Background
Hepatobiliary cancers, including hepatocellular carcinoma (HCC), intrahepatic and extrahepatic cholangiocarcinoma (iCCA, eCCA), gallbladder cancer (GBC), and ampulla of Vater cancer (AoV), are aggressive malignancies with poor prognosis due to late-stage diagnosis and limited treatment options. Major risk factors such as gallstone disease, primary sclerosing cholangitis (PSC), and metabolic dysfunction–associated liver disease/steatohepatitis (MASLD/MASH) typically develop years before cancer onset, offering a window for early detection. Circulating metabolites, reflecting altered metabolic pathways in these conditions, may serve as biomarkers for hepatobiliary cancer risk stratification.

Objectives
To identify plasma metabolites associated with hepatobiliary cancer risk using a two-phase strategy leveraging their links to clinical risk factors, with independent validation in UK Biobank and EPIC.

Methods
We analysed targeted NMR metabolomics data in UK Biobank participants (batches 1–2: ~274,000 for discovery; batch 3: ~228,000 for validation). In discovery phase one, we identified metabolites associated with seven hepatobiliary cancer risk factors (gallstones, cholecystitis, cholecystectomy, PSC, MASLD, MASH, cirrhosis) using multivariable Cox regression with data-shared lasso regularization and 1,000 bootstraps. In discovery phase two, these metabolites were tested for associations with incident hepatobiliary cancers (GBC, HCC, iCCA, eCCA, AoV) using lasso-penalized Cox models adjusted for time-dependent risk factors.

Validation analyses applied identical Cox models in batch 3 without penalization. We compared hazard ratios with discovery results and assessed for HCC predictive performance of metabolomic scores versus polygenic risk scores. Scores were dichotomized at the top decile, and combined risk was evaluated using multivariable Cox models. Absolute risks by age were estimated using a competing-risks framework. Amino acid associations were externally validated in nested case–control studies within EPIC.

Results
Discovery phase one identified 27 metabolites linked to gallstone-related conditions, 11 to PSC, and 34 to metabolic liver disease, with some overlapping associations. In discovery phase two, 3 metabolites were associated with GBC, 18 with HCC, 5 with iCCA, and 2 with eCCA; none for AoV. Strongest signals were detected for HCC, notably phosphatidylcholine and XS-VLDL-P clusters, which showed increasing association strength with disease severity and cancer risk. For GBC, XS-VLDL-P and HDL-TG clusters were notable.

Independent UK Biobank validation confirmed most associations, particularly for HCC and iCCA. Amino acid associations detected for HCC (alanine, glutamine, tyrosine, creatinine, Fischer’s ratio) were validated in EPIC. Absolute-risk analyses showed clear stratification for HCC: top 10% metabolomic score had substantially higher risk, and combined high metabolomic and polygenic risk yielded the greatest risk. For other cancers, metabolomic scores identified modestly elevated risk groups. Sensitivity analyses confirmed robustness and sex-stratified analyses revealed marked differences for some metabolites.

Conclusions/Implications
Metabolomic profiling identified risk signatures for hepatobiliary cancers, particularly HCC, years before diagnosis. These findings suggest a feasible path toward blood-based risk stratification, complementing genetic scores and enabling targeted surveillance in high-risk groups. While clinical translation requires replication and cost-effectiveness studies, the potential to detect risk decades earlier could transform prevention strategies for hepatobiliary cancers that currently lack screening options.