Decoding sepsis: A technical blueprint for an algorithm-driven system architecture.

Area of Science:

• Medical Informatics
• Machine Learning Operations
• Healthcare Systems Engineering

Background:

• Sepsis is a critical healthcare challenge with high mortality rates.
• Early detection and intervention are crucial for improving patient outcomes.
• Current methods often lack real-time capabilities, especially in emergency department (ED) settings.

Purpose of the Study:

• To present a scalable, serverless machine learning (ML) operations architecture for near real-time sepsis risk-stratification.
• To enable timely sepsis treatment by identifying high-risk patients in ED waiting rooms.
• To overcome limitations of unavailable pathology data in ED settings.

Main Methods:

• Integration of HL7 message processing via MuleSoft within Amazon Web Services (AWS).
• Utilization of AWS Lambda for real-time data processing and AWS SageMaker for ML model deployment.
• Development of an Extreme Gradient Boosting model, evaluated using Receiver Operating Characteristic (ROC) curves across different age groups.

Main Results:

• The system demonstrated high reliability with 99.7% HL7 message processing success.
• The Extreme Gradient Boosting model achieved an overall accuracy of 0.84 and an F1-score of 0.80.
• Strong performance was observed across age cohorts, with Area Under the Curve (AUC) values ranging from 0.806 to 0.867.

Conclusions:

• The developed architecture provides a robust solution for near real-time sepsis risk-stratification in ED waiting rooms.
• The system has the potential to significantly enhance early sepsis detection and intervention.
• Further optimization is identified for peak load scenarios and code set management.