Registry of Aortic Diseases to Model Adverse Events and Progression (ROADMAP) in Uncomplicated Type B Aortic Dissection: Study Design and Rationale

Study Design

The Registry of Aortic Diseases to Model Adverse Events and Progression (ROADMAP) is a collaboration between 10 academic aortic centers, initiated in 2017, to assemble a large enough study cohort combining imaging and clinical data to test and validate morphologic predictors and improve risk stratification for patients with aortic diseases. This international registry currently includes seven North American and three European centers with long-standing expertise in aortic diseases (Table 1). The current investigation is designed as a retrospective, observational, longitudinal, multicenter study to externally validate a previously derived risk prediction model for late adverse events in patients with uTBAD (4). An overview of the study design flowchart is available in Figure 1. The study protocol has been reviewed and approved by the institutional review boards of each participating institution, and the requirement for written informed consent has been waived.

Table 1: List of Aortic Centers Included in ROADMAP Study in Patients with uTBAD

Figure 1:

Download as PowerPoint

Study Objective and End Points

The objective of this study is to strengthen the evidence for imaging-based risk stratification and individualized treatment of patients with uTBAD by externally validating a previously derived risk prediction model for late adverse events (4). The predictors based on the derivation cohort include one clinical parameter (connective tissue disease) and four imaging parameters (maximum aortic diameter, in millimeters; FL circumferential angle, in degrees; FL outflow, in milliliters per minute; and the number of identifiable intercostal arteries, n) (4).

The main study end point will be the first occurrence of any late adverse event, defined as fatal or nonfatal aortic rupture, organ or limb ischemia, aortic aneurysm formation, rapid aortic growth, new refractory hypertension, and pain uncontrollable by medical management (Table 2). Late adverse events will therefore be derived from two data streams: clinical outcomes and serial imaging-based morphologic outcomes. Clinical adverse events will be reported by each participating center. Morphologic outcomes will be derived from baseline and follow-up imaging data and will be centrally analyzed at the imaging core laboratory.

Table 2: List of Late Adverse Events

Patients

The validation cohort will include patients retrospectively identified from institutional databases of eight aortic centers: five centers in the United States (Ann Arbor, Michigan; Atlanta, Georgia; Boston, Massachusetts; Houston, Texas; Pittsburgh, Pennsylvania), one center in Canada (Toronto, Ontario), and two in Europe (Bologna, Italy; Zurich, Switzerland). Patients enrolled in the original derivation cohort (Stanford, California; Maastricht, the Netherlands) will not be included (Table 1) (4).

We will include patients aged 18 years or older who presented to any of the participating centers with acute Stanford type B aortic dissection between January 1, 2001, and December 31, 2013, and who survived the initial hospitalization without complications necessitating endovascular or surgical intervention. Complications are defined as the following: aortic rupture or signs of impending rupture, organ or limb ischemia, refractory pain, and uncontrollable hypertension. Patients will be included if at least one CT angiography (CTA) scan was performed at the time of admission and initial hospitalization and if at least one follow-up CT scan or MRI scan is available. No minimum time period between hospitalization and follow-up is required for inclusion to capture patients with adverse events early after discharge. Based on the combined institutional experiences of the participating aortic centers, we expect that the median follow-up time of patients in the validation cohort will be similar to or greater than that in the development cohort (850 days; range, 247–1824 days). At baseline, only CTA scans will be included. Both CT and MRI studies with or without contrast enhancement will be included for the follow-up. Patients will be followed up until a late adverse event occurs or until December 31, 2020. Exclusion criteria will be iatrogenic or traumatic dissection, intramural hematoma, penetrating atherosclerotic ulcer, rupturing aortic aneurysm, and dissection involving exclusively the abdominal aorta, which has a different natural history and fewer endovascular treatment options (13). We will also exclude patients with inadequate or incomplete imaging data, as summarized in Table 3.

Table 3: Inclusion and Exclusion Criteria

Sample Size Estimation

To estimate the minimum sample size for external validation of our predictive model, we follow the recommendations of Steyerberg et al (14,15) suggesting that for a five-parameter prediction model, a minimum of 100 events is necessary to establish validity in an external cohort. Assuming a similar event rate as in the derivation cohort of 0.4, we aim for a sample size of 250 patients. To account for incomplete or corrupted data, 300 patients will be initially included.

Data Collection, Transfer, and Storage

Clinical and imaging data will be collected independently at each aortic center. Data will be locally coded and anonymized following institutional guidelines. Clinical data will be transferred to the core laboratory (Stanford 3D and Quantitative Imaging Laboratory) by using secure electronic transfer protocols. CT and MRI data will be securely transferred electronically or by using encrypted hard drives, in accordance with institutional and core laboratory security protocols. After transfer, clinical and imaging data will be screened for completeness and consistency and will be uniquely recoded. Clinical data will be stored in a local Research Electronic Data Capture (REDCap) database. Image data will be stored on a dedicated medical image processing server (iNtuition; TeraRecon) with a secure backup.

