肿瘤标志物预测孤立性肺结节恶性概率模型的建立与初步评价

Establishment and preliminary evaluation of a model for predicting the malignant probability of solitary pulmonary nodule with tumor markers

目的利用肿瘤标志物建立预测孤立性肺结节(SPN)恶性概率的数学模型,并评价其临床价值。方法选取250例SPN患者,考察其年龄、性别、吸烟史、症状、结节最大径、结节部位、病理,以及血清癌胚抗原(CEA)、细胞角蛋白19片段抗原(CYFRA21-1)、神经元特异性烯醇化酶(NSE)水平,采用二分类Logistic回归法作影响因素筛选,并建立Logistic回归模型。绘制受试者工作特征曲线(ROC)并计算曲线下面积(AUC)以评价模型准确性,并与梅奥模型比较以评价模型的临床价值。结果 CEA(P=0.002,OR=5.921,95%CI=1.968~17.819),CYFRA21-1(P=0.046,OR=2.500,95%CI=1.018~6.142),症状(P=0.010,OR=2.384,95%CI=1.234~4.607),结节最大径(P=0.001,OR=2.331,95%CI=1.441~3.773)与SPN的良恶性有关;由此建立预测模型:P=ex/(1+ex),X=-1.991+0.869×症状+0.846×结节最大径+1.779×CEA+0.916×CYFRA21-1;采用Hosmer-Lemeshow检验模型的拟合度较好(P=0.691);当截点为0.636时,灵敏度为63.5%,特异度为71.2%;预测模型(AUC:0.734±0.033)与指南推荐的梅奥模型(AUC:0.792±0.047)进行比较,差异无统计学意义(P>0.05)。结论 CEA、CYFRA21-1、症状和结节最大径为恶性SPN的独立危险因素;由此建立的Logistic回归模型准确性较高,有较好的临床价值。

Objective To establish a model for predicting the malignant probability of solitary pulmonary nodule （SPN） with the tumor markers and to evaluate its clinical value. Methods A retrospective cohort study in Yantai Yu- huangding Hospital of Shandong Province included 250 patients with definite pathological diagnosis of SPN from Jan. 2010 to Oct. 2015. Clinical data included age, gender, quantity of smoking history, symptoms, site, maximum diame- ter, the levels of CEA, CYFRA21-1 and NSE. By means of Logistic regression, 9 factors were analyzed to establish the model. Receiver operating characteristic curve （ROC） was performed and the area under the curve （AUC） was cal- culated. Our model was compared with the Mayo model to evaluate its clinical value. Results CEA level（ P = 0. 002, OR=5.921, 95%CI= 1.968-17.819）, CYFRA21-1 level（P =0.046, OR=2.500, 95%CI=1.018-6.142）, symptom （ P = 0. 010, OR = 2. 384, 95 % CI = 1. 234-4. 607 ） and maximum diameter（ P = 0. 001, OR = 2.331, 95 % CI = 1. 441- 3.773 ） were associated with malignant SPN. Our prediction model ： P = ex/（ 1 ＋ ex ）, X = - 1.991 ＋ 0.869 × symptom ＋0. 846 × maximum diameter ＋ 1. 779 × CEA ＋ 0. 916 × CYFRA21-1. The goodness-of-fit of the model was fairly good. When the optimal cut-off point was 0. 636, the sensitivity was 63.5% and specificity was 71.2%. The differ- ence of AUC between our model and the Mayo clinic model was not statistically significant（ P 〉 0.05）. Conclusion CEA, CYFRA21-1, symptom and maximum diameter are independent risk factors of malignant SPN. Our research shows that the Logistic regression model has high accuracy and clinical value.