Development And Validation Of A Prediction Model For Aortic Adverse Events After Endovascular Repair For Acute Type B Aortic Dissection Involving Abdominal Visceral Segment

BackgroundAcute aortic dissection is characterized by a rapid onset of blood flow into the false lumen through a tear in the intima,which may cause aortic rupture and lead to sudden death in severe cases.Thoracic endovascular aortic repair(TEVAR),which reduces blood flow in the false lumen and promotes positive aortic remodeling by sealing the proximal tear with a covered stent,has become the main treatment for Stanford type B aortic dissection(TBAD).When the aortic dissection involves abdominal visceral segment,TEVAR only by covering the proximal tear may bring hidden dangers to the long-term prognosis of patients,and the risk of postoperative adverse events increases.The development of a prediction model for adverse events may identify high-risk patients,provide a basis for improving treatment plans and follow-up strategies,and help to improve the efficacy of endovascular treatment and long-term prognosis.Objective(1)to development and validate a clinical prediction model for aortic adverse events after TEVAR for acute TBAD involving abdominal visceral segment by collecting clinical information,laboratory examination results and imaging measurements;(2)To perform the radiomics analysis of abdominal visceral segment of acute TBAD,development and validate the clinical-radiomics comprehensive prediction model of aortic adverse events after TEVAR,and compare the performance of the comprehensive prediction model and the clinical prediction model.MethodsPatients with acute TBAD involving visceral region who underwent the first TEVAR in our hospital from January 2011 to June 2020 were consecutively enrolled.According to the inclusion and exclusion criteria,a total of 309 patients were selected in this study,and they were randomly divided by 8:2,including 247 patients in the training cohort and 62patients in the test cohort.The patients’general clinical information and laboratory examination results were collected through the electronic medical record system,and the DICOM format data of preoperative CT enhanced angiography of total aorta were obtained through PACS.The Aquarius image workstation was used to measure the length and area of aortic dissection.ITK-SNAP software was used to segment the visceral region of the abdominal aorta,and the radiomics features were extracted using Pyradiomics.Patients were followed up through the electronic medical record system and telephone,and the end point was aortic adverse events.Univariate and multivariate COX regression analysis were used to screen predictors of clinical prediction model,and LASSO-Cox regression was used to screen radiomics features and establish radiomics score.COX regression was used to develop the clinical prediction model and clinical-radiomics comprehensive prediction model.The discrimination of the model was evaluated by C-index and AUC.The calibration was evaluated by calibration curve,and the clinical applicability was evaluated by DCA.The NRI and IDI were used to evaluate the performance of the clinical model and the comprehensive model.Results1.Development and validation of clinical prediction model:There was no significant difference in baseline between the training set and the test set.A total of 61 aortic adverse events occurred,including 50 in the training set and 11 in the test set.The most common type of adverse event was distal aortic dilatation.The variables selected in the clinical model were chronic kidney disease,pericardial effusion,maximum area of ascending aorta>1025mm~2 and maximum area of descending aorta>1253 mm~2.The C-index of the training set was 0.73(95%CI:0.661-0.799),and the AUC of 1 year,2 years,and 3 years was 0.766,0.750,and 0.737,respectively.The C-index of the test set was 0.694(95%CI:0.537-0.851),and the AUC of 1 year,2 years,and 3 years was 0.699,0.739,and 0.746,respectively.The1-,2-and 3-year calibration curves showed that the predicted and observed values were close to the ideal line,and the curve did not deviate significantly from the ideal line.DCA curve showed that when the threshold probability of the training set was 0.04-0.5,and the threshold probability of the test set was 0.05-0.25,the clinical prediction model could bring net benefit to patients.According to the clinical prediction model,the patients were divided into high-risk group and low-risk group.KM curve between the two groups showed that the event-free survival rate of high-risk group was significantly lower in the training set(P<0.0001),but there was no significant difference in the test set(P=0.059).2.Development and validation of clinic-radiomics prediction model:A total of 428radiomics features were extracted using Pyradiomics,including 72 first-order features(16.8%),56 shape features(13.1%)and 300 texture features.Pearson correlation analysis filtered out 168 independent radiomics features,and LASSO COX regression screened out13 features closely related to adverse events.The comprehensive model was constructed based on the radiomics score and clinical predictors.The C-index of the training set was0.812(95%CI:0.749-0.875),and the AUC of 1 year,2 years and 3 years were 0.848,0.839and 0.838,respectively.The C-index of the test set was 0.74(95%CI:0.591-0.889),and the AUC of 1 year,2 years and 3 years was 0.828,0.741 and 0.670,respectively.The predicted and observed values of the calibration curve of the comprehensive model in the training set and the test set at 1,2,and 3 years were close,and the curve did not significantly deviate from the ideal line.DCA curve showed that when the threshold probability of the training set was 0.05-0.5,and the threshold probability of the test set was 0.05-0.3,the clinical-radiomics comprehensive prediction model could bring net benefit to patients.According to the comprehensive model,the KM curve between the two groups showed that the event-free survival rate of the high-risk group was significantly lower in the training set and the test set,and the difference was statistically significant(P<0.0001 and P=0.023).When the threshold of prediction probability was 0.2-0.4,the NRI of the comprehensive model and the clinical model were greater than 0 at 1 year,2 years and 3 years.The IDI values of the two models at 1 year,2 years and 3 years were greater than 0.ConclusionsChronic kidney disease,pericardial effusion,maximum area of ascending aorta,maximum area of descending aorta and the radiomics score of abdominal visceral segment are the predictive factors of the aortic adverse events of acute TBAD.The clinical prediction model and the clinical-radiomics comprehensive prediction model have good discriminations,and the comprehensive model is more efficient,can be used to identify high-risk patients,provide the basis for individualized treatment and follow-up strategies,and help improve the treatment effect and the prognosis of patients.