A Machine Learning Model Based on Ultrasound and Serological Markers for Predicting Postoperative Stenosis of Rex Shunt Procedure in Children With Extrahepatic Portal Venous Obstruction.

PURPOSE: To evaluate the value of combining ultrasound parameters with serological markers for predicting postoperative Rex shunt stenosis in children with extrahepatic portal venous obstruction (EHPVO). METHODS: The training cohort and temporal validation cohort consisted of 114 and 38 children with EHPVO after Rex shunt procedure, respectively. The study variables included clinical characteristics, ultrasound parameters, and serological markers. Five machine learning models were developed to predict postoperative stenosis after the Rex shunt procedure at different follow-up periods, with performance evaluated using the area under the receiver operating characteristic curve (AUROC). RESULTS: In the temporal validation cohort, the optimal Logistic regression model integrating ultrasound parameters and serological markers achieved AUROCs of 0.61 (95% CI: 0.38-0.84) at 1 month, 0.67 (95% CI: 0.45-0.90) at 3 months, and 0.82 (95% CI: 0.67-0.98) at 6 months for predicting stenosis. CONCLUSIONS: A model combining serological markers and ultrasound parameters serves as a promising exploratory prototype for predicting Rex shunt stenosis in children with EHPVO, pending further independent multicenter validation.