A nomogram for predicting overall survival in advanced hepatocellular carcinoma patients receiving radiotherapy combined with targeted therapy: a multicenter retrospective study

Introduction

Hepatocellular carcinoma (HCC) ranks as the sixth most common cancer and the third leading cause of cancer-related mortality worldwide (1). Despite advances in surveillance and early detection, a significant proportion of patients are diagnosed at an advanced stage, where curative treatments such as resection, transplantation, or ablation are no longer viable (2). For these patients, systemic therapies—including tyrosine kinase inhibitors, immune checkpoint inhibitors, and their combinations—have become the mainstay of treatment (3,4). Nevertheless, many patients either fail to respond or develop resistance to systemic agents, underscoring the need for effective locoregional strategies.

Radiotherapy (RT), particularly advanced modalities such as stereotactic body RT (SBRT) and intensity-modulated RT (IMRT), has emerged as a promising treatment option for patients with unresectable or advanced-stage HCC (5,6). RT can offer durable local control and symptomatic relief, especially in cases involving vascular invasion or extrahepatic metastasis (7,8). Despite these advantages, the overall prognosis of patients receiving RT remains unsatisfactory due to tumor heterogeneity and variable treatment responses. This highlights the urgent need for accurate prognostic tools to guide clinical decision-making and personalize treatment strategies.

The integration of machine learning techniques such as least absolute shrinkage and selection operator (LASSO) regression with Cox proportional hazards modeling provides a robust framework for identifying relevant prognostic variables from high-dimensional clinical data (9,10). LASSO is particularly useful in handling multicollinearity and improving model interpretability by selecting a parsimonious set of predictors. When combined with the Cox model, it allows for the construction of clinically meaningful survival prediction tools, such as risk scores or nomograms, which can be validated across independent cohorts (11,12).

In this study, we aimed to develop and validate a LASSO-Cox-based prognostic model for advanced HCC patients treated with RT combined with targeted therapy. Our goal was to identify key clinical factors associated with overall survival (OS) and to construct a predictive tool that can support individualized treatment planning and improve patient stratification in clinical practice. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-967/rc).

Methods

Study design

This retrospective study was conducted using data from The Affiliated Hospital of Southwest Medical University, Chongqing General Hospital, and The Affiliated Traditional Chinese Medicine Hospital of Southwest Medical University in China between April 2016 and March 2024. Eligible patients were consecutively enrolled based on the following inclusion criteria: (I) a confirmed diagnosis of HCC by either histopathological or clinical criteria according to established guidelines; (II) Barcelona Clinic Liver Cancer (BCLC) stage B/C; (III) receipt of RT combined with targeted therapy was the primary treatment modality; and (IV) availability of complete clinical, laboratory, and imaging data.

Patients were excluded if they met any of the following criteria: (I) Child-Pugh C; (II) coexisting malignancies other than HCC; (III) presence of severe hepatic encephalopathy or refractory ascites; or (IV) inability to tolerate RT due to poor performance status or other medical contraindications.

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of The Affiliated Hospital of Southwest Medical University (approval No. KY2020254). All participating institutions were also informed and agreed to the study. Given the retrospective nature of the study, the requirement for informed consent was waived by the ethics committee.

Data

All patients received IMRT combined with targeted therapy. The median radiation dose was 48 Gy, delivered in daily fractions of 3 Gy. The dose of the drug was determined based on the height and weight of the patient. In this retrospective study, clinical data were extracted from the electronic medical records of three tertiary hospitals. The following baseline characteristics were collected: sex, age [mean ± standard deviation (SD)], hepatitis B virus (HBV) infection status, tumor number (with ≥2 classified as multiple tumors), tumor size (mean ± SD), white blood cell (WBC) count (mean ± SD), and alpha-fetoprotein (AFP) level (with ≥400 ng/mL as the cutoff). Liver function was evaluated using the Child-Pugh classification, and tumor staging was assessed using the BCLC system. Additional variables included portal vein tumor thrombosis (PVTT), lymph node metastasis (N) stage, and distant metastasis (M) stage.

