Conformance-Aware Predictive Process Monitoring for Early Detection of Sepsis Deterioration Using Incomplete Care...

Area of Science:

• Healthcare Informatics
• Clinical Decision Support Systems
• Process Mining

Background:

• Sepsis is a critical condition with high mortality, often due to delayed detection and variable care.
• Current predictive models for sepsis risk lack temporal and process-related factors, overlooking care pathway inefficiencies.
• Early prediction of sepsis deterioration requires incorporating deviations from standard care protocols.

Purpose of the Study:

• To propose a Conformance-Aware Predictive Process Monitoring (CAPPM) framework for early sepsis deterioration detection.
• To leverage incomplete care pathways for improved predictive accuracy.
• To integrate process mining techniques with machine learning for enhanced clinical decision support.

Main Methods:

• Discovered the reference care pathway from the Sepsis Cases Event Log.
• Engineered temporal, behavioral, and conformance-based features from ongoing patient cases.
• Trained and evaluated supervised learning models (Adaptive Boosting, Gradient Boosting) using these features, assessing performance via AUROC.

Main Results:

• Models incorporating conformance and pathway features outperformed those using only traditional attributes.
• Adaptive Boosting achieved an AUROC of 0.744, and Gradient Boosting achieved 0.731.
• These results demonstrate enhanced early detection capabilities for sepsis deterioration.

Conclusions:

• Early deviations in patient care pathways and temporal progression are significant predictors of sepsis deterioration.
• The integration of process mining and machine learning offers a powerful approach for proactive clinical interventions.
• The CAPPM framework shows promise for time-critical clinical decision support in sepsis management.