Use of multi-omics biomarkers to predict benefit from immune checkpoint inhibitors in biliary tract cancers: A systematic review and meta-analysis.

e14587 Background: Predictive biomarkers for immune checkpoint inhibitors (ICIs) in biliary tract cancers (BTC) are urgently needed. Multi-omics platforms including genomics, transcriptomics, spatial profiling and proteomics have emerged as promising tools. We evaluated the pooled predictive performance of omics-based biomarkers for ICI benefit in BTC. Methods: We searched PubMed, Embase and Scopus (2020-2025) for studies assessing pre-treatment omics biomarkers in ≥20 adult BTC patients receiving PD-1/PDL1 therapy; four studies used multi-omics profiling and two used single-omics. Primary outcomes were progression free survival (PFS), overall survival (OS) and predictive accuracy (AUC), analyzed using random-effects meta-analysis with Knapp-Hartung (KH) adjustment. Hazard ratios (HR) were harmonized so biomarker-positive groups consistently represented superior outcomes; regimens included ICI monotherapy and ICI combined with chemotherapy and pooled estimates used each study’s own biomarker cut-offs. Pooled AUC was derived from the three ROC-reporting studies and survival from the four HR-reporting studies, with subgroups (immune-activation, genomic-instability, negative-predictor and liquid vs tissue biomarkers), leave-one-out sensitivity analyses and a reviewer derived pathway convergence map linking reported genes and proteins to canonical immune pathways. RoB and certainty evaluated using QUIPS and GRADE. Results: Six studies including 443 participants met criteria. For PFS, the pooled HR was 0.23 (95%CI 0.094–0.558; KH CI 0.087–0.602; I 2 = 58%). For OS, the pooled HR was 0.24 (95%CI 0.084–0.711; KH CI 0.097–0.628; I 2 = 47%) with leave-one-out analyses showing stable estimates. Funnel plot suggested possible small study/publication bias and formal tests were limited by the number of studies. Predictive accuracy values ranged 0.831–0.867; pooled AUC (from the three studies with ROC data) was 0.858 (95%CI 0.797–0.918). Subgroup analyses showed no differences across immune-activation, genomic-instability or negative-predictor biomarkers (P > 0.7) and no difference between liquid vs tissue biomarkers for PFS or OS (P > 0.3). Pathway mapping showed convergent IFN-γ signaling, chemokine-mediated T/NK recruitment, antigen processing pathways and spatial immune niches. RoB was moderate in four studies and high in two; overall GRADE certainty was moderate. Conclusions: Across omics platforms, biomarker-positive BTC patients showed markedly improved PFS and OS (~75% risk reduction) and strong predictive accuracy. Despite heterogeneous biomarker definitions and regimens, a convergent immune-responsive phenotype identifies ICI-sensitive BTC and supports prospective biomarker stratified trials using standardized assays to validate these signals and prioritize patients for ICI-containing regimens pending validation.