Effect of neoadjuvant radiotherapy combined with surgery on long-term prognosis of patients with centrally located hepatocellular carcinoma: A machine learning and propensity score weighting analysis.

Abstract

e16259 Background: Primary liver cancer is the sixth most common malignancy globally and the third leading cause of cancer-related deaths. Centrally located hepatocellular carcinoma (HCC) is a distinct subtype of liver cancer, typically arising in the central liver segments. It is further defined as tumors adherent to or within 1 cm of major intrahepatic vascular or biliary structures. This anatomical location makes surgical resection technically challenging, with a higher risk of recurrence. Neoadjuvant radiotherapy (NeoRT) has been increasingly applied in the treatment of HCC; however, its therapeutic efficacy remains uncertain. Methods: This study retrospectively collected data on HCC patients who underwent liver resection (LR) from January 2018 to June 2024. The inclusion criteria as follows: (1) age ≥18 years; (2) diagnosed with centrally located HCC; (3) Child-Pugh A; (4) BCLC stage 0 or A. The exclusion criteria as follows: (1) portal venous tumor thrombus; (2) undergoing other forms of radiation therapy; (3) multiple liver tumors. Patients were divided into two groups based on whether they received NeoRT. Directed acyclic graph (DAG) was constructed to identify potential confounders. Propensity scores (PS) were calculated using seven machine learning algorithms, including decision tree, KNN, LDA, logistic regression, random forest, SVM and xgboost. The best algorithm was selected according to standardized mean difference. Subsequently, the optimal weighting method of four propensity score weighting was used to balance confounders, including entropy, inverse probability, matching and overlap weighting. Weighted Kaplan-Meier, Cox regression, E-value calculation, and landmark analysis were performed to evaluate the association between NeoRT and recurrence-free survival (RFS). Registration number: ChiCTR2300072378. Results: This study included 480 patients, comprising 430 in the LR group and 50 in the NeoRT group. 22 potential confounders were identified using DAG. PS were estimated via an optimal logistic regression based machine learning. Group imbalances were addressed using overlap weighting. Before weighting, the 1-year, 3-year, and 5-year RFS rates were 80%, 63%, and 47% in the NeoRT group versus 60%, 36%, and 24% in the LR group. Weighted Kaplan-Meier analysis showed that the NeoRT group had significantly better RFS (P < 0.01). Weighted Cox regression indicated that NeoRT was an independent protective factor for RFS (HR = 0.42, 95% CI: 0.23-0.76, P = 0.005). The E-value for unmeasured confounding was 3.06, supporting the robustness of findings. Landmark analysis further confirmed the superior RFS in the NeoRT group (P < 0.05). Conclusions: NeoRT combined with surgery is a safe and effective strategy for the treatment of centrally located HCC, contributing to improved RFS.