Objective: To examine a survival prognostic model applicable for patients with intrahepatic cholangiocarcinoma (ICC) based on Bayesian network. Methods: The clinical and pathological data of ICC patients who underwent curative intent resection in ten Chinese hepatobiliary surgery centers from January 2010 to December 2018 were collected.A total of 516 patients were included in the study.There were 266 males and 250 females.The median age(M(QR)) was 58(14) years.One hundred and sixteen cases (22.5%) with intrahepatic bile duct stones,and 143 cases (27.7%) with chronic viral hepatitis.The Kaplan-Meier method was used for survival analysis.The univariate and multivariate analysis were implemented respectively using the Log-rank test and Cox proportional hazard model.One-year survival prediction models based on tree augmented naive Bayesian (TAN) and naïve Bayesian algorithm were established by Bayesialab software according to different variables,a nomogram model was also developed based on the independent predictors.The receiver operating characteristic curve and the area under curve (AUC) were used to evaluate the prediction effect of the models. Results: The overall median survival time was 25.0 months,and the 1-,3-and 5-year cumulative survival rates was 76.6%,37.9%,and 21.0%,respectively.Univariate analysis showed that gender,preoperative jaundice,pathological differentiation,vascular invasion,microvascular invasion,liver capsule invasion,T staging,N staging,margin,intrahepatic bile duct stones,carcinoembryonic antigen,and CA19-9 affected the prognosis(χ2=5.858-54.974, all P<0.05).The Cox multivariate model showed that gender,pathological differentiation,liver capsule invasion,T stage,N stage,intrahepatic bile duct stones,and CA19-9 were the independent predictive factors(all P<0.05). The AUC of the TAN model based on all 19 clinicopathological factors was 74.5%,and the AUC of the TAN model based on the 12 prognostic factors derived from univariate analysis was 74.0%,the AUC of the naïve Bayesian model based on 7 independent prognostic risk factors was 79.5%,the AUC and C-index of the nomogram survival prediction model based on 7 independent prognostic risk factors were 78.8% and 0.73,respectively. Conclusion: The Bayesian network model may provide a relatively accurate prognostic prediction for ICC patients after curative intent resection and performed superior to the nomogram model.
目的： 探讨基于贝叶斯网络建立肝内胆管癌患者根治性切除术后生存预测模型的临床价值。 方法： 回顾性分析中国10家中心2010年1月至2018年12月收治的经意向性根治切除治疗的516例肝内胆管癌患者的临床资料。男性266例，女性250例，年龄［M（QR）］58（14）岁；116例（22.5%）合并肝内胆管结石，143例（27.7%）合并慢性病毒性肝炎。使用Kaplan-Meier法绘制生存曲线，单因素生存分析采用Log-rank检验，多因素生存分析使用Cox回归模型。根据不同变量特征分别采用树增益朴素贝叶斯网络（TAN）及朴素贝叶斯网络算法，应用BayesiaLab软件建立以12个月为目标节点的生存预测模型。同时建立列线图模型，采用受试者工作特征曲线和曲线下面积（AUC）评价模型预测效果的优劣。 结果： 516例患者的总体中位生存时间为25.0个月，1、3、5年累积总体生存率分别为76.6%、37.9%、21.0%。单因素分析结果显示，性别、术前有无黄疸、肿瘤分化程度、有无血管侵犯、有无脉管侵犯、有无肝包膜侵犯、T分期、N分期、切缘、有无结石、癌胚抗原水平、CA19-9水平是患者的预后因素（χ2=5.858~54.974，P值均<0.05）。Cox多因素回归模型分析结果显示，性别、肿瘤分化程度、有无肝包膜侵犯、T分期、N分期、有无结石及CA19-9水平为患者的独立预后因素（P值均<0.05）。依据19个临床病理学因素建立的TAN模型的AUC为74.5%，依据单因素分析得到的12个预后因素所建立的TAN模型的AUC为74.0%，依据多因素分析得到的7个独立预后因素所建立的朴素贝叶斯网络模型的AUC为79.5%，依据多因素分析得到的7个独立预后因素所建立的列线图生存预测模型的AUC为78.8%，C-index为0.73。 结论： 基于贝叶斯网络建立的肝内胆管癌根治术后生存预测模型具有较高的准确性，较列线图具有一定的优势。.