Real time automatic risk prediction in ICU patients treated with ECMO

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Data Science

Background:

• The COVID-19 pandemic highlighted the critical role of intensive care units (ICUs) and advanced life support technologies.
• Extracorporeal membrane oxygenation (ECMO) is vital for severe respiratory failure, generating extensive patient data.
• Machine learning (ML) offers potential for analyzing complex ICU data to improve patient outcomes.

Purpose of the Study:

• To develop and evaluate ML models for predicting clinical deterioration and improvement in COVID-19 patients receiving ECMO.
• To identify an optimal model for real-time clinical assessment and early warning of significant patient status changes.

Main Methods:

• Utilized 81 labeled multivariate time series from COVID-19 pneumonia patients on ECMO.
• Applied Support Vector Machine (SVM) and Random Forest (RF) models to predict clinical changes.
• Assessed model performance using metrics like Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• The optimal Random Forest model demonstrated exceptional performance and calibration.
• The developed Clinical Assessment Score accurately predicted significant clinical deterioration and improvement.
• Achieved high AUROC values (0.9176, 0.8944, 0.8556) for 4, 8, and 12-hour prediction intervals, respectively.

Conclusions:

• Machine learning, particularly Random Forest, is effective for real-time patient monitoring in critical care.
• The Clinical Assessment Score can serve as a valuable tool for anticipating critical events in ECMO patients.
• This approach supports proactive clinical decision-making and potentially improves patient management in ICUs.