Prediction Of Progression-free Survival Of First-generation EGFR-TKI In Advanced Lung Adenocarcinoma Based On CT Radiomics

Objective:To establish a combined model of radiomics features,clinical indicators and conventional CT features based on computed tomography(CT)images to predict the early occurrence of acquired targeted resistance to epidermal growth factor receptor(EGFR)mutations in lung adenocarcinoma(Adenocarcinoma of the lung,LUAD)patients,and help clinicians to evaluate the initial efficacy of patients before treatment.Timely control the effect of targeted therapy.Materials and methods:Retrospectively collected LUAD patients with EGFR mutations with genetic testing and pathological results.Based on the median progression-free survival(PFS)of 11 months of the firstgeneration oral targeted drugs reported in most literatures,they were divided into early-generation targeted acquired resistance and late-generation targeted acquired resistance.In this study,the enrolled patients were randomly divided into training cohort and test cohort at a ratio of 6 : 4.Firstly,the region of interest(ROI)of solid tumors was delineated from the axial lung window and mediastinal window images of CT scan within 1 month before targeted therapy,and then the radiomics features in ROI were extracted.Then the features were screened by variance method,Spearman correlation analysis,rank sum test and LASSO algorithm.Radiomics scores were calculated according to the LASSO algorithm to establish a radiomics model.Univariate analysis was used to screen out features with statistical differences(P < 0.05)in clinical indicators and CT features.Subsequently,multivariate logistic regression analysis was performed on Radiomics scores,clinical indicators and conventional CT features.In this study,three combinations and three individual models were constructed and a nomogram was constructed,including the Clinical + CT+ Radio-scores model,the Clinical + Radio-scores model,the Clinical + CT model,the radiomics model,the conventional CT feature model,and the clinical model.Receiver operating characteristic curve(ROC)and Hosmer-Lemeshow test were used to analyze the discrimination and calibration of each model in the two data sets.The area under the curve(AUC),sensitivity,specificity and Youden index of ROC were calculated.The Delong test was used to compare the differences in AUC between the six models.The calibration curve and decision curve(DCA)were used to evaluate the diagnostic efficacy and clinical effectiveness of different models.Results:A total of 267 patients were included in this study.Among them,there were 141 cases in the early-generation targeted drug resistance group and 126 cases in the late-generation targeted drug resistance group.The median PFS(months)of the two groups were 8.5(6.3,11)and 18(15,21),respectively,with statistical differences(P < 0.001).Among the six models established in this study,the Clinical + CT + Radio-scores model had the best diagnostic performance.Among the indicators included in the model,Radio-Scores was the highest independent influencing factor(OR = 191014.900).In addition,the model also included four factors : gender,total bilirubin,direct bilirubin and pleural retraction.In the training cohort,the AUC value of the Clinical + CT + Radioscores model(AUC = 0.961)was higher than that of the Radio-scores model(AUC = 0.923),CT+ Clinical model(AUC = 0.735),Clinical + Radio-scores model(AUC = 0.951),Clinical model(AUC = 0.666)and CT-Chatacter model(AUC = 0.673).The De Long test showed that except for the Clinical + Radio-scores model,the P values of the Clinical + CT + Radio-scores model were less than 0.05 compared with the other four models.In the test cohort,the diagnostic performance of the Clinical + CT + Radio-scores model was also superior to the other groups.The AUC value was 0.844,which was higher than the Clinical + Radio-scores model(AUC = 0.813),Radio-scores model(AUC = 0.834)and Clinical + CT model(AUC = 0.683),Clinical model(AUC = 0.584)and CT-Character model(AUC = 0.670).The De Long test results showed that the Clinical + CT +Radio-scores model was statistically different from the other four models except the Radio-scores model.Secondly,whether in the training cohort or the test cohort,the sensitivity,specificity and Youden index of the combined model are also higher than the other five models.Finally,the calibration curve and decision curve analysis of the combined model also showed its high clinical effectiveness.Therefore,in this study,the Clinical + CT + Radio-scores model has the strongest discrimination ability and is the optimal model.Conclusion:The combined model(Clinical + CT + Radio-scores)based on clinical indicators,conventional CT features and radiomics features can distinguish early and late acquired firstgeneration targeted drug-resistant patients,which is conducive to clinicians ’ preliminary evaluation of patients before treatment and individual precision treatment.