Research questions

The research questions for this systematic review are twofold:

1. What are the effects of RT on stroke recovery?
2. Are prescribed RT parameters (frequency, intensity, type, and duration) associated with stroke recovery?

Review team

The review team will consist of 9 academic members (5 trainees and 4 faculty members) and up to 7 community members. The academic members have research expertise in several areas including stroke rehabilitation, aging, musculoskeletal health, statistical analysis, clinical trial, and systematic review methodology. Four academic members are physical therapists with clinical backgrounds in stroke and aging.

We will also recruit 5–7 community members with lived experience of stroke who will form our community advisory group and will engage with the review in several capacities. Community advisory group members will be recruited by invitation from local stroke groups and will be diverse with respect to stroke experience (i.e., level of disability and type of stroke), biological sex, race, and age. All advisory group members will be involved continuously during the review according to the ACTIVE (Authors and Consumers Together Impacting on eVidencE) Framework, described by Pollock and colleagues [17]. The framework details 12 stages in the review process, where partners may look to contribute. Generally, they will be involved in: planning methods, selecting evidence, analysis, interpretation, writing, and knowledge translation activities. Details of the engagement are described below. Community advisory group members will also be invited as co-authors on the final manuscript.

Eligibility criteria

We will include randomized trials examining the effects of RT compared to no exercise or usual care in people living with stroke, transient ischemic attack (TIA), or cerebrovascular accident. Studies will be included if they 1) included only adults (≥18 years old) with a primary diagnosis of stroke, TIA or cerebrovascular accident, 2) used any form of RT (functional, circuit, or traditional RT) as defined by the American College of Sports Medicine [18] (i.e., a form of exercise designed to improve muscular fitness by exercising a muscle or muscle group against external resistance), 3) compared RT to no exercise, usual care, or conventional therapy, and 4) reported at least 1 exercise prescription parameter (frequency, intensity, type, duration of intervention and/or sessions).

The exclusion criteria are as follows: i) not primary research articles (e.g., thesis/dissertations, systematic reviews, secondary data analyses), and ii) trials that combined RT with another form of therapy or co-intervention that is currently recommended [19] to improve physical function in people with stroke. Ineligible co-interventions include: Mental imagery, functional electrical stimulation, constraint-induced movement therapy, mirror therapy, sensory stimulation, virtual reality, bilateral arm training, non-invasive brain stimulation, treadmill-based gait training, robot-assisted training, Tai-Chi, other forms of exercise therapy. Nutritional and/or pharmacological co-interventions will also be ineligible. Aerobic and stretching exercises will be permitted if they are only used for warm-up and/or cool-down purposes. Trials that included balance and functional exercises will not be considered co-interventions, as they are commonly used in functional strength training interventions.

There will be no restrictions for time post-stroke. We will include studies from the acute (0 to 1 week post-stroke), subacute (1 week to 6 months post-stroke), and chronic phases of stroke recovery (>6 months post-stroke) [20]. However, given the heterogeneity in definitions of “usual care” or “conventional therapy” during the chronic phases of stroke recovery, we will only include studies comparing RT to “usual care” or “conventional therapy” if the study population was in the acute to subacute phase of stroke recovery (i.e., less than 6 months), or if usual care was otherwise explicitly defined as no intervention. There will be no restrictions on the types of activities included in usual care or conventional therapy.

Information sources & search strategy

Eight electronic databases (MEDLINE, EMBASE, EMCARE, AMED, PsychINFO, CINAHL, SPORTDiscus, and Web of Science) will be searched from inception for articles published in English. The search strategy was developed in consultation with a research librarian from McMaster University with experience in exercise-based interventions. A sample of the search strategy for MEDLINE can be found in Table 1. Additionally, we will search 2 clinical trial registries (ClinicalTrials.gov, and WHO International Clinical Trials Registry Platform) and perform hand-searches through Google Scholar’s “cited by” function, and through reference lists of included studies and relevant systematic reviews. We will search for studies using terms related to “Stroke or Transient Ischemic Attack” and “Resistance Training”. The construct of interest for this systematic review is stroke recovery, which will be defined by our partners before data extraction. As such we did not include any key terms related to specific outcomes of interest.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Search strategy for MEDLINE.

https://doi.org/10.1371/journal.pone.0295680.t001

Study records

Four independent pairs of authors (KSN and one other) will pilot the study selection process with 100 titles and abstracts (i.e., 25 per independent pair) and 20 full-text articles (i.e., 5 per independent pair) using a pre-populated Microsoft Excel spreadsheet (Microsoft Corporation, Redmond, WA, USA). Next, independent pairs of authors will screen titles and abstracts and full-text articles using online systematic review management software (Covidence, Veritas Health Innovation Ltd, Melbourne, Victoria, Australia). Disagreements will be managed by consensus discussion or consultation with an author not involved in the disagreement.

