Prediction of early treatment response to drug-eluting beads chemoembolization for hepatocellular carcinoma using
Jian Li1,2,3, Ruopeng Liang1,2,3, Weijie Wang1,2,3
1Department of Hepatobiliary and Pancreatic Surgery, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan, China.
Background:
Drug-eluting bead trans-arterial chemoembolization (DEB-TACE) has been extensively employed as a locoregional therapy for hepatocellular carcinoma (HCC). In this work, machine-learning algorithms are used to forecast early treatment response in HCC patients undergoing DEB-TACE.
Methods:
We collected data on patients with HCC who underwent anthracycline-loaded DEB-TACE at our institution over two periods: July 2023 to November 2024 and January to September 2025. The treatment response was evaluated at 1, 3, and 6 months after therapy by dynamic contrast-enhanced computed tomography (CT) based on the modified Response Evaluation Criteria in Solid Tumors (mRECIST). Univariable and multivariable logistic regression analyses were conducted. Independent predictors identified were used to build the predictive model. Machine learning models were evaluated using multiple performance metrics, including AUC, accuracy, sensitivity, and specificity, across the internal validation set and the temporal validation cohort.
Results:
A total of 299 HCC patients were included (196 internal cohort, 103 temporal validation cohort), with objective response rates at 1, 3, and 6 months of 56.1%, 51.2%, and 45.9% for the internal cohort and 57.3%, 52.4%, and 49.5% for the temporal validation cohort. Risk factor analysis identified independent predictors of treatment response at various time points, including age, BCLC stage, tumor distribution, and largest tumor diameter. LR showed the highest AUC across all time points, with values of 0.830, 0.840, and 0.854 in the internal validation set and 0.817, 0.832, and 0.839 in the temporal validation cohort at M1, M3, and M6. Subgroup analyses suggested generally consistent predictive performance of the LR model across selected treatment-related subgroups.
Conclusion:
In the treatment of hepatocellular carcinoma with DEB-TACE, machine learning models demonstrated promising performance in predicting early treatment response, supporting the development of personalized treatment strategies in the future.
