Background. Several inflammatory and metabolic biomarkers have been associated with hepatocellular carcinoma (HCC) risk in phase I and II biomarker studies but not in longitudinal studies. We conducted a study of promising existing biomarkers using PRoBE (Prospective-specimen collection, retrospective-blinded-evaluation) guidance for conducting a phase III biomarker study, and developed a robust metabolic biomarker panel predictive of HCC.

Methods. We used data and banked serum from a prospective multicenter cohort (Texas HCC Consortium Cohort) of adult patients with cirrhosis but without HCC at the time of enrollment (started in 2016) who were followed until the development of HCC, death or 12/2022. All participants had a liver imaging negative for HCC at baseline. We custom designed a FirePlex immunoassay to measure baseline serum levels of 39 potential HCC biomarkers and established an optimal set of biomarkers with the highest discriminatory ability for HCC. We performed bootstrapping to evaluate the predictive performance using C-index, the time-dependent area under the receiver operating characteristic curve (AUROC), and calibration curve. We quantified the incremental predictive value of the biomarker panel when added to two previously validated models that included clinical factors and two established protein biomarkers.

Results. We developed the model in 2266 patients with cirrhosis of whom 126 (5.6%) patients progressed to HCC during median follow up of 39.9 months. The mean age was 59.6 years, 37.6% were women, 51% White, 31% Hispanic, 16% African American. The risk factors for cirrhosis were alcohol-related liver disease (17%), MASLD (32%), active HCV (13%), cured HCV (24%), or HBV (1%) We identified a 9-biomarker panel (P9) with a C-index of 0.67 (95% CI, 0.66, 0.67), including insulin growth factor 1, interleukin 10, transforming growth factor beta 1, adipsin, fetuin-A, interleukin-1 beta, macrophage stimulating protein alpha chain, serum amyloid A, and tumor necrosis factor alpha. Adding P9 to our previous clinical model with 10 factors (age, sex, HCV status, ALT, platelet count, albumin, AFP, alcohol use, smoking, body mass index) improved AUROC at 1 and 2 years by 4.8% and 2.7%, respectively (p<0.01), and a greater increase (3.2% and 3.6%) in the non-viral hepatitis group. aMAP score (age, male, albumin-bilirubin, and platelets) had a lower predictive value than our base model (0.651 vs 0.715). Adding P9 to aMAP score improved AUROC at 1 and 2 years by 14.2% and 7.6%, respectively (p<0.01). Adding AFP L-3 and DCP did not change the predictive ability of the P9 model.

Conclusions. We identified a panel of 9 circulating serum biomarkers that is independently associated with HCC risk among patients with cirrhosis, and that improved the predictive ability of risk stratification models containing clinical factors.

Performance characteristics of predictive models for HCC risk stratification in patients with cirrhosis.

Hao Duong

Appearances