e16183 Background: Immune checkpoint inhibitors (ICIs) have improved outcomes for advanced biliary tract cancer (BTC), but survival benefits remain modest and predictive biomarkers to guide clinical practice are lacking. Methods: We analyzed 588 advanced BTC patients receiving ICIs from a multicenter retrospective cohort and two prospective clinical trials. To address substantial multicollinearity among inflammation-, nutrition-, and liver function–related biomarkers, we adopted a rigorous three-stage feature selection pipeline integrating Boruta, LASSO regression, and akaike information criterion (AIC)-based best-subset selection. Cox proportional hazards regression and machine learning survival models were constructed, and an integer-based bedside scoring system (BICCAPS) was derived. Dynamic changes in prognostic nutritional index (PNI) and systemic inflammation response index (SIRI) from baseline to cycle 2–3 were evaluated for additional prognostic value. Results: Seven variables (ICI treatment line, carbohydrate antigen 19-9 (CA19-9), carcinoembryonic antigen (CEA), peritoneal metastasis, SIRI, PNI and bilirubin) were incorporated into Cox proportional hazards regression model and random survival forest models, which showed good discrimination and calibration for OS across all cohorts. The BICCAPS score stratified patients into low-, intermediate- and high-risk groups with clearly separated median overall survival (OS) in the training (18.9, 10.5 and 6.8 months) and validation cohorts. Worsening PNI or SIRI during early treatment was associated with significantly shorter OS and progression-free survival. Conclusions: This study provides the first clinically implementable decision-making framework for immunotherapy in BTC. By translating complex statistical modeling into a simple bedside scoring system and incorporating dynamic biomarker assessment, the BICCAPS score offers a practical, low-cost, and accessible tool for individualized ICI management.
Zheng et al. (Thu,) studied this question.