摘要： 目的·建立并评价肝硬化患者并发轻微型肝性脑病的筛查模型。方法·以2017年6月—2019年11月住院的404例肝硬化患者为研究对象，采集其临床资料。基于Logistic回归分析和人工神经网络分别建立轻微型肝性脑病筛查模型，对2种模型的判别能力进行评价和比较。结果·Logistic回归分析提示，肝硬化并发轻微型肝性脑病的独立危险因素为年龄、糖尿病史、感染、肾功能不全、营养风险、总胆红素>24 μmol/L、血氨>47 μmol/L、国际标准比值≥1.5（均
P<0.05）。人工神经网络模型与Logistic回归模型的受试者操作特征曲线（receiver operator characteristic curve，ROC curve）的曲线下面积（area under the curve，AUC）分别为0.814和0.737（
Z=4.208，
P=0.000），灵敏度分别为72.4%、69.9%，特异度分别为76.7%、67.8%。结论·人工神经网络模型对轻微型肝性脑病的筛查效能优于Logistic回归模型。
关键词: 轻微型肝性脑病, 人工神经网络, 筛查模型 Abstract: Objective·To establish and evaluate a screening model of liver cirrhosis patients complicated with minimal hepatic encephalopathy (MHE).
Methods·A total of 404 patients with liver cirrhosis who were hospitalized from June 2017 to November 2019 were selected as the research subjects, and the clinical data of them were collected. Based on Logistic regression analysis and artificial neural network (ANN), the MHE screening models were established, and the discriminant ability of the two models was evaluated and compared.
Results·The Logistic regression model showed that age, history of diabetes mellitus, infection, renal insufficiency, nutritional risk, total bilirubin>24 μmol/L, blood ammonia>47 μmol/L and international normalized ratio (INR)≥1.5 were the significant risk factors (all P<0.05). The area under the curve (AUC) of receiver operator characteristic curve (ROC curve) of ANN model and Logistic regression model were 0.814 and 0.737 (Z=4.208, P=0.000), respectively. The sensitivities were 72.4% and 69.9%, and the specificities were 76.7% and 67.8%, respectively.
Conclusion·The ANN model is more effective than the Logistic regression model in MHE screening.
Key words: minimal hepatic encephalopathy (MHE), artificial neural network (ANN), screening model PDF全文下载地址:
点我下载PDF 