Introduction Glycolysis-driven hepatocellular carcinoma (HCC) has been associated with resistance to immune checkpoint inhibitors; however, clinical implementation of glycolysis-targeted therapies remains limited by the reliance on transcriptomic subclassification. Methods We developed a logistic regression model using clinical and pathological features to predict transcriptome-defined glycolysis subclass (training cohort, n = 113). An integer-based scoring system was derived for bedside application. Model performance was assessed using classification metrics, with external validation in TCGA-LIHC cohort. Imaging correlations were analyzed with 18F-fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography/computed tomography (PET/CT) and magnetic resonance imaging (MRI) in a clinical cohort (n = 133) lacking transcriptomic data. In addition, pathology-free LASSO model was also developed using 18F-FDG PET/CT, contrast-enhanced CT/Gd-MRI, and serum biomarkers. Results The integer logistic model (cutoff ≥3), incorporating serum α-fetoprotein, serum albumin, arterial enhancement pattern, and histopathological phenotypes (macrotrabecular-massive, compact, and pseudoglandular), achieved an AUROC of 0.905, accuracy of 0.850, and F1 score of 0.80 in the training cohort, with consistent performance across subgroups. External validation using the fully annotated subset of the TCGA-LIHC cohort (n = 70) demonstrated preserved model performance (AUROC = 0.825). When applied to the clinical cohort, predicted glycolysis-enhanced HCCs demonstrated significantly higher 18F-FDG uptake and distinct hepatobiliary phase defect on Gd-EOB-DTPA–enhanced MRI. Patients classified as the glycolysis subclass showed a trend toward shorter overall survival (median 95 months, p = 0.079) and significantly shorter recurrence-free survival (median 14.4 vs. 37.7 months, p = 0.017). Visual 18F-FDG PET positivity was strongly associated with the glycolysis subclass (p < 0.001) and corresponded to a tumor-to-liver uptake ratio ≥ 1.51. A pathology-free LASSO model using four non-invasive variables demonstrated robust predictive performance (AUROC = 0.893). Elevated glycolytic activity was exploratorily correlated with lower CD8⁺ T-cell infiltration and increased PD-1⁺CTLA-4⁺CD8⁺ T cells. Conclusion We provide a transcriptomics-independent logistic regression model for identifying glycolysis-enhanced HCC. This approach enables practical metabolic subclassification of HCC using routinely available clinical data. Macrotrabecular-massive and compact histopathological patterns, inhomogeneous arterial hyperenhancement, elevated serum α-fetoprotein levels, and 18F-FDG PET positivity are characteristic features of this subclass and may serve as potential indicators to inform the applicability of glycolysis–targeting therapies.
Aoki et al. (Thu,) studied this question.