Development and validation of machine learning-based model for predicting early recurrence for patients with HBV-associated hepatocellular carcinoma after curative hepatectomy - 06/12/25
Abstract |
Aims |
Early recurrence (ER) is strongly associated with poor long-term survival in patients with hepatocellular carcinoma (HCC). This study aimed to explore a prediction model based on machine learning (ML).
Methods |
Six ML algorithms were constructed and compared. The top-performing model was further compared with a conventional logistic regression model. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) values at both global and individual levels.
Results |
Among 903 patients included, 351 (38.9 %) experienced ER within two years. As a result, the random forest (RF) model was selected and demonstrated superior discrimination (AUC 0.917) compared with the logistic regression model (AUC 0.853). SHAP analysis identified multiple tumors, microvascular invasion, and tumor size > 5 cm as the key contributors to ER. An online calculator based on the RF model is accessible at: HCCEarlyRecurrencePredictor/ .
Conclusion |
The RF model offers a clinically interpretable and deployable tool to support individualized postoperative surveillance strategies.
Le texte complet de cet article est disponible en PDF.Highlights |
• | Significant advance in prognosis prediction models with high predictive efficiency for early-stage recurrence (ER) of hepatocellular carcinoma (HCC). |
• | Systematic comparison of traditional logistic regression analysis and machine learning (ML) for ER of HCC. |
• | Robust ML model performance in patients with HBV-related HCC after curative hepatectomy. |
• | Potential for statistical software advancement in online calculator. |
Keywords : Hepatocellular carcinoma, Hepatectomy, Early recurrence, Machine learning, Random forest, SHapley additive exPlanations
Abbreviations : HCC, ER, HBV, MVI, AFP, CA19-9, ALB, GGT, AST, PT, PLT, ML, RF, XGBoost, SVM, KNN, DT, NB, LR, ROC, AUC, PR, DCA, SHAP
Plan
Vol 251
Article 116716- janvier 2026 Retour au numéro
Article précédent
- A simple five-factor risk model for predicting new-onset diabetes after distal pancreatectomy
- Wei-Hsun Lu, Ting-Kai Liao, Ping-Jui Su, Chih-Jung Wang, Ying-Jui Chao, Yan-Shen Shan
- Hands-on training for preoperative skin preparation improves compliance with manufacturer instructions for use
- Diego Schaps, Michael Schultz, Regina Woody, Sabran Masoud, Christopher Mantyh
Bienvenue sur EM-consulte, la référence des professionnels de santé.
L’accès au texte intégral de cet article nécessite un abonnement.
Déjà abonné à cette revue ?