A machine learning algorithm for detecting abnormal patterns in continuous capnography and pulse oximetry monitoring

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Respiratory Physiology

Background:

• Continuous capnography is vital for monitoring patient ventilation but prone to artifacts causing alarm fatigue.
• Smart algorithms are needed to accurately detect abnormal ventilation, enabling timely intervention.
• Distinguishing true ventilation abnormalities from artifacts is crucial for effective patient monitoring.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for classifying combined capnography and pulse oximetry waveforms.
• To differentiate between normal and abnormal ventilation patterns using multimodal physiological data.
• To enhance the accuracy of respiratory monitoring systems and reduce false alarms.

Main Methods:

• Utilized data from the prospective PRODIGY trial, combining capnography and pulse oximetry.
• Expert review established ground truth for abnormal ventilation segments (60s before, 30s after event).
• Trained five ML models on extracted features, optimizing for an Fβ score (β=2), with XGBoost showing the highest performance.

Main Results:

• Achieved high inter-rater agreement (>87%) among experts, validating 7,858 sequences (2,944 abnormal).
• The optimized XGBoost model achieved an Fβ score of 0.94, with a recall of 0.98 and precision of 0.83.
• Demonstrated the algorithm's reliability in distinguishing normal from abnormal respiratory waveforms.

Conclusions:

• Machine learning offers a promising approach to improve respiratory monitoring accuracy.
• The developed algorithm effectively distinguishes normal from abnormal ventilation waveforms, potentially reducing alarm fatigue.
• Further research is required to differentiate artifactual signals from true abnormal ventilation patterns.