Associations Between Time After Stroke and Exercise Training Outcomes: A Meta-Regression Analysis.

Background Knowledge gaps exist regarding the effect of time elapsed after stroke on the effectiveness of exercise training interventions, offering incomplete guidance to clinicians. Methods and Results To determine the associations between time after stroke and 6-minute walk distance, 10-meter walk time, cardiorespiratory fitness and balance (Berg Balance Scale score [BBS]) in exercise training interventions, relevant studies in post-stroke populations were identified by systematic review. Time after stroke as continuous or dichotomized (≤3 months versus >3 months, and ≤6 months versus >6 months) variables and weighted mean differences in postintervention outcomes were examined in meta-regression analyses adjusted for study baseline mean values (pre-post comparisons) or baseline mean values and baseline control-intervention differences (controlled comparisons). Secondary models were adjusted additionally for mean age, sex, and aerobic exercise intensity, dose, and modality. We included 148 studies. Earlier exercise training initiation was associated with larger pre-post differences in mobility; studies initiated ≤3 months versus >3 months after stroke were associated with larger differences (weighted mean differences [95% confidence interval]) in 6-minute walk distance (36.3 meters; 95% CI, 14.2-58.5), comfortable 10-meter walk time (0.13 m/s; 95% CI, 0.06-0.19) and fast 10-meter walk time (0.16 m/s; 95% CI, 0.03-0.3), in fully adjusted models. Initiation ≤3 months versus >3 months was not associated with cardiorespiratory fitness but was associated with a higher but not clinically important Berg Balance Scale score difference (2.9 points; 95% CI, 0.41-5.5). In exercise training versus control studies, initiation ≤3 months was associated with a greater difference in only postintervention 6-minute walk distance (baseline-adjusted 27.3 meters; 95% CI, 6.1-48.5; fully adjusted, 24.9 meters; 95% CI, 0.82-49.1; a similar association was seen for ≤6 months versus >6 months after stroke (fully adjusted, 26.6 meters; 95% CI, 2.6-50.6). Conclusions There may be a clinically meaningful benefit to mobility outcomes when exercise is initiated within 3 months and up to 6 months after stroke.

cardiorespiratory fitness

exercise training

6‐minute walk distance

10 meter walk time

Clinical Perspective

What Is New?

Given that early initiation of exercise after stroke is often advocated, and there is little clinical evidence to support this, we conducted the first meta‐regression analysis with the primary objective of examining the association between time elapsed after stroke to initiation of exercise training and clinical outcomes.

In fully adjusted randomized studies, there was a clinically important benefit to 6‐minute walk distance when starting exercise training within 3 months, with a similar weighted mean difference when starting within 6 months of stroke compared with later, with no significant time effect on cardiorespiratory fitness, balance, or 10‐meter walking speed.

What Are the Clinical Implications?

The time window for improved outcome in 6‐minute walk distance related to exercise training may span longer time periods than previously thought, and may fall within distinct post‐stroke phases, with no time association for other outcomes; yet the number of adverse events in studies that were started within the first month after stroke was concerning, suggesting careful application of exercise training in the early phases.

Stroke is the leading cause of adult neurological disability, and the aging population and accumulating risk factors lead some countries to project marked increases in stroke prevalence. , At least one‐third of those who suffer a stroke will be left with functional impairment and disability. Therefore, it is not surprising that following a stroke, physical activity falls well below recommended levels within the first 2 weeks after stroke and persists into the chronic phases of stroke >6 months later. This pattern of inactivity leads to cardiorespiratory deconditioning that is half of age‐ and sex‐predicted normative values for sedentary adults, falling below the necessary criterion for independent living. , This deconditioning can compound the effects of stroke impairments affecting independence in carrying out activities of daily living , and is also associated with increased risk of morbidity, mortality, and stroke hospitalizations. , , , , A recent Cochrane review of randomized controlled studies that included aerobic and circuit training interventions (published up to 2018) revealed that exercise training (ET) not only results in improved cardiorespiratory fitness (CRF) but also yields gains in other important domains of stroke recovery, including functional mobility measured by 6‐minute walk distance (6MWD) and fast and comfortable short‐distance gait speed and balance. Improving walking capacity (endurance and independence) is one of the most frequently stated goals of people following stroke, and poor balance is associated with a greater risk of falls, which can lead to hip fracture and other injuries. Therefore, determining strategies to optimize CRF is of great importance from both a functional and quality‐of‐life perspective. ,

