ScholarWorks@동국대학교

초록

Endoscopic retrograde cholangiopancreatography (ERCP) remains the standard for common bile duct (CBD) stone removal. However, current guidelines often result in unnecessary procedures. This study aimed to develop and validate a machine learning model using synthetic data augmentation to improve CBD stone prediction. Electronic health records from patients with suspected CBD stones were analyzed from three independent tertiary centers (733 patients for internal validation, 348 for external validation). A large language model (LLM) generated curated synthetic data to augment the training dataset. The ExtraTrees classifier was selected after evaluating multiple algorithms. Model performance was assessed using area under the receiver-operating characteristic curve (AUROC), calibration curves, and decision curve analysis. The model incorporated 11 routinely available variables and achieved an AUROC of 0.982 (95% CI 0.952-1.000) in internal validation and 0.957 (95% CI 0.937-0.974) in external validation. Compared to existing clinical guidelines, the model substantially reduced unnecessary ERCPs (0% vs 11.9-22.9% in internal validation; 6.7% vs 29.2-35.9% in external validation) while maintaining low false-negative rates (17.6% vs 28.2-45.6% in external validation). Calibration curves demonstrated good alignment between predicted and observed outcomes. Decision curve analysis confirmed greater net clinical benefit than existing guidelines. Feature importance analysis identified CBD dilatation >= 10 mm as the most influential predictor, followed by liver function parameters. This machine learning model, enhanced by LLM-generated synthetic data, significantly outperformed existing clinical guidelines for CBD stone prediction. The model provides a practical risk stratification tool using routinely available emergency department variables and may reduce unnecessary ERCP procedures while maintaining diagnostic safety.

키워드