A machine-learning model integrating glypican-3 expression with clinical features showed promising performance for estimating hepatocellular carcinoma recurrence at different intervals after ablation in a retrospective study of 447 patients. The study included people with treatment-naïve, solitary HCC who underwent ablation at one center. Its Extreme Gradient Boosting model was evaluated for overall recurrence and for recurrence within three post-ablation time windows.
The investigators selected five variables for the final model: tumor size, alpha-fetoprotein level, gender, glypican-3 expression and Barcelona Clinic Liver Cancer stage. Feature selection combined logistic regression, Minimum Redundancy Maximum Relevance and Mutual Information. Multiple models were constructed, and Extreme Gradient Boosting, abbreviated XGB, was selected as the best-performing option. The cohort was divided into training, validation and test sets in a 3:1:1 ratio, allowing performance to be assessed across all three partitions rather than in the training data alone.
For overall recurrence, the XGB model achieved area-under-the-curve values of 0.819 in the training set, 0.813 in the validation set and 0.790 in the test set. In the test set, the corresponding value was 0.809 for recurrence during the first year after ablation, 0.776 for recurrence between one and two years, and 0.781 for recurrence after two years. These results indicate that the model’s measured discrimination remained similar across the three time windows in the internal test cohort. They do not by themselves establish how likely an individual patient is to experience recurrence.
Subgroup analyses showed stable performance across gender, age and tumor-size categories, with interaction p-values above 0.7 for all three subgroup assessments. In Kaplan–Meier analysis, the model also separated patients classified as high risk from those classified as low risk. The findings support further evaluation of the GPC-3-integrated XGB model as a possible approach to individualized, time-specific recurrence-risk assessment after HCC ablation. Its broader clinical value remains unconfirmed because the research was retrospective, limited to a single center and assessed through an internal data split. The authors state that prospective multicenter validation is needed to confirm whether the model generalizes beyond the study population.
Subject of Research: Time-specific recurrence-risk prediction after ablation for hepatocellular carcinoma using glypican-3 and clinical features
Article Title: Machine learning-based prediction of time-dependent recurrence risk after ablation for hepatocellular carcinoma integrating glypican-3 and clinical features: a retrospective study
Article References: Li, H., Li, X., Li, J., Sun, Q., Bian, L., Zhang, Z., Wang, L., & Gao, Y. (2026). Machine learning-based prediction of time-dependent recurrence risk after ablation for hepatocellular carcinoma integrating glypican-3 and clinical features: a retrospective study. BMC Cancer. https://doi.org/10.1186/s12885-026-16804-7
Image Credits: AI Generated
DOI: 10.1186/s12885-026-16804-7
Keywords: hepatocellular carcinoma, recurrence prediction, tumor ablation, glypican-3, machine learning, XGB, clinical features, retrospective study, time-specific risk, Machine, learning-based, prediction
News Source: Nathaniel Bowman. (October 22, 2026). Retrospective study tests time-specific HCC recurrence prediction after ablation. Scienmag.
