Improving Mechanical Ventilator Clinical Decision Support Systems with a Machine Learning Classifier for Determining...

Area of Science:

• Biomedical Informatics
• Critical Care Medicine
• Machine Learning

Background:

• Clinical decision support systems (CDSS) are crucial for improving care quality in critical care settings.
• Current CDSS often lack comprehensive data from physiologic monitoring devices, limiting their effectiveness.
• Specifically, mechanical ventilation (MV) CDSS lack knowledge of critical settings like ventilation mode.

Purpose of the Study:

• To develop a machine learning model for accurate per-breath classification of mechanical ventilation modes.
• To enhance the capabilities of CDSS for critically ill patients on mechanical ventilation.
• To address data limitations in current informatics infrastructure for MV.

Main Methods:

• Developed a high-performance machine learning model for classifying ventilation modes.
• The model performs per-breath classification of five common US ventilation modes.
• Evaluated model robustness against missing data due to software or sensor errors.

Main Results:

• Achieved an average F1-score of 97.52% for classifying ventilation modes.
• Demonstrated methodologic improvements over previous approaches in the field.
• Showcased high robustness to missing data, indicating reliability in real-world scenarios.

Conclusions:

• The developed machine learning model enables accurate identification of mechanical ventilation modes.
• This technology can significantly improve the data available to CDSS for mechanical ventilation.
• The approach offers a robust solution for enhancing critical care decision support, even with data imperfections.