Data from included records will be extracted using online systematic review management software (Covidence, Veritas Health Innovation Ltd, Melbourne, Victoria, Australia). Two authors will pilot the data extraction process with 5 included studies, and fields will be refined after discussion. Thereafter, 2 independent pairs of authors (KSN, and EW, AM or EH) will extract data using the updated extraction template. Any missing data will be handled by contacting the corresponding author(s), then by imputation, whenever possible. Discrepancies between authors will be resolved by consensus or consultation with a third author if disagreement persists.

Data items

Risk of bias assessment

We will use the Cochrane Collaboration revised risk of bias assessment (RoB-2) tool [21] to assess risk of bias in each study across 5 domains of: the randomization process, deviations from intended intervention, missing outcome data, measurement of the outcome, selection of the reported result. Risk of bias will be presented with either a summary plot or “traffic light” plot, generated by an application from McGuinness & Higgins [22]. Studies with an overall high risk of bias will be removed in sensitivity analyses. Two independent authors will use the RoB-2 assessment. Disagreements will be resolved by consensus discussion or by third author in disagreement persists.

Data synthesis

Results will be synthesized qualitatively in-text and in tabular format. The review team will work with the community advisory group members to identify research gaps emerging from the included studies, with interest in equity-related gaps such as recruited participant’s sex, race and age. The gaps will be described in-text and in tabular format. All quantitative data will be synthesized using Stata SE, version 17.0 (StataCorp LLC, College Station, TX, USA). The accepted significance level for all analyses is set to α = 0.05.

To answer our primary research question of examining the effect of RT on stroke recovery, we will use random-effects meta-analyses, comparing 1) RT to no exercise and 2) RT to usual care across each outcome measure. Variables will be treated as continuous and effect estimates will be presented as mean differences whenever possible. Whenever possible, we will group measures by test. Otherwise, standardized mean differences (SMD) will be presented using Glass’s delta (Δ) for pooled outcome measures. For instance, if there are insufficient studies represented for certain outcomes, we will combine measures by upper- and lower-body outcomes and by construct (e.g., composite versus single-construct measures of physical function, muscle strength, muscle power, etc.). Cluster randomized controlled trials will be treated by multiplying the standard errors of treatment effects by the square root of the design effect, as outlined by the Cochrane Handbook for Systematic Reviews of Interventions [23]. If standard deviations (SD) of change scores are not provided, they will be imputed using the following formulae, per the Cochrane Handbook [23]: (1) (2)

For our secondary research question of examining the influence of RT parameters on treatment effects, we will conduct separate, univariable meta-regression analyses across 4 covariates: frequency, intensity, type, and duration of the RT program. Frequency (sessions per week) and duration (number of weeks) will be treated as continuous, the intensity will be treated as categorical using contemporary definitions of low, moderate, high intensity, and high velocity (i.e., power focused) as defined by the American College of Sports Medicine [18], and type (separated by traditional, functional, or circuit-based RT) also as categorical. Definitions for each parameter category are in Table 2. The credibility of statistically significant meta-regression analyses will be evaluated using the Instrument to assess the Credibility of Effect Modification Analyses (ICEMAN) [24].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Definitions for each parameter category for meta-regression analyses.

https://doi.org/10.1371/journal.pone.0295680.t002

Heterogeneity will be assessed first by visual inspection of the forest plots and chi-square test (i.e., Q statistic), where a statistically significant test indicates the presence of heterogeneity between studies. As the Cochrane Handbook for Systematic Reviews of Interventions [23] outlines, we will consider low, moderate, substantial, or considerable heterogeneity if I² <30%, 30–60%, 60–75%, and >75%, respectively. For our primary research question, we will explore heterogeneity using univariable meta-regression analyses with 3 covariates: time post-stroke (3 levels: acute, sub-acute, and chronic stroke), age (continuous variable), and sex (3 levels, tertiles of low, moderate, and high proportion of females). Missing covariate data will be handled first by contacting the corresponding authors, then excluding trials if data is not present. Publication bias will also be examined by visual inspection of the funnel plot and Egger test, and sensitivity analyses will be performed by removing studies found to have small-study effects, defined as having high influence on funnel plot asymmetry (point lays outside of 95% confidence intervals).

Certainty of evidence

We will use the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach to assess the certainty of evidence based on 8 criteria: Risk of bias, inconsistency of results, indirectness of evidence, imprecision, large magnitude of effect, confounding effects, presence of a dose-response relationship, and publication bias [25]. Disagreements will be resolved by consensus discussion or by a third author in disagreement persists. For each outcome and analysis, the certainty of evidence will be considered either high, moderate, low, or very low.

Three review team members (KSN, KM, AM), who each have experience with RT program design and delivery, will perform the initial GRADE assessment. Thereafter, the community advisory group members with lived experience will be invited to provide their importance ranking. Each member will rate the importance of all included outcomes on a scale from 1 (of least importance) to 9 (of most importance). The median score across the 8 assessments will be taken. Scores for each outcome will be used in the GRADE assessment to label outcomes: of limited importance (median score: 1 to 3), important but not critical (median score: 4 to 6), or critical (median score: 7 to 9) [26].