Prediction Model Of Inadequate Response To Ursodeoxycholic Acid In Patients With Primary Biliary Cholangitis

Objective:To establish prediction models about treatment response of ursodeoxycholic acid(UDCA)in patients with primary biliary cholangitis(PBC)and try to find potential biomarkers for patients with poor response to UDCA.Method:1.We included patients diagnosed as PBC in Xiangya Hospital of Central South University from January 2010 to December 2020 retrospectively,and the clinical data of patients were collected as the baseline data to construct prediction model.2.Patients with PBC were prospectively included,and their serum and fecal bile acids were analyzed to find potential markers that could be used to identify patient incomplete response to UDCA preliminarily.Result:1.Random forest importance ranking showed that poor response to UDCA was closely related to C-reactive protein,D-dimer,alkaline phosphatase,bile acid and gamma-glutamyl transferase levels.The area under the curve(AUC)of the random forest model was 0.81(95 % CI0.776-0.847),and the accuracy was 0.78,compared to the logistic regression model was 0.64(95 % CI 0.596-0.685),the accuracy was 0.64.2.Metabolomics results showed that there had significant differences in serum bile acids between the two groups,while no significant difference in fecal bile acids.Conclusion:1.A prediction model based on clinical features that can identify UDCA non-responder early is established through machine learning.Compared with the logistic regression model,the random forest model has better prediction performance.2.The baseline levels of C-reactive protein,D-dimer,alkaline phosphatase,bile acid and gamma-glutamyltransferase in PBC patients were significantly associated with PBC patients responding to UDCA.3.There were significant differences in some serum bile acids between two groups.Serum bile acids can be used as potential biomarkers for non-response with UDCA in PBC patients.