A machine learning-based choledocholithiasis prediction tool to improve ERCP decision making: a proof-of-concept...

Area of Science:

• Gastroenterology
• Medical Informatics
• Machine Learning in Medicine

Background:

• Existing guidelines for predicting choledocholithiasis show limited accuracy.
• This leads to overuse of endoscopic retrograde cholangiopancreatography (ERCP), a procedure with associated risks.
• Improved patient stratification is needed to optimize ERCP selection and consider alternative, lower-risk interventions.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting choledocholithiasis.
• To compare the model's performance against current ASGE and ESGE guidelines.
• To assess the potential of the model to improve ERCP utilization and patient outcomes.

Main Methods:

• A gradient boosting model (GBM) was developed using patient data from two cohort studies.
• The model utilized eight identified predictors of choledocholithiasis.
• Performance was assessed using 10-fold cross-validation and AUC, comparing against ASGE and ESGE guidelines.

Main Results:

• The GBM achieved an accuracy of 71.5% (AUC 0.79), outperforming ASGE (62.4% accuracy, AUC 0.63) and ESGE (62.8% accuracy, AUC 0.67) guidelines.
• The model correctly reclassified 22% of patients recommended for unnecessary ERCP by ASGE guidelines.
• It also identified 48% of ERCPs incorrectly rejected by ESGE guidelines as appropriate.

Conclusions:

• A machine learning tool provides real-time, personalized choledocholithiasis probability and ERCP recommendations.
• This tool offers more accurate guidance for ERCP use compared to current ASGE and ESGE guidelines.
• The model has the potential to decrease ERCP-related morbidity and avoid missed diagnoses of choledocholithiasis.