Objective: To develop an XGBoost machine learning model incorporating the SAT score to predict prognosis in patients with intermediate/advanced HCC receiving targeted immunotherapy combined with TACE, enabling personalized treatment.
Methods: We retrospectively analyzed 186 advanced HCC patients treated from July 2016 to February 2023. Patients were stratified using the "6 and 12" SAT score. Subgroup analysis assessed its predictive value for progression-free survival (PFS) and overall survival (OS). Feature selection and modeling (predicting 2-year survival) employed XGBoost, Random Forest (RF), and K-Nearest Neighbors (KNN), incorporating SAT score, treatment mode, clinical features, and hematological indicators. Model performance was evaluated via cross-validation, ROC curves, and decision curve analysis (DCA). The SHAP package explained feature importance in the optimal model.
Results: SAT score, high serum AFP, tumor invasion of the liver surface, MVI, multiple lesions, and gender were independent risk factors for PFS and OS (P < 0.05). For SAT score ≤6, TACE alone yielded higher 2-year survival than combination therapy. For SAT score=7, OS did not differ significantly between treatments (P=0.86). For SAT score>7, OS significantly improved with combination therapy (P < 0.05). The XGBoost model demonstrated excellent prognostic prediction, showing high accuracy and clinical utility.
Conclusions: The SAT score has significant value in selecting treatments for advanced HCC. The XGBoost-based machine learning model incorporating the SAT score provides high predictive value for the prognosis of advanced HCC patients receiving targeted immunotherapy combined with TACE.