Machine learning-based clustering differentiates bilateral hepatocellular carcinoma characteristics and prognosis.

Abstract

Bilateral hepatocellular carcinoma represents a biologically heterogeneous disease with uncertain optimal surgical selection criteria. Although hepatic resection can provide survival benefit in selected patients, outcomes remain variable, and phenotypic determinants of prognosis have not been clearly defined. Patients who underwent curative-intent hepatectomy for bilateral hepatocellular carcinoma from a multi-institutional database from 2000 to 2023 were reviewed. k-means clustering to identify phenotypic subgroups. Overall survival and recurrence-free survival were evaluated using Kaplan-Meier product-limit methods and multivariable Cox regression. Among 347 patients, unsupervised clustering identified 2 distinct phenotypes: cluster 1 (n = 95) characterized by larger tumors, higher tumor burden, non-cirrhotic livers, and frequent major hepatectomy; and cluster 2 (n = 252) characterized by multiple smaller lesions, treated with greater use of minimally invasive surgery and concomitant ablation. Cluster 1 demonstrated more aggressive pathologic features and worse survival (5-year overall survival 34.9 vs 63.2%; P < .001) and higher early recurrence (53.7 [n = 51] vs. 38.1% [n = 96]; P = .02). Cluster classification remained independently associated with overall survival (hazard ratio [HR], 2.15; 95% CI, 1.48-3.13; P < .001) and recurrence-free survival (HR 1.52; 95% CI, 1.11-2.06; P < .01) even after adjustment for margin status and microvascular invasion. Bilateral hepatocellular carcinoma encompasses 2 clinically relevant phenotypes with distinct tumor burden, operative strategies, and outcomes. Phenotype-based stratification using preoperative clinical and radiologic variables may help refine surgical selection. Low-burden multifocal disease achieved favorable survival, whereas dominant plus satellites disease had inferior prognosis, supporting multimodal strategies for this subgroup.