基于logistic回归分析构建肝功能不全临床分型预测模型

Establishment of the model predicting clinical classification of hepatic insufficiency based on logistic regression analysis

目的探讨肝功能不全临床分型的影响因素并建立预测模型。方法选取75例肝功能不全患者作为研究对象,采用国际医学科学组织委员会的分型标准将患者分为肝细胞损伤型与胆汁淤积型,比较两组一般资料与生化指标的差异。应用logistic回归方法建立合适的模型,并绘制相应的列线图。最后,绘制受试者工作特征曲线(ROC)与临床决策曲线进一步验证预测模型的准确性。结果75例患者中肝细胞损伤型41例,胆汁淤积型34例。多因素回归筛选出谷草转氨酶与血红蛋白为临床分型的影响因素(OR分别=0.98、0.92,P均<0.05)。基于这2个因素构建预测模型列线图。验证模型准确性的ROC曲线下面积为0.90,临床决策曲线分析表明,当阈值概率在0.1~0.9范围内时,该模型在临床上的净获益最高。结论谷草转氨酶、与血红蛋白影响肝功能不全患者的临床分型,基于这2个因素构建预测模型列线图的预测模型准确性较高。

Objective To explore the influencing factors of clinical classification of hepatic insufficiency and establish a predictive model.Methods Seventy-five patients with hepatic dysfunction were selected as the subjects.They were di-vided into hepatocellular injury group and cholestasis group according to the classification standard of the council for in-ternational organizations of medical sciences,and the differences in general data and biochemical indexes between the two groups were compared.Logistic regression analysis was used to establish a suitable model and construct the corresponding nomogram.Finally,the receiver operating characteristic(ROC)curve and clinical decision curve were drawn to further verify the accuracy of the prediction model.Results Among 75 patients,41 cases were enrolled in the hepatocyte injury group,and 34 cases were enrolled in the cholestasis group.Multivariate logistic regression analysis showed that aspartate aminotransferase and hemoglobin were independent factors for clinical classification(OR=0.98,0.92,P<0.05).The nom-ogram prediction model was constructed based on the two independent factors.The area under the curve of the model was 0.90,and the clinical decision curve analysis showed that when the threshold probability was in the range of 0.1 to 0.9,the model had the highest net benefit.Conclusion Aspartate aminotransferase and hemoglobin independently affect the clinical classification of patients with liver insufficiency,and the accuracy of the nomogram prediction model based on these two factors is relatively high.