A total of 3,471 consecutive patients with ischemic stroke were admitted during the study period. Of them, 594 patients met the inclusion criteria and 335 of the 594 were eligible but excluded. Reasons for exclusion were refused consent (88), death and/or comfort measure only (CMO) (47), no acute lesion on MRI (44), inability to attend follow-up (27), new claustrophobia or anxiety (26), MRI technique failure (26), loss to follow-up (19), hemicraniectomy (18), lack of a follow-up image due to protocol restrictions (17), inability to obtain consent (11), severe cerebral edema (6), subacute lesion on MRI (3), and acute hemorrhagic infarction (3). Thus, the total study population consisted of 259 patients. Of these, 153 patients were prospectively collected, and 106 patients were retrospectively collected applying identical criteria.
Within the resulting patient population, the study group was defined a priori as the natural history cohort: therefore, 72 were excluded due to intravenous tissue plasminogen activator administration, 38 due to investigational drug studies [hyperoxia (22), induced hypertension (10), and desmoteplase (6)], 6 due to intra-arterial tissue plasminogen activator, 4 due to emergency carotid bypass surgery, 3 due to intra-arterial mechanical thrombolysis, and 1 due to an investigational interventional NeuroFlo device. Further analyses included the subsequent 135 patients. Table summarizes the baseline characteristics, and clinical and imaging features of this cohort. Cerebral infarcts were located within the middle cerebral artery territory in 96 patients, anterior cerebral artery territory in 4 patients, posterior cerebral artery territory in 11 patients, internal carotid artery territory in 5 patients, and basilar artery and/or cerebellar artery territories in 19 patients. Ischemic lesion volumes on acute and follow-up images and volume of growth are also listed in table .
Baseline characteristics and clinical/imaging features
Hemoglobin level demonstrated a statistically significant inverse correlation with acute DWI volume (r = −0.20, p = 0.02) and volume of infarct growth (r = −0.20, p = 0.02). Admission blood glucose was associated with infarct growth (r = 0.18, p = 0.04), but not with initial acute DWI volume (p > 0.20), which is consistent with prior reports [1
]. CCS-TOAST stroke subtypes demonstrated a statistically significant association with initial infarct volume [11
] (p = 0.04), but not with infarct expansion (p = 0.11). NIHSS scores were available for a subset of patients (n = 85), and analysis of these patients showed correlation with acute DWI volume (r = 0.50, p < 0.001) and infarct growth (r = 0.35, p < 0.001). Time to follow-up scan was also correlated with the volume of infarct growth (r = −0.27, p < 0.01). Other comparisons, including age, gender, and time-to-acute MRI, did not demonstrate significant correlations with the above imaging outcomes.
We also evaluated the correlation between hemoglobin and the imaging outcomes, stratified by sex, since women have lower hemoglobin levels than men. We found the inverse correlation was consistent for men (n = 90; r = −0.27, p = 0.01 for acute DWI volume, and r = −0.22, p = 0.04 for infarct growth) and also present for women (n = 45; r = −0.26, p = 0.08 for acute DWI volume, and r = −0.42, p = 0.004 for infarct growth). Because hemoglobin may be a marker for chronic disease, we evaluated whether there was an association between hemoglobin level and Framingham stroke risk score [12
]. In patients in whom a risk score could be calculated (n = 94), there was no significant correlation with hemoglobin (r = −0.03, p = 0.77). We also evaluated the association between hemoglobin level and individual stroke risk factors including type II diabetes (p = 0.46), coronary heart disease (p = 0.25), atrial fibrillation (p = 0.32), hypertension (p = 0.03), hyperlipidemia (p = 0.56), smoking (p = 0.67) and CCS (p = 0.08). Hypertensive subjects had higher hemoglobin levels (14.3 vs. 13.7 g/dl), but hypertension was not statistically significantly associated with either acute DWI volume or infarct growth. Together, these data argue that hemoglobin is not a marker for chronic disease based on other known risk factors for stroke.
We next constructed a multivariable regression model, which demonstrated that hemoglobin was an independent predictor for acute DWI volume (p < 0.005). Table lists the covariables included in the model, including age, gender, admission glucose, time to acute MRI and CCS stroke subtype. Because the effect of hemoglobin may be influenced by either age or sex, we explored the interaction of gender·hemoglobin and age·hemoglobin post hoc in the model. However, neither interaction term had a statistically significant effect on the model (both p > 0.20). We also included admission NIHSS score into the model post hoc because this value was not available for all patients (n = 85). In this model with fewer total subjects, hemoglobin remained an independent predictor of acute DWI volume (p = 0.03, table ).
Multivariate relationships with acute infarct volume as the dependent variable
We also explored the independent effect of hemoglobin on infarct growth (table ). We evaluated absolute infarct growth, rather than its percentage, which has been shown to be associated with neurological outcome [13
]. We included the volume of the perfusion-diffusion mismatch, the time to the follow-up scans, and all other variables used previously. Hemoglobin was independently associated with infarct growth (p = 0.02). Additional variables that exhibited independent effects were gender (p = 0.01), stroke subtype (p = 0.03) and NIHSS score (p < 0.01). Based on this model, for each drop in hemoglobin of 1 g/dl, there is a 5.5 ± 2.4 cm3
increase in the volume of infarct growth (fig. ).
Multivariate relationships with volume of stroke growth as the dependent variable
Hemoglobin level versus infarct growth.
Because elevated hematocrit has been reported to associate with infarct expansion [14
], we explored whether a U-shaped polynomial association existed with hemoglobin levels. We found that in bivariate analysis, quadratic regression provided a statistically significantly improved fit compared to linear regression (p < 0.01). However, when we analyzed this relationship in multivariable regression, incorporating hemoglobin with a quadratic term did not meet statistical criteria for a better fit than linear regression (p = 0.06), perhaps due to the few number of patients at the edges of the hemoglobin range.
Finally, we performed a subset analysis only in patients who had oligemic brain tissue ‘at risk’. Although perfusion MRI is imprecise in differentiating between benign oligemia and true penumbra, we reasoned that selecting all patients with a perfusion-diffusion mismatch would provide a population of patients with greater risk for infarct growth; in 110 patients, perfusion volume was ≥120% of diffusion volume, a previously defined selection criterion [10
]. Hemoglobin had a statistically significant inverse correlation with infarct growth (r = −0.20, p = 0.04), consistent with the entire cohort. In a multivariable analysis (table ), hemoglobin remained independently significant in this subgroup (p = 0.01), as did gender (p = 0.02), CCS stroke subtype (p = 0.03) and NIHSS (p = 0.02). Conversely, in patients who did not have a mismatch on perfusion MRI (n = 25), hemoglobin did not correlate with the volume of infarct expansion.
Multivariate relationships with volume of stroke growth as the dependent variable (only in those patients with perfusion-diffusion mismatch >120%)