Interpretable Machine Learning Model for Predicting Sepsis and Septic Shock Among Patients with Documented Fever at...

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Machine Learning in Healthcare

Background:

• Sepsis and septic shock are critical conditions requiring early detection.
• Existing scoring systems like qSOFA and MEWS have limitations in predicting sepsis at emergency department triage.
• There is a need for improved, interpretable tools for early sepsis identification.

Purpose of the Study:

• To develop and validate an interpretable machine learning-based scoring system for predicting sepsis and septic shock in febrile patients at ED triage.
• To compare the performance of the novel scoring system against established scores (qSOFA, MEWS).

Main Methods:

• Retrospective, single-center study of adult febrile patients in the ED (January 2016 - December 2021).
• Development of a novel scoring system using the AutoScore framework, incorporating nine key variables.
• Performance evaluation using Area Under the Receiver Operating Characteristic Curve (AUROC), comparing with qSOFA and MEWS.

Main Results:

• The developed scoring system demonstrated strong predictive performance: AUROC of 0.844 for septic shock and 0.703 for sepsis.
• The novel score significantly outperformed qSOFA (AUROC: 0.605) and MEWS (AUROC: 0.678) in predicting septic shock.
• For equivalent specificity, the new system showed higher sensitivity than MEWS.

Conclusions:

• The developed machine learning scoring system is an interpretable and practical tool for predicting sepsis and septic shock at ED triage.
• This novel system offers superior predictive capabilities compared to qSOFA and MEWS.
• Early and accurate identification of sepsis using this tool can potentially improve patient outcomes.