Hypoxemia prediction model based on XGBoost during sedation for gastrointestinal endoscopy

Area of Science:

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

Background:

• Hypoxemia is a frequent complication of sedated gastrointestinal endoscopy, posing significant risks.
• Predicting and preventing hypoxemia remains a clinical challenge.
• Artificial intelligence (AI) offers potential for accurate hypoxemia prediction using integrated clinical data.

Purpose of the Study:

• To develop a robust, interpretable, and generalizable Machine Learning (ML) model for predicting hypoxemia during sedated gastrointestinal endoscopy.
• To evaluate the performance of AI in identifying patients at risk of hypoxemia.
• To enhance patient safety during endoscopic procedures.

Main Methods:

• A prospective study of 647 adult patients undergoing sedated gastrointestinal endoscopy.
• Utilized statistical analyses and ML techniques including SHapley Additive exPlanations (SHAP) and eXtreme Gradient Boosting (XGBoost).
• Model performance evaluated using Accuracy, Precision, Recall, F1-score, and ROC-AUC; feature importance analyzed.

Main Results:

• The XGBoost model achieved high performance with an accuracy, recall, and F1-score of 0.91 and ROC-AUC of 0.74.
• Identified 29 key features, including BMI, waist circumference, neck circumference, age, and baseline SpO2, as significant predictors.
• Model performance improved with a larger, balanced dataset, highlighting the importance of sample size.

Conclusions:

• A robust XGBoost-based hypoxemia prediction model was developed for sedated gastrointestinal endoscopy.
• The AI model demonstrates potential to enhance patient safety and clinical decision-making.
• Future research should focus on larger, diverse datasets and advanced methods like latent-space analysis to improve model accuracy and applicability.