基于LI-RADS分类的CT征象构建0-A期乙型肝炎相关肝细胞癌预后预测模型的研究

Research on the Development of a Prognostic Model for 0-A Stage Hepatitis B-Related Hepatocellular Carcinoma Based on LI-RADS CT Imaging Features

摘要: 目的：探讨基于LI-RADS分类的CT征象联合临床特征构建预测模型，评估其预测乙型肝炎相关肝细胞癌(HBV-HCC) BCLC 0-A期患者术后2年无复发生存率的价值。方法：回顾性纳入200例接受射频消融(Radiofrequency ablation, RFA, n = 99)或手术切除(Surgical resection, SR, n = 101)的BCLC 0-A期HBV-HCC患者。通过单因素及多因素Cox回归分析筛选独立预后因素，分别构建RFA组和SR组预后预测模型。采用时间依赖ROC曲线、校准曲线、决策曲线(Decision curve, DCA)及一致性指数(C-index)评估模型性能，并通过Kaplan-Meier生存分析验证风险分层能力。采用倾向评分匹配(Propensity score matching, PSM)平衡基线，并进行“治疗互换式交叉验证”，比较高风险/低风险亚组患者接受不同治疗的2年无复发生存期(Recurrence-Free Survival, RFS)差异。结果：单因素及多因素Cox回归分析结果显示，LDH、动脉期瘤周强化、周围血管位置及肿瘤位置是RFA组RFS的独立预测因子；年龄、ALT、CA199及肿瘤位置为SR组RFS的独立预测因子。RFA组模型预测2年RFS的AUC为0.86 (C-index = 0.806, 95% CI: 0.75~0.86)，SR组模型预测2年RFS的AUC为0.79 (C-index = 0.691, 95% CI: 0.61~0.76) PSM后交叉验证显示：46.4%的RFA高风险患者若接受SR治疗，2年RFS可从0.22 ± 0.15提升至0.41 ± 0.05 (P < 0.001)；60.7%的SR高风险患者若接受RFA治疗，2年RFS可从0.65 ± 0.10提升至0.84 ± 0.26 (P < 0.001)。结论：基于LI-RADS的CT影像联合临床指标构建的预测模型可有效识别高危患者，并通过跨治疗分层指导个体化治疗决策。

Abstract: Purpose: To explore the value of a prediction model combining CT imaging features based on LI-RADS classification and clinical characteristics in predicting 2-year recurrence-free survival (RFS) after treatment for patients with hepatitis B virus-related hepatocellular carcinoma (HBV-HCC) at BCLC stage 0-A. Methods: A retrospective cohort of 200 patients with BCLC stage 0-A HBV-HCC who underwent radiofrequency ablation (RFA, n = 99) or surgical resection (SR, n = 101) was included. Independent prognostic factors were screened using univariate and multivariate Cox regression analyses. Separate prognostic prediction models were then constructed for the RFA group and the SR group. Model performance was evaluated using time-dependent receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and the concordance index (C-index). The risk stratification capability was validated using Kaplan-Meier survival analysis. Propensity score matching (PSM) was employed to balance baseline characteristics. “Treatment-swapped cross-validation” was performed to compare the differences in actual 2-year recurrence-free survival (RFS) between high-risk and low-risk subgroups when receiving the alternative treatment. Results: Univariate and multivariate Cox regression analysis identified LDH, arterial peritumoral enhancement, perivascular location, and tumor location as independent predictors of RFS in the RFA group. Age, ALT, CA199, and tumor location were independent predictors of RFS in the SR group. For predicting 2-year RFS, the RFA group model achieved an AUC of 0.86 (C-index = 0.806, 95% CI: 0.75~0.86), while the SR group model achieved an AUC of 0.79 (C-index = 0.691, 95% CI: 0.61~0.76). After PSM, cross-validation showed: For the 46.4% of RFA high-risk patients, if they had received SR instead, their 2-year RFS could have increased from 0.22 ± 0.15 to 0.41 ± 0.05 (P < 0.001). For the 60.7% of SR high-risk patients, if they had received RFA instead, their 2-year RFS could have increased from 0.65 ± 0.10 to 0.84 ± 0.26 (P < 0.001). Conclusion: The prediction model based on LI-RADS CT imaging features combined with clinical indicators can effectively identify high-risk patients. Furthermore, risk stratification across treatments can guide individualized treatment decisions.