Patients with acute ischemic stroke, who were admitted to University Hospital of Cleveland, a tertiary medical center, between February and November of 2000 were identified. Of this cohort, patients who underwent CGCT to evaluate AA were selected. Over the interval of the study, there were no consistent guidelines for the use of CGCT in looking for the etiology of ischemic stroke. The CGCT procedure was selected for the patients who had technically inadequate TEE studies, or for those in whom TEE had been unrevealing. To assess for a selection bias, the study sample was compared to the remaining patients with TEE procedures who did not have CGCT.
(1) Adult 18years or older.
(2) Acute ischemic stroke diagnosed by a stroke neurologist.
(3) Completed CGCT for the purpose of evaluating AA.
(1) Stroke patients who did not undergo the CGCT.
(2) Patients who could not complete the CGCT.
(3) The image quality was not adequate for reliable interpretation.
The explanatory variable was the presence or absence of AA on the CGCT and the response variables were stroke severity at presentation and 3-month outcome. The initial stroke severity was assessed using the National Institute of Health Stroke Severity Scale (NIHSS). The NIHSS is routinely recorded in ischemic stroke patients admitted to our institution. The NIHSS was dichotomized to less or equal to 6 points for mild and moderate to severe stroke.
The second clinical outcome of interest is the 3-month modified Rankin disability scale (modified Rankin Scale, mRS). The mRS ranges between 0 and 6, where 0 represents normal function without any residual deficit following a stroke and 6 indicates death. The mRS was dichotomized to 0–1 and 2–6 for good and bad outcomes, respectively.
The statistical analyses were performed using SPSS V11.0.1 (2001, SPSS Inc.), JMP V3 (SAS Institute, Cary, NC, USA, 1995), and Analyse-it (by Analyse-It Software Inc., England, UK, 2002).
Demographics, clinical, and imaging baseline data were presented in a table format, giving the mean with standard deviations (M
SD) and median with inter-quartile ranges. For categorical variables, frequencies, proportions, and percentages were provided. Student’s two sample t
-tests were used to compare the continuous normally distributed data with unequal variances. Fisher’s exact test and chi-square were used to compare the proportions among the binary variables. Level of significance was selected at the conventional and standard type I error of α
Associations between atheroma presence and stroke severity as estimated by NIHSS, or of atheroma presence and 3-month mRS were examined using multiple logistic regression to adjust for other known stroke risk factors, including: age, sex, race, hypertension (HTN), diabetes (DM), smoking, and coronary artery disease (CAD). To allow adjustment for other covariates, several models were built with up to five covariates in each model in accordance with parsimonies rule. Factors known to be associated with stroke severity at presentation were included to adjust for confounders or effect modifiers. Prior to incorporating these factors into the model, a univariate analysis was performed to asses the relation of the baseline variables and AA (independent variable) to initial Stroke Severity (response variable).
Variables were incorporated in the model if they were deemed clinically significant or had a p-value of <0.5 on the univariate analysis. We allowed for a large p-value because of the sample size and the nature of the exploratory analysis. Regression coefficients were estimated using the maximum likelihood. Goodness of fit was assessed using the likelihood ratio test with chi-square statistics between the full and reduced models. Co-linearity between covariates were assessed by scatter plots and comparing the regression models with and without the variable of interest.
Due to the small sample size, there was limited ability to adjust for other covariates simultaneously using the standard logistic regression, and propensity score (PS) analysis was used to confirm the findings. The PS technique was implemented to incorporate all baseline variables into one continuous score based on the probability of having AA. The PS is the probability between 0 and 1 for a subject to have atheroma present given all his baseline variables, and is estimated using standard logistic regression. The PS model was evaluated for its ability to discriminate between subjects who had atheromas and those who did not, using the area under the receiver operating characteristic (ROC) curve (C Statistic). After calculating the probability for every patient, the PS was incorporated in the regression model as a continuous covariate in addition to the AA to adjust for confounders (Joffe and Rosenbaum, 1999