Predicting early postoperative PONV using multiple machine-learning- and deep-learning-algorithms.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Clinical Prediction Models

Background:

• Postoperative nausea and vomiting (PONV) significantly impacts patient satisfaction and increases healthcare costs due to extended hospital stays.
• Early identification and management of PONV are crucial for improving patient outcomes and optimizing resource utilization.

Purpose of the Study:

• To develop and evaluate a preliminary artificial intelligence (AI) algorithm model for the early prediction of PONV.
• To identify key clinical factors contributing to the occurrence of early PONV.

Main Methods:

• Utilized R for statistical analysis and Python for developing the machine learning prediction model.
• Engineered features to identify the top contributing factors for early PONV.
• Evaluated multiple AI algorithms including CNNRNN, Decision Tree, SVC, and adaboost for prediction accuracy, precision, and AUC.

Main Results:

• Haloperidol administration, patient sex, age, smoking history, and previous PONV history were identified as the top 5 predictors of early PONV.
• The CNNRNN algorithm demonstrated the highest accuracy (0.872), while CNNRNN also showed the highest precision (1.000).
• Logistic Regression, SVC, and adaboost achieved the top AUC scores, indicating strong predictive performance.

Conclusions:

• AI algorithms are capable of accurately predicting early PONV.
• Logistic Regression, SVC, and adaboost algorithms exhibited the best overall performance in predicting PONV.
• Established a publicly accessible online tool for predicting early PONV using a Streamlit app.