基于肝脾及门静脉MRI形态学特征的Logistic回归模型预测乙肝肝硬化继发食管胃底静脉曲张出血风险

A Logistic regression model based on morphology of liver, spleen and portal vein system depicted on MRI to predict the risk of oesophagogastric variceal bleeding secondary to hepatitis B-related cirrhosis

目的 探讨基于肝脾及门静脉MRI形态学特征的新模型预测乙肝肝硬化继发食管胃底静脉曲张出血(OVB)的价值。方法 前瞻性分析210例连续性乙肝肝硬化病人的肝脾及门静脉MRI形态学特征及临床资料,采用随机数字表法按7∶3的比例将病人分为训练集(147例)和验证集(63例)。2年随访期内,训练集和验证集分别有47例和21例发生OVB。在训练集,采用卡方检验、t检验或Mann-Whitney U检验比较有和无OVB病人间肝脾及门静脉MRI形态学特征及临床资料的差异。将差异有统计学意义的指标纳入多因素Logistic回归分析,获得OVB的独立预测因素。基于独立预测因素构建OVB预测模型,并通过受试者操作特征(ROC)曲线下面积(AUC)评估模型的预测效能,最后在验证集通过Kappa检验验证模型的预测效能。结果 训练集中,OVB病人肝右叶及全肝体积小于无OVB病人(均P<0.05),左内叶、左外叶及尾叶体积在有和无OVB的病人间差异无统计学意义(均P>0.05)。OVB病人脾体积、脾体积与各肝叶体积的比值、门静脉系统直径及腹水发生率均大于无OVB病人(均P<0.05)。多因素Logistic回归分析显示肝右叶体积、胃左静脉直径、门静脉直径以及腹水(比值比分别为0.994、2.24、1.571及3.983,均P<0.05)是OVB的独立预测因素,基于独立预测因素建立的Logistic模型预测OVB的效能极佳(AUC=0.934)。在验证集,预测模型效能也极佳(κ=0.828)。结论 本研究建立的Logistic模型预测继发于乙肝肝硬化的OVB具有较高的预测效能。

Objective To explore the value of a novel model based on the morphology features of liver, spleen and portal vein system depicted on MRI in predicting oesophagogastric variceal bleeding(OVB) secondary to hepatitis B-related cirrhosis. Methods Morphology features of liver, spleen and portal vein system depicted on MRI and clinical data of 210 consecutive cirrhotic patients with hepatitis B were analyzed prospectively. The patients were divided into training cohort(147 cases) and validation cohort(63 cases) according to the ratio of 7∶3 using the random number table. OVB occurred in 47 and 21 individuals in the training and validation cohorts during the two years’ follow-up period, respectively. In the training cohort, the differences in the morphology features on MRI and clinical data between the patients with and without OVB were compared using chi-square, t test or Mann-Whitney U tests. The features with statistical difference were fed into a multivariate logistic regression to determine independent predictors of OVB. The OVB prediction model was constructed based on the independent predictors. Predictive performance of the model was evaluated by the area under receiver operating characteristic curve(AUC), and was ultimately validated by Kappa test in the validation cohort. Results In the training cohort,the volumes of right liver lobe and total liver in the patients with OVB were smaller than that in the patients without OVB(all P<0.05), and the volumes of the left medial, left lateral and caudate lobes were not significantly different between the patients with and without OVB(all P>0.05). The spleen volume, ratios of spleen volume to individual liver lobe volumes, diameters of portal venous system, and incidence of ascites in the patients with OVB were higher than without OVB(all P <0.05).Multivariate Logistic regression analyses revealed that the right lobe volume, left gastric vein and portal vein diameters, and ascites(odds ratios 0.994, 2.24, 1.571, and 3.983, respectively, all P<0.05) were independent predictors of OVB. The logistic model based on independent predictors had excellent performance in predicting OVB(AUC=0.934), and the performance of prediction model was also excellent(κ=0.828) in the validation cohort. Conclusion The logistic model established in this study has excellent performance in predicting OVB secondary to hepatitis B-related cirrhosis.