Abstract
Radiotherapy (RT) significantly improved prognosis in advanced hepatocellular carcinoma (HCC) patients treated with immunotherapy and targeted therapy. Among multiple algorithms, the StepCox (forward) + Ridge model achieved superior predictive performance, supporting its potential value in individualized prognostic assessment.
Introduction
Hepatocellular carcinoma (HCC) remains a global health challenge, ranking as the third leading cause of cancer-related mortality. While the combination of immunotherapy and targeted therapy has improved outcomes, patient responses are highly heterogeneous. Recent evidence suggests that Radiotherapy (RT) can act as an "in situ vaccine," enhancing the systemic effects of immunotherapy. However, clinicians lack reliable tools to predict which patients will benefit most from this triple-threat "immunoradiotherapy" approach. This study validates the benefit of adding RT and develops an AI-driven model to predict overall survival (OS) using standard clinical data.
Methods
The researchers conducted a retrospective study of 175 HCC patients from three tertiary hospitals in China. To ensure a fair comparison between those who received RT (n=115) and those who did not (n=60), Propensity Score Matching (PSM) was employed to balance baseline characteristics such as age, sex, and tumor stage.
To identify the most accurate predictive framework, the study screened 101 machine learning (ML) algorithm combinations. These combinations paired various feature selection methods (e.g., Lasso, StepCox, Random Survival Forest) with different survival models (e.g., Ridge, Cox, Elastic Net). The dataset was randomly divided into a training cohort (n=105) and a validation cohort (n=70) at a 6:4 ratio. Models were built using 11 clinical variables, including Child-Pugh class, BCLC stage, tumor size, and serum AFP levels.
Key Results
Clinical Synergy of Radiotherapy
The addition of RT to the immunotherapy and targeted therapy regimen significantly improved patient outcomes. Patients receiving RT achieved a median overall survival of 15.4 months, compared to 8.5 months for those in the non-RT group (p = 0.003).
Optimal Machine Learning Framework
Out of 101 contenders, the combination of StepCox (forward) for variable selection and Ridge regression for modeling emerged as the superior framework. This model achieved a C-index of 0.68 in the training set and 0.65 in the validation set. Time-dependent ROC analysis demonstrated robust performance, with AUC values ranging from 0.72 to 0.75 for 1-, 2-, and 3-year survival predictions.
The final model relies on four key factors: Child-Pugh class, BCLC stage, tumor size, and treatment modality. By using these standard clinical metrics, the model is highly deployable in real-world hospital settings without requiring expensive genomic sequencing.
Discussion and Limitations
While the StepCox + Ridge model successfully stratified patients into high- and low-risk groups, the authors acknowledge several limitations. The retrospective nature of the study and the relatively small sample size (175 patients) necessitate further validation in larger, multi-center prospective cohorts. Additionally, the current model relies solely on clinical data; integrating radiomics or genomics could potentially enhance its predictive accuracy in future iterations.
Conclusion
This study demonstrates that adding RT to immunotherapy and targeted therapy significantly improves survival in advanced HCC patients. The developed AI-based StepCox + Ridge model provides a reliable, non-invasive tool for individualized prognostic assessment, supporting more personalized treatment strategies in clinical practice.
Summary updated