Constraint-induced movement therapy for upper extremities in stroke patients (Review)

Transcription

1 Constraint-induced movement therapy for upper extremities in stroke patients (Review) Sirtori V, Corbetta D, Moja L, Gatti R This is a reprint of a Cochrane review, prepared and maintained by The Cochrane Collaboration and published in The Cochrane Library 2009, Issue 4

2 T A B L E O F C O N T E N T S HEADER ABSTRACT PLAIN LANGUAGE SUMMARY BACKGROUND OBJECTIVES METHODS RESULTS DISCUSSION AUTHORS CONCLUSIONS ACKNOWLEDGEMENTS REFERENCES CHARACTERISTICS OF STUDIES DATA AND ANALYSES Analysis 1.1. Comparison 1 Constraint versus control: primary outcome, Outcome 1 Disability post-intervention.. 45 Analysis 2.1. Comparison 2 Constraint versus control: primary outcome, Outcome 1 Disability - three to six-month follow up Analysis 3.1. Comparison 3 Constraint versus control: subgroup analysis on primary outcome, Outcome 1 Amount of task practice Analysis 3.2. Comparison 3 Constraint versus control: subgroup analysis on primary outcome, Outcome 2 Anatomical region restraint Analysis 3.3. Comparison 3 Constraint versus control: subgroup analysis on primary outcome, Outcome 3 Time since stroke Analysis 4.1. Comparison 4 Constraint versus control: secondary outcomes, Outcome 1 Arm Motor Function Analysis 4.2. Comparison 4 Constraint versus control: secondary outcomes, Outcome 2 Perceived Arm Motor Function (Amount of Use) Analysis 4.3. Comparison 4 Constraint versus control: secondary outcomes, Outcome 3 Perceived Arm Motor Function (Quality of Use) Analysis 4.4. Comparison 4 Constraint versus control: secondary outcomes, Outcome 4 Arm Motor Impairment.. 53 Analysis 4.5. Comparison 4 Constraint versus control: secondary outcomes, Outcome 5 Quality of life ADDITIONAL TABLES APPENDICES HISTORY CONTRIBUTIONS OF AUTHORS DECLARATIONS OF INTEREST SOURCES OF SUPPORT DIFFERENCES BETWEEN PROTOCOL AND REVIEW INDEX TERMS i

3 [Intervention Review] Constraint-induced movement therapy for upper extremities in stroke patients Valeria Sirtori 1, Davide Corbetta 1, Lorenzo Moja 2, Roberto Gatti 3 1 Unita Operativa di Recupero e Rieducazione Funzionale, Fondazione Centro San Raffaele del Monte Tabor, Milan, Italy. 2 Italian Cochrane Centre, Mario Negri Institute for Pharmacological Research, Milan, Italy. 3 School of Physiotherapy, University Vita-Salute San Raffaele, Milan, Italy Contact address: Valeria Sirtori, Unita Operativa di Recupero e Rieducazione Funzionale, Fondazione Centro San Raffaele del Monte Tabor, via Olgettina, 58, Milan, Lombardia, 20132, Italy. sirtori.valeria@hsr.it. Editorial group: Cochrane Stroke Group. Publication status and date: New, published in Issue 4, Review content assessed as up-to-date: 18 May Citation: Sirtori V, Corbetta D, Moja L, Gatti R. Constraint-induced movement therapy for upper extremities in stroke patients. Cochrane Database of Systematic Reviews 2009, Issue 4. Art. No.: CD DOI: / CD pub2. Background A B S T R A C T In stroke patients, upper limb paresis affects many activities of daily life. Reducing disability is therefore a major aim of rehabilitation programmes for hemiparetic patients. Constraint-induced movement therapy (CIMT) is a current approach to stroke rehabilitation that implies the forced use and the massed practice of the affected arm by restraining the unaffected arm. Objectives To assess the efficacy of CIMT, modified CIMT (mcimt), or forced use (FU) for arm management in hemiparetic patients. Search methods We searched the Cochrane Stroke Group trials register (last searched June 2008), the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library Issue 1, 2008), MEDLINE (1966 to June 2008), EMBASE (1980 to June 2008), CINAHL (1982 to June 2008), and the Physiotherapy Evidence Database (PEDro) (June 2008). Selection criteria Randomised control trials (RCTs) and quasi-rcts (qrcts) comparing CIMT, mcimt or FU with other rehabilitative techniques, or none. Data collection and analysis Two review authors independently classified the identified trials according to the inclusion and exclusion criteria, assessed methodological quality and extracted data. The primary outcome was disability. Main results We included 19 studies involving 619 participants. The trials included participants who had some residual motor power of the paretic arm, the potential for further motor recovery and with limited pain or spasticity, but tended to use the limb little if at all. Only five studies had adequate allocation concealment. The majority of studies were underpowered (median number of included patients was 15) and we cannot rule out small-trial bias. Six trials (184 patients) assessed disability immediately after the intervention, indicating 1

