A Machine Learning Approach Yields a Multiparameter Prognostic Marker in Liver Cancer

Abstract A number of staging systems have been developed to predict clinical outcomes in hepatocellular carcinoma (HCC). However, no general consensus has reached regarding the optimal model. New approaches such as machine learning (ML) strategies are powerful tools for incorporating risk factors from multiple platforms. We retrospectively reviewed baseline information, including clinicopathologic characteristics, laboratory parameters, and peripheral immune features reflecting T-cell function, three HCC cohorts. gradient-boosting survival (GBS) classifier was trained with prognosis-related variables training dataset validated two independent constructed a 20-feature GBS model one feature, 14 five function parameters obtained blood mononuclear cells. The model–derived scores demonstrated high concordance indexes (C-indexes): 0.844, 0.827, 0.806 set validation sets 1 2, respectively. could separate patients into high-, medium- low-risk subgroups respect death all datasets (P < 0.05 comparisons). higher score positively correlated stage presence portal vein tumor thrombus (PVTT). Subgroup analyses Child–Pugh class, Barcelona Clinic Liver Cancer stage, PVTT status supported prognostic relevance GBS-derived algorithm conventional system. In summary, multiparameter ML signatures offers different approach identify greatest HCC-related death.