Criteria for considering studies for this review
Types of studies
We will include randomised controlled trials (RCTs), including cluster-RCTs. We will also include RCTs with a cross-over design. However, we will interpret these RCTs as having a parallel group design up to the point of cross-over.
One arm of the trial must include caregiver-mediated exercises and we will consider this arm as the experimental group for the purpose of this review.
The other (control) arm can receive usual treatment, no treatment or any other type of rehabilitation intervention or attention-control group as long as it is not caregiver-mediated. We will accept usual treatment when it is described as usual care in the setting of the patient. Therefore, it is important that caregiver-mediated exercises are not explicitly asked for in the control group.
Types of participants
People, at least 18 years old, who have suffered a stroke. Stroke is defined by the World Health Organization as "a clinical syndrome typified by rapidly developing signs of focal or global disturbance of cerebral functions, lasting more than 24 hours or leading to death, with no apparent causes other than of vascular origin" (WHO 1989). We will include RCTs regardless of timing after stroke and setting.
Types of interventions
One arm of the RCT must include exercises done together with a caregiver, whereas the caregiver involvement is not explicitly asked for in the other arm of the RCT.
We will include trials in which the patient and their caregiver are trained or instructed together, as well as trials in which the caregiver alone is trained or instructed.
There will be no limit to the number of sessions or to the frequency of delivery. We will include all types of exercises as long as they are aimed at improving motor function. Therefore, we will exclude RCTs of speech, swallowing or cognitive interventions done together with a caregiver.
A caregiver will be defined as someone who is willing and able to do exercises together with the patient. This can be someone close to the patient, for example a partner, family member or friend. This can also be a volunteer. A caregiver is not a professional and is not paid for their efforts.
We will include interventions delivered at any location, for example at home, in hospital or in a rehabilitation setting.
We will carefully judge trials that combine caregiver-mediated exercises with other interventions. If the interventions and outcomes can be separated, we will include them. We will exclude trials where the influence of the caregiver-mediated exercises cannot be isolated.
We will contact trial authors if it is unclear whether a trial meets our definition.
Types of outcome measures
Patient: ADL measures, such as the Barthel index (BI) (Mahoney 1965; Collin 1988), Functional Independence Measure (FIM) (Dodds 1993), modified Rankin Scale (mRS) (De Haan 1995; Dromerick 2003), Nottingham Extended ADL index (NEADL) (Nouri 1987), or Frenchay Activities Index (FAI) (Wade 1985). We will combine the BI and FIM in our analyses as they are both measurements of basic ADL. The mRS is often dichotomised into good or poor outcomes. When this is done we will analyse the scale separately. When the mRS is not dichotomised we will combine it with the BI and FIM. We will combine the NEADL and FAI because they are both measurements of extended ADL.
Caregiver: measures of mood, burden and QoL, for example Caregiver Strain Index (CSI) (Robinson 1983), Hospital Anxiety and Depression Scale (HADS) (Zigmond 1983; Herrmann 1997; Aben 2002; Bjelland 2002), or CarerQoL (Brouwer 2006; Hoefman 2011). These scales measure different constructs and we will not combine them.
We will distinguish between caregivers who are family or friends and other volunteer caregivers for the above-mentioned measures of outcome.
Gait and gait related measures: walking speed, walking distance, Timed-Up-and-Go test (TUG) (Collen 1990; Flansbjer 2005), Rivermead Mobility Index (RMI) (Collen 1991; Hsieh 2000; Hsueh 2003), Berg Balance Scale (BBS) (Berg 1992; Berg 1995; Stevenson 2001; Mao 2002). These scales measure different constructs and we will not combine them.
Measures of motor impairment: Motricity Index (MI) (Collin 1990), Fugl-Meyer Assesment (FMA) (Duncan 1983; Sanford 1993; Shelton 2001). We will combine the MI and FMA in our analyses.
Measures of mood and QoL of the patient, for example, measured by the Stroke Impact Scale (SIS) (Duncan 1999; Duncan 2002; Duncan 2003) and HADS (Zigmond 1983; Herrmann 1997; Aben 2002; Bjelland 2002). We will only combine measurement scales if they measure the same construct, e.g. mood or QOL.
Measures of fatigue of the patient, for example, measured by the Fatigue Severity Scale (FSS) (Valko 2008).
Adverse outcomes, for example, pain, injury, or falls. If possible, we want to compare the total number of falls between groups, and the number of patients experiencing at least one fall between groups.
Measures of upper limb activities of function, for example, Action Research Arm Test (ARAT) (Hsieh 1998; Platz 2005; Chen 2012).
Length of stay in hospital, rehabilitation centre or nursing home, or treatment in an outpatient clinic.
We will not exclude trials with outcome measures that differ from the ones mentioned above. We will verify if they measure the same construct. If this is the case, we will pool them, if they do not measure the same construct, we will systematically report these outcomes.
Search methods for identification of studies
See the 'Specialized register' section in the Cochrane Stroke Group module. We will search for trials in all languages and arrange translation of papers published in languages other than English.
We will search the Cochrane Stroke Group Trials Register and the following electronic databases and trials registers:
Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library, latest issue);
Cochrane Database of Systematic Reviews (CDSR) (The Cochrane Library, latest issue);
Cochrane Methodology Register (CMR) (The Cochrane Library, latest issue);
Database of Abstracts of Reviews of Effects (DARE) (The Cochrane Library, latest issue);
Health Technology Assessment Database (HTA) (The Cochrane Library, latest issue);
NHS Economic Evaluation Database (NHS EED) (The Cochrane Library, latest issue);
MEDLINE (Ovid) (from 1946) (Appendix 1);
EMBASE (Ovid) (from 1980);
CINAHL (Cumulative Index of Nursing and Allied Health Literature) (EBSCO) (from 1982);
SPORTDiscus (EBSCO) (from 1985);
AMED (Alternative and Complementary Medicine) (Ovid) (from 1985);
Physiotherapy Evidence Database (PEDro) (http://www.pedro.org.au/);
EU Clinical Trials Register (https://www.clinicaltrialsregister.eu);
Stroke Trials Registry (www.strokecenter.org/trials/);
Current Controlled Trials (www.controlled-trials.com);
WHO International Clinical Trials Registry Platform (http://www.who.int/ictrp/en/);
Australian New Zealand Clinical Trials Registry (www.anzctr.org.au/).