4 a significant standard mean difference (SMD) of 0.36, 95% confidence interval (CI) 0.06 to For the most frequently reported outcome, arm motor function (11 studies involving 373 patients), the SMD was 0.72 (95% CI 0.32 to 1.12). There were only two studies that explored disability improvement after a few months of follow up and found no significant difference, SMD (95% CI to 0.40). Authors conclusions CIMT is a multifaceted intervention: the restriction to the normal limb is accompanied by a certain amount of exercise of the appropriate quality. It is associated with a moderate reduction in disability assessed at the end of the treatment period. However, for disability measured some months after the end of treatment, there was no evidence of persisting benefit. Further randomised trials, with larger sample sizes and longer follow up, are justified. P L A I N L A N G U A G E S U M M A R Y Constraint-induced movement therapy for upper extremities in stroke patients Paralysis of an arm after a stroke makes arm movements, such as reaching, grasping, and manipulating an object difficult. In turn, this causes many difficulties in activities of daily life, such as bathing, dressing, eating, and toileting. Constraint-induced movement therapy (CIMT) is a type of rehabilitation therapy in which the patient is obliged to use the paralysed arm. The normal arm and hand are prevented from moving with a glove and a special arm rest. We found 19 studies involving 619 participants, which assessed whether CIMT could reduce disability in stroke patients with a paralysed arm. Patients were included in the studies if they had good potential for recovery but tended not to use the arm. At the end of the treatment period, compared with exercise without constraint, CIMT improved the patient s ability to manage activities of daily living, but there was no evidence that this improvement was maintained over the next six months. Further larger trials to assess whether CIMT provides lasting benefit are justified. B A C K G R O U N D Cerebrovascular disease is the third leading cause of death in the USA and Western Europe, and an important cause of disability (Frizzell 2005; Gerzeli 2005; McCarron 2006). Upper extremity paresis is a leading cause of functional disability after stroke as it causes difficulties in everyday life (e.g. self-care, feeding, dressing) (Gresham 1975; Whitall 2000). The principal guidelines about the management and recovery of upper extremity movement in stroke patients include treatment with biofeedback (Crow 1989; Moreland 1994; Rathkolb 1990; Sathian 2000), electrical stimulation/functional electrical stimulation (de Kroon 2002; de Kroon 2005; Kimberley 2002; Peurala 2002; Wang 2002), thermal stimulation (Chen 2005), mental practice (Page 2005b; Page 2007a), robot assistance (Hesse 2003; Lum 2002; Masiero 2007), repetitive task training (French 2007), and constraint-induced movement therapy (CIMT) (Miltner 1999; Page 2001; Page 2002a; Taub 1993; Taub 1994; Taub 1999). CIMT, as described by the first authors (Miltner 1999; Taub 1994; Taub 1999), is based on two fundamental principles. 1. Forced use of the affected arm by restraining the unaffected arm, with a sling or a hand splint, during dedicated exercise sections or usual activities of daily living (90% of waking hours). 2. Massed practice (several hours of exercise) of the affected arm through a shaping method, where shaping involves a commonly operant conditioning method in which a behavioural (in this case movement ) objective is approached in small steps of progressively increasing difficulty. The participant is rewarded with enthusiastic approval for improvement, but never blamed or punished for failure. The initial report of the use of CIMT proposed extensive and intensive training (six to eight hours per day) (Miltner 1999; Taub 1994; Taub 1999); over the years, though, others have developed different forms of constraint therapy, reducing the training during the period of restraint (Page 2001; Page 2002a; Page 2002b), or concentrating only on the use of restraint (forced use), with no additional treatment of the affected arm (Burns 2007; Ploughman 2004). The rationale is based on the theory of learning non-use from experiments on monkeys. Researchers observed that after upper 2

