The "A-B-C" Classification Reveals Outcome Trajectories in Patients With Hepatocellular Carcinoma Under...

We aimed to identify subgroups of patients with unresectable hepatocellular carcinoma (HCC) using a non a priori, data-driven machine learning approach to uncover subgroups with distinct profiles across a range of clinical and tumor-related endpoints.

This international study across 12 centers included patients with unresectable HCC receiving first-line atezolizumab/bevacizumab. Baseline clinical, biological, and tumor characteristics were collected. Clusters were correlated with clinical and tumor-related outcomes, including RECIST 1.1 response, histology, hepatic decompensation, patterns of progression, overall survival (OS) and progression-free survival (PFS).

Among 1399 patients (85% male; 75% cirrhotic; 39% with portal vein invasion [PVI]; 35% with metastasis), divided into derivation (n = 958) and validation (n = 409) cohorts, we identified 3 subgroups: Cluster A (47.5% patients): older age, higher body mass index (BMI), preserved liver function, multiple small tumors; Cluster B (11%): VP1/VP2 PVI, higher hepatitis B virus (HBV), and lower metabolic syndrome; Cluster C (41.2%): moderate liver dysfunction, elevated alpha-fetoprotein, and either VP3/VP4 PVI or high tumor burden. Cluster C had poor tumor differentiation (P = .02) and more macrotrabecular massive subtype (P = .02). Cluster A had longer OS (median not reached) vs Clusters B (18.2 months) and C (14.1 months; P < .0001), lower progression rates (25.1% vs 34.3% and 41.2%; P < .0001), and longer PFS (median 10.93 vs 7.73 and 6.10 months; P < .0001). Clusters correlated with prognosis across Barcelona Clinic Liver Cancer stages, occurrence of early hepatic decompensation and patterns of progression (P < .001).

The Atezolizumab-Bevacizumab-Cluster (A-B-C) classification for unresectable HCC identified subgroups with histology, hepatic decompensation, and patterns of progression, and prognostic trajectories, informing hypothesis generation for trial design and patient selection.