While guidelines endorse physical activity and exercise across all phases of stroke recovery, the optimal time between stroke and initiation of ET to support improvements in CRF, functional mobility, gait speed, and balance has not been well established. The Stroke Roundtable Consortium advocated to focus recovery trials on the first week to the first month after stroke (acute and early subacute phases). The rationale for early interventions is that rapid changes and most behavioral recovery is reported to occur within this time frame, which is a critical time for neural plasticity and brain repair processes, and patients are most responsive to treatment. Specifically, evidence from preclinical studies indicates that key molecular, genetic, and cellular changes occur in this window, triggering elevated dendritic sprouting, changes in gene expression, and the suppression of neuronal apoptosis. , However, there is some evidence that time‐dependent recovery may fall within distinct post‐stroke phases. For example, some studies report that most patients reach their peak walking function between 2 and 3 months after stroke. , , Other studies report no further improvements after 6 months, and others have estimated it to extend beyond a year following the stroke event. , Yet there is little clinical evidence to show that starting an exercise intervention earlier yields an advantage.

There is a dearth of controlled studies introducing ET at different initiation points with direct comparisons that would inform best‐practice guidelines on timely initiation. However, there have been numerous observational and controlled studies that initiated ET, each at different points in the recovery period. These studies can be combined quantitatively through meta‐regression analyses. This study used meta‐regression analyses to determine whether time elapsed from stroke to the start of ET was associated with the pre‐ versus postintervention outcomes in CRF, balance, 6MWD, and 10‐meter comfortable and fast walk time (10MWT), and with greater differences in those outcomes between ET versus control groups. Time since stroke was examined as a continuous variable with log‐transformation to estimate the general trend and dichotomized to consider whether the magnitudes of those differences at 3‐ or 6‐month thresholds might be clinically meaningful. Examining the data in distinct phases may provide a more clinically useful measure to guide healthcare professionals as to when to initiate ET throughout the continuum of care. Specifically, while the transitions in care after a stroke are variable, patients are in acute care/inpatient and outpatient rehabilitation for up to 3 months. , This also coincides with the timing (3 months) of when neurobiological protective mechanisms have recovered sufficiently to allow for higher‐intensity ET at or above the anaerobic threshold. After the 3‐month period, some patients will be referred to cardiac rehabilitation for a further 3 months of treatment (up to 6 months after stroke), and others are discharged into the community.

Methods

This meta‐analysis was conducted according to our predefined protocol, and the reporting of findings followed the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines. The Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines compliance check list is presented in Table S1. All data and supporting materials have been provided with the published article.

Study Eligibility Criteria

Studies were included that were (1) original research articles studying patients following stroke, (2) consisting of at least 1 study group receiving an exercise intervention with an aerobic component but without external stimuli or robotic assistance (For the purpose of this study, aerobic training was defined as planned, structured, and repetitive exercise [excluding incidental exercise that occurs during physical therapy] that is progressed in duration or intensity or both. Examples of aerobic training include walking, stationary cycling [arm or leg], stepping machine, and treadmill exercise. Examples of activities that are not considered aerobic training include sensorimotor or task‐related training for the purpose of improving function [excluding therapeutic activities that would not induce an appropriate aerobic stimulus]; (3) reporting time since stroke or defining an interval of time since stroke in their subject inclusion; and (4) measuring the outcomes of interest. Articles that performed secondary analyses from other studies or reused data from previous studies were excluded.

Literature Search

Seven electronic databases were searched from inception to June 30, 2020: Medline (Ovid), Embase (Ovid), APA PsycINFO (Ovid), PubMed (non‐Medline), Cochrane Controlled Trials Register, Cochrane Database of Systematic Reviews, and CINAHL (EbscoHost).

The search strategies were developed in collaboration with an information specialist using a modified population intervention comparison outcome (PICO) framework. The Population comprised stroke (any type); the Intervention was aerobic exercise; and the Outcomes included varied functional mobility measures. These results were limited to the specific study types and humans. No date or language restrictions were applied. The reference lists of included studies were also checked for relevant materials not identified through database searching. The Medline search strategy is shown in Table S2.

