Machine learning-based clustering differentiates bilateral hepatocellular carcinoma characteristics and prognosis
Yutaka Endo1, Jun Kawashima2, Odysseas Chatzipanagiotou3
1Department of Surgery, The Ohio State University Wexner Medical Center and James Comprehensive Cancer Center, Columbus, OH; Division of Transplant Surgery, University of Rochester Medical Center, Rochester, NY.
Surgery
|April 9, 2026
Summary
Bilateral hepatocellular carcinoma presents two distinct phenotypes impacting patient survival. Identifying these phenotypes preoperatively can guide surgical selection for better outcomes in liver cancer treatment.
Area of Science:
- Hepatobiliary surgery
- Surgical oncology
- Liver cancer research
Background:
- Bilateral hepatocellular carcinoma (BHCC) is heterogeneous, complicating surgical strategy.
- Optimal criteria for surgical selection in BHCC remain unclear.
- Phenotypic factors influencing BHCC prognosis are not well-defined.
Purpose of the Study:
- To identify distinct phenotypic subgroups within bilateral hepatocellular carcinoma.
- To evaluate the prognostic implications of these phenotypes after surgical resection.
- To refine surgical selection criteria for BHCC.
Main Methods:
- Retrospective review of 347 patients undergoing curative-intent hepatectomy for BHCC (2000-2023).
- k-means clustering to define phenotypic subgroups based on clinical and radiological data.
- Survival and recurrence rates analyzed using Kaplan-Meier and Cox regression.
Main Results:
- Two phenotypes identified: Cluster 1 (high tumor burden, larger tumors, non-cirrhotic liver) and Cluster 2 (multiple smaller lesions, cirrhotic liver).
- Cluster 1 showed worse survival (5-year OS 34.9% vs 63.2%) and higher recurrence rates.
- Phenotype independently predicted overall survival (HR 2.15) and recurrence-free survival (HR 1.52).
Conclusions:
- BHCC comprises two clinically significant phenotypes with differing outcomes.
- Preoperative stratification by phenotype can aid surgical decision-making.
- Low-burden multifocal disease has favorable outcomes, while dominant plus satellite lesions require multimodal approaches.
