Machine Learning Approach to Optimize Sedation Use in Endoscopic Procedures.

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Machine Learning in Healthcare

Background:

• Colonoscopy sedation decisions (moderate vs. deep) lack clear guidelines, causing practice variability.
• Moderate sedation is cost-effective but often poorly tolerated, leading to patient discomfort and procedure abandonment.
• Deep sedation is effective but incurs higher costs.

Purpose of the Study:

• To develop a Machine Learning (ML) model predicting successful colonoscopy completion with moderate sedation.
• To assist physicians in selecting appropriate sedation levels, optimizing patient care and resource utilization.

Main Methods:

• An XGBoost ML model was trained and validated using data from 10,025 colonoscopies.
• The model utilized patient demographics, comorbidities, and medication data for prediction.
• A 70%-30% split was used for training and testing the model.

Main Results:

• The XGBoost model achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.762.
• The model demonstrated strong performance with an F1-score of 0.85 for predicting moderate sedation needs.
• Precision and recall scores were 0.81 and 0.89, respectively, indicating reliable predictions.

Conclusions:

• The developed ML model can serve as an effective decision support tool for physicians.
• It helps enhance confidence in choosing between moderate sedation and anesthesia for colonoscopies.
• The model has the potential to standardize sedation practices and improve patient outcomes.