Methodological Quality Assessment and Risk of Bias

Risk of bias was evaluated on the basis of criteria adapted from the Newcastle Ottawa Scale and the Cochrane Collaboration’s Risk of Bias Assessment Tool. , Each paper was assessed by 2 independent raters, and disagreement was resolved by consensus or by a third rater.

Data Extraction and Characteristics of the Exercise Intervention

Means and SDs of preintervention and postintervention outcomes were extracted. Means and SDs were estimated when descriptive statistics were reported in other formats. The mean time after stroke was extracted or estimated as the main independent variable of interest. Other relevant study characteristics, including study group age, sex, adverse event proportion, stroke severity/motor recovery level, proportion of intervention completers, and data spread (eg, SD or quantiles) of post‐stroke time were also extracted.

Characteristics of the intervention were also extracted. Exercise modality was stratified into walking/ambulatory or non–weight bearing/seated, as walking is more likely to improve walking speed and endurance than non–weight bearing modalities because of task specificity. Exercise dose was calculated by the number of training sessions per week×minutes per session×total weeks. When a range was given, the higher value was used. Dose was stratified as 1000 or less versus more than 1000 “units” as previously described. Intensity was stratified into moderate (40%–59% heart rate reserve or VO2R (oxygen uptake reserve) or 46%–63% of VO2max (maximal oxygen uptake), or 64%–76% of HRmax (maximal heart rate) or rating of perceived exertion of 12–13/20) or at least vigorous (greater than or equal to the following: 60%–89% heart rate reserve or VO2R or 64%–90% of maximal oxygen uptake (VO2max) and 77% to 95% of maximal heart rate (HRmax) or rating of perceived exertion of 14–17/20).

Statistical Analysis

To investigate the relationship between exercise outcomes and time elapsed between stroke and intervention, meta‐regression analyses were conducted. Outcome estimates from each study included in the meta‐regression were obtained as weighted mean differences and 95% CIs using random‐effects models with Knapp‐Hartung adjustment. We chose a priori a random‐effects model because of methodological differences between studies that were expected to contribute to different underlying true effects, and we used a restricted maximum likelihood estimator to minimize the influence of nuisance parameters. A set of analyses compared postintervention outcomes with preintervention outcomes, adjusted for the preintervention outcome measure, since it may influence post‐stroke improvement. A second set of analyses compared postintervention outcomes between intervention and control groups (reference group), adjusting for the preintervention mean difference between groups and preintervention performance in the intervention group. Time after stroke was modeled as a logarithmically transformed continuous variable. To provide estimates for specific time frames, analyses were conducted dichotomizing time after stroke into binary variables, using a 3‐month or 6‐month cutoff. The reference levels in each analysis were >3 months and >6 months, respectively. Where the number of included studies permitted, additional models were further adjusted for age, female proportion, exercise intensity (binary), exercise dose (binary), and whether exercise was ambulatory (binary). Unstandardized meta‐regression coefficients (B) and their 95% CIs were obtained using the metafor package in R 3.5.1. Bubble plots were depicted using the ggplot2 package. Risk of publication bias was assessed using Begg’s rank correlation test.

Results

Overall, 148 studies and 5987 patients with stroke were included in this meta‐regression analysis. , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , The flow diagram, study characteristics, and risk of bias assessment table are presented in Figure S1 and Tables S3 and S4. Of 148 studies, 86 studies had an appropriate control group and were included in the analyses comparing postintervention outcomes between intervention and control groups. Ambulatory exercise as an intervention was prescribed in 118 studies. In addition, 53 studies reported vigorous intensity or greater was prescribed, and 73 studies had an exercise dose >1000 units. Only 70 studies reported on adverse events, and 96 reported stroke severity/motor recovery level using a diversity of scales.

Time After Stroke and Differences Between Intervention Versus Control

When time to start ET was a continuous variable (Table S5), there were no significant associations with greater benefit of the intervention versus control over time in 6MWD, 10MWT (comfortable or fast), Berg Balance Scale score, or peak oxygen uptake in baseline‐adjusted or in 6MWD fully adjusted analyses (Figures 1A, 2A and 2C, 3A and 3C, and Table S6).

Figure 1

Meta‐regression of 6‐minute walk distance (meters) by time after stroke of controlled comparisons.

