Characterizing drivers of change in intraoperative cerebral saturation using supervised machine learning.

Area of Science:

• Anesthesiology
• Neuroscience
• Medical Informatics

Background:

• Regional cerebral oxygen saturation (rSO2) monitoring is crucial for assessing cerebral perfusion.
• Optimizing rSO2 may enhance neurological and non-neurological patient outcomes.
• Understanding factors influencing rSO2 is essential for effective intraoperative management.

Purpose of the Study:

• To establish a hierarchy of hemodynamic and ventilatory variables impacting intraoperative rSO2 changes.
• To apply supervised machine learning for predicting rSO2 trends.
• To quantify the influence of physiological variables on cerebral oxygenation.

Main Methods:

• A post-hoc analysis of 146 high-risk surgery patients.
• Machine learning classification (XGBoost) trained on hemodynamic and ventilatory data.
• Feature importance analysis using information gain and permutation importance.

Main Results:

• The XGBoost classifier achieved 77.1% accuracy and 0.86 AUC.
• Key predictors of rSO2 changes included ETCO2 slope, SpO2 slope, and MAP.
• Carbon dioxide (CO2) significantly mediates intraoperative rSO2 fluctuations.

Conclusions:

• This study quantifies the hierarchy of variables affecting rSO2 during surgery.
• Insights into CO2's role in cerebral blood flow provide a better understanding of cerebral oxygenation.
• Findings contribute to optimizing intraoperative management for improved patient care.