Regression Model for Predicting the Glycolysis-Driven-HCC Subclass

Objective: Enhanced glycolysis in gastric and lung cancers raises lactate levels, upregulating PD-1 in Tregs and inducing anti-PD-1/PD-L1 inhibitor resistance. This study aimed to create a model to predict the glycolysis-driven-HCCs without genetic analysis.
Methods: (1) Genetic analysis of 136 HCCs identified the glycolysis subclass and evaluated the exclusivity using PCA/silhouette analysis. Its generalizability was validated in 916 HCCs. (2) A logistic regression model to predict the glycolysis-driven-HCC using non-genetic biomarkers was developed and evaluated for accuracy.
Results: (1) Hierarchical clustering identified a highly excluded glycolysis subclass with increased lactate production (KEGG pathway), upregulated cell cycle genes (GO analysis), frequent TP53 mutations, poor differentiation, high vascular invasion, elevated AFP/DCP levels, and poor prognosis. Inhomogeneous/rim-APHE were characteristic on CECT/MRI, and the deconvolution-algorithm revealed heterogeneous TIME. (2) Cut-off were determined using ROC analysis and Chi-square/Fisher's tests showed that tumor diameter>5.75cm, NLR>1.95, Cre>0.755, AFP>19.5, DCP>1824, inhomogeneous-APHE(Li-RADS), poorly differentiated, Compact/MTM, and non-Pseudoglandular/non-Microtrabecular types were associated with glycolysis subclass. Logistic regression analysis with gender, albumin, and total bilirubin levels indicated that AFP>19.5, inhomogeneous-APHE(Li-RADS), Compact type, non-Pseudoglandular, and albumin>3.15 were significant factors (p<0.05). The model demonstrated an AUROC of 0.893, accuracy of 0.898, kappa coefficient of 0.788, positive predictive value of 0.883, and negative predictive value of 0.907 for diagnosing the glycolysis subclass.
Conclusion: A glycolysis-driven HCC subclass, characterized by high lactate production, can be diagnosed through clinical, imaging, and pathological findings without requiring comprehensive RNA/DNA analysis.