A, Time as a continuous variable (in log scale±95% CI). B, ≤3 months vs >3 months after stroke.

Figure 2

Meta‐regression of 10‐meter walk time (m/s) by time after stroke of controlled comparisons (A and B = 10‐meter fast walk speed and C and D = 10‐meter comfortable walk speed (m/s)).

A and C, Time as a continuous variable (in log scale±95% CIs). B and D, ≤3 months vs >3 months after stroke.

Figure 3

Meta‐regression of balance and cardiorespiratory fitness outcomes by time after stroke of controlled comparisons. (A and B = Berg Balance Scale, and C and D = Cardiorespiratory Fitness, mL·kg−1∙min−1).

A and C, Time as a continuous variable (in log scale±95% CI). B and D, ≤3 months vs >3 months after stroke.

ET initiated within 3 months versus >3 months after stroke showed a greater difference in postintervention 6MWD between ET and controls (baseline‐adjusted B=27.289 meters; 95% CI, 6.065–48.513; t=2.59; P=0.013; fully adjusted B=24.942 meters; 95% CI, 0.820–49.064; t=2.10; P=0.043) (Tables 1 and 2, Figure 1B). No other significant associations in other outcomes were observed (Figure 2B and 2D and Figure 3B and 3D).

Table 1

Summary of Meta‐Regressions Between Time After Stroke ≤3 vs >3 Months and Change in Outcome Measures (Pre‐Post and Intervention vs Control)*

Outcome						Begg’s rank test †
Weighted mean difference	Number of studies	Estimate [95% CI]	t‐value	DF	P value	tau	P value
Post‐ vs preintervention ‡
6‐minute walk distance, m	111	−34.456 [−50.835 to −18.077]	−4.17	108	<0.001	0.15	0.018
10‐meter walk test, comfortable speed, m/s	75	−0.102 [−0.153 to −0.051]	−3.96	72	<0.001	0.15	0.057
10‐meter walk test, fast speed, m/s	63	−0.171 [−0.264 to −0.079]	−3.71	60	<0.001	−0.01	0.953
V˙O_2peak, mL·kg−1∙min−1	57	−0.943 [−2.129 to 0.242]	−1.60	54	0.116	0.14	0.125
Berg Balance Scale score	47	−3.549 [−6.579 to −0.519]	−2.36	44	0.023	0.20	0.052
Intervention vs control §
6‐minute walk distance, m	48	−27.289 [−48.513 to −6.065]	−2.59	44	0.013	0.20	0.043
10‐meter walk test, comfortable speed, m/s	28	−0.062 [−0.185 to 0.062]	−1.03	24	0.312	0.10	0.465
10‐meter walk test, fast speed, m/s	23	−0.125 [−0.252 to 0.003]	−2.05	19	0.054	0.15	0.346
V˙O_2peak, mL·kg−1∙min−1 ||	27	0.052 [−1.629 to 1.732]	0.06	23	0.950	0.07	0.620
Berg Balance Scale score ¶	13	0.761 [−2.216 to 3.738]	0.58	9	0.577	0.33	0.129

DF indicates degrees of freedom; and eak oxygen uptake.

The reference group is ≤3 months.

Significance in Begg’s rank test indicates significant risk of publication bias.

Estimate was controlled for baseline value.

Estimate was controlled for baseline between‐group difference and baseline value in the intervention group.

There were only 6 studies in the group of ≤3 months.

There were only 4 studies in the group of ≤3 months.

Table 2

Summary of Meta‐Regressions Between Time After Stroke ≤3 vs >3 months and Change in Outcome Measures (Pre‐Post and Intervention vs Control) With Additional Covariates*

Outcome						Begg’s rank test †
Weighted mean difference	Number of studies	Estimate [95% CI]	t‐value	DF	P value	tau	P value
Post‐ vs preintervention ‡
6‐minute walk distance, m	103	−36.331 [−58.499 to −14.162]	−3.25	95	0.002	0.14	0.042
10‐meter walk test, comfortable speed, m/s	67	−0.128 [−0.193 to −0.063]	−3.92	59	<0.001	0.15	0.067
10‐meter walk test, fast speed, m/s	59	−0.163 [−0.299 to −0.026]	−2.40	51	0.02	0.00	0.958
V˙O_2peak, mL·kg−1∙min−1	51	−0.823 [−1.96 to 0.313]	−1.46	43	0.141	0.14	0.149
Berg Balance Scale score	40	−2.940 [−5.472 to −0.408]	−2.37	32	0.024	0.20	0.075
Intervention vs control §
6‐minute walk distance, m	44	−24.942 [−49.064 to −0.820]	−2.10	35	0.043	0.15	0.155

DF indicates degrees of freedom; and , peak oxygen uptake.

