Abstract
Intrahepatic cholangiocarcinoma (iCCA) is an aggressive malignancy characterized by profound molecular heterogeneity and poor prognosis. Programmed cell death (PCD) regulates tumor progression and shapes the tumor immune microenvironment (TME), yet the roles of distinct PCD subtypes in iCCA remain elusive. Here, we integrated bulk and single-cell transcriptomic datasets derived exclusively from intrahepatic cholangiocarcinoma (iCCA) in TCGA, GEO, and the fan_match cohort, and curated genes representing 21 PCD subtypes. Among 117 machine-learning algorithm combinations, a backward stepwise Cox regression (StepCox) combined with random survival forests (RSF) constructed a robust nine-gene prognostic signature (ATF6, ACVR1, ACAP2, C6orf136, CD4, ABCB9, ABCC1, CSNK2A2, ABCG1) that consistently stratified patients into high- and low-risk groups with distinct outcomes across independent cohorts. High-risk patients exhibited inferior survival, greater immunosuppressive features, elevated TME scores, and enrichment of inflammatory pathways. The PCD score demonstrated predictive value for immunotherapy response across multiple independent cohorts. Single-cell analyses further revealed increased regulatory T-cell infiltration and more complex intercellular communication in high-risk tumors, whereas low-risk tumors displayed stronger collagen signaling and a less immunosuppressive tumor immune microenvironment. Collectively, these findings underscore the importance of PCD-related genes for prognosis and for predicting immunotherapy response in iCCA, and provide a practical framework for risk stratification and therapeutic decision-making. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1038/s41598-026-44332-8.