Abstract 3378: Targeted serum proteomics identifies a signature for use in multivariable logistic regression modeling to predict HCC

Abstract Background: Hepatocellular carcinoma (HCC) is curable in its early stages. However, early stage HCC is difficult to detect without invasive surgical procedures. Additionally, current blood-based biomarkers have low sensitivity and specificity in detecting HCC. Human data and animal models support a functional role for the TGF-β pathway in HCC initiation and progression. The goal of this study is to determine whether TGF-β pathway markers can predict HCC in cirrhotic patients that are at high risk for HCC. Methods: We used an aptamer-based high-throughput proteomics assay to detect a selected set of proteins associated with fibrosis or TGF-β signaling in serum samples from 101 patients with only cirrhosis and 33 patients with cirrhosis and HCC. We applied multivariable logistic regression modeling to identify a biomarker panel predictive of HCC and then integrated this model with clinical variables to develop and a final predictive HCC model. The predictive power, sensitivity, and specificity of the final model were determined. Results: We identified a protein signature in patient serum that differentiated patients with only cirrhosis from those with HCC. We also identified a potential high-risk cirrhosis patient group (n = 14) with a similar signature to that of HCC patients. The logistic regression model identified that a panel of 15 proteins was independently predictive of HCC. By adding clinical predictors into the model, we developed a model that required only 5 of the proteins to achieve high predictive power. Analysis of the receiver operating characteristic (ROC) curve of the final model showed an area under the curve (AUC) of 0.96. The mean predicted probability of having HCC was 8.3% in the non-HCC group [95% confidence interval (CI) 5.6 - 11.1%] and 75% in the HCC group (95% CI 63 - 86%). At a threshold of 0.30, sensitivity was 0.88, specificity was 0.88, and positive predictive value (PPV) was 0.71. At a threshold of 0.04, the sensitivity increased to 1.0 at the expense of specificity (0.64) and PPV (0.48). Conclusions: The final 5-protein model that includes clinical variables represents a potential non-invasive method to detect early HCC. Furthermore, follow up studies of patients with the 15-protein signature that resembles the signature of HCC patients is warranted to determine if this signature represents patients in the earliest stages of HCC or who subsequently are diagnosed with HCC. Citation Format: Shuyun Rao, Richard Amdur, Xiyan Xiyan, Kirti Shetty, Herbert Yu, Linda L. Wong, Wilma Jogunoori, Karan Amin, Patricia S. Latham, Lopa Mishra. Targeted serum proteomics identifies a signature for use in multivariable logistic regression modeling to predict HCC [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3378.