Establishment Of A Prediction Model For Malignant Probability Of Pulmonary Nodules

Background: Lung cancer is the most common malignancy in China and worldwide.With a very poor prognosis,5-year relative survival rate for lung cancer is only 19.4 % in China.According to the eighth edition of TNM staging of lung cancer published by IASLC,the 5-year survival rate of stage IA lung cancer is more than 80%,while the 5-year survival rate of stage IV lung cancer is only 10%.Therefore,the early diagnosis of lung cancer may be the best way to improve the current situation of lung cancer in China.With the implementation of the cancer screening program in China,the number of pulmonary nodules found by chest CT examinations is increasing.Because of its complex shape,it is difficult to diagnose pulmonary nodules only by imaging.In this study,we will establish a diagnostic model,which combines characteristics of clinical,imaging and serum tumor markers of patients with pulmonary nodules,to provide the basis for clinical diagnosis.Methods: Referring to the 2018 Lung cancer screening guideline with LDCT in China,408 patients with pulmonary nodules who underwent surgical treatment and had definite pathological diagnosis in Tianjin Medical University General Hospital from January 2017 to January 2019 were selected in our study.According to the distribution of benign and malignant,the patients were divided into two groups: group A(308 cases)and group B(100 cases).Using SPSS 20.0 software,we first screen out the factors with significant difference between benign and malignant pulmonary nodules by univariate analysis of the above clinical data.Then,we performed Logistic regression analysis on this factors in group A,and screen out the final predictors of malignant pulmonary nodules and established a probability prediction model for malignant pulmonary nodules.Finally,we use the data of group B to verify the model,and draw the ROC curve according to the results in order to evaluate the accuracy of the model prediction.Results: In this study,univariate and multivariate analysis showed that age,nodule density,maximum diameter,unclear boundary,lobulation,lymphadenopathy,CEA and CYFRA 21-1 were independent risk factors for malignant pulmonary nodules;Clear boundary,calcification and wide base shape outwardly close to the pleura are protective factors suggesting that pulmonary nodules tend to be benign;Gender,history of extrathoracic malignancy,family history of malignancy,smoking history,multiple nodules,SCC and NSE were not independent risk factors for malignant pulmonary nodules.There were significant differences between benign and malignant groups in pulmonary nodules with spiculation and pleural retraction sign.However,multivariate analysis confirmed that they were not predictors of malignant pulmonary nodules,but the diagnostic accuracy of the model decreased after they were removed from the equation,so they were retained.The malignant probability prediction model of pulmonary nodules was established according to the clinical data of 308 patients with pulmonary nodules in group A: Malignant probability of pulmonary nodules P=ex/(1+ex),X=-6.311+(0.040*Age)+(4.443* Ground glass nodule)+(3.236* Part solid nodule)+(0.090* Maximum nodule diameter)+(-1.031* boundary)+(1.236* lobulation)+(0.826* spiculation)+(-2.086* calcification)+(-2.108* wide base shape outwardly close to the pleura)+(0.677* pleural retraction sign)+(1.078* lymphadenopathy)+(0.233*CEA)+(0.458*CYFRA 21-1).Validation of the model with data of 100 patients with pulmonary nodules in group B.According to the verification results,the sensitivity of the model was 89.4%,the specificity was 85.3%,the positive predictive value was 92.2%,the negative predictive value was 80.6%,the Kappa value was 0.736(P<0.01),and the AUC was 0.901±0.033(95%CI: 0.837~0.966).It is suggested that the model diagnosis is accurate.Conclusion: We established a diagnostic model by combining the characteristics of clinical,imaging and serum tumor markers of patients with pulmonary nodules.And we verified that the model was accurate in the diagnosis of benign and malignant pulmonary nodules.By observing the coefficients of predictors,we found that nodule density played an important role in the prediction of benign and malignant pulmonary nodules.Moreover,the imaging features of malignant pulmonary nodules of different densities are dissimilar,such as lobulation and spiculation are rarely observed on ground glass nodules.Therefore,to establish equations according to different density of pulmonary nodules may improve the diagnostic efficiency of the model.