Machine Learning Model for Predicting Biliary Complications After Liver Transplantation.

Area of Science:

• Hepatology
• Transplant Surgery
• Medical Informatics

Background:

• Biliary complications (BCs) are significant risks after liver transplantation, with poorly defined risk factors and variable onset times.
• Machine learning (ML) shows promise in analyzing complex medical data for prediction in liver transplantation.

Purpose of the Study:

• To determine if ML can efficiently predict BCs after liver transplantation.
• To identify specific risk factors for BCs at 3, 6, and 12 months post-transplant.

Main Methods:

• Utilized data from 517 patients across two centers, randomly split into training and validation sets.
• Employed K-fold cross-validation, synthetic minority oversampling technique, and SHapley Additive exPlanation (SHAP) values for model validation and interpretation.
• Developed and evaluated seven ML algorithms for predicting BCs at 3, 6, and 12 months post-transplant.

Main Results:

• Support vector machine (SVM) demonstrated the highest predictive accuracy for BCs, with AUC values of 0.916 (3-month), 0.892 (6-month), and 0.885 (12-month).
• Identified key 3-month risk factors: donor age, Model for End-Stage Liver Disease (MELD) score, neoplastic disease, diabetes, hypertension, and intraoperative blood transfusion.
• Identified 6-month risk factors: recipient age, donor and recipient BMI, and diabetes. 12-month risk factors included recipient age, diabetes, and basiliximab.

Conclusions:

• ML algorithms effectively identified risk factors for BCs across all postoperative periods.
• The findings provide valuable insights for optimizing patient management and potentially mitigating BCs after liver transplantation.