5 limb de-afferentation, monkeys did not use their affected limb even though their motor ability was nearly normal (Knapp 1963; Taub 1977; Taub 1980). This non-use was an acquired behaviour learned during the spinal shock period and, as a consequence of its origin, could be reversed by behavioral measures such as, for example, constraint of the sound limb. Thus the learned nonuse theory predicts that patients have, in fact, greater movement ability than they show in their everyday tasks. If this is correct, constraint of the unaffected arm would be a useful tool for realising this ability in everyday activities (Sterr 2006). During the last 10 years, the neuroplasticity and cortical reorganisation of the central nervous system (CNS) has been observed and described in trials with stroke patients undergoing CIMT. These studies have associated noticeable brain changes with functional improvements related to CIMT. To measure changes in cortical activation and enlargement of cerebral areas, some of these studies have included transcranial magnetic stimulation (TMS) (Kim 2004; Schaechter 2002; Szaflarski 2006), or functional magnetic resonance imaging (fmri) (Levy 2001; Liepert 2000; Liepert 2001; Ro 2006); the remaining studies included other approaches (Kopp 1999). O B J E C T I V E S The purpose of this systematic review is to evaluate the efficacy of CIMT, mcimt or FU as rehabilitative techniques in the management of affected upper limb in hemiparetic patients after stroke. M E T H O D S Types of interventions The studies included all used CIMT or mcimt or FU for the treatment of the affected upper limb compared with other rehabilitative techniques (occupational therapy or physiotherapy) or none. For the purpose of this review we used the following definitions (as described in Hoare 2007): CIMT: restraint of the unaffected upper limb, with more than three hours per day of therapy. mcimt: restraint of the unaffected upper limb, with three hours or less per day of therapy. FU: restraint of the unaffected upper limb but no specific treatment of the affected upper limb. We will consider all interventions, irrespective of: number of hours of training per day; number of hours of constraint per day; duration of treatment; type of exercise used in training sessions. Types of outcome measures If a study presented more than one measure for the same outcome category, we included the measure most frequently used across studies in the analysis. Primary outcome Disability Functional Independence Measure (FIM), Barthel Index (BI). Secondary outcomes Criteria for considering studies for this review Types of studies Randomised controlled trials (RCTs) and quasi-rcts (qrcts) comparing CIMT or mcimt or FU with other rehabilitative techniques (occupational therapy or physiotherapy), or none. Types of participants We examined trials enrolling adults (aged over 18 years) with a clinical diagnosis of stroke, either ischaemic or haemorrhagic (World Health Organization definition, Hatano 1976), with paresis of an arm. Arm motor function Wolf Motor Function Test (only score) (WFMT), Arm Research Arm Test (ARAT), Arm Motor Ability Test (AMAT), Emory Function Test (EMF), Assessment of motor and process skills (AMPS). Perceived arm motor function Motor Activity Log (MAL): Amount of Use (AoU) and Quality of Use (QoU). Arm motor impairment Fugl Meyer Assessment (FMA), Chedoke McMaster Impairment Inventory (CMII), hand strength. 3

