Dynamic DBil and TBA Trajectories for Predicting Native Liver Survival in Biliary Atresia: A Multicenter Study with an Open-Access Online Tool.

AIM: This study aimed to construct a multicenter native liver survival (NLS) prediction model for biliary atresia (BA) using dynamic postoperative DBil and TBA trajectories, with an online individualized tool. METHODS: We retrospectively enrolled 473 children with type III BA from five centers between January 2016 and December 2025. Restricted cubic spline (RCS) analyses characterized DBil and TBA associations with NLS at four postoperative time points. Group-based trajectory modeling (GBTM) identified trajectory subgroups, and Cox proportional hazards models evaluated prognostic stratification and predictive performance with bootstrap-based internal validation. RESULTS: RCS analyses confirmed that both biomarkers exhibited time-dependent, nonlinear associations with NLS. GBTM identified five clinically distinct trajectory subgroups for both DBil and TBA. DBil trajectory groups demonstrated significant stratification of both NLS and overall survival (OS), whereas TBA trajectory groups stratified NLS alone. The time-dependent AUC of the DBil trajectory model exceeded that of the TBA model at all time points from 1 to 6 years, and both models showed good calibration and net clinical benefit on decision curve analysis. The combined model (DBil + TBA trajectories) significantly outperformed the DBil trajectory model alone (ΔC = 0.024, P = 0.002), the TBA trajectory model alone (ΔC = 0.071, P < 0.001), and the single time-point DBil reference model (ΔC = 0.085, P < 0.001). CONCLUSION: The combined DBil and TBA trajectory Cox model, developed based on a multicenter cohort, outperforms static indicators for post-KPE NLS prediction. An open-access online tool is provided to support individualized prognostic assessment, pending external validation in independent cohorts.