Comprehensive prognostic risk prediction of hepatocellular carcinoma (HCC) after surgical treatment is particularly important for guiding clinical decision-making and improving postoperative survival. Hence, we aimed to build prognostic models based on serum metabolomics data, and assess the prognostic risk of HCC within 5 years after surgical resection. A pseudotargeted gas chromatography-mass spectrometry (GC-MS)-based metabolomics method was applied to analyze serum profiling of 78 HCC patients. Important metabolic features with discriminant ability were identified by a novel network-based metabolic feature selection method based on combinational significance index (N-CSI). Subsequently, phenylalanine and galactose were further identified to be relevant with mortality by the Cox regression analysis, while galactose and tyrosine were associated with recurrence and metastasis. Two models to predict risk of mortality (risk score of overall survival, RSOS) and risk of recurrence and metastasis (risk score of disease-free survival, RSDFS) were generated based on two panels of metabolites, respectively, which present favorable ability to predict prognosis of HCC, especially when combined with clinical staging system. The performance of models was further validated in an external independent cohort from 91 HCC patients. This study demonstrated that metabolomics is a powerful tool for risk screening of HCC prognosis.
Keywords: Hepatocellular carcinoma; Metabolomics; Network-based metabolic feature selection; Prognosis; Risk score.