Explainable Machine Learning for Prediction of Early Postoperative Nausea and Vomiting After General Anesthesia.

Area of Science:

• Anesthesiology
• Artificial Intelligence
• Predictive Analytics

Background:

• Postoperative nausea and vomiting (PONV) is a frequent complication of anesthesia.
• Predicting PONV risk is crucial for effective perioperative management.
• Current prediction methods often lack precision or rely on invasive data.

Purpose of the Study:

• To assess the feasibility of using machine learning (ML) models for early PONV risk prediction.
• To identify key predictors of PONV using routinely available, non-invasive clinical indicators.
• To apply explainable artificial intelligence (XAI) for clinical interpretability.

Main Methods:

• Retrospective analysis of 927 patient cases.
• Evaluation of various ML models (traditional, deep learning, ensemble) using 16 non-invasive indicators.
• Application of SHAP (SHapley Additive exPlanations) values for feature importance analysis.

Main Results:

• Models showed moderate discriminative performance; Random Forest achieved 83.5% accuracy.
• Key predictors identified include neostigmine, history of nausea/vomiting, cardiovascular disease, gender, pain, surgery type, and anesthesia duration.
• Dexamethasone use was associated with reduced PONV risk.

Conclusions:

• ML models trained on non-invasive data can effectively predict early PONV risk.
• XAI techniques provide valuable insights into PONV predictors, supporting clinical decision-making.
• This approach enhances perioperative care by enabling informed risk assessment and management.