We developed the MEDLINE search strategy with the help of the Cochrane Stroke Group Trials Search Co-ordinator (Brenda Thomas) and we will adapt this for the other databases (Appendix 1).
Searching other resources
We will carry out the following additional searches to identify further published, unpublished and ongoing studies:
search the reference lists of all included articles;
contact experts in the field;
contact authors of conference abstracts for full-text publications, if available; and
use Science Citation Index Cited Reference Search for forward tracking of important articles.
Data collection and analysis
Selection of studies
Two review authors (JV, JMV) will independently screen the titles and abstracts of the records obtained from the electronic searches and exclude obviously irrelevant studies. We will obtain the full text of the remaining studies and the same two authors will select studies for inclusion in the review based on the inclusion criteria described previously. We will resolve any disagreements by discussion and, where necessary, in consultation with a third review author (EvW).
Data extraction and management
Two review authors (JV, JMV) will conduct data extraction and will review the risk of bias of the eligible trials. The review authors will not be blinded to study authors, journals or outcomes. We will resolve any disagreements by discussion. If no consensus is reached, a third author (EvW) will make the final decision. One review author (JV) will extract the data and a second review author (JMV) will cross check the extracted data using a standard checklist, including randomisation method, study population, intervention methods and delivery, outcome measures, and follow-up.
Assessment of risk of bias in included studies
We will use the tool for assessing risk of bias in included RCTs as described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We will assess all forms of bias (selection bias, performance bias, attrition bias, detection bias, reporting bias and other bias) with their domains (random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data, selective reporting and other sources of bias). We will present the results in 'Risk of bias' tables. For each entry our judgement ('low risk', 'high risk' or 'unclear risk') will be given, followed by a description of the judgement. We will make our judgements transparent, and will use comments or quotes when necessary.
Measures of treatment effect
For continuous outcomes using similar measurement scales, we will use the mean difference (MD) with 95% confidence intervals (CIs). If similar outcomes are measured on different scales we will use Hedges’ g and calculate the 95% CI.
We will report the direction of the effect for every scale in order to align the treatment effects between outcomes scales. Scales in which a low score reflects a favourable outcome and a high score an unfavourable outcome, will be multiplied by -1.
We will extract means and standard deviations (SDs) of post-intervention scores and follow-up scores. Where available, we will also extract means and SDs of change from baseline.
For dichotomous outcomes (e.g. falls: yes/no) we will calculate summarised odds ratios (ORs) with 95% CIs.
We will use Review Manager 5.2 for all quantitative analyses (RevMan 2012).
Unit of analysis issues
We will take into account that studies can have different levels of randomisation, e.g. at the level of a participating person, or at the level of a group of participating people (cluster randomisation).
When there are multiple observations because of follow-up measurements, we will analyse outcomes for the different follow-up categories (short-term: zero to three months, medium-term:four to six months, and long-term: six months and more).
If we identify studies with multiple intervention groups, we will make multiple pair-wise comparisons between all possible pairs of intervention groups. We will make sure that participants are not double-counted in the analysis.
Dealing with missing data
If data are missing or not in a form suitable for quantitative pooling, we will contact the trial authors for additional information.
Assessment of heterogeneity
Heterogeneity between the included RCTs will be inevitable (Higgins 2011). Therefore, we will assess the impact of heterogeneity in the meta-analysis for each outcome with the I² statistic. When there is substantial statistical heterogeneity (I² > 50%) we will use a random-effects model, otherwise we will use a fixed-effect model for meta-analysis.
Assessment of reporting biases
We will assess reporting bias by a funnel plot in which the effect estimates and precision (standard error) of each individual RCT will be plotted.
When we find two or more RCTs with a low risk of bias in which study population, intervention and outcome measures are the same, we will perform a meta-analysis of the comparison caregiver-mediated exercises versus the control group (usual care, no intervention, or any other intervention).
We will include a 'Summary of findings' table using the Cochrane template. The following seven outcomes will be included: ADL measures, burden of the caregiver, walking speed, walking distance, mood of the patient, length of stay, and adverse events.
For each outcome we will include, the number of participants, the overall quality of the evidence using GRADE levels of evidence, the magnitude of the effect, a measure of burden of the outcome and comments.
If there are not enough studies for a meta-analysis we will synthesise the findings of the included studies systematically in a table. Hereby, we will use the GRADE levels of evidence to define the quality of the studies (Higgins 2011).
In the same way we will systematically report other outcome measures that we cannot include in a meta-analysis because they do not measure the same construct as our predefined outcome measures.
We will use Review Manager 5.2 for the analyses (RevMan 2012).
Subgroup analysis and investigation of heterogeneity
If we find sufficient studies (e.g. two or more for a subgroup) we will perform subgroup analyses for type of experimental intervention (caregiver-mediated exercises in addition to regular therapy versus caregiver-mediated exercises done alone), type of control group (control intervention or no intervention) and timing post-stroke ((sub)acute stage: less than six months versus chronic stage: more than six months).
Where we use a fixed-effect model, we will apply a random-effects model to assess the robustness of the results to the method used.