The reference group is ≤3 months

Significance in Begg’s rank test indicates significant risk of publication bias.

Estimate was controlled for baseline value, age, female proportion, exercise intensity (binary), exercise dose (binary), and ambulatory exercise (binary).

Estimate was controlled for baseline between‐group difference, baseline value, age, female proportion, exercise intensity (binary), exercise dose (binary), and ambulatory exercise (binary).

Considering a 6‐month post‐stroke time cutoff, a similar trend was seen for ET initiated ≤6 months versus >6 months after stroke for 6MWD (baseline‐adjusted B=21.89 meters; 95% CI, 1.660–42.119; t=2.18; P=0.035; fully adjusted B=26.608 meters; 95% CI, 2.644–50.572; t=2.25, P=0.031) (Tables S7 and S8). There were no significant associations in other outcomes (Table S8 and Figure S2).

Time After Stroke and Postintervention Versus Preintervention Differences

When time to start ET was a continuous variable (Table S5), with respect to preintervention performance, earlier post‐stroke ET intervention was associated with a greater difference in postintervention 6MWD (B=10.55 meters per log unit of time; 95% CI, 5.72–15.44; t=4.32, P<0.001; Figure S3A), 10MWT with a comfortable speed (B=0.04 m/s per log unit of time; 0.02–0.06; t=4.02; P<0.001; Figure S3G), fast‐speed 10MWT (B=0.036 m/s per log unit of time; 95% CI, 0.007–0.065; t=2.47; P=0.016; Figure S3D) and Berg Balance Scale score (B=0.896 units per log unit of time; 95% CI, 0.023–1.769; t=2.07; P=0.045; Figure S4A); however, an association was not observed with peak oxygen uptake (Figure S4D).

When intervention time post‐stroke was dichotomized at ≤3 months versus >3 months, studies initiated within 3 months after stroke showed an association favoring greater improvement in 6MWD (B=34.456 meters; 95% CI, 18.08–50.835; t=4.17; P<0.0001; Table 1 and Figure S3B) and 10MWT with a comfortable speed (B=0.102 m/s; 95% CI, 0.051–0.153; t=3.96; P<0.001; Figure S3H) with respect to baseline performance. A similar trend was observed in fast 10MWT (B=0.171 m/s; 95% CI, 0.079–0.264; t=3.71; P<0.001; Figure S3E) and Berg Balance Scale (B=3.549 score units; 95% CI, 0.519–6.579; t=2.36; P=0.023; Figure S4B). Associations between time and improvement were not observed in peak oxygen uptake (Figure S4E). Adjustment for additional covariates did not change the main results (Table 2).

Considering a 6‐month post‐stroke time cutoff, similar results were seen for all outcomes, except for comfortable 10MWT and Berg Balance Scale scores (Tables S7 and S8 and Figures S3 and S4).

Discussion

To our knowledge, this is the first study to be conducted with the primary objective of examining the associations between elapsed time to initiate ET after stroke and CRF, mobility, or balance using meta‐regression analyses. In randomized studies, there was a moderate and clinically important additional benefit to 6MWD observed when starting ET within 3 months, with a similar weighted mean difference when starting within 6 months of stroke compared with later. However, there was no significant time association for CRF, balance, or short‐distance walking speed when compared with control conditions. When time to initiate ET following stroke was treated as a continuous variable, there were no significant associations with any of the outcome measures. This suggests that time‐dependent recovery of functional mobility may fall within distinct post‐stroke phases. Nevertheless, the augmented outcome in 6MWD is of clinical importance, given that improving mobility and walking capacity represent the biggest unmet physical activity needs of people following stroke. ,

