乙型肝炎肝硬化上消化道出血无创性预测模型的建立

Noninvasive prediction model for upper gastrointestinal bleeding in patients with hepatitis B related cirrhosis

目的研究乙肝肝硬化患者发生上消化道出血(UGB)的危险因素,并构建乙肝肝硬化患者并发UGB的无创性预测模型。方法采用回顾性病例对照的研究方法,收集了2017年1月至2019年6月徐州医科大学附属医院收治的255例乙肝肝硬化患者的临床病例资料(其中以乙肝肝硬化UGB者为试验组,未出血者为对照组)。通过Logistic回归分析,构建乙肝肝硬化患者UGB的无创性预测模型,并最终绘制预测模型及各个指标的ROC曲线,计算曲线下面积。结果结果显示合并肝细胞癌、白蛋白、纤维蛋白原(FIB)、白细胞计数、食管胃底静脉曲张(EGV)程度、血小板计数/脾脏长径(PC/SD)在模型中均有统计学意义(P<0.05)。基于危险因素构建的模型对于筛查乙肝肝硬化UGB的敏感度为80.7%,特异度为86.8%。ROC曲线下面积(AUC)为0.904(0.867~0.941),优于其他任意单一指标。结论本研究基于6项独立影响因素建立的无创性预测模型具有较好的准确性,有待进一步验证和模型调优。

Objective To study on the risk factors and to construct a noninvasive prediction model for upper gastrointestinal bleeding(UGB)in patients with hepatitis B related cirrhosis.Methods The clinical data of 255 patients with hepatitis B related cirrhosis admitted to the Affiliated Hospital of Xuzhou Medical University from January 2017 to June 2019 were collected.They were devided into two groups and studied by retrospective case-control method.119 Patients with UGB were in the experimental group and 136 patients without UGB were in the control group.A non-invasive prediction model for UGB in patients with hepatitis B related cirrhosis was established through Logistic regression analysis.The receiver operating characteristic curve(ROC curve)of the prediction model with various indicators were drawn and the area under the curve(AUC)was calculated.Results It was shown that hepatocellular carcinoma,albumin,fibrinogen,leukocyte count,esophageal and gastric varices,platelet/spleen length were independent influencing factors for UGB in the model(P<0.05).The sensitivity and specificity of the risk factors-based model for screening UGB in hepatitis B related cirrhosis were 80.7%and 86.8%,respectively.The area under the ROC curve(AUC)was 0.904(0.867~0.941),which was better than any other single index.Conclusion The established non-invasive prediction model based on six independent influencing factors in this study has good accuracy for UGB,which warrants further verification and model tuning.