Exploring Features Contributing to the Early Prediction of Sepsis Using Machine Learning.

Area of Science:

• Medical Informatics
• Computational Biology
• Health Data Science

Background:

• Sepsis is a life-threatening condition requiring timely diagnosis.
• Electronic health records and healthcare technologies are advancing medical informatics.
• Early sepsis detection improves patient survival and outcomes.

Purpose of the Study:

• To explore variables associated with sepsis development using SHAP analysis.
• To evaluate supervised learning models for sepsis classification.
• To assess feature contributions at different admission time points.

Main Methods:

• Utilized electronic health records and administrative data.
• Applied SHapley Additive exPlanations (SHAP) for variable importance.
• Developed and evaluated supervised learning models using data from hour 1 and hour 5.
• Analyzed feature contributions at different time intervals.

Main Results:

• Missing data in early admission stages can be effectively used for sepsis prediction.
• SHAP analysis identified key predictors for sepsis development.
• Significant inconsistency in contributing features was observed between early and later admission stages.

Conclusions:

• Machine learning models can effectively predict sepsis early, even with incomplete initial data.
• Understanding feature importance dynamics is crucial for robust sepsis prediction models.
• The findings support the integration of advanced analytics in critical care for sepsis management.