All data were reviewed and verified independently by two experienced investigators to ensure accuracy and completeness.

Statistical analysis

Categorical variables were compared using the Chi-squared test, while continuous variables were analyzed with Student’s t-test. OS was estimated using the Kaplan-Meier method, and survival differences between groups were assessed with the log-rank test. To identify independent prognostic factors for OS, the LASSO regression was first applied for variable selection, followed by stepwise multivariable Cox proportional hazards modeling. A prognostic nomogram was developed based on the training cohort and subsequently validated in the validation cohort using receiver operating characteristic (ROC) curve analysis, calibration plots, and decision curve analysis (DCA). All statistical analyses were conducted using R software (version 4.4.3), with a two-sided P value <0.05 considered statistically significant.

Results

Patient characteristics

A total of 248 patients were enrolled in the analysis. Males accounted for the majority (83.1%), and the mean age at diagnosis was 54.3 years. HBV infection was present in 66.9% of the cohort. The majority of patients (67.3%) exhibited multiple tumors (≥2 nodules), with an average tumor diameter of 7.29 cm. The mean WBC count was 6.00 ×10⁹/L. Elevated AFP (≥400 ng/mL) was detected in 43.1% of patients. In terms of liver function, 21.8% of patients were classified as Child-Pugh class B. According to the BCLC staging system, most individuals (78.2%) were categorized as stage C. PVTT was identified in 64.9% of cases. Additionally, N stage and M stage were observed in 48.4% and 25.8% of patients, respectively (Table 1). The cohort was randomly divided into a training cohort (n=148) and a validation cohort (n=100) in a 6:4 ratio. There were no significant differences in baseline characteristics between the two groups (Table 1).

Response, progression-free survival, OS

Among the 248 patients included, 44 (17.7%) achieved complete response (CR), 110 (44.4%) had partial response (PR), 88 (35.5%) exhibited stable disease (SD), and 6 (2.4%) experienced progressive disease (PD). A total of 91 patients died. The median progression-free survival (mPFS) was 7.8 months [95% confidence interval (CI): 7.0–9.1], and the median OS (mOS) was 27.5 months (95% CI: 23.8–not reached). The 1-, 2-, and 3-year OS rates were 70.1%, 55.3%, and 44.6%, respectively (Figure 1A).There was no significant difference in median OS between the training and validation cohorts (27.5 vs. 26.3 months, P=0.45, Figure 1B).

Figure 1 Kaplan-Meier curves for OS. (A) OS curve for all patients (n=248). (B) OS comparison between training (n=148) and validation (n=100) cohorts, with no significant difference (P=0.45). OS, overall survival.

LASSO-Cox

LASSO regression analysis initially selected seven potential prognostic variables: HBV status, Child-Pugh classification, BCLC stage, PVTT, M stage, AFP level, and tumor size (Figure 2A,2B). Subsequent stepwise Cox regression in the training cohort identified four independent predictors of OS: Child-Pugh class, PVTT, M stage, and AFP level (Table 2, Figure 3).

Figure 2 LASSO regression for variable selection. (A) LASSO coefficient profiles of the 12 candidate variables plotted against log(λ). (B) Ten-fold cross-validation for tuning parameter selection. The optimal λ value was selected at the minimum partial likelihood deviance. LASSO, least absolute shrinkage and selection operator.

Figure 3 Forest plot of selected prognostic factors based on multivariate Cox regression. AFP, alpha-fetoprotein; CI, confidence interval; HR, hazard ratio; M, metastasis; PVTT, portal vein tumor thrombosis.

Model construction and validation

A nomogram incorporating Child-Pugh classification, PVTT, M stage, and AFP level was constructed based on the training cohort to predict 1-, 2-, and 3-year OS (Figure 4). In the validation cohort, the model demonstrated favorable discrimination performance, with area under the ROC curve (AUC) values of 0.688, 0.817, and 0.847 for predicting 1-, 2-, and 3-year OS, respectively (Figure 5A). DCA further confirmed the clinical utility of the nomogram, showing net benefit across a range of threshold probabilities for 1-year (Figure 5B), 2-year (Figure 5C), and 3-year (Figure 5D) survival predictions. The calibration plots (Figure S1) also demonstrated good agreement between the predicted and observed survival probabilities at all evaluated time points.

