Machine Learning Risk Model Outperforms TNM Staging in Post-Surgical HCC Recurrence

Article Impact Level: HIGH
Data Quality: STRONG
Summary of  Gut  https://doi.org/10.1136/gutjnl-2026-338227  
Dr. Ying Zhang  et al.

Points

Clinician-scientists in Singapore developed a multi-omics machine-learning tool that accurately predicts post-surgical recurrence risks in hepatocellular carcinoma patients.
Genomic evaluation of the PLANet cohort revealed that 64.2% of post-resection patients experienced cancer recurrence across intrahepatic or distant extrahepatic metastatic sites.
Intrahepatic relapses frequently developed through polyclonal seeding, creating early recurrences characterized by high cell plasticity and a regulatory T cell-enriched microenvironment.
Extrahepatic distant metastases primarily originated through monoclonal seeding driven by an aggressive C5/C6 subclone existing within the primary liver tumor.
Incorporating a 15-gene signature into the predictive model yielded an 86% area under the curve score, outperforming conventional TNM staging systems.

Summary

the clonal evolution and predictive risk factors for post-surgical recurrence in patients with hepatocellular carcinoma (HCC). Hepatocellular carcinoma constitutes the third leading cause of cancer-related mortality globally, with intrahepatic recurrence accounting for 70% to 80% of post-resection relapse cases. Led by researchers across the National Cancer Center Singapore, Duke-NUS Medical School, and A*STAR Genome Institute of Singapore, the investigation analyzed clinical outcomes and multi-omics profiles within the PLANet cohort to elucidate the molecular mechanisms driving tumor recurrence and improve post-surgical risk stratification.

Comprehensive genomic and transcriptomic profiling of 106 post-resection HCC patients identified recurrence in 68 individuals (64.2%), comprising 48 intrahepatic recurrences, 11 distant metastases, and 9 co-occurring local and distant recurrences. Genomic analysis demonstrated two distinct clonal dissemination patterns: polyclonal seeding and monoclonal seeding. More than half of intrahepatic recurrences originated via polyclonal seeding, exhibiting early relapse, marked phenotypic plasticity, and a regulatory T cell-enriched immunosuppressive microenvironment. Conversely, monoclonal seeding drove delayed recurrences and distant extrahepatic metastases, originating predominantly from an aggressive dominant C5/C6 subclone within the primary tumor.

To improve prognostic precision, investigators developed a machine learning-based multi-omics risk stratification model incorporating clinical parameters, tumor dimensions, alpha-fetoprotein levels, and a 15-gene expression signature. Validated across three independent cohorts including the TCGA-LIHC dataset, the multi-omics tool achieved an area under the receiver operating characteristic curve (AUC) of 86%, significantly outperforming traditional TNM staging (AUC 56% to 68%). The authors conclude that integrating clonal seeding dynamics with multi-omics profiling enables precise post-resection recurrence risk stratification, offering a robust framework for personalizing adjuvant therapy trials and surveillance strategies in HCC.

Link to the article: https://gut.bmj.com/content/early/2026/07/21/gutjnl-2026-338227

References

Zhang, Y., Sekar, K., Phua, C. Z. J., Yap, C. K., Shuen, T. W. H., Kaya, N. A., Liu, M., Lu, B., Tan, C. Y. L., Chong, S. L., Seshachalam, V. P., Chew, S. C., Wu, L., Chen, J., Kendarsari, R. I., Lim, J. Q., Zheng, Q., Chiu, J. Y. H., Liu, Z., … Tam, W. L. (2026). Clonal diversity underpins distinct modes of recurrence in hepatocellular carcinoma: The PLANet cohort study. Gut, gutjnl-2026-338227. https://doi.org/10.1136/gutjnl-2026-338227