肝硬化门静脉血栓形成风险预测列线图模型的建立与验证

Establishment and validation of a risk prediction model for portal vein thrombosis in liver cirrhosis by nomogram

目的探讨肝硬化门静脉血栓形成(portal vein thrombosis,PVT)的独立危险因素,依据独立危险因素尝试建立肝硬化患者并发PVT风险的预测模型并评估其预测能力。方法2019年12月—2021年10月,在武汉大学人民医院住院治疗的295例肝硬化病例,采用随机数字表法分成建模集(n=207)和内部验证集(n=88);另收集同期宜昌市中心人民医院、武汉市普仁医院、阜阳市第二人民医院、三峡大学人民医院住院治疗的肝硬化病例作为外部验证集(n=92)。建模集按有无PVT分成PVT组(n=56)和非PVT组(n=151),首先采用单因素分析初步筛选PVT的相关指标,再进行多因素Logistic回归分析(向前逐步回归法)筛选出PVT的独立危险因素,最后根据获得的独立危险因素构建列线图预测模型。内部验证集和外部验证集用于模型预测能力验证,使用区分度(ROC曲线分析)评价模型区分肝硬化患者有无PVT的能力,使用Hosmer-Lemeshow拟合优度检验评价模型预测风险与实际风险的一致性。结果单因素分析发现,吸烟史、脾切除史、经颈静脉肝内门体分流术(trans-jugular intrahepatic portosystemic shunt,TIPS)治疗史、消化道出血史、静脉曲张内镜治疗史、血红蛋白含量、丙氨酸转氨酶水平、天冬氨酸转氨酶水平、D-二聚体水平在PVT组与非PVT组间差异均有统计学意义(P<0.05)。多因素Logistic回归分析发现,吸烟史(P=0.020,OR=31.21,95%CI:1.71~569.40)、D-二聚体水平(P=0.003,OR=1.12,95%CI:1.04~1.20)、血红蛋白含量(P=0.039,OR=0.99,95%CI:0.97~1.00)、TIPS治疗史(P=0.011,OR=18.04,95%CI:1.92~169.90)、静脉曲张内镜治疗史(P=0.001,OR=3.21,95%CI:1.59~6.50)均是肝硬化患者并发PVT的独立危险因素。ROC曲线分析发现,内部验证集的ROC曲线下面积(AUC)为0.802(95%CI:0.709~0.895)(P<0.001),外部验证集的AUC为0.811(95%CI:0.722~0.900)(P<0.001),两者AUC均大于0.75。Hosmer-Lemeshow拟合优度检验发现,内部验证集(χ^(2)=3.602,P=0.891)和外部验证集(χ^(2)=11.025,P=0.200)的P值均大于0.05。结论吸烟史、TIPS治疗史、静脉曲张内镜治疗史、D-二聚体水平、血红蛋白含量是肝硬化患者并发PVT的独立危险因素,基于以上因素建立的PVT风险预测列线图模型具有较强的预测能力。

Objective To explore the independent risk factors of portal vein thrombosis(PVT)in liver cirrhosis,and to establish and evaluate a risk prediction model for PVT in patients with cirrhosis.Methods A total of 295 cases of cirrhosis hospitalized in Renmin Hospital of Wuhan University from December 2019 to October 2021 were divided into a modeling set(n=207)and an internal validation set(n=88)by the random number table.In addition,patients with cirrhosis hospitalized in Yichang Central People's Hospital,Wuhan Puren Hospital,No.2 People's Hospital of Fuyang City and People's Hospital of China Three Gorges University during the same period were collected as an external validation set(n=92).The modeling set was divided into PVT group(n=56)and non-PVT group(n=151).Univariate analysis was used to preliminarily screen the related indicators of PVT,and then multivariate logistic regression analysis with forward stepwise regression was used to determine independent risk factors for PVT.A nomogram prediction model was constructed based on the independent risk factors obtained.The internal and external validation set were used to verify the predictive ability of the model.Distinction degree was used to evaluate the ability of the model to distinguish patients with or without PVT.Hosmer-Lemeshow goodness-of-fit test was used to evaluate the consistency between predicted risk and the actual risk of the model.Results Univariate analysis showed that smoking,history of splenectomy,trans-jugular intrahepatic portosystemic shunt(TIPS),gastrointestinal bleeding and endoscopic variceal treatment,and levels of hemoglobin,alanine aminotransferase,aspartate aminotransferase and D-dimer were significantly different between the PVT group and the non-PVT group(P<0.05).Multivariate logistic regression analysis found that smoking(P=0.020,OR=31.21,95%CI:1.71-569.40),levels of D-dimer(P=0.003,OR=1.12,95%CI:1.04-1.20)and hemoglobin(P=0.039,OR=0.99,95%CI:0.97-1.00),history of TIPS(P=0.011,OR=18.04,95%CI:1.92-169.90)and endoscopic variceal treatment(P=0.001,OR=3.21,95%CI:1.59-6.50)were independent risk factors for PVT in patients with liver cirrhosis.Receiver operator characteristic(ROC)curve analysis showed that the area under the ROC curve(AUC)for the internal validation set was 0.802(95%CI:0.709-0.895)(P<0.001),and the AUC for the external validation set was 0.811(95%CI:0.722-0.900)(P<0.001).Both AUC were larger than 0.75.The calibration curve of Hosmer-Lemeshow goodness-of-fit test showed that the P values of both internal validation set(χ^(2)=3.602,P=0.891)and the external validation set(χ^(2)=11.025,P=0.200)were larger than 0.05.Conclusion Smoking,history of TIPS or endoscopic variceal treatment,levels of D-dimer and hemoglobin are independent risk factors for PVT in patients with liver cirrhosis.The prediction nomogram model based on the above factors has strong predictive ability.