【摘要】 目的 建立并验证肝癌患者术后1、2、3年预后模型。方法 获取2010—2016年SEER数据库内1 526例肝癌手术治疗患者信息，按7:3随机分为建模组（n=1 070）和内部验证组（n=456）；获得TCGA数据库148例肝癌患者术后信息为外部验证组1(TCGA队列)，收集黄山市人民医院2012—2021年102例肝癌患者术后信息为外部验证组2（黄山队列）。建模组用单因素Cox回归、Lasso回归和多因素Cox回归分析预后因素，并构建列线图模型。用Brier评分、C指数、ROC曲线下面积（AUROC）及校正曲线评估模型的预测效能。最后验证CNLC分期在该模型的效能。结果 多因素Cox回归分析显示，AFP水平、分化程度、TNM分期、手术方式和系统治疗是影响肝癌患者术后预后的独立因素。将上述因素与性别、放疗纳入构建总生存期（OS）列线图模型。列线图在建模组的C指数为0.712,1、2、3年Brier评分分别为0.085、0.135和0.171,1、2、3年AUROC分别为0.745、0.742和0.758；在内部验证组的C指数为0.727,1、2、3年Brier评分分别为0.084、0.124和0.160,1、2、3年AUROC分别为0.732、0.763和0.742；在TCGA队列的C指数为0.705,1、2、3年Brier评分分别为0.089、0.140和0.135,1、2、3年AUROC分别为0.684、0.706和0.789；在黄山队列的C指数为0.700;1、2、3年Brier评分分别为0.101、0.110和0.129;1、2、3年AUROC分别为0.755、0.734和0.731。校正曲线显示模型预测上述4组肝癌患者术后1、2、3年OS与实际OS有较好的一致性。CNLC分期列线图的预测效能优于TNM分期。结论 本研究建立的列线图预测模型可有效预测肝癌患者术后OS，可为黄山肝癌患者个体化的诊疗提供参考。更多还原

【Abstract】 objective To develop an innovative and valuable prognostic model for hepatocellular carcinoma(HCC) patients after surgery. Methods A total of 1 526 HCC patients after surgery in the SEER database from 2010 to 2016 were analyzed. Patients were randomly divided into training cohort(n=1 070) and internal validation cohort(n=456) based on the ratio of 7:3. One hundred and forty-eight HCC patients from TCGA database and 102 HCC patients from Huangshan People’s Hospital in China between 2012—2021 were included as external cohort 1 and cohort 2, respectively. A prognostic nomogram was established with regression analyses in the training cohort, which was validated by other 3 independent validation cohorts. Area under the receiver operating characteristic curves(AUROC) and calibration plots were drawn, Brier scores, and C-index were calculated to assess the predictive ability. Finally, the prognostic model incorporating the stage of China Liver Cancer Staging(CNLC) system was tested. Results Serum AFP level, histological differentiation, TNM stage,surgical approach, and systemic therapy were independent factors affecting the prognosis of HCC after surgery,which were used to establish a overall survival(OS) nomogram. In the training cohort, the C-index was 0.712;the Brier scores at 1-, 2-, and 3-year were 0.085, 0.135, and 0.171; the AUROC values for OS at 1-, 2-, and 3-year were 0.745, 0.742, and 0.758. In the internal validation group, the C-index was 0.727, the Brier scores at 1-, 2-, and 3-year were 0.084, 0.124, and 0.160; the AUROC values for OS at 1-, 2-, and 3-year were 0.732, 0.763, and 0.742. In the TCGA cohort, the C-index was 0.705, the Brier scores at 1-, 2-and 3-year were 0.089, 0.140, and 0.135, the AUROC values for OS at 1-, 2-, and 3-year were 0.684, 0.706, and 0.789. In the Huangshan cohort, the C-index was 0.700, the Brier scores at 1-, 2-, and 3-year were 0.101, 0.110, and 0.129; the AUROC values for OS at 1-, 2-, and 3-year were 0.755, 0.734, and 0.731. Meanwhile, the calibration curves showed that the model predicted OS at 1-, 2-, and 3-year were concordant with the actual OS in above 4 cohorts. Finally, the nomogram of CNLC system had better predictive ability than that of TNM stage. Conclusion Nomogram could accurately predict the outcomes of HCC patients after surgery in this study.更多还原