Figure 4 Nomogram for predicting 1-, 2-, and 3-year overall survival. AFP, alpha-fetoprotein; M, metastasis; PVTT, portal vein tumor thrombosis.

Figure 5 Evaluation of the nomogram model performance. (A) Time-dependent ROC curves for predicting 1-, 2-, and 3-year overall survival, with AUCs of 0.688, 0.817, and 0.847, respectively. (B-D) DCA showing the net benefit of the nomogram model (coxm) for predicting 1-year (B), 2-year (C), and 3-year (D) survival, compared to the “treat-all” and “treat-none” strategies. AUC, area under the curve; DCA, decision curve analysis; ROC, receiver operating characteristic.

Discussion

In this multicenter retrospective study, we developed and validated a prognostic model for OS in patients with advanced-stage HCC treated with IMRT combined with targeted therapy. Using a combination of LASSO regression and Cox proportional hazards modeling, we identified four independent prognostic factors—Child-Pugh classification, PVTT, M stage, and elevated AFP levels—which were incorporated into a clinically applicable nomogram. The model demonstrated strong discriminatory ability and calibration, offering a reliable tool to aid in individualized risk stratification and therapeutic planning.

The prognostic significance of liver function status, vascular invasion, tumor dissemination, and tumor biology in advanced HCC has been well-documented. Our findings reinforce these associations in the context of RT-treated patients. Specifically, Child-Pugh class B was independently associated with worse OS, reflecting the detrimental impact of impaired liver reserve on both treatment tolerance and tumor control (13). PVTT, another key factor, is a hallmark of aggressive disease and often limits the feasibility of curative approaches; its identification as an independent predictor highlights the need to refine treatment strategies in this subgroup (14). Additionally, the presence of extrahepatic metastases and high AFP levels—both markers of systemic disease burden and tumor aggressiveness—were consistently linked to poor survival outcomes (15-17). These variables are not only prognostically informative but also biologically meaningful, indicating distinct tumor phenotypes that may respond differently to RT combined with targeted therapy.

Our study contributes meaningfully to the evolving role of RT in HCC management. While traditionally reserved for palliative care or bridging therapy, advances in delivery techniques such as IMRT and SBRT have expanded RT’s role into definitive treatment, particularly for patients who are ineligible for surgery or ablation (18,19). However, the response to RT is highly heterogeneous and difficult to predict using conventional staging systems alone. The nomogram developed in this study addresses this gap by providing a nuanced, individualized risk assessment tool that integrates key clinical features. Importantly, the model maintained predictive accuracy in an independent validation cohort, underscoring its potential for broader clinical application (20).

While LASSO regression is traditionally employed in high-dimensional data settings, we selected this approach to address multicollinearity among clinically relevant variables, such as BCLC stage and M stage. Its use enabled the construction of a parsimonious and stable model, even within a moderate-dimensional framework. Although the final model incorporated four established prognostic factors—Child-Pugh classification, PVTT, M stage, and AFP level—its contribution lies in integrating these parameters into a validated and clinically applicable nomogram tailored for advanced HCC patients undergoing RT combined with targeted therapy. This specific patient subgroup has been underrepresented in prior prognostic modeling efforts, and our study addresses an important gap by providing a practical tool for individualized survival prediction in this treatment context.

The predictive performance of our model, as reflected by time-dependent AUC values of 0.688, 0.817, and 0.847 for 1-, 2-, and 3-year OS, respectively, compares favorably with existing prognostic tools in advanced HCC. Moreover, the DCA and calibration plots confirmed both the clinical utility and accuracy of the nomogram. These strengths suggest that the model may facilitate better patient counseling, assist in clinical trial stratification, and support more rational treatment planning in multidisciplinary settings.