6 Dexterity Nine Hole Peg Test* (NHPT), Grooved Pegboard Test (GPT). Quality of life Stoke Impact Scale (SIS). These items are measures (or subgroups) added during the systematic review process and were not presented in the original protocol of this review. These items were perceived as being of importance for physiotherapists and patients and offer a more complete picture about the efficacy of this technique. * These items are measures where a low score indicates a positive outcome. The results were expressed as negative values for these studies. In all other measures, a high score indicates a good outcome, and findings were expressed as positive values. methods (e.g. randomisation, blindness, completeness of follow up, reliability and validity of scales), details of participants (e.g. age, sex, time since stroke, side affected), interventions, inclusion and exclusion criteria, and all assessed outcomes. We resolved disagreements by consensus and consulted a third review author (RG) if disagreements persisted. We contacted study authors for clarification or to request missing data, if possible. If we could not contact an author, we estimated the standard deviation of the pre-post treatment differences in the treatment and control groups from the variances, as needed for data analysis. Quality assessment We assessed the quality of studies using the criteria in the Cochrane Handbook for Systematic Reviews of Interventions (Cochrane Handbook 2008). Search methods for identification of studies See the Specialized register section in the Cochrane Stroke Group module. We searched the Cochrane Stroke Group trials register, which was last searched by the Review Group Co-ordinator on 5 June We also searched the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library Issue 1, 2008) (Appendix 1), MEDLINE (1966 to June 2008) (Appendix 1), EMBASE (1980 to June 2008) (Appendix 2), CINAHL (1982 to June 2008) (Appendix 3), and the Physiotherapy Evidence Database (PEDro) ( (June 2008) (Appendix 4). In an effort to identify further published, ongoing and unpublished studies, we searched reference lists of relevant papers and contacted researchers in the field of CIMT. Data collection and analysis Methods of randomisation We regarded a randomisation method as appropriate if it meant that each study participant had the same chance of receiving each intervention. We considered the following methods of allocation appropriate: using random number tables, a computer random number generator, coin tossing, or card shuffling. Allocation concealment (when the investigators cannot predict which treatment comes next) We scored this as: A - adequate, when the method of allocation is clearly described (e.g. central randomisation, serially numbered opaque, sealed envelopes); B - unclear, when the authors do not report any allocation concealment approach at all, or report an approach that does not satisfy point A; C - inadequate, when the method of allocation is not concealed. Identification of relevant trials One review author (DC) read the titles of identified references and eliminated obviously irrelevant studies. We obtained abstracts for the remaining studies and then, on the basis of the inclusion criteria, two review authors (DC & VS) independently ranked these as relevant, irrelevant or unsure. We retrieved and reviewed the full text articles for those ranked as unsure. We resolved disagreements by consensus, and consulted a third review author (RG) if disagreements persisted. Data extraction Two review authors (VS and DC) independently extracted data. We recorded all data on a standardised checklist, incorporating: Potential for selection bias after allocation We scored this as: A - trials where an intention-to-treat analysis is possible and there are few losses to follow up; B - trials reporting exclusions (as listed in A but less than 10% exclusions); C - no reporting of exclusions, or more than 10% exclusions, or wide differences in exclusions between groups. Blinding with reference only to the outcome assessor We scored this as: A - blind; B - not blind. 4

7 Follow up We scored this as: A - adequate: if the numbers and reasons for dropouts and withdrawals in all intervention groups are described and if 90% or more of the randomised participants were included in the analysis, or if it is specified that there were no dropouts or withdrawals; B - unclear: if the report gives the impression there were no dropouts or withdrawals, but it is not specifically stated; C - inadequate: if less than 90% of the randomised participants were included in the analysis or the number or reasons for dropouts and withdrawals were not described. where constraint was not accompanied by additional exercise or the number of hours and type of treatment in the control and constraint groups are the same). 4. Time since stroke: we used mean time since stroke at recruitment to classify trials as zero to three months, three months to six months, and over nine months. To investigate differences between subgroups, we used the approach for a significance test described by Deeks This method is implemented in the Cochrane Review Manager software, RevMan 5.0, for fixed-effect analyses based on the inversevariance method (RevMan 2008). Scales to measure outcomes Scales had to be supported by studies about their psychometric properties. We classified the scales as: A - adequate: if studies support the reliability and validity of the scale; B - unclear: if supporting data are not provided or the scale has never been tested; C - inadequate: if there is evidence of insufficient reliability or validity. Data analysis We did a statistical summary of treatment effects only if there was no major clinical heterogeneity. We assessed the degree of heterogeneity among the trials by the I 2 statistic for each outcome. We judged I 2 > 50% as substantial heterogeneity (Cochrane Handbook 2008). We converted continuous data to mean difference (MD) and, if different scales were used, we first computed a standardised mean difference (SMD), and second, an overall MD and overall SMD. We calculated overall estimates using the fixedeffect or random-effects model, depending on the I 2 heterogeneity test results. Amendments to protocol In 2003, based on the available evidence about CIMT, we published our original protocol in The Cochrane Library (Sirtori 2003). We subsequently found that the protocol did not reflect the increasing variability among potentially relevant primary studies and was out of date in terms of systematic review methodology. The main shortcomings in the original protocol related to: the inclusion criteria in terms of participants, interventions and outcome measures, as they were too restrictive and narrowly focused, being de facto a subgroup analysis (Cochrane Handbook 2008); and the Methods section, which did not provide enough detail to ensure replicability (for more details see the Differences between protocol and review section). R E S U L T S Subgroup analysis There were four possible post-hoc subgroup analyses (Table 1). 1. Dosage of task practice : we calculated this by multiplying the number of weeks by the number of sessions per week by the session duration in hours. We divided trials into those providing more than 30 hours of training, and those providing 30 hours of training or less. We based the cut-off of 30 hours on the cut-off of three hours between CIMT and mcimt per single treatment (see Types of interventions ) and subsequently computing a total volume of work during the rehabilitation period. 2. Anatomical region restraint: we divided studies in to those constraining the unaffected arm only at the hand by a mitt, and those constraining both hand and arm by a sling and mitt. 3. Constraint effect: we included only the studies where the independent variable between groups was only constraint (e.g. Description of studies See: Characteristics of included studies; Characteristics of excluded studies. Results of the search The search strategy identified over 9000 articles. On the basis of information in titles and abstracts, we identified 91 studies as potentially relevant and we obtained the full text papers. Seventy-three studies did not meet at least one of our inclusion criteria: firstly, most studies were not randomised; and secondly, they did not consider CIMT or mcimt. We excluded one RCT (Suputtitada 2004) because the data were not available (see Characteristics of excluded studies ). 5