Subsequent meta‐regression analyses were conducted to examine the association of time to initiate ET on outcome measures in single group pre‐post studies. Results revealed that there was an augmented improvement associated with ET when initiated earlier for 6MWD, 10MWT, and balance but not CRF ≤3 months versus later and when time was expressed as a continuous variable. Extending the time threshold to 6 months, the weighted mean differences were less favorable than at ≤3 months except for comfortable 10WMT, which was similar. As in controlled studies, time had no association with CRF. Yet given the finding that when compared with a control condition there was no advantage of earlier training, except for 6MWD, the additional benefit of early ET for short‐distance walking speed and balance may be accounted for, at least in part, by spontaneous recovery and concomitant usual care rehabilitation in the pre‐post studies. However, regarding usual care rehabilitation, a recent Cochrane review of studies examining effects of aerobic and circuit training following stroke reported slightly higher effect sizes when ET was introduced after usual care than when initiated during usual care for change in CRF, balance, gait speed, and 6MWD. This suggests that spontaneous recovery may be a more influential driver of the earlier initiation advantage in all but 6MWD outcomes in pre‐post studies, requiring further investigation.

Meta‐Regression of Randomized Studies Demonstrated an Association Between Time and 6‐Minute Walk Distance Outcome

Six‐Minute Walk Distance

The augmented outcome in 6MWD translated into a weighted mean difference advantage of 24.9 meters (95% CI, 0.82–49.1 when starting ET within 3 months of a stroke compared with later (P=0.04) and a 26.6 meter (95% CI, 2.6–50.6 difference when starting ET within 6 months compared with later (P=0.03). The similar augmented outcome in 6MWD at 3 and 6 months suggests that the time window for enhanced recovery from an ET intervention can extend past 3 months when considering the potential effect on 6MWD. The magnitude of the augmented outcome represents a moderate difference given that the minimal clinically important difference has been estimated at 20 to 50 meters. , Similar results from a previous meta‐analysis conducted by Boyne et al, of 16 studies (published up to 2015) were reported, where there was a larger effect size for 6MWD when ET was started <6 months after stroke compared with ≥6 months of 25 meters (95% CI, −4 to 53). The results were not adjusted for covariates, while the current meta‐regression included more studies, and adjusted for 3 exercise parameters (intensity, dose, and modality), as well as age, sex, and the control intervention baseline mean differences.

Short‐Distance Walking Speeds

Meta‐regression analyses revealed a clinically meaningful advantage for fast 10MWT when ET was initiated ≤3 months compared with >3 months after stroke (0.125 m/s; 95% CI, −0.003 to 0.25; P=0.054) but not within 6 months compared with later (0.079 m/s; 95% CI, −0.024 to 0.182; P=0.13). While the 3‐month analysis was not statistically significant, the estimate was clinically meaningful given that the minimal clinically important difference for gait speed has been estimated at 0.1 m/s to 0.175 m/s. , , There was no association between time and 10MWT at comfortable speed. These results are similar to results from the meta‐analysis conducted by Boyne et al; despite combining fast and comfortable 10MWT speed data (n=13 studies). Specifically, there was a nonsignificant but borderline clinically important difference of 0.09 m/s (95% CI, −0.00 to 0.18) when ET was started <6 months versus ≥6 months after the stroke event. Collectively, these results indicate a weaker association between time to start ET and 10MWT than between time and 6MWD outcome. This may be related to previous reports of a stronger positive correlation between CRF and 6MWD than between CRF and 10MWT, but a lack of a time‐CRF association suggests a complex series of factors accounting for the association between time and mobility observed in this study, that requires further investigation.

The underlying mechanisms for these earlier improvements in function have not been fully elucidated. While some of the neurotrophic effects mentioned previously in people following stroke are thought to benefit cognition, brain‐derived neurotrophic factor has been shown to contribute in part to post‐stroke improvements in mobility. Brain‐derived neurotrophic factor has been linked to neuroplastic changes, such as dendritic growth. , Aerobic exercise interventions following stroke in rodents can enhance brain‐derived neurotrophic factor levels in the brain, likely contributing to improvements in mobility function. Thus, starting ET during this critical period may enhance spontaneously occurring regenerative processes and yield greater gains in mobility than exercise initiated in the later phases.

Balance and CRF

Finally, there was no association between time to start ET and postintervention CRF or balance when ET groups were compared with controls. Boyne et al, also reported no time association with change in CRF when introduced <6 months versus ≥6 months (−0.1 mL·kg−1∙min−1; 95% CI, −3.2 to 2.9) similar to the 0.052 mL·kg−1∙min−1 (95% CI, −1.6 to 1.7) difference in the current study (≤3 months versus >3 months). There were no studies conducted >3 to 6 months that measured balance.

