.
• Development of the InSight sepsis early warning algorithm using vital signs.
• Validation of the algorithm's predictive performance against Systemic Inflammatory Response Syndrome (SIRS) episodes.

Main Results:

• The InSight algorithm achieved a sensitivity of 0.90 and specificity of 0.81 in predicting sepsis up to three hours prior to a five-hour SIRS episode.
• The algorithm demonstrated an average area under the ROC curve of 0.83 for predictions up to three hours before SIRS.
• Coevolution of multiple risk factors was found to be more critical for advanced sepsis prediction than isolated factors.

Conclusions:

• Sepsis can be predicted at least three hours in advance using commonly available vital signs with the InSight algorithm.
• The algorithm's performance surpasses current standard practice methods for early sepsis identification.
• High-order correlations of vital sign measurements are crucial for improving early sepsis prediction accuracy.