The value of the serum cytokine-to-immune ratio model in predicting the progression-free survival after interventional therapy for hepatocellular carcinoma

To develop and validate a prognostic nomogram for predicting progression-free survival (PFS) in patients with hepatocellular carcinoma (HCC) following transcatheter arterial chemoembolization (TACE) combined with radiofrequency ablation (RFA). The model incorporates serum transforming growth factor-β1 (TGF-β1), interleukin-8 (IL-8), and the peripheral blood CD4+/CD8 + ratio to enable precise risk stratification and inform adjuvant therapy decisions. This retrospective study enrolled 138 patients with HCC who underwent initial TACE/RFA between January 2020 and December 2023. Patients were randomly divided into training (n = 97) and internal validation (n = 41) sets at a 7:3 ratio. PFS was the primary outcome. Independent prognostic factors were identified via univariate and multivariate Cox regression analyses in the training set and were used to construct a nomogram. Concurrently, random forest (RF), k-nearest neighbors, and gradient boosting models were developed, with hyperparameters optimized by ten-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). The SHAP (SHapley Additive exPlanations) framework was applied to interpret the optimal model. Patients were stratified into risk groups based on the median nomogram score; survival differences were evaluated using Kaplan-Meier curves and the log-rank test. Multivariate Cox analysis confirmed tumor number, BCLC stage, serum transforming growth factor-β1 (TGF-β1), serum interleukin-8 (IL-8), and the CD4+/CD8 + ratio as independent prognostic factors for PFS (all P < 0.05). Multiple tumors, advanced BCLC stage (B/C), and elevated TGF-β1 and IL-8 levels were independent risk factors, whereas a higher CD4+/CD8 + ratio was a protective factor. Among the machine learning models, RF demonstrated superior predictive performance, with AUCs of 0.846 (95% CI: 0.742–0.949) in the training set and 0.791 (95% CI: 0.619–0.964) in the validation set. Calibration curves for the RF model indicated excellent agreement between predicted and observed outcomes. DCA showed a favorable net clinical benefit across a wide range of threshold probabilities. SHAP analysis identified serum TGF-β1 as the most influential predictor; its effect direction was consistent with the Cox model, with high TGF-β1, high IL-8, advanced BCLC stage, and multiple tumors increasing the risk score, and a high CD4+/CD8 + ratio decreasing it. A nomogram was subsequently constructed based on these factors. The risk stratification derived from the nomogram effectively distinguished patients with significantly different PFS outcomes in both cohorts (P < 0.05, log-rank test). The prognostic nomogram integrating TGF-β1, IL-8, CD4+/CD8 + ratio, and clinicopathological factors demonstrates robust discrimination, calibration, and clinical utility for patients with HCC after TACE/RFA. The RF algorithm enhanced predictive accuracy, while SHAP improved model interpretability. This model effectively identifies high-risk patients, offering an objective foundation for individualized surveillance and adjuvant treatment strategies.