Post-Anesthesia Care Unit (PACU) readiness predictions using machine learning: a comparative study of algorithms.

Area of Science:

• Anesthesiology and Perioperative Medicine
• Artificial Intelligence in Healthcare
• Health Informatics

Background:

• Timely Post-Anesthesia Care Unit (PACU) discharge is crucial for patient safety and efficient hospital operations.
• Premature discharge risks complications, while delays strain resources.
• Machine learning (ML) offers a novel approach to predict optimal discharge timing using patient data.

Purpose of the Study:

• To evaluate the efficacy of multiple ML models in predicting PACU discharge readiness.
• To compare ML model performance against traditional methods like staff evaluations and the Aldrete checklist.
• To identify the most accurate ML algorithms for optimizing PACU discharge decisions.

Main Methods:

• A cross-sectional study involving 830 patients undergoing general anesthesia.
• Collected patient demographics, surgical details, and Aldrete scores.
• Tested various ML models (RF, SVM, LR, DT, KNN, ANN, XGBoost) using two prediction approaches: 15-minute intervals and binary classification.
• Evaluated models based on accuracy, precision, recall, F1 score, and AUC, comparing against staff and Aldrete scores.

Main Results:

• The Random Forest (RF) algorithm demonstrated high performance across both prediction approaches.
• RF achieved an AUC of 0.87 and accuracy of 0.71 when benchmarked against Aldrete scores.
• In binary classification, RF achieved an AUC of 0.85 and accuracy of 0.86 against staff evaluations, with ANN also showing strong results.
• No statistically significant differences were found between the top-performing models due to overlapping confidence intervals.

Conclusions:

• Random Forest (RF) and Artificial Neural Network (ANN) models show significant potential for predicting PACU discharge readiness.
• These ML tools may offer a more consistent and data-driven alternative to current staff assessments and the Aldrete checklist.
• Further research is needed to validate and implement these ML models for improved patient outcomes and hospital efficiency.