We used data from the Ischemic Stroke Genetics Study (ISGS). ISGS is a multicentre inception cohort study of subjects with first-ever ischaemic stroke. The protocol has been described previously.21
All participants provided written informed consent to participate in ISGS, and data collection was approved by each site’s institutional review board. Participating centres were instructed to screen patients consecutively.
All available medical records pertaining to the stroke evaluation were compiled in standardised fashion, stripped of personal health identifiers and centrally reviewed by a single neurologist for the purpose of assessing stroke subtype diagnoses using multiple standardised criteria. The medical records reviewer completed a stroke work-up checklist, focusing on brain imaging, structural and electrical cardiac evaluation, and cervical and intracranial vascular imaging. Physical activity was assessed using a standardised questionnaire with three categories to determine the patient’s activity over the preceding year. Physical activity was defined by actions that produced sweating or a noticeable increase in heart rate analogous to what has been done in the National Health and Nutrition Examination Survey.22
Leisure-time physical activity levels were divided into low (vigorous activity sufficient to break a sweat or noticeably raise heart rate less than once a week), moderate (vigorous activity sufficient to break a sweat or noticeably raise heart rate one to three times a week) or high (vigorous activity sufficient to break a sweat or noticeably raise heart rate four or more times a week). Infarct size was determined on the basis of magnetic resonance (MR) imaging scan when available and the computed tomographic scan otherwise. At enrolment (date consent signed) and approximately 3 months thereafter, functional status was assessed using the Oxford Handicap Scale (OHS), 23
Barthel Index (BI) 24
and Glasgow Outcome Scale (GOS).25
Neurological impairment at enrolment was assessed using the National Institutes of Health Stroke Scale (NIHSS).26
Baseline assessments were done face to face, and 3-month assessments were done by telephone interview.
Patient and stroke characteristics are summarised using frequencies and percentages, and χ2
tests were used to assess differences in these variables between patients in the three physical activity levels. Stroke severity was quantified by the NIHSS, BI, OHS, GOS and infarct size at baseline and by the BI, OHS and GOS at follow-up. We dichotomised each measure into “good” and “bad” groups according to the following cut-points. For BJ (possible range of values, 0–100), values of 95 and higher were considered good; for GOS (range 1–5), a value of 1 was considered good; for OHS (range 0–5), values of 0 and 1 were considered good; and for NIHSS (range 0–42), values between 0 and 5 were considered good. The selection of values of 0 to 5 as good outcomes on NIHSS was based on prior studies in which these values predicted discharge disposition after hospitalisation.27
These same cut-points were used at follow-up, except for OHS, where a value of 0 was considered good. Infarct size was determined on the basis of maximal cross-sectional diameter of the zone of restricted diffusion as seen on diffusion-weighted imaging (or zone of hyperintensity seen on fluid-attenuated inversion recovery, if late presentation caused resolution of the area of restricted diffusion) for patients who had MR imaging. For patients who did not have MR imaging, infarct size was determined by the maximal cross-sectional diameter of the zone of hypointensity representing acute infarction. Infarcts smaller than 1.5 cm were considered small, and those 1.5 cm and larger were considered large. The dichotomisations were selected to summarise findings in clinically interpretable ways and were selected before statistical analysis. A χ2
test was used to assess the association between physical activity level and the number of bad outcomes (of NIHSS, GOS, OHS and BI). Logistic regression was then used to assess the association between physical activity level and functional status, adjusted for patient and stroke characteristics. A hierarchical approach was used where we included progressively more covariates in subsequent models. Initially, we adjusted for age, body mass index, site, race, sex, primary and secondary smoking exposure; alcohol use; sibling history of stroke; history of diabetes, transient ischaemic attack (TIA), congestive heart failure, myocardial infarction, atrial fibrillation, and hypertension; prior aspirin and anticoagulant use; and glucose levels. Sibling history of stroke was included because it had previously been shown to influence stroke severity in this population.26
These variables were included in the model as categorical variables, as shown in supplementary table 1
. The next model also included the prestroke OHS score. We believed that the prestroke OHS score would be highly correlated with the physical activity level, and which came first would not be clear. Stroke characteristics were then added to the previous model. These included the Trial of Org 10172 in Acute Stroke Treatment (TOAST) and Oxfordshire subtypes and thrombolysis within 24 h after the stroke. A backward-stepping algorithm was then used on the last model to remove variables from the model that were not significant. Results for physical activity are reported as odds ratios (ORs) and 95%, CIs. In all models, the low physical activity group was used as the reference cell. SAS Version 9.1 for Windows was used for all analyses (SAS Institute, Cary, North Carolina).