Performance of AI-based machine learning models for overall survival prediction in advanced hepatocellular carcinoma patients receiving immunoradiotherapy

Shanbao Ke^1*Feng Xiao¹Xiaonan Wang²Shengyuan Luo³Jianwei Zhou^1*

• ¹Henan Provincial People's Hospital, Zhengzhou, China
• ²Boxing People's Hospital, Binzhou, China
• ³Changzhi Medical College, Changzhi, China

Background: Hepatocellular carcinoma (HCC) remains a leading cause of cancer-related mortality worldwide. Although immunotherapy and targeted therapy have improved survival in advanced HCC, outcomes remain heterogeneous. Radiotherapy (RT) may enhance systemic treatment efficacy through local control and immunomodulation. Artificial intelligence (AI) offers opportunities to integrate multimodal data for individualized prognostic assessment. Methods: A total of 175 HCC patients were included in this study: 115 in the RT group (RT + immunotherapy + targeted therapy) and 60 in the non-RT group (immunotherapy + targeted therapy). Baseline characteristics were analyzed with chi-square and Mann–Whitney U tests. Overall survival (OS) was compared using the Kaplan–Meier method and log-rank test. Patients were randomly divided into a training cohort and a validation cohort (6:4 ratio). Prognostic factors were identified in the training cohort and incorporated into 101 machine learning (ML) algorithms. Model performance was assessed using the concordance index (C-index), receiver operating characteristic (ROC) curves, and risk score stratification. Results: The RT group achieved significantly longer OS than the non-RT group (median OS: 15.4 vs. 8.5 months, P = 0.003). Four variables ("Child," "BCLC stage," "Size," and "Treatment") were identified as prognostic factors. Among 101 ML models, the StepCox (forward) + Ridge model showed the best performance (C-index: 0.68 in training, 0.65 in validation). Time-dependent ROC analysis demonstrated AUC values of 0.72, 0.75, and 0.74 at 1-, 2-, and 3-year OS in the training cohort, and 0.72, 0.75, and 0.73 in the validation cohort, respectively. Conclusion: RT significantly improved prognosis in advanced 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.