联合肝癌患者病理和预后相关蛋白数据构建肝癌预后决策树模型  

Tree-structured survival analysis of patients with hepatocellular carcinoma based on clinicopathological features and prognosis-related protein expression data

作　　者：汪泓 阳霞[2] 张红雁 WANG Hong;YANG Xia;ZHANG Hongyan(The First Affiliated Hospital of USTC,Division of Life Sciences and Medicine,University of Science and Technology of China,Hefei 230001,China)

机构地区：[1]中国科学技术大学附属第一医院,安徽合肥230001 [2]中山大学肿瘤防治中心病理科,广东广州510060

出　　处：《安徽医学》2020年第3期261-268,共8页Anhui Medical Journal

基　　金：国家自然科学基金(项目编号:81902979);安徽省自然科学基金(项目编号:903257950007)。

摘　　要：目的利用肝癌患者临床病理信息和预后相关蛋白数据,采用决策树方法构建肝癌预后模型。方法从TCGA数据库中获得肝癌组织反相蛋白质阵列,使用LASSO逻辑回归筛选Tubulinα-1B、PAI-1和B-raf蛋白为肝癌候选预后标志物。采用免疫组化分析中山大学肿瘤防治中心775例肝癌患者癌组织中Tubulinα-1B、PAI-1和B-raf蛋白表达。使用决策树方法鉴定肝癌患者临床病理信息和预后相关蛋白表达数据在预测肝癌预后中的重要性并进行预后分层。结果决策树模型的分叉节点分别为TNM分期、Tubulinα-1B、肿瘤大小和血管浸润。该模型可进行肝癌预后高、中、低风险分层,其在544例肝癌患者的训练队列中特异度为72.8%,在231例肝癌患者的验证队列中特异度为74.1%。结论联合肝癌患者病理和预后相关蛋白数据,利用决策树方法构建的肝癌预后模型可区分肝癌患者预后风险,有助于临床制定个体化医疗方案。Objective To establish and validate an easy-to-use prognostic model to predict the prognosis ofpatients with hepatocellular carcinoma(HCC).Methods Tubulinα-1B,PAI-1 and B-raf were selected from reverse-phase protein arrays(RPPA)profiles obtained from The Cancer Genome Atlas(TCGA)data using Least Absolute Shrinkage and Selection Operator(LASSO)logistic regression.Classification and regression tree(CART)was employed to determine variable importance and estimate prognostic risk of 775 HCC patients from Sun Yat-sen University Cancer Center(SYSUCC),China,in this study.Tubulinα-1B,PAI-1 and B-raf were chosen as three most powerful predictive markers from TCGA HCC RPPA dataset and detected by immunohistochemical analysis in 775 HCC sample.A total of 775 patients with HCC were assigned to a training cohort and a validation cohort by computer-generated random number assignment.Results CART analysis identified TNM stage,Tubulinα-1B,tumor size and vascular invasion as discriminative nodes,and generated a tree model for a training group of 544 patients in order to classify them as high-,medium-or low-risk,with accuracy of 72.8%.The accuracy,sensitivity and specificity was 74.1%in validated cohort of 231 HCC patients.Conclusions Combined with clinicopathological variables and tumor IHC score,a simple,clinically relevant CART prognostic model is developed and validated to identify patients with HCC at high-,medium-and low-risk using routinely available variables.The CART model providesclinicians with a practical bedside tool for HCC risk stratification and facilitated medical decision-making for individualized therapy.

关 键 词：Tubulinα-1B PAI-1 B-RAF 临床病理信息 决策树 预后模型 肝细胞癌 

分 类 号：R735.7[医药卫生—肿瘤]