Explainable machine-learning predictions for the prevention of hypoxaemia during surgery

Area of Science:

• Anesthesiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Intraoperative hypoxemia (low oxygen during surgery) is a significant risk for patients.
• Current methods for predicting hypoxemia during general anesthesia are unreliable.
• Developing predictive tools is crucial for improving patient safety and outcomes.

Purpose of the Study:

• To develop and evaluate a machine learning-based system for real-time prediction of intraoperative hypoxemia risk.
• To provide interpretable risk factors contributing to hypoxemia predictions.
• To assess the system's impact on anesthesiologists' ability to anticipate hypoxemia events.

Main Methods:

• Development of a machine learning system trained on minute-by-minute electronic medical record data from over 50,000 surgeries.
• Real-time risk prediction and explanation generation during general anesthesia.
• Evaluation of the system's performance and its effect on anesthesiologists' predictive capabilities.

Main Results:

• The machine learning system significantly improved anesthesiologists' ability to predict hypoxemia events, nearly doubling anticipation rates from 15% to 30%.
• The system provided interpretable explanations for hypoxemia risks, consistent with existing medical literature and expert knowledge.
• Identified modifiable factors associated with hypoxemia risk, suggesting potential for early intervention.

Conclusions:

• Machine learning offers a promising approach to real-time prediction and understanding of intraoperative hypoxemia.
• The developed system enhances anesthesiologists' awareness of hypoxemia risks and contributing factors.
• This technology has the potential to improve clinical decision-making and patient care during anesthesia.