孤立性肺结节的良恶性预测因子及预测模型的建立与验证

Establishment and validation prediction models of benign and malignant predictors for solitary pulmonary nodules

摘要:

目的： 探讨良恶性孤立性肺结节（solitary puhnonary nodule，SPN）的预测因子并建立预测模型，验证模型效能，并与经典模型进行诊断效能比较，以期提高早期肺癌的非侵入性诊断准确率。方法： 选取2014年1月至2021年1月经重庆医科大学附属第二医院诊治的522例SPN患者，分为试验组（432例）和验证组（90例），回顾性分析其临床及CT影像学特征，筛选出恶性SPN的独立预测因子，并建立临床预测模型。将验证组资料代入模型进行验证，并与Mayo模型及北大模型进行诊断效能比较，绘制受试者工作特征（receiver operating characteristic，ROC）曲线。结果： 二元logistic分析显示，上叶、毛刺征、分叶征、血管集束征、边界不清及结节最大径是判断良恶性结节的独立危险因子，建立的预测模型为P=e^x/（1+e^x），X=-3.742+（0.185×结节最大径）+（1.423×毛刺征）+（1.143×分叶征）+（3.783×血管集束征）+（2.526×边界不清）+（0.730×上叶）。本研究模型 ROC曲线下面积（0.875）高于Mayo模型（0.776）及北大模型（0.779）（P<0.05），提示本研究模型诊断效能可能优于Mayo模型及北大模型。结论： 上叶、毛刺征、分叶征、血管集束征、边界不清、结节最大径是预测良恶性SPN的独立预测因子，本预测模型诊断效能可能优于国内外经典模型，对早期肺癌的诊断有更大的借鉴意义。

Abstract:

Objective: To analyze the prdictors of solitary pulmonary nodules（SPN） and establish a mathematical prediction model to verify the accuracy of the model and compare the diagnostic efficiency with the classical models，in order to improve the accuracy of non-invasive diagnosis of early lung cancer. Methods: A total of 522 SPN patients treated by The Second Affiliated Hospital of Chongqing Medical University from January 2014 to January 2021 were selected and divided into case group（432 cases） and validation group（90 cases），and their clinical and CT imaging features were retrospectively analyzed to screen out independent predictors of malignant SPN and establish a clinical prediction model. The validation group data were inserted into the model for verification，the diagnostic efficacy was compared with the Mayo model and Peking University model，and the receiver operating characteristic（ROC） curve was drawn. Results: Binary Logistic analysis showed that upper lobe，spiculation，lobulation，vascular aggregation sign，unclear boundary and the maximum diameter of nodules were independent risk factors for benign and malignant SPN. The established prediction model was P=e^x/（1+e^x），X=-3.742+（0.185×the maximum diameter of nodule）+（1.423×spiculation）+（1.143×lobulation）+（3.783×vascular aggregation sign）+（2.526×unclear boundary）+（0.730×upperlobe）. The area under ROC curve（AUC） of our model（0.875） was significantly higher than that of the Mayo model（0.776） and Peking University model（0.779）（P<0.05）. It was suggested that the diagnostic efficiency of this model might be better than that of Mayo model and Peking University model. Conclusion: Upper lobe，spiculation，lobulation，vascular aggregation sign，unclear boundary and the maximum diameter of nodules are independent risk factors for benign and malignant SPN，and the diagnostic efficiency of our prediction model might be better than that of Mayo model and Peking University model，which is of great significance for the diagnosis of early lung cancer.