Predicting anesthetic infusion events using machine learning.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Anesthesiology

Background:

• Anesthesiologist shortage necessitates innovative solutions.
• Machine learning offers potential for automating and predicting clinical decisions.
• Accurate prediction of anesthetic drug administration is crucial for patient safety.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting anesthesiologist decisions regarding remifentanil administration.
• To formulate the prediction of analgesic flow rate adjustments as a supervised binary classification problem.
• To assess the performance of various machine learning models in forecasting clinical actions.

Main Methods:

• Formulated anesthesiologist decision-making as a supervised binary classification task.
• Evaluated six machine learning models: logistic regression, SVM, random forest, LightGBM, ANN, and LSTM.
• Utilized data from 210 actual surgical cases for model training and validation.
• Employed Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• The Long Short-Term Memory (LSTM) model achieved promising predictive performance for 1-minute ahead remifentanil flow rate increases.
• LSTM model achieved sensitivity of 0.659, specificity of 0.732, and ROC-AUC of 0.753.
• Feature importance analysis revealed trends partially consistent with established clinical knowledge.

Conclusions:

• Machine learning, particularly LSTM, demonstrates potential for predicting anesthesiologist decisions in real-time during surgery.
• This predictive capability could aid in managing anesthesia and addressing anesthesiologist shortages.
• Model interpretability using SHAP provides insights into clinical decision-making processes.