Imaging Analysis Plan

Imaging data will be evaluated by using commercial three-dimensional (3D) image processing software (iNtuition version 4.4.13; TeraRecon). We will employ custom-built workflows, consisting of standardized procedures for manipulation of imaging data sets, identification and annotation of anatomic features (eg, branch vessel location relative to the true lumen and FL), and obtaining measurements. Key postprocessing steps along with the completed set of annotations and measurements will be recorded and saved for each imaging study, at every time point, and for each observer on the TeraRecon 3D server as “scene files” in Digital Imaging and Communications in Medicine (DICOM) format.

Morphologic assessment of baseline imaging data.— Baseline images—obtained during the initial hospitalization of each patient—will be reviewed to localize, classify, and measure features of the four predefined morphologic predictors: maximum aortic diameter, FL circumferential degree, FL outflow, and the number of intercostal arteries (Fig 2). If more than one CTA scan was performed during initial hospitalization, the scan with the highest image quality will be selected for analysis. The steps of the workflow are described in detail by Sailer et al (4) and are summarized in Figure 2. Aortic diameter measurements are obtained from double-oblique reformations, orthogonal to the aortic centerline. We include the aortic wall (“outer-to-outer” wall), which facilitates measurements if the aortic wall is thickened or if thrombus is present and also allows inclusion of follow-up scans performed without intravenous contrast enhancement. Baseline morphologic assessment will be performed by one of two readers (D.M., M.J.W.), each with more than 8 years of experience in cardiovascular imaging. Both readers will be blinded to outcomes. A subset of 30 of an expected 250 baseline imaging studies (12%) will be independently analyzed by both readers to determine the interobserver agreement.

Figure 2:

Download as PowerPoint

Morphologic assessment of follow-up imaging data.— Follow-up imaging data—numbering approximately three times as many as baseline scans—will be reviewed to establish morphologic outcomes. By tracking each patient’s maximum aortic diameter over time, we will determine aortic growth and the maximum aortic diameter attained by each patient within the study period. Morphometric analysis of serial follow-up imaging data will be performed by the same two experts in cardiovascular imaging (D.M., M.J.W.), in addition to one expert 3D technologist (Shannon Walters, RT, MS) with 12 years of experience in cardiovascular image processing.

Adjudication of abnormal imaging findings.— Any imaging findings on baseline scans considered inconsistent or unusual for patients with uTBAD will also be reviewed by a senior cardiovascular radiologist (D.F.), with more than 20 years of experience, to ascertain the diagnosis. Patients with involvement of the ascending aorta and patients with an aneurysmal (≥6 cm) transverse or descending thoracic aorta at baseline will be excluded. We will also exclude patients with alternative diagnoses, such as limited intimal tears, intramural hematoma, penetrating atherosclerotic aortic ulcer, and rupturing aortic aneurysm, by using established criteria (16,17). New findings suggestive of an adverse event at follow-up imaging—such as signs of rupture, new dissection, or redissection—will also undergo expert review and adjudication.

Data Extraction, Quality Check, and Outlier Analysis

A dedicated Python script will be used to automatically extract all labels, annotations, and measurements recorded by the observers. Content stored in the DICOM headers of the “scene” series on TeraRecon will be converted to extensible markup language files and saved as comma-separated values files and imported into the REDCap database. All combined clinical and imaging data will be checked for consistency, and an outlier analysis will be performed. Each patient will be ultimately categorized as positive or negative for the study end point—the occurrence of a late adverse event—on the basis of clinical and/or imaging data.

Statistical Analysis

Baseline demographics, clinical data, and imaging measurements will be presented as means ± SDs or medians with IQRs in parentheses for continuous variables, depending on data distribution, and as numbers with percentages in parentheses for categorical variables. Demographic, clinical, and imaging data differences between patients with and without late adverse events will be investigated by using unpaired tests. Interobserver variability of morphologic imaging variables will be evaluated by using intraclass correlation coefficients and Bland-Altman plots. The proportional hazards assumption will be assessed by determining the correlation between the Schoenfeld residuals and time. If the proportional hazards assumption fails, we will evaluate proportional odds or accelerated failure time models.

We will externally validate the original five-parameter prediction model by means of multivariable Cox regression analysis with the same single clinical variable (presence of connective tissue disease) and four imaging variables (maximum aortic diameter, FL circumferential angle, FL outflow volume, and number of identifiable intercostal arteries in the dissected portion of the aorta) (4). Patients will be classified into three risk categories based on their individually calculated risk scores by using the original model (4): low (<6), intermediate (6–7), and high (>7) risk for late adverse events. Survival analyses—based on freedom from any of the predefined late adverse events (Table 2)—will be assessed after 1 year, 2 years, and 5 years of follow-up time after the initial dissection. Standard right censoring will be used for patients without the occurrence of late adverse events within the follow-up time (including patients who drop out before any adverse effects occur). Cases of informative censoring will be recorded. Concordance statistics will be assessed to evaluate model performance. Data analysis will be performed by using R Studio version 1.2.1335 (R Foundation for Statistical Computing) and SPSS for Windows, version 26 (IBM). A P value less than .05 will be considered to indicate statistical significance.