Metabolic and Inflammatory Risk Stratification in Resected Cholangiocarcinoma: A Retrospective Exploratory Analysis.

Background: Cholangiocarcinoma (CCA) frequently recurs after curative-intent resection. Metabolic alterations and systemic inflammation may contribute to aggressive disease, but their combined prognostic relevance remains unclear. Methods: We retrospectively reanalyzed 88 patients with resected CCA and documented recurrence, including 37 with early and 51 with late recurrence. A metabolite-based risk score was developed from recurrence-associated serum metabolites identified previously. Model development and internal validation used all 88 patients, whereas analyses involving preoperative neutrophil-to-lymphocyte ratio (NLR), clinicopathological variables, and survival outcomes included 85 patients with complete clinical data. Performance was assessed using repeated nested cross-validation, bootstrap validation, discrimination, calibration, and survival analyses. Results: The 10-metabolite risk score (MRS-10) showed moderate internally validated discrimination for distinguishing early from late recurrence, with a pooled patient-level out-of-fold AUC of 0.828 (95% CI, 0.725-0.911), with limited calibration. High MRS was associated with shorter disease-free survival (DFS) and overall survival (OS) (both p < 0.001) and remained independently associated with DFS events (HR, 7.92; 95% CI, 3.94-15.92) and mortality (HR, 4.05; 95% CI, 2.08-7.89). High MRS was also associated with elevated NLR (p < 0.001). The combined MRS-NLR framework improved prognostic discrimination and stratified patients into groups with stepwise differences in DFS and OS. Exploratory subgroup analyses showed broadly consistent adverse associations across selected clinicopathological strata. Conclusions: In this retrospective exploratory analysis, a metabolite-based risk score combined with preoperative NLR was associated with recurrence and survival outcomes in resected CCA. This metabolic-inflammatory framework may provide complementary prognostic information alongside conventional clinicopathological factors. However, the findings remain hypothesis-generating and require validation in larger independent cohorts before clinical application.