Association Between Time and 6MWD but Not Between Time and CRF or Balance

Given that balance and CRF are predictors of 6MWD and 10MWT outcomes, it was unexpected that the association of early training with improved 6MWD and 10MWT did not occur concurrently with improved CRF and balance. , , , The underlying reasons for this may be multifactorial. During measurement of CRF, patients in the earlier phase following stroke may have failed to reach a physiological maximum or reached a lower percentage of their physiological maximum on the exercise stress test than patients later in recovery. In a study of 98 consecutively enrolled patients in the chronic stroke phase (22±44 months after stroke), only 18.4% reached a true physiological maximum, with most discontinuing early for noncardiovascular reasons such as leg weakness or pain. For studies that included patients earlier following stroke, the addition of elevated blood pressure, cardiac arrhythmia, deconditioning, or other issues that can be more common early after stroke may also lead to earlier test termination. , , If the tests were stopped because of motor performance and not cardiorespiratory end points, including meeting sufficient respiratory exchange ratio values, then VO2peak may not be capturing the true effect of the intervention. Although oxygen uptake achieved at the anaerobic threshold may be a more metabolically uniform measure, fewer studies reported these data or the proportion of patients who reached an appropriate respiratory exchange ratio value. It is also possible that ET resulted in earlier improved gait economy so that patients required less oxygen when walking at the same speed, allowing a faster sustained walking pace. However, in a recent well‐designed, multicenter, randomized study conducted by Nave et al, 4 weeks of aerobic exercise initiated a median of 28 days after stroke resulted in no difference in gait economy versus relaxation sessions after intervention, or at 3‐ and 6‐months follow‐up.

Clinical Implications: Evaluating Risks and Benefits of Early Initiation of ET

Given the magnitude and clinical importance of the additional gain in 6MWD, initiation of ET should be considered within 3 and up to 6 months after stroke to take advantage of the augmented priming effect of ET. However, several barriers to including ET during inpatient and outpatient stroke rehabilitation have been identified previously and would need to be addressed. These include insufficient time during the therapy session, insufficient length of stay in rehabilitation, interference with other therapy schedules, and comorbid cardiac conditions. , This is not surprising given the significant time requirement reported in the earlier intervention studies of 20 to 30 minutes, 5 session/wk of treadmill exercise. , ,

Medical complexity of patients, such as cardiac conditions, may be associated with increased risk during ET. Therefore, when evaluating when to initiate an exercise intervention, the type and rate of adverse events with respect to elapsed time from stroke should be evaluated against clinical benefits. Unfortunately, only 47% (70/148) of the studies included in this meta‐regression analysis reported on adverse events, prohibiting a meaningful risk‐benefit analysis. However, it is important to explore this issue, at least qualitatively. There was a concerning number of adverse events reported in studies that were started within the first month following stroke. A single group study was conducted in 20 people with mild to no disability. Over half of the participants developed nonserious adverse events (noninjurious falls, dizziness, pain in lower extremities, tiredness) occurring in 14% of all 224 treadmill training sessions; however, no neurological deterioration was detected. Participants attained the target exercise intensity in only 31% of sessions. Nave et al randomized 200 patients a median of 28 days after moderate to severe stroke, to either 4 weeks of relaxation sessions or body weight–supported treadmill aerobic exercise (25 minutes, 5 times/wk at 50%–60% of the predicted maximal heart rate). Adverse events were higher in the exercise compared with the control condition. Specifically, there were increased falls during the treatment period and a higher number of acute hospital admissions and recurrent strokes in the ET group compared with the control group. The authors stated, “For clinical practice, the results of this pragmatic trial do not support the use of aerobic physical fitness training in moderately or severely affected adults in the subacute phase of stroke.” Moreover, ET when compared with relaxation control, did not result in additional benefit to maximal walking speed, Barthel index, but a moderate nonsignificant benefit was noted for the 6MWD after intervention (19 meters; 95% CI, −8 to 46) and persisted at the 6‐month follow‐up at 26 meters (95% CI, −1 to 53). A subsequent safety analysis of this study revealed that the association of aerobic training with serious adverse event incidence rates were related to comorbid atrial fibrillation and diabetes. A review from our group have advocated for delaying moderate to higher intensity exercise for people with diabetes/hyperglycemia, given the higher mortality rates in those with hyperglycemia at the time of stroke, the altered time course of recovery of blood‐brain barrier function, the potential effect on orthostatic hypotension, and that impaired cerebral autoregulation may intensify risk in people with type 2 diabetes. , , , Specifically, we suggested delaying higher intensity exercise for those with a blood glucose level of ≥160 mg/dL measured within the first 48 hours of stroke and including this as part of the preparticipation screening criteria. Furthermore, atrial fibrillation may reduce cardiac output that has the potential to result in cerebral hypoperfusion episodes associated with activity, , especially in the presence of impaired cerebral autoregulation, which could lead to symptoms such as dizziness. Therefore, it is recommended that light‐intensity exercise should be maintained in these patients until the expected recovery of cerebral autoregulation.