Importantly, the prognostic model enables early identification of high-risk patients who may benefit from more aggressive or combined treatment strategies. For patients identified as high-risk based on the nomogram-derived score, they may require closer surveillance and earlier intervention upon disease progression. Integrating the risk score into multidisciplinary treatment planning may facilitate more tailored care pathways, ultimately aiming to prolong survival and improve quality of life in this challenging patient population (21-23).

Furthermore, unlike traditional staging systems such as BCLC or tumor-node-metastasis (TNM), which often provide only coarse risk stratification, our nomogram integrates individualized clinical features to produce a continuous risk score (24,25). This enables a more refined assessment of prognosis, particularly in patients who may not be clearly categorized into conventional high- or low-risk groups. By quantifying survival probability on a continuous scale, the model can serve as a dynamic decision-support tool within multidisciplinary treatment planning, including tumor board discussions or real-time outpatient decision-making (26,27).

Future research should focus on prospective validation of this model in broader populations, as well as on integrating dynamic and molecular data to improve prognostic granularity. The development of web-based or app-based platforms to facilitate bedside use of the nomogram could further enhance its translational value. Furthermore, exploring the predictive utility of the model in the context of novel treatment strategies, such as RT combined with immune checkpoint inhibitors or anti-angiogenic agents, may provide insights into optimal sequencing and patient selection (28-30). In addition, a recent nationwide cohort study demonstrated a significant association between statin use and improved survival following HCC resection, highlighting the need to incorporate medication history and broader clinical context into future prognostic models (31). Finally, recent advances in ensemble learning techniques—such as multi-model voting and hybrid architectures—offer promising opportunities to enhance model robustness and predictive performance. These methods have shown success in other complex medical prediction domains and may represent a valuable direction for refining future HCC survival models (32,33).

Several limitations should be acknowledged. First, the retrospective nature of the study inherently introduces potential selection bias and unmeasured confounding, despite the use of standardized inclusion criteria and rigorous data verification across participating centers. Second, although the model was validated using a temporally independent cohort within the same study, the overall sample size (n=248) and the number of observed events (n=91 deaths) remain relatively limited, particularly in the validation cohort (n=100). This constraint may affect the statistical power and the precision of survival estimates. Third, all patients were recruited from tertiary hospitals within China, and although data were obtained from multiple centers, the validation remains semi-external rather than fully independent. This limits the generalizability of our findings to different geographic regions, healthcare systems, and patient populations with varying etiologic backgrounds or clinical practices. Lastly, the current model was constructed using baseline clinical parameters alone. The incorporation of dynamic or multi-dimensional data, such as radiomic features, genomic signatures, treatment response, or circulating biomarkers, may further improve prognostic accuracy and should be investigated in future prospective studies.

Conclusions

In conclusion, we developed and externally validated a robust LASSO-Cox-based prognostic model for predicting OS in HCC patients treated with RT combined with targeted therapy.

Acknowledgments

The authors would like to thank all the reviewers who participated in the review and MJEditor (www.mjeditor.com) for its linguistic assistance during the preparation of this manuscript.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-967/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-967/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-967/prf

