A Dynamic Prognostic and Adaptive Treatment Framework for Advanced Biliary Tract Cancer

Area of Science:

• Oncology
• Biostatistics
• Machine Learning in Medicine

Background:

• First-line immuno-chemotherapy is standard for advanced biliary tract cancer (BTC), but patient outcomes vary significantly.
• There is a need for dynamic monitoring and treatment adaptation based on evolving clinical data.
• Current prognostic models do not adequately capture longitudinal changes in patient status.

Purpose of the Study:

• To develop dynamic models that update survival predictions for advanced BTC patients receiving first-line immuno-chemotherapy.
• To enable real-time prognostic stratification and support treatment adaptation.
• To integrate baseline and longitudinal clinical, imaging, biomarker, and molecular data.

Main Methods:

• Developed the Individualized Dynamic Risk Estimation Model for biliary tract cancer (iDREM-BTC) using Bayesian joint modeling.
• Integrated baseline clinical, imaging, and longitudinal biomarkers (C-reactive protein, CA19-9, total bilirubin).
• An enhanced version, iDREM(Pro)-BTC, incorporated immunohistochemical and genomic data.
• Validated models across multiple development, internal, and external cohorts (total 2314 patients).

Main Results:

• Machine learning identified age, ECOG status, tumor burden/stage, and longitudinal biomarkers as key mortality predictors.
• iDREM-BTC demonstrated robust dynamic discrimination (overall AUCs 0.705-0.755) with improved performance over follow-up (AUCs up to 0.810 at 6 months).
• iDREM(Pro)-BTC showed higher discrimination (AUC 0.807 in development) and retained external validation performance (AUC 0.718).
• iDREM-BTC-defined high-risk patients showed improved overall survival separation.

Conclusions:

• iDREM-BTC and iDREM(Pro)-BTC provide dynamically updated survival estimates for advanced BTC patients.
• These models support individualized prognostic stratification and can inform treatment adjustments during first-line immuno-chemotherapy.
• The models offer valuable bedside tools for routine monitoring and treatment of BTC patients.