Early mobilization studies not included in the current meta‐regression analysis have introduced sitting, standing, and walking within 24 hours of a stroke. These studies have raised safety concerns while revealing little evidence of a favorable functional outcome. , , , The results of the most influential study, A Very Early Rehabilitation Trial After Stroke, demonstrated deleterious effect of mobilization initiated within 24 hours. , Specifically, there was an increased risk of death in the intervention group at 14 days after stroke. This is largely consistent with the preclinical evidence indicating greater risk when ET is initiated very early following stroke. , , , The underlying mechanisms for these adverse events are unknown but may be related to neurobiological protective mechanisms such as cerebral autoregulation, which take up to 2 to 3 months to recover sufficiently to fully protect the brain from the increase or fluctuations in blood pressure that occur with exercise (see review ). Safety, preparticipation screening, and exercise prescription guidelines for early exercise interventions should be a priority. Future studies should include a risk‐benefit analysis given that cerebral protective mechanisms may not have fully recovered in the subacute stages of stroke.

Limitations

Some study quality issues and risk of publication bias were detected, but it is unclear how this might affect the meta‐regression analyses. Some studies had small sample sizes, unbalanced groups related to recovery potential, or a lack of nonactive controls. Because of inconsistency in reporting, anthropometric and disability/stroke severity measurements (eg, Fugl‐Meyer score) could not be included as covariates in the study. Some studies that started remotely from stroke may have had different cohort characteristics that may have contributed to heterogeneity and widened CIs. For completeness, we opted to include these data. Several studies did not report time since stroke or report data in a usable manner, which may reduce the comprehensiveness of the meta‐analysis. We were not able to differentiate between compensation, true motor recovery, and/or therapy‐induced recovery and could not control for the cumulative dose of usual care rehabilitation. Although several outcomes showed an association with timing of ET initiation following stroke, causality cannot be inferred in the current study. There were few studies randomized on the basis of time after stroke ; further controlled studies introducing ET at different initiation points with groups balanced for recovery potential would be needed to establish precise estimates of timing effects to initiate ET to optimize outcomes and their true benefits.

Conclusions

The results of this study reflect the complex relationship between time to initiate ET and postintervention physiological outcomes. There may be varying time windows for augmented responses and no time association for some outcomes. The time windows for augmented outcomes related to ET may span longer time periods for 6MWD than previously thought. Initiating exercise earlier (within 6 months) appears to be associated with a greater improvement in 6MWD and to a lesser extent in fast‐speed 10MWT (within 3 months), but not with CRF, balance, or comfortable 10MWT. Spontaneous recovery and accompanying usual care rehabilitation may account in part for the advantage of earlier ET initiation in pre‐post 10MWT and balance outcomes, requiring further investigation. The early phases after stroke are a dynamic and volatile time, necessitating careful application of ET.

Sources of Funding

Drs Swardfager and Marzolini gratefully acknowledge support from the Heart and Stroke Foundation Canadian Partnership for Stroke Recovery. Dr Swardfager acknowledges support from the Canadian Institutes of Health Research (PJT‐159711) and from the Canada Research Chairs Program. Dr Marzolini acknowledges support from the Heart and Stroke Foundation of Canada.

Disclosures

None.

Tables S1–S8

Figures S1–S4

Click here for additional data file.