Development Of Prognostic Models For Malignant Pleural Mesothelioma Based On Clinical,CT Radiomics And BAP1

Objectives:1.To development and validate the prognostic model of Malignant pleural mesothelioma(MPM)based on routine clinical indicators.2.To investigate the prognostic factors with CT morphological features,tumor size,and radiomics in MPM patients,and development and validate the clinical and CT imaging combined prognostic model.3.To explore the feasibility of clinical,CT morphology,and radiomics for prediction of BAP 1 mutation status in MPM patients;and investigate the effect of BAP1 mutation status on prognosis of MPM patients.Methods:1.210 patients pathologically confirmed as MPM were enrolled in this retrospective multicenter study from 2007 to 2020,and divided into 112 cases of training set(105 cases with outcome events)and 98 cases of test set(76 cases with outcome events)according to the admission time.Cox proportional risk model was used to select the clinical prognostic factors,and built the nomogram for risk stratification in MPM patients.Harrell’s Concordance Index(C-Index)and calibration curve were used to evaluate the model’s discrimination and consistency respectively.2.A total of 164 MPM patients(derived from Part 1)underwent pre-treatment CT scan,and were divided into 82 training set(with 76 outcome events)and 82 test set(with 68 outcome events)according to the admission time.The association between overall survival(OS)and CT morphological features,three quantitative methods of tumor size(maximum pleural thickness,2D and 3D volume of tumor),2D and 3D radiomics were analyzed,respectively.The Cox proportional risk model was conducted by combining the optimal imaging factor with the clinical model to establish the nomogram for prognostic prediction and stratification.C-index and calibration curve were performed to evaluate the predictive performance of the combined model.3.Seventy-four MPM patients(derived from Part 1)were randomly divided into the training set(n=51)and test set(n=23)at the ratio of 7:3,whose the paraffin-embedded tumor tissues were detected by Sanger sequencing for BAP1 status(22 cases of mutant and 52 cases of wild-type).The 2D and 3D radiomic features extracted from unenhanced CT images were selected to construct the predictive model for BAP1 status using Logistic regression model.Synthetic minority over-sampling technique(SMOTE)was employed to augment the training set data.The receiver operating characteristic(ROC),calibration curves and Decision Curve Analysis(DCA)were used to evaluate the model’s differentiation,calibration and clinical utility,respectively.The correlation between BAP1 status and prognosis was assessed in 67 patients received follow-up.Results:1.Based on routine clinical indicators,Cox univariate analysis indicated age(P=0.008),residence(P=0.001),age-adjusted Charlson Comorbidity Index(aCCI)(P=0.030),serum albumin(P=0.023),CA153(P=0.010),D2-40(P=0.012),clinical stage Ⅲ(P=0.009),stage Ⅳ(P=0.006),surgery(P=0.005)and chemotherapy(P=0.041)were the prognostic factors.Cox Multivariate analysis showed that residents in Chuxiong area(HR=2.127),albumin reduction(HR=1.583),clinical stage Ⅳ(HR=3.073)and chemotherapy(HR=0.476)were independent factors for OS prediction.The C-index of the nomogram established by the above four factors in the training and test sets were 0.662 and 0.613,respectively.Patients were stratified according to the median risk score of nomogram,and the low-risk group had better outcomes than the high-risk group in both the training(P<0.001)and test(P=0.003)sets.2.For CT morphological features,Cox univariate analysis revealed that OS of patients was significantly correlated with interlobar fissure thickening(P=0.046)and its quantitative value(P=0.013),tumoral encasement of lung(P=0.001)and invasion(P=0.038),and Cox multivariate analysis showed that only tumoral encasement of lung was an independent prognostic factor(HR=1.986)in the training set,but without significance in the test set.The C-index of training and test sets based on tumoral encasement of lung were 0.572 and 0.544,respectively.For three quantitative methods of tumor size,there was a cross-correlation among the maximum pleural thickness,2D and 3D volume of tumor(P<0.001).Patients with the maximum pleural thickness>14.5 mm,2D volume>500 mm3,and 3D volume>111000 mm3 in the training set had a worse prognosis(P<0.05).None of the three methods was a prognostic factor in the test set.Maximum pleural thickness had the highest predictive accuracy in the training set(C-index=0.642),while 3D volume had the highest predictive accuracy in the test set(C-index=0.546),and the best calibration in both training and test sets.For unenhanced CT radiomics,3D radiomic features were more stable than 2D features(P<0.001).The C-index of 2D and 3D radiomic signatures were respectively 0.642 and 0.674 in the training set and correspondingly 0.612 and 0.574 in the test set.The median risk score based on 3D radiomic model can be used to make prognostic stratification in both training(P<0.001)and test(P=0.04)sets.However,there was no significant difference in OS between the low and high risk groups with 2D radiomic model(P=0.18).The predictive performance of the clinical and 3D radiomic combined model was higher in training(C-index=0.709)and test(C-index=0.624)sets than that of the separate clinical or 3D radiomic model.The patients in the training(P<0.001)and test(P=0.007)sets can be stratified into low and high risk groups with the combined model based on the median risk scores.3.The AUCs of 3D and 2D radiomic models for the prediction of BAP1 mutation status were 0.786&0.683 in the training and 0.768&0.652 in the test sets,respectively.The calibration curve and DCA curve also indicated that 3D model was better than 2D in predicting the BAP1 mutation status.There was no significant difference in OS between the BAP1 mutant and wild-type groups(P=0.337),but the median survival time of the mutant group was greater than that of the wild group(503 days&367 days).Conclusions:1.The nomogram established based on routine clinical indicators provides a reliable tool for prognostic prediction and risk stratification in MPM patients.2.The clinical and CT 3D radiomics combined model can improve the predictive performance,which is a new and optimized method to predict the prognosis and make risk stratification for MPM patients.3.CT 3D radiomics model can be used as a noninvasive imaging marker to predict the mutation status of BAP 1.