Funding: This work was financially supported by the Medical Research Foundation of Chongqing General Hospital (No. Y2023YXYJMSXM04 to K.X.), and Science and Technology Research Program of Chongqing Municipal Education Commission (No. KJQN202400116 to K.X.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-967/coif). K.X. reports that this study was supported by Medical Research Foundation of Chongqing General Hospital (No. Y2023YXYJMSXM04), and Science and Technology Research Program of Chongqing Municipal Education Commission (No. KJQN202400116). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of The Affiliated Hospital of Southwest Medical University (approval No. KY2020254). All participating institutions were also informed and agreed to the study. Given the retrospective nature of the study, the requirement for informed consent was waived by the ethics committee.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Tan EY, Danpanichkul P, Yong JN, et al. Liver cancer in 2021: Global Burden of Disease study. J Hepatol 2025;82:851-60. [Crossref] [PubMed]
2. Su K, Liu Y, Wang P, et al. Heat-shock protein 90α is a potential prognostic and predictive biomarker in hepatocellular carcinoma: a large-scale and multicenter study. Hepatol Int 2022;16:1208-19. [Crossref] [PubMed]
3. Fan HL, Chen JL, Liu ST, et al. Remimazolam induced cytotoxicity mediated through multiple stress pathways and acted synergistically with tyrosine kinase inhibitors in hepatocellular carcinoma. Redox Rep 2025;30:2475696. [Crossref] [PubMed]
4. Su K, Guo L, Ma W, et al. PD-1 inhibitors plus anti-angiogenic therapy with or without intensity-modulated radiotherapy for advanced hepatocellular carcinoma: A propensity score matching study. Front Immunol 2022;13:972503. [Crossref] [PubMed]
5. Kirichenko AV, Lee D, Wagner P, et al. Image-Guided Stereotactic Body Radiotherapy (SBRT) with Enhanced Visualization of Tumor and Hepatic Parenchyma in Patients with Primary and Metastatic Liver Malignancies. Cancers (Basel) 2025;17:1088. [Crossref] [PubMed]
6. Su K, Wang F, Li X, et al. Effect of external beam radiation therapy versus transcatheter arterial chemoembolization for non-diffuse hepatocellular carcinoma (≥ 5 cm): a multicenter experience over a ten-year period. Front Immunol 2023;14:1265959. [Crossref] [PubMed]
7. Lee SM, Choi JH, Yoon JH, et al. Efficacy and safety of image-guided hypofractionated radiotherapy for hepatocellular carcinoma with portal vein tumor thrombosis: a retrospective, multicenter study. BMC Cancer 2025;25:736. [Crossref] [PubMed]
8. Xu K, Gu T, Su K, et al. Stereotactic body radiation therapy (SBRT) increases anti-PD-1 antitumor activity by enhancing the tumor immune microenvironment in mice with metastatic hepatocellular carcinoma. Discov Oncol 2025;16:1081. [Crossref] [PubMed]
9. Jiang L, Su K, Wang J, et al. The established of a machine learning model for predicting the efficacy of adjuvant interferon alpha1b in patients with advanced melanoma. Front Immunol 2024;15:1495329. [Crossref] [PubMed]
10. Li X, Cui Y, Gao S, et al. Development and validation of a score model for predicting the risk of first esophagogastric variceal hemorrhage and mortality in patients with hepatocellular carcinoma. Ann Med 2025;57:2490210. [Crossref] [PubMed]
11. Abdulrazzaq MM, Ramaha NTA, Hameed AA, et al. Consequential Advancements of Self-Supervised Learning (SSL) in Deep Learning Contexts. Mathematics 2024;12:758. [Crossref]
12. Ashrafi MR, Amanat M, Garshasbi M, et al. An update on clinical, pathological, diagnostic, and therapeutic perspectives of childhood leukodystrophies. Expert Rev Neurother 2020;20:65-84. [Crossref] [PubMed]
13. Su K, Gu T, Xu K, et al. Gamma knife radiosurgery versus transcatheter arterial chemoembolization for hepatocellular carcinoma with portal vein tumor thrombus: a propensity score matching study. Hepatol Int 2022;16:858-67. [Crossref] [PubMed]
