Machine-Learning Based Prognostic Model for Predicting Early Recurrence in HCC Patients After Hepatectomies: An
Heng-Yuan Hsu1, Jiunn-Chang Lin2,3,4, Chun-Wei Huang1,5
1Division of General Surgery, Department of Surgery, New Taipei Municipal Tucheng Hospital, Built and Operated by Chang Gung Medical Foundation, New Taipei, 23652, Taiwan.
Journal of Hepatocellular Carcinoma
|July 1, 2026
Summary
Machine learning models can predict hepatocellular carcinoma (HCC) recurrence after surgery. Explainable AI identifies high-risk patients for closer monitoring, improving prognosis despite calibration shifts across different centers.
Area of Science:
- Hepatocellular Carcinoma Research
- Machine Learning in Oncology
- Clinical Risk Stratification
Background:
- Early recurrence (within 24 months) post-resection is a major challenge in hepatocellular carcinoma (HCC) management.
- Lack of standardized adjuvant therapies necessitates accurate postoperative risk stratification.
- Explainable machine learning (ML) offers a novel approach to optimize risk modeling using accessible patient parameters.
Purpose of the Study:
- To evaluate explainable machine learning (ML) architectures for optimizing risk stratification in hepatocellular carcinoma (HCC) patients post-resection.
- To identify key predictors of early HCC recurrence using ML models.
- To develop a tool for improved patient risk-tiering and clinical decision-making.
Main Methods:
- Retrospective analysis of 1,681 HCC patients undergoing curative-intent hepatectomy (training cohort) and 251 patients (external validation cohort).
- Comparison of four ML algorithms: random survival forest, Cox-nnet, LASSO, and extreme gradient boosting (XGBoost).
- Feature significance determined using SHAP values; model performance assessed by concordance index (C-index).
Main Results:
- The XGBoost model demonstrated high discrimination (training C-index: 0.98; external validation C-index: 0.72).
- Key predictors identified: sex, preoperative treatment, tumor size, satellite lesions, and vascular invasion.
- A derived nomogram effectively stratified patient risk (p < 0.0001), though absolute recurrence probabilities were overestimated in the external cohort.
Conclusions:
- The XGBoost model provides robust, cross-center generalizability for categorical risk stratification in HCC recurrence.
- This explainable AI tool reliably identifies high-risk patients for enhanced surveillance, rather than estimating absolute recurrence probabilities.
- Prospective validation across diverse populations is needed before widespread clinical implementation.