8 Included studies A total of 19 published RCTs met the inclusion criteria (Alberts 2004; Atteya 2004; Boake 2007; Dahl 2008; Dromerick 2000; Lin 2007; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004; Taub 1993; Wittenberg 2003; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c). In five studies, patients were randomised to one of the following three interventions: mcimt, traditional rehabilitation (training without restriction of the sound limb), and no treatment (Atteya 2004; Page 2001; Page 2002b ; Page 2004; Page 2008). In order to reduce the heterogeneity among studies and to preserve the equipoise principle (Edwards 1998), we considered only the data from arms comparing mcimt with traditional rehabilitation. For more details, see the Characteristics of included studies table. The majority of studies were conducted in the USA (11 studies) and Taiwan (four studies). Six were identified as pilot RCTs (Alberts 2004; Dromerick 2000; Myint 2008; Page 2002b; Page 2005a; Ploughman 2004), although it is not clear whether pilot refers to examination of the new CIMT, to the feasibility of the study, or to the small sample and lack of sample size calculation. Thirteen studies were multicentric. Eleven studies enrolled outpatients, four trials enrolled inpatients, and four trials enrolled both. Participants A total of 619 participants were enrolled in the 19 trials. There were more men (N = 387, 62%) than women (N = 232, 37%). The mean age of the patients ranged from 37 years (Page 2004) to 87 years (Wu 2007c), with the majority between 55 and 70 years. Time since stroke was zero to three months for five trials (Boake 2007; Dromerick 2000; Myint 2008; Page 2005a; Ploughman 2004); three to nine months for five trials (Alberts 2004; Atteya 2004; Page 2001; Page 2002b; Wolf 2006), and more than nine months for five trials (Lin 2007; Page 2004; Page 2008; Taub 1993; Wittenberg 2003). Three trials considered patients in which the stroke onset varied between one to 37 months (Wu 2007a, Wu 2007b, Wu 2007c), and one trial considered patients in which the stroke onset varied between one to 92 months (Dahl 2008). Eleven studies, with a total of 387 participants, reported the number of patients with ischaemic stroke (clinical or instrumental diagnosis) (N = 344, 88%) (Alberts 2004; Atteya 2004; Boake 2007; Dromerick 2000; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Wittenberg 2003; Wolf 2006). The remaining 12% had haemorrhagic stroke. Sixteen studies, with a total of 351 participants, reported the number of right-side-affected patients (N = 186, 51%) (Alberts 2004; Atteya 2004; Boake 2007; Dromerick 2000; Lin 2007; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004; Taub 1993; Wu 2007a; Wu 2007b; Wu 2007c). Eight studies, with a total of 410 participants, reported the number of