Logistic回归结合ROC曲线在HBV相关肝硬化合并肝癌患者中的诊断价值

Diagnostic value of Logistic regression combined with ROC curve analysis model in patients with hepatitis B cirrhosis and liver cancer

研究以治疗前170例HBV相关肝硬化合并肝癌、90例HBV相关肝硬化和90例慢性乙型肝炎(chronic hepatitis B, CHB)患者为研究对象,检测异常凝血酶原(abnormal prothrombin,PIVKA-Ⅱ)、高尔基体蛋白73(Golgi protein 73,GP73)、AFP和甲胎蛋白异质体(alpha fetoprotein heterogeneity,AFP-L3)的血清水平,建立单项检测的ROC曲线,确定各指标的最佳诊断临界值(cutoff值),再通过Logistic回归评价多变量检测模型的诊断价值。结果表明,单项检测时,PIVKA-Ⅱ的ROC曲线下面积(area under the curve, AUC)最大(0.920),敏感度和特异度分别为91.2%和85.6%。多项检测时,PIVKA-Ⅱ联合AFP的Logistic回归方程为最佳诊断模型,ROC AUC最大(0.951),敏感度和特异度分别为87.6%和91.1%,优于两项指标的并联和串联实验。肝癌风险概率值P=1/[1+e^-(-5.284+9.087×PIVKA-Ⅱ+4.756×AFP)],在早期(Ⅰ期和BCLC A期)肝癌患者中的敏感度为66.7%,在小肝癌(肿瘤直径<3 cm)患者中的敏感度为67.7%。因此,PIVKA-Ⅱ联合AFP的Logistic回归方程诊断模型在HBV相关肝硬化合并肝癌患者中具有良好的诊断价值,利用肝癌风险概率值可从良性肝病患者中及时诊断出早期肝癌和小肝癌。

170 patients with hepatitis B cirrhosis and liver cancer, 90 patients with hepatitis B cirrhosis, 90 patients with chronic hepatitis B were enrolled in the study. Serum abnormal prothrombin(PIVKA-Ⅱ)(chemiluminescent enzyme immunoassay), GP73(Golgi protein 73)(ELISA), the serum levels of AFP(electrochemiluminescence) and AFP-L3(alpha-fetoprotein heterogeneity,AFP-L3)(microcentrifugation column method) were measured to establish ROC curve in order to determine the optimal diagnostic threshold(cutoff value) for each index. The diagnostic index of the multivariate detection model was then evaluated by using the Binary Logistic process by means of SPSS for Logistic regression. The results showed that the ROC AUC of PIVKA-Ⅱ was the largest(0.920), and the sensitivity and specificity were 91.2% and 85.6%, respectively. In the multiple test, the Logistic regression equation of PIVKA-Ⅱ combined with AFP was the best diagnostic model. The ROC AUC was the largest(0.951) and the sensitivity and specificity were 87.6% and 91.1%, respectively. The diagnostic accuracy was better than parallel test and series test of the indicators. The risk probability of liver cancer was P=1/[1+e^-(-5.284+9.087×PIVKA-Ⅱ+4.756×AFP)], with a sensitivity of 66.7% in early liver cancer patirents(stageⅠand BCLC stage A) patients and 67.7% in small liver cancer patients(tumor diameter < 3 cm). Therefore, the Logistic regression equation of PIVKA-Ⅱ combined with AFP was the best diagnostic model for hepatitis B cirrhosis with liver cancer. The risk probability of liver cancer could be used to diagnose early liver cancer and small liver cancer from patients with benign hepatopathy.