Interpretable cancer risk prediction remains challenging because models with high accuracy are often opaque, while transparent tools tend to underperform. Biliary tract cancer (BTC) is an aggressive malignancy that is frequently diagnosed at an advanced stage, in part because early symptoms overlap with benign biliary tract disease (BTD). As a case study of our approach, we developed and validated an interpretable risk prediction framework that identifies high-risk BTC candidates within a BTD population using routinely available clinical data. In this retrospective multicenter case?control study, we analyzed electronic medical records from four tertiary academic centers in South Korea. A high-capacity black-box teacher model was first trained to distinguish BTC from BTD using data from three institutions (n=1,439). We then applied a two-stage knowledge distillation framework to train an interpretable, non-differentiable student model―a decision-rule based risk tree―using both ground-truth labels and the teacher’s calibrated predictions. External validation was performed on an independent cohort from a fourth institution (n=245). The resulting risk tree defines transparent BTC risk strata among patients with BTD and maintains strong performance even when CA19-9 is normal or unavailable, supporting applicability to Lewis antigen?negative individuals and illustrating how knowledge distillation can bridge the gap between black-box accuracy and interpretable cancer risk prediction.