Vessel Wall Enhancement on Black-Blood MRI Predicts Acute and Future Stroke in Cerebral Amyloid Angiopathy

Study Design and Cohort

Institutional review board approval was obtained before this retrospective study at our academic center from 2015 to 2019 in patients undergoing vwMRI for acute neurologic deficits concerning for ischemic stroke. Because of the retrospective nature, informed consent was not required by the institutional review board. In this protocol, all patients with CAA included had probable or possible CAA (based on modified Boston criteria and microhemorrhages detected before vwMRI) and documented acute focal neurologic deficits warranting vwMRI during the stroke work-up. All patients with CAA admitted under the neurology team with suspected stroke during this time period underwent vwMRI work-up per protocol. Although a few scans exhibited mild motion artifacts, none was sufficient to exclude any participants from interpretation. Chart review was used to determine age, sex, and other potential cerebrovascular confounders at the time of vwMRI.

CAA Diagnosis

Patients diagnosed with probable or possible CAA were included in this study. Modified Boston criteria were used by a vascular neurologist to diagnose CAA.¹⁴ Briefly, modified Boston criteria for probable CAA included age 55 years or older, appropriate clinical history, and MR imaging findings demonstrating either 1) multiple hemorrhages restricted to lobar, cortical, or corticosubcortical regions of varying sizes or ages without another cause or 2) a single lobar, cortical, or corticosubcortical hemorrhage and focal (≤3 sulci) or disseminated (>3 sulci) cortical superficial siderosis without another cause. Cases of possible CAA were also included based on the modified Boston criteria definition: single lobar, cortical, or corticosubcortical hemorrhage without other cause. Twelve of our CAA cases were also confirmed with biopsy.

vwMRI Protocol

All MR imaging was obtained at 3T on Verio, Trio, or Prisma platforms (Siemens) with standard head coils and a standard contrast dose of MultiHance (Bracco), 0.1 mmol/kg. Sequences included those in the Online Supplemental Data. The sampling perfection with application-optimized contrasts by using different flip angle evolutions (SPACE; Siemens) sequences were acquired with delay alternating with nutation for tailored excitation flow suppression.²⁴ Images were obtained in this order—precontrast: DWI, SWI, TOF, T2 SPACE, T1 SPACE and postcontrast: TOF and T1 SPACE.

Brain Parenchymal Findings

Both acute and future ischemic stroke were determined using the American Heart Association definition of CNS infarction as previously described.²⁵ Briefly, acute or future ischemic stroke was defined by brain cell death attributable to ischemia based on 1) imaging evidence of cerebral infarction or 2) clinical symptoms persisting ≥24 hours, with other causes excluded. We also reviewed neurovascular clinic or inpatient charts to determine presence of AIS occurring at the time of imaging for concurrent stroke and at follow-up visits or neuroimaging for future stroke. Although all patients with CAA in this study had the clinical suspicion of stroke, only DWI-positive and low-ADC value acute infarcts were included as positive for AIS. In addition, the DWI positive infarct did not have to necessarily explain the patient's presentation and could be included even if clinically silent. The variable time between symptom onset and MR imaging was included as a potential confounder in the statistical analysis. Acute infarct was defined using DWI derived from DTI trace images described previously with parameters in the Online Supplemental Data.²⁶ DTI trace outperforms conventional diffusion-weighted sequences in detecting acute infarcts.²⁷ Brain DWI was interpreted by a subspecialty-trained, Certificate of Added Qualification–certified neuroradiologist blinded to vascular imaging. Follow-up imaging was only performed if patients developed an additional new, acute neurologic deficit concerning for ischemic stroke. DWI was the primary imaging determinant for both acute and future ischemic infarct in all patients. Although we had access to both CT and FLAIR imaging for all patients, these did not show additional interval infarcts not detected by DWI. Acute hemorrhage was also diagnosed on a CT scan by a neuroradiologist blinded to vascular imaging.

Vessel Wall Enhancement

Two separate board-certified radiologists evaluated VWE using T1 SPACE pre- and postcontrast sequences on vwMRI, blinded to additional imaging or clinical information. In cases of disagreement, a third radiologist was used as a tiebreaker. A signal threshold was used for VWE positivity using signal at or above the pituitary infundibulum as previously described.^1,3 VWE was graded in a binary manner—either positive or negative for the entire set of brain images of each patient. This was done for the whole brain without regard to vascular segment, nor was it necessarily coupled with the area of DWI positivity. In addition to evaluating whole-brain VWE, we evaluated territorial VWE in patients with acute ischemic infarcts. VWE was evaluated in medium to small artery walls and only in vessels with confirmed arterial flow visible on TOF (including the first- through fourth-order branches of the circle of Willis). The TOF was only used to confirm the vessel was an artery and not a vein, and the VWE readers did not simultaneously grade the TOF for stenosis (see later discussion). Enhancement along a single vessel wall was considered positive, though in all positive patients, >1 vessel was involved.

Stenosis on vwMRI

Maximum percent diameter stenosis was determined per the Warfarin-Aspirin Symptomatic Intracranial Disease trial measurement of intracranial stenosis.²⁸ The subspecialty-trained, Certificate of Added Qualification–certified neuroradiologist determined stenosis blinded from DWI and VWE images at the time of TOF evaluation (and performed after a 2-month washout period). Of all intracranial arteries visible on TOF, the artery with maximum stenosis was selected using a combination of axial imaging, multiplanar reformats, maximum intensity projections, and 3D reformats. Then the diameter (b) at the level of maximal stenosis and diameter (a) of the downstream vessel distal to the stenosis were used to calculate percent diameter stenosis using the formula [(a-b)/a] × 100%. Stenosis was measured at the narrowest segment of the vessel (b) perpendicular to the long axis of the vessel on multiplanar reformats using a submillimeter measurement tool. From the Warfarin-Aspirin Symptomatic Intracranial Disease trial, in cases in which the downstream segment of the vessel was diseased, a proximal normal vessel segment was used for the comparison diameter (a). In cases in which the entire vessel was diseased, the contralateral vessel was used for the comparison diameter (a).

Interrater Reliability

For interrater reliability, we evaluated an equal number of negative control participants alongside the patients with CAA. This group of CAA-negative patients were age-matched (older than 55 years old) and had undergone vwMRI during the same time period, with the same protocol, and in the same scanners. These patients had no evidence of probable or possible CAA, vascular stenosis, aneurysm, or acute stroke or hemorrhage. Interrater reliability for the binary vwMRI measurement of positive or negative enhancement was calculated in the combined group of CAA and negative control participants blinded to all other imaging and clinical covariates using prevalence- and bias-adjusted kappa statistics.

Statistical Analysis

Univariable binary Poisson regression, with robust standard errors, was used in the patients with CAA to determine the prevalence ratio (PR) of vwMRI factors, as well as clinical confounders for concurrent AIS as determined by retrospective chart review. These confounders included maximum percent stenosis, age, female sex, hypertension, hyperlipidemia, diabetes, smoking, antiplatelet use, anticoagulation use, antihypertension use, and statin use. Systemic inflammatory markers (erythrocyte sedimentation rate and C-reactive protein) were also recorded in 30 of the 50 patients with CAA. The time between symptom onset and MR imaging was also evaluated in relation to AIS. Next, all potential confounding variables with P < .20 from the univariable analysis were placed in an initial multivariable binary Poisson regression model with robust standard errors for the prediction of AIS; then variables were eliminated in a backward fashion until all remaining variables met the threshold P < .20.

Accuracy of the prediction of AIS was determined using the area under the receiver operating characteristic curve, also called the c-statistic. Lastly, we investigated whether the presence of VWE predicted future ischemic stroke using Kaplan-Meier survival analysis to estimate the ischemic event rate. Ischemic event rates per 100 person-years were calculated for each outcome, and the formula annual risk = 1(exp[-event rate × time]) was used to estimate the absolute annual risk. Time to event was further analyzed for stroke by use of univariable followed by multivariable Cox regression analysis. After univariable analysis, factors with P < .20 were placed in an initial multivariable model; then variables underwent backward elimination to a threshold P < .20. All patients were followed up by chart review for interval stroke symptoms to the end of the study in 2019. Cox regression was used to account for variable times to event.

In binary outcome models, 5 outcome events for every predictor are sufficient to avoid overfitting.²⁹ Our final model for future stroke met this. The final Poisson regression model, however, did not, and this is known to increase the likelihood of finding false-positive predictors.³⁰ To assess the stability of the final Poisson model, the bootstrap inclusion fraction was computed for each predictor, which is the percentage of times the variable remains in the final model in a large number of bootstrap resamples in which the variable selection is repeated.^31,32 Predictors with bootstrap inclusion fractions <50% were dropped from the final model as unreliable because these would not likely remain as significant predictors in future datasets.

All statistical analyses were performed using Stata-15.1 statistical software (StataCorp).