14. Zeng H, Su K, Chen X, et al. A propensity score matching study on survival benefits of radiotherapy in patients with inoperable hepatocellular carcinoma. Sci Rep 2023;13:6879. [Crossref] [PubMed]
15. Johnson P, Zhou Q, Dao DY, et al. Circulating biomarkers in the diagnosis and management of hepatocellular carcinoma. Nat Rev Gastroenterol Hepatol 2022;19:670-81. [Crossref] [PubMed]
16. Abouzied MM, Alhinti N, AlMuhaideb A, et al. Extrahepatic metastases from hepatocellular carcinoma: multimodality image evaluation. Nucl Med Commun 2021;42:583-91. [Crossref] [PubMed]
17. Wang Y, Jian W, Yuan Z, et al. Deep learning with attention modules and residual transformations improves hepatocellular carcinoma (HCC) differentiation using multiphase CT. Precision Radiation Oncology 2025;9:13-22. [Crossref]
18. Wang Y, Guan F, Wang S, et al. Radiation induced liver injury (RILI) evaluation using longitudinal computed tomography (CT) in image-guided precision murine radiotherapy. Precis Radiat Oncol 2024;8:182-90. [Crossref] [PubMed]
19. Wang Q, Qiao W, Zhang H, et al. Nomogram established on account of Lasso-Cox regression for predicting recurrence in patients with early-stage hepatocellular carcinoma. Front Immunol 2022;13:1019638. [Crossref] [PubMed]
20. Wang G, Ding F, Chen K, et al. CT-based radiomics nomogram to predict proliferative hepatocellular carcinoma and explore the tumor microenvironment. J Transl Med 2024;22:683. [Crossref] [PubMed]
21. Li H, Li T, Hu J, et al. A nomogram to predict microvascular invasion in early hepatocellular carcinoma. J Cancer Res Ther 2021;17:652-7. [Crossref] [PubMed]
22. Lu Z, Sun Z, Liu C, et al. Prognostic nomogram for hepatocellular carcinoma with radiofrequency ablation: a retrospective cohort study. BMC Cancer 2021;21:751. [Crossref] [PubMed]
23. Llovet JM, Brú C, Bruix J. Prognosis of hepatocellular carcinoma: the BCLC staging classification. Semin Liver Dis 1999;19:329-38. [Crossref] [PubMed]
24. Gao TM, Bai DS, Qian JJ, et al. The growth rate of hepatocellular carcinoma is different with different TNM stages at diagnosis. Hepatobiliary Pancreat Dis Int 2021;20:330-6. [Crossref] [PubMed]
25. Su K, Shen Q, Tong J, et al. Construction and validation of a nomogram for HBV-related hepatocellular carcinoma: A large, multicenter study. Ann Hepatol 2023;28:101109. [Crossref] [PubMed]
26. Pan YX, Chen JC, Fang AP, et al. A nomogram predicting the recurrence of hepatocellular carcinoma in patients after laparoscopic hepatectomy. Cancer Commun (Lond) 2019;39:55. [Crossref] [PubMed]
27. Liu X, Li H, Wang F, et al. Transhepatectomy combined with arterial chemoembolization and transcatheter arterial chemoembolization in the treatment of hepatocellular carcinoma: a clinical prognostic analysis. BMC Gastroenterol 2023;23:299. [Crossref] [PubMed]
28. Varghese TP, John A, Mathew J. Revolutionizing cancer treatment: The role of radiopharmaceuticals in modern cancer therapy. Precis Radiat Oncol 2024;8:145-52. [Crossref] [PubMed]
29. Liu HY, Lee YD, Sridharan S, et al. Stereotactic body radiotherapy in the management of hepatocellular carcinoma: An Australian multi-institutional patterns of practice review. J Med Imaging Radiat Oncol 2021;65:365-73. [Crossref] [PubMed]
30. Jeon D, Cha HR, Chung SW, et al. Association between statin use and the prognosis of hepatocellular carcinoma after resection: a nationwide cohort study. EClinicalMedicine 2023;65:102300. [Crossref] [PubMed]
31. Abbas Z, Kim S, Lee N, et al. A robust ensemble framework for anticancer peptide classification using multi-model voting approach. Comput Biol Med 2025;188:109750. [Crossref] [PubMed]
32. Zaidi SAJ, Ghafoor A, Kim J, et al. HeartEnsembleNet: An Innovative Hybrid Ensemble Learning Approach for Cardiovascular Risk Prediction. Healthcare (Basel) 2025;13:507. [Crossref] [PubMed]
33. Xu J, Wang T, Li J, et al. A multimodal fusion system predicting survival benefits of immune checkpoint inhibitors in unresectable hepatocellular carcinoma. NPJ Precis Oncol 2025;9:185. [Crossref] [PubMed]