Expert Augmented Prediction of Circulatory and Respiratory Instability from High Resolution Vital Signs

Area of Science:

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

Background:

• Existing early warning systems for circulatory and respiratory instability (CRI) in intensive care units (ICUs) have limitations.
• These systems often use single vital-sign parameters or delayed clinical data, hindering timely detection.
• There is a need for advanced systems that leverage high-frequency vital-sign data for improved accuracy and interpretability.

Purpose of the Study:

• To develop and validate an interpretable, expert-augmented early warning system (EAEWS) for detecting CRI in ICUs.
• To utilize high-frequency, 1-second resolution vital-sign data (heart rate, blood pressure, respiratory rate, oxygen saturation) for enhanced predictive performance.
• To improve the clinical interpretability of machine learning models through expert-augmented learning.

Main Methods:

• Developed machine learning models using over 627,000 hours of continuous vital-sign data from 1702 ICU patients.
• Incorporated trend-based, reference, and statistical features derived from vital-sign trajectories.
• Validated models internally and externally (MIMIC-III cohort), transforming tree-based models into physiologically meaningful decision rules refined by expert input.

Main Results:

• The developed models achieved strong predictive performance (Area Under the Receiver Operating Characteristic curve > 0.8) in both internal and external validation.
• Performance surpassed conventional single-parameter indices and was comparable to models using laboratory and demographic data.
• Trend-based features derived from high-resolution vital-sign data significantly improved predictive accuracy.

Conclusions:

• The Expert-Augmented Early Warning System (EAEWS) provides accurate, low-frequency alerts with transparent explanations for CRI detection in ICUs.
• The system effectively leverages high-frequency vital-sign data and expert knowledge for improved clinical decision support.
• EAEWS offers a scalable framework for real-time CRI detection, potentially enhancing patient outcomes in critical care settings.