Abstract 4915: A multi-protein score to predict hepatocellular carcinoma occurrence

Abstract Background: Hepatocellular carcinoma (HCC) is the most common type of liver cancer, frequently developing on a background of chronic liver disease. Our understanding of the totality of HCC etiology remains incomplete. Early detection of HCC remains a critical challenge in clinical practice as prognosis remains poor for individuals diagnosed at advanced stage. Recent advances in proteomic technologies enable systematic analysis of protein-disease associations, which has potential to improve mechanistic understanding of HCC etiology and aid in the discovery of novel biomarkers for early detection. Aims: We aimed to investigate the associations of serum proteins with subsequent risk of HCC and create a multi-protein score to predict disease occurrence. Methods: We conducted a nested case-control study within the PLCO cohort, analyzing 118 incident HCC cases individually matched to 118 controls (median time to HCC diagnosis:10 years). Using the Olink HT platform, we measured 5,400 proteins. Multivariable conditional logistic regression models were used to estimate protein-HCC risk associations, and we developed a multi-protein risk score using LASSO regression, which was evaluated using the area under the curve (AUC). All analyses underwent Bonferroni correction for multiple testing. We tested for replication our findings in the UK Biobank, a prospective cohort of 50,000 participants with protein data from the Olink Explore 3072 platform. During a median follow-up of 11 years, 28 participants developed HCC. We estimated protein associations using Cox proportional hazards regression and evaluated the multi-protein score's predictive performance in the overall cohort and among participants with liver disease—a population at elevated risk of HCC—by calculating the C-index. Results: Among PLCO participants, we identified 127 proteins significantly associated with HCC risk. Of these, 126 proteins were positively associated, with the strongest association observed for TNFRSF11B (OR per 1 SD: 2.85, 95% CI 1.90-4.28). IGFBP-3 was the only protein inversely associated with risk (0.36, 0.24-0.54). Notably, 118 of the 127 proteins were available in the UK Biobank, and 117 of these associations were successfully replicated. LASSO regression revealed 8 key protein predictors. The resulting protein score exhibited high discriminatory performance in the PLCO cohort (AUC = 0.90) and generalizability in the UK Biobank (C-Index = 0.89 overall, 0.92 among high-risk individuals). Conclusion: Our findings highlight strong potential of a multi-protein predictive score for HCC, evident years before diagnosis across 2 distinct cohorts. Citation Format: Eleanor Watts, Steven Moore, Belynda Hicks, Wen-Yi Huang, Erikka Loftfield, Mitchell Machiela, Katherine McGlynn, Jianxin Shi, Rachael Stolzenberg-Solomon, Cody Watling. A multi-protein score to predict hepatocellular carcinoma occurrence [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4915.