We performed a retrospective chart review of patients admitted to two metropolitan hospitals staffed by the same stroke team: Suburban Hospital, Bethesda, MD, and Washington Hospital Center, Washington, DC. Patients were identified through a stroke registry/database. Data were available from Suburban Hospital from 2000 to 2007, and from Washington Hospital Center from 2004 to 2007. Suburban Hospital serves a predominantly middle class white population, while Washington Hospital Center serves a predominantly lower middle class black population that has been officially designated as medically underserved. All potential patients with stroke at both hospitals underwent routine MRI studies including a GRE sequence, unless a contraindication to MRI existed. The study was performed in accordance with the Institutional Review Boards of both hospitals.
Patients were included in the study if 1) their diagnosis was primary intracerebral hemorrhage (not limited to first ICH), 2) an MRI with GRE and fluid-attenuated inversion recovery (FLAIR) sequences was performed within 1 month of their admission for primary ICH, and 3) their race was specified as black or white. There were insufficient numbers of other races/ethnicities to provide a meaningful analysis. A diagnosis of primary (nontraumatic) ICH was given after each patient underwent routine imaging and diagnostic evaluations to identify an underlying cause for the hemorrhage including MRI in all patients, and typically cerebral angiography for patients with unexplained lobar hemorrhages or absence of vascular risk factors.
The following clinical information was extracted from the medical record of each patient: age, sex, risk factors for ICH (prior stroke, hypertension, coronary artery disease [CAD], diabetes mellitus, hyperlipidemia, tobacco use), medications at time of admission, heavy alcohol use (>2 drinks/day), and active treatment with antiplatelet or anticoagulant therapy at the time of admission. Race (using US Census Bureau classifications) was assigned by a physician incorporating information provided by the medical record and confirmed with the subject. Vascular risk factors were identified by prior history at the time of admission. Hypertension was defined by prior medical history. Untreated hypertension was defined as absence of antihypertensive medication at the time of admission in any patient with a history of hypertension. There was insufficient information in the medical record to extract information about socioeconomic status (SES).
Images were acquired using a gradient recalled echo sequence at Suburban Hospital on a 1.5 T (General Electric) scanner and at Washington Hospital Center on a 3.0 T (Philips) scanner. For the 1.5 T system: repetition time (TR)/echo time (TE) 800/20 msec, field of view (FOV) = 24 cm, 256 × 192 matrix, 20 7-mm-thick contiguous but interleaved axial-oblique slices aligned with the AC-PC, 30° flip angle, NEX = 1, SENSE = 1. For the 3.0 T system: TR/TE 875/11 msec, FOV = 22.4 cm, 224 × 112 matrix, 35 4-mm-thick contiguous but interleaved axial-oblique slices aligned with the AC-PC, 40° flip angle, NEX = 1/2, SENSE = 2.
A related but separate study was performed to ensure that differences in magnet strength did not impact the number of microbleeds detected.15
For this secondary study 31 patients (independent from the patients included in the current study) underwent scanning with gradient echo imaging on both a 1.5 T and 3.0 T scanner a median of 3 days apart (range 0–280). The scan parameters (TR/TE/slice thickness) for the 1.5 T and 3.0 T systems were the same as those employed for the current 87-subject study. Scans were independently interpreted by an experienced stroke neurologist and a neuroradiologist blinded to field strength and patient identity. There was no difference between the number of microbleeds detected on 1.5 T (p
≤ 0.8) compared to 3.0 T MRI (p
≤ 0.9) for the two readers. A total of 35% of scans were microbleed positive on the 1.5 T vs 32% on the 3.0 T. There were 10 cases in which there was a discrepancy in number of microbleeds between field strengths (range of discrepancies 1–4). In half of these cases, more microbleeds were seen on the 1.5 T scans. Inter-rater agreement for the presence of microbleeds yielded a kappa of 0.86.
One investigator (C.K.) interpreted the imaging data for the current analysis blinded to clinical information. The following data points were assessed employing the gradient echo sequences: ICH location (lobar, subcortical/deep, infratentorial), total number of microbleeds, number of microbleeds by location (lobar, subcortical/deep, infratentorial), and total number of chronic hematomas. Microbleeds were defined as rounded, punctate, homogenous hypointensities generally <5–10 mm in size located within the parenchyma. Hypointensities appearing in sulci consistent with vessels visualized end-on were not counted as microbleeds. Symmetric hypointensities in the basal ganglia most likely represent calcifications or iron deposition and were not counted as microbleeds. Chronic hematomas were defined as slit like regions of hypointensity. FLAIR sequences were used to rate leukoaraiosis using the four-point Fazekas scale.16
ICH volumes were calculated by outlining regions of interest on each slice using semiautomated image segmentation tools (Cheshire, Perceptive Informatics, Inc., Boulder, CO).
Differences in dichotomous variables between the black cohort and the white cohort were analyzed using χ2
analysis or Fisher exact test. The Student t
test or the Mann–Whitney U
test was used to analyze differences in the mean or median of continuous variables between groups. Logistic regression analysis was performed to test the predictive value for microbleed occurrence (presence of 1 or more) employing the SPSS forced entry method. Risk factors that were significant on univariate analysis were included; age and hypertension were also forced into the model based on prior studies suggesting these items are significant risk factors.17,18
A second analysis replaced history of hypertension with untreated hypertension. All statistical analyses were performed using the Statistical Package for the Social Sciences version 14.0 (SPSS, Chicago, IL).