Machine learning-based prediction of recurrence in primary Budd-Chiari syndrome

Introduction This study aimed to identify important predictors of first recurrence in patients with primary Budd-Chiari syndrome (BCS) and to develop machine learning models for predicting recurrence. Methods A total of 754 patients with primary BCS who received treatment at the Affiliated Hospital of Xuzhou Medical University between January 2015 and July 2022 were retrospectively included. Four prediction models, including random forest (RF), logistic regression (LR), decision tree (DT), and light gradient boosting machine (LightGBM), were developed and evaluated. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC) and other classification metrics. SHapley Additive exPlanations (SHAP) analysis was performed to identify influential predictors. Results The RF model achieved AUROC values of 0.7297 in the test set and 0.8028 in the validation set. In the independent test set, LR, DT, and LightGBM achieved AUROC values of 0.7017, 0.6654, and 0.6522, respectively. The RF model showed the highest AUROC, with an accuracy of 0.7273, sensitivity of 0.5946, specificity of 0.7857, positive predictive value of 0.5500, and negative predictive value of 0.8148. SHAP analysis identified age, prothrombin time, and neutrophil count as influential predictors in the RF model. Discussion The RF model demonstrated the best predictive performance among the evaluated models in the independent test set. These findings suggest that machine learning approaches may provide useful information for recurrence risk stratification in patients with primary BCS. However, further external validation is required before clinical implementation..