Analysis Of Invasive Risk Factors And Establishment Of Predictive Models For Early Lung Adenocarcinoma Shown As Pure Ground Glass Nodules

Purpose: A visual prediction model for the identification of pre-invasive(atypical adenomatous hyperplasia and adenocarcinoma in situ)and invasive lesions(minimally invasive adenocarcinoma and invasive adenocarcinoma)of early lung adenocarcinoma was established by analyzing the clinical and imaging features of pure ground glass nodules(pGGNs).Method: Clinical features and CT images of 146 patients with pGGNs,enrolled in the First Affiliated Hospital of China Medical University from June 2016 to November 2018,confirmed by surgical pathology were analyzed retrospectively as modeling group.Age,gender,BMI,smoking history,drinking history,exposure history of dust,past tumor history,family history of tumor,tumor location,lesion size,lobulated sign,burr sign,pleural indentation,bronchial aeration sign,vascular bundle sign,lobulation and common tumor markers were compared between preinvasive lesions and invasive lesions through univariate analysis and receiver-operating characteristic curve(ROC).Logistic regression analysis was performed to analyze independent risk factors,and R software was used to establish a visual prediction model.Clinical and imaging data of 34 patients with pure ground glass nodules from December 2018 to February 2019 were collected as validation group for external validation.Result: Of the 146 patients,44 had pre-invasive lesions including 16 atypical adenomatous hyperplasia and 28 in situ adenocarcinoma,102 with invasive lesions,including 26 minimally invasive adenocarcinoma and 76 invasive adenocarcinoma.Univariate analysis showed: gender(P=0.007),CYFRA21-1(P=0.037),lesion size(P<0.001),bronchial aeration sign(P=0.001),vascular bundle sign(<0.001),burr sign(P= 0.037),lobulated sign(P=0.004)had a statistical difference.There were no significant differences between Age(P=0.062),BMI(P=0.429),smoking history(P=0.07),exposure history of dust(P=0.745),drinking history(P=0.714),family history of tumor(P=0.474),CEA(P=0.106),NSE(P=0.226),SCC(P=0.745),lesion location(P=0.876),and pleural indentation(P=0.091).The best cut-off size for identifying preinvasive and invasive lesions was 11.85 mm,with sensitivity and specificity of 90.2% and 68.2%,respectively,and the area underthe curve was 0.866.Multivariate analysis suggested that lesion size(OR=28.950,95%CI:3.856~217.365,P=0.001),vascular bundle sign(OR=10.799,95%CI: 1.960~59.506,P=0.006),lobulated sign(OR=372.292,95%CI: 3.396~40818.795,P=0.014)and CYFRA21-1(OR=30.794,95%CI: 2.469~382.846,P=0.008)are independent predictors of pGGNs invasiveness.Using these variables,a nomogram plot of the clinical prediction model was drawn using R software with a C-index of 0.902.The calibrated C-index of internal validation is 0.89,the C-index of external validation is 0.86.Conclusion: Lesion size,vascular bundle sign,lobulated sign and CYFRA21-1 are independent predictors of pGGNs invasiveness.And we propose a nomogram to enable clinicians to better estimate the invasiveness of pGGNs...