Selection criteria

We included studies evaluating adults (aged >18 years) who had suffered a stroke (ischaemic or haemorrhagic) or TIA according to the WHO definition.23 Only RCTs assessing the effects of exercise or physical fitness interventions compared with usual care or sham were included. Exercise therapy included cardiorespiratory (aimed at improving fitness), resistance (aimed at improving muscle strength or endurance) or mixed components. Interventions termed ‘aerobic’ involved a clear aim to enhance physical fitness by stimulating heart rate and respiratory rate, for example, running, walking, circuits, cycling. Studies of physical therapy (therapies used to promote, maintain or restore physical function, eg, walking, balance, etc), were only included if they involved cardiorespiratory or resistance components. Only studies reporting on systolic blood pressure (SBP), diastolic blood pressure (DBP), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), fasting glucose (FG), BMI or secondary cerebrovascular events and cardiovascular death were included.

Search strategy

Electronic databases search included OVID MEDLINE, PubMed, Web of Science, The Cochrane Library, The National Institute for Health and Care Excellence, TRIP Database, CINAHL, PsycINFO, SCOPUS, UK Clinical Trials Gateway and the China National Knowledge Infrastructure, from 1966 to October 2017. Selected medical subject headings were combined with free text terms relating to stroke (eg, cerebrovascular accident, cerebral infarction), TIA (eg, transient ischaemic attack, mini-stroke), exercise (eg, physical activity, aerobic exercise) and secondary prevention (eg, outcomes, vascular events, death, BP, lipid profiles, glucose, weight or BMI), to create a search strategy finalised for MEDLINE (see online supplementary material). This was adjusted for use in other databases using appropriate Boolean operators and search symbols. Chinese translations were used for Chinese databases (see online supplementary material). Reference lists of all relevant studies and systematic reviews were scanned for additional potentially relevant studies. Our search did not extend to grey literature.

Study selection and data extraction

Two review authors (CW and ANA) independently eliminated duplicate and irrelevant studies from title and abstract. Full texts of the remaining studies were obtained and critically reviewed by both authors independently for inclusion. Both reviewers independently extracted data into a prepiloted spreadsheet. Data included participant demographics, study setting and timeframe, intervention arms and their descriptors and outcome measures. The Consensus on Exercise Reporting Template (CERT)24 was used to characterise the reporting of exercise specifics (eg, equipment required, group vs individual, level of supervision, etc) (see online supplementary material). Data required for meta-analysis were collected, including mean differences and associated SD between baseline and postintervention outcome assessments. Basic unit conversions for lipid and glycaemic profiles were undertaken to ensure international system of units for analysis (see online supplementary material). Discrepancies in study selection or data extraction were discussed by both reviewers and adjudicated by a third reviewer (MS).

Quality appraisal

All studies meeting inclusion criteria for the review were assessed independently by both reviewers (CW and ANA) for risk of bias using a checklist developed by the Cochrane Collaboration establishing bias attributable to patient selection, randomisation, blinding, attrition and reporting.25 Studies at high risk of bias in one or more domains were classed as high risk overall, while those with unclear information in any domain were classed as moderate risk. Low-risk studies met low-risk criteria for each domain. Quality assessment of evidence for each outcome measure that pooled data (SBP, DBP, TC, LDL-C, HDL-C, FG and BMI) was also undertaken using the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) approach.26 An initial grading of ‘high quality’ for RCTs was downgraded by one level for serious concerns (or by two levels for very serious concerns) about the risk of bias, inconsistency, indirectness, imprecision or publication bias relating to combined evidence for each outcome (see online supplementary material).

Data analysis

A qualitative synthesis of study characteristics was undertaken. Statistical analyses of outcome data were performed using RevMan 5 (V.5.3 from the Cochrane Collaboration). Meta-analyses for continuous variables (SBP, DBP, TC, LDL-C, HDL-C, FG and BMI) used the mean difference and the SD of the mean difference, between intervention and control groups. Not all trials reported changes in the SDs preintervention and postintervention. In these instances, trial authors were contacted directly to request missing data, and where this was not possible, we used indirect methods to estimate missing values.27 We calculated the I²  statistic to assess statistical heterogeneity.28 A random effects model was applied due to anticipated heterogeneity in population (stroke severity, TIA) and exercise interventions. For meta-analysis of outcomes reported by 10 trials or more, we assessed publication bias using tests by Egger et al 29 and Begg and Mazumdar30 in STATA statistical software (V.14.0, StataCorp), although we acknowledge that funnel plot asymmetry can be due to reasons other than publication bias. Prespecified subgroup analyses were undertaken to estimate the effects of variables such as inclusion of studies at high risk of bias, the effect of early (within 6 months) versus late (after 6 months) exercise initiation, incorporation of health education and participant type (TIA/non-disabling stroke vs severe stroke). Further details on the complete statistical tests used are available in the online supplementary material.

Quality appraisal

Ten studies were at low risk of bias,33 34 37 39 43–46 48 50 seven were at moderate risk of bias31 and three were at high risk of bias35 38 42 (table 1). GRADE assessment of the quality of evidence for each meta-analysed outcome ranged from moderate to very low (see online supplementary material).

Study characteristics

Twenty selected RCTs recruited 1031 patients from 10 countries, details of which are summarised in table 2. Average ages of study participants ranged from 52 to 69 years. Sixteen trials involved only patients who had suffered stroke,31–36 38–40 42 44–46 48–50 two included only patients with TIA37 47 and two included both stroke and TIA.41 43 Eleven trials recruited patients to interventions within 6 months32 36–39 41–45 47 and nine trials after 6 months31 33–35 40 46 48–50 of the index event. The majority of studies recruited only ambulant participants (with or without walking aids), while five studies recruited patients with more severe stroke.31–33 44 45 Five studies excluded patients on the basis of cognitive impairment.35 38 40 43 48 Exercise interventions were compared with usual care in 12 studies,32 34 36–38 40–42 47–49 therapeutic activities in 4 studies (eg, standing and balance training),35 44–46 stretching in 3 studies31 33 50 and an education programme in 1 study.39 Exercise interventions varied considerably as reported according to the CERT (see online supplementary material). Overall programme duration ranged from 6 weeks to 6 months and weekly frequency of sessions from 1 to 5, with each session ranging in duration from 30 to 90 min. Most exercise interventions involved 3×1 hourly sessions per week (mean weekly minutes=175, median=180), but minutes of weekly programmed exercise activity ranged from 60 to 350 min. Eleven studies involved graded increases in exercise duration or intensity31 32 35 37 39 40 46–50; only three did not include target heart rate changes.39 41 49 Four interventions included health educational,34 37 38 43 and only two involved any home programme content.37 39 Studies reported poorly on qualifications of supervising health professionals, intervention fidelity and whether exercise was delivered individually or in groups. All studies reported outcome measures relevant to secondary vascular risk, most commonly blood pressure (13 studies), with lipid profile, fasting glucose and BMI reported in at least eight studies. One study reported on vascular events rates.37 Two studies referred to the same intervention and group of patients, but reported on different outcome measures,44 45 thus were viewed as one study for statistical purposes.

Blood pressure

Data on SBP and DBP were pooled from 12 studies including 606 patients (intervention n=305).31 32 34–37 39 40 43 47 48 50 Blood pressure data from one study40 did not include measures of variability and was excluded from meta-analysis. There were significantly greater reductions in SBP with exercise interventions compared with controls (−4.30 mm Hg, 95% CI −6.77 to −1.83 mm Hg, I²=33%; GRADE of evidence: moderate level) (figure 2). Both tests by Begg and  Mazumdar (Pr>|z|=0.373) and Egger et al (P>|t|=0.226) suggested low risk for publication bias.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Forest plots of the effects of exercise vs control interventions on (A) systolic blood pressure, including effects of (B) early (<6 months) vs late initiation (>6 months), (C) inclusion of an educational component and (D) population condition (stroke vs transient ischaemic attack).

There were significantly greater reductions in DBP with exercise interventions compared with controls (−3.12 mm Hg, 95% CI −4.89 to −1.34 mm Hg; GRADE of evidence: low level) (figure 3). There was significant heterogeneity (I²=68%) and evidence of publication bias (test by Begg and Mazumdar Pr>|z|=0.047, test by Egger et al P>|t|=0.033, 95% CI −8.87 to −0.47).

Exercise interventions initiated within 6 months32 36 37 39 43 47 of index stroke or TIA appeared to have a larger effect on SBP (−8.46 mm Hg, 95% CI −12.18 to −4.75, I²=0%) than those initiated after 6 months (−2.33 mm Hg, 95% CI −3.94 to −0.72, I²=0%).31 34 35 40 48 50 The benefits for DBP became non-significant with exercise initiated after 6 months (−2.08 mm Hg, 95% CI −4.82 to 0.66, I²=70%). Exercise interventions involving an education component34 37 39 43 also appeared to have greater effects on SBP (−7.81 mm Hg, 95% CI −14.34 to −1.28, I²=40%) compared with those with no education component (−2.78 mm Hg, 95% CI −4.33 to −1.23, I²=0%).31 32 35 36 47 48 50 Exclusion of one study at high risk of bias35 ³⁵ did not significantly effect overall results (see online supplementary material).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Forest plots of the effects of exercise vs control interventions on (A) diastolic blood pressure, including effects of (B) early (<6 months) vs late initiation (>6 months), (C) inclusion of an educational component, and (D) population condition (stroke vs transient ischaemic attack).

Lipid profiles

Nine RCTs reported on total cholesterol (n=370 patients; intervention n = 185)33–35 42 43 45 46 49 50 (figure 4A). Pooled analysis showed significant overall reductions in total cholesterol with exercise compared with control (−0.27 mmol/L, 95% CI –0.54 to 0.00 mmol/L, I² =70%, p=0.05; GRADE of evidence: low). This beneficial effect however disappeared with the exclusion of two studies that were at high risk of bias35 42 (−0.13 mmol/L, 95% CI −0.40 to 0.14, online supplementary material).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Forest plots of the effects of exercise vs control interventions on (A) total cholesterol, (B) low-density lipoprotein cholesterol, (C) high-density lipoprotein cholesterol, (D) fasting glucose and (E) body mass index.

Levels of LDL-C have stronger correlations with risk of future stroke and are the main target of lipid-lowering interventions.51 Seven RCTs reported on LDL-C (n=303 patients; intervention n=151)35 39 42 44 46 49 50 (figure 4B). There was no effect of exercise compared with control (−0.28 mmol/L, 95% CI −0.63 to 0.07 mmol/L, I²=85%; GRADE of evidence: very low), except among interventions involving health education.

Higher levels of HDL-C have been associated with reduced risk of stroke among diverse populations.52 Nine RCTs reported on HDL-C (n=394 patients; intervention n=197)33 35 39 42 43 45 46 49 50 (figure 4C). There was no effect of exercise compared with control (+0.08 mmol/L, 95% CI −0.02 to 0.17 mmol/L, I²=77%; GRADE of evidence: very low).

Fasting blood glucose

Seven RCTs measured fasting glucose changes (n=364 patients; intervention n=135).33 37 38 43 44 46 49 There was no effect of exercise on fasting glucose compared with control (−0.14 mmol/L, 95% CI −0.29 to 0.01, moderate heterogeneity I²=42%; GRADE of evidence: moderate; figure 4D). Subgroup analyses did not alter these findings.

Body mass index

EIght RCTs (n=446 patients; intervention n=225) measured BMI at baseline and follow-up.34 35 37 39 43 49 50 There was no effect of exercise on BMI compared with control (−0.0 kg/m², 95% CI –0.26 to 0.25 kg/m², I²=32%; GRADE of evidence: moderate; figure 4E), nor any effects from subgroups.

Secondary vascular events

One study randomised 70 patients with non-cardioembolic stroke to a 24-week programme of exercise including salt reduction and dietary education or to usual care with health education.39 It was terminated early after results showed significantly fewer hospitalisations or deaths secondary to stroke, myocardial infarction or peripheral arterial disease in the intervention group (adjusted HR 0.194, 95% CI 0.121 to 0.737, p=0.005). The intervention in this trial also resulted in greater reductions in SBP, DBP and increased levels of HDL-C.

Adverse events

Three patients from the 20 RCTs suffered adverse events. All adverse events were unrelated to the exercise interventions (see online supplementary material). One trial41 reported benign ECG abnormalities on exercise testing prior to exercise interventions, and one patient went on to have percutaneous coronary intervention for underlying CHD.

Effect of exercise on vascular risk factors

Exercise interventions can lead to clinically meaningful reductions in SBP and DBP after stroke or TIA. This is important because SBP and DBP may be equally important predictors of secondary stroke risk.53 While exercise causes immediate elevations in heart rate, cardiac output and peripheral vasodilatation through nitric oxide release,54 the long-lasting effects that may mediate blood pressure reduction include enhanced endothelial sensitivity to nitric oxide, a reduction in sympathetic tone and salt load, all ultimately reducing systemic vascular resistance.55

Antihypertensive therapy reduces the risk of secondary stroke by 24% and myocardial infarction by 21% when compared with placebo.56 A recent meta-analysis of 10 RCTs evaluating antihypertensive therapy after stroke showed that these effects occurred with reductions in SBP of 5.1 mm Hg and DBP of 2.5 mm Hg across a range of antihypertensive classes.57 Thus, the blood pressure reductions associated with exercise interventions in our meta-analysis (SBP 4.3 mm Hg, DBP 2.5 mm Hg) compare favourably with pharmacological approaches. The effects of exercise may be additive to pharmacological intervention. For example, a trial evaluating the effects of an exercise programme in patients with drug-resistant hypertension (patients already taking three antihypertensives) revealed mean SBP and DBP reductions of 6 (±12) mm Hg and 3 (±7) mm Hg, respectively in the exercise group compared with controls (p=0.03).58

Exercise is associated with increased calorific expenditure, peripheral insulin sensitivity59 and enhanced muscle metabolism of cholesterol.60 Our analysis suggested that exercise may have beneficial effects on lipid profiles in patients after stroke, although heterogeneity in these data means the results should be interpreted cautiously.

Exercise programmes initiated within 6 months after stroke or TIA had greater blood pressure reductions compared with programmes initiated after 6 months. This may be because individuals are more likely to engage with behaviour modification soon after illness compared with later, often referred to as the ‘sentinel event effect’ as is evidenced for smoking cessation interventions following myocardial infarction.61 Early intervention may be of increased benefit after stroke and TIA as the risks of recurrence are highest within the first few months.62 However, significant reductions in SBP were still observed when interventions were started later. Therefore, promoting exercise, even in the chronic phase after stroke, may still lead to clinically important benefits. The inclusion of a health education component alongside exercise interventions also appeared beneficial. Indeed, meta-analyses of education interventions targeting diet result in clinically meaningful in blood pressure and lipids,63 64 and may be an important part of a holistic secondary stroke prevention programme.

Secondary vascular events

One study reported on vascular outcomes, reporting that exercise participants were approximately 80% less likely to suffer a vascular death or hospitalisation over a median follow-up of 2.9 years. The early initiation (1 month postevent), long intervention duration (24 weeks), inclusion of salt reduction education and the use of activity monitors (which may themselves improve overall levels of physical activity65) may have contributed to the observed effectiveness and the large reductions in SBP (13.0 mm Hg) and DBP (6.3 mm Hg). In addition, by including patients with non-cardioembolic ischaemic stroke only, the investigators may have selected a group of patients standing to benefit most from classical vascular risk reduction. Nevertheless, this was a small, single-centre study and further research to establish the link between exercise and reduced secondary vascular events remains a priority.

Future research

Another priority area for research relates to enhancing the uptake of exercise as an intervention. In the UK, despite the strong evidence base for cardiovascular mortality reduction with exercise-based cardiac rehabilitation, and its provision freely within the National Health Service, fewer than half of eligible patients take up the intervention.66 67 This may be due to a number of factors such as multimorbidity, problems with transport and poor social support.68 To extend the potential benefits of exercise, future research should evaluate facilitators and barriers to participation in exercise interventions among patients with stroke, and the role of emerging assistive technologies in enhancing intervention delivery, for example, smartphones, fit-bits, tablet PCs. Future trials should also include full descriptions of intervention characteristics (ie, following the CERT) so clinicians know which types and intensities of exercise should be prescribed in clinical practice.

Safety of exercise

Patients and clinicians may have concerns about engaging in physical activity soon after stroke or TIA. Our results suggest that exercise interventions are safe to undertake in patients with stroke or TIA, even in the very early phases. A Cochrane review of 58 trials of physical fitness training for functional recovery after stroke also reported excellent safety data with no reports of severe adverse events.18

Limitations

Owing to the significant variation in the type of exercise intervention, it is difficult to ascertain the ideal design for effectiveness. The relatively small number of studies and participants precluded investigating the effects of specific components of exercise interventions (eg, single vs group sessions). However, exercise interventions initiated early rather than late, and those combined with health education appeared more effective for reducing SBP. It is still uncertain if the effects of the interventions relate to behaviour change outside of the exercise sessions. Only one study used activity monitors to link exercise with increased overall step counts.39

Since vascular risk factors were measured at the end of the study interventions, we do not know if the beneficial effects of exercise extend beyond the intervention duration. Studies of exercise in other patient cohorts have suggested a decline in beneficial effects soon after intervention termination,69 although data from cardiac rehabilitation studies demonstrate reductions in cardiovascular mortality and hospitalisation that persist at 12 months follow-up.70 71 The majority of participants in included trials were relatively young (average age 69 years) with mildly disabling stroke or TIA who were ambulant. This may limit the generalisability of our findings to older, more disabled patients with stroke. Our analysis suggested only modest SBP reductions among exercise studies that included more disabled patients with stroke, with no changes in DBP. These patients, who often experience greater fatigue, functional and cognitive impairment, may find it more difficult to comply with the frequency, duration and target heart rates of prescribed exercise that lead to beneficial effects.

The quality of evidence (GRADE) for the outcomes ranged from moderate to very low, primarily due to inconsistency and imprecision of results. All of the included studies were small pilot trials, none powered independently to detect significant differences in any of the vascular risk factors measured. These small studies may also be subject to inherent exaggeration of effect sizes due to variations in methodological quality and heterogeneity of patient cohorts. Although we performed funnel plots for publication bias, we did not search grey literature nor included manuscripts outside of the English and Chinese language. Therefore, there is a risk of publication and language bias.

Our results suggest that clinicians managing patients with stroke or TIA should recommend aerobic exercise (eg, brisk walking, cycling, stepping), aiming for target heart rate reserves of between 50% and 70%, for around 150–180 min per week (divided over 2–3 sessions). Exercise sessions should be prioritised to start early (ie, within 6 months) after stroke or TIA. One way to deliver the suggested aerobic exercise interventions in combination with health education, may be to refer to existing local phase III/IV cardiac rehabilitation services, some of which across the UK are already accepting stroke referrals.43 However, in order for this to become an established service, high-quality RCTs, recruiting larger, less selected groups of stroke survivors, and incorporating longer follow-up, are desperately needed to investigate the effects of exercise on cardiovascular events and mortality.