Prognostic Models for Small Hepatocellular Carcinoma Using Inflammatory Indices and Machine Learning: A Propensity
1Hepatobiliary & Hydatid Disease Department, Digestive & Vascular Surgery Center, First Affiliated Hospital of Xinjiang Medical University, State Key Laboratory of Pathogenesis, Prevention and Treatment of High Incidence Diseases in Central Asia, Urumqi, People's Republic of China.
Background:
The prognosis for patients with small hepatocellular carcinoma (HCC) after curative resection is variable. Early recurrence (within 2 years) remains a significant clinical challenge, closely associated with poor long-term outcomes. Although inflammatory biomarkers have shown prognostic value, integrated models for predicting both early recurrence and long-term survival are lacking. This study developed and validated prognostic models for overall survival (OS) and recurrence-free survival (RFS) in HCC patients post-resection, employing propensity score matching (PSM) to control for confounders between small (≤3 cm) and non-small HCC, with an emphasis on assessing model performance within the small HCC subgroup.
Methods:
A retrospective analysis of HCC patients who underwent hepatectomy was performed. PSM (1:1 nearest-neighbor with replacement) was applied to balance baseline characteristics between small and non-small HCC groups. After matching, a balanced cohort was used for survival analysis. Independent prognostic factors were identified through Cox regression. A traditional nomogram and risk score models were developed and compared against three machine learning models (LASSO-Cox, Random Forest, XGBoost) using time-dependent AUC.
Results:
Following PSM, 165 patients (90 with small HCC and 75 with non-small HCC) were included. Small HCC patients demonstrated significantly better OS and RFS (both p < 0.01). Multivariate analysis identified tumor size, SII, and AAR as independent predictors for OS, and tumor size, PAR, NLR, and GPR for RFS. The LASSO-Cox model exhibited the best overall performance, achieving the highest accuracy for early recurrence (2-year RFS AUC = 0.727) and competitive accuracy for long-term survival (5-year OS AUC = 0.698). The nomogram retained good interpretability (5-year OS AUC = 0.691).
Conclusion:
This study establishes a comprehensive prognostic framework for small HCC, integrating tumor size with systemic inflammation (SII, NLR), liver function (AAR, GPR), and nutritional-coagulation status (PAR). The LASSO-Cox model is recommended for predicting both early recurrence and long-term survival, offering refined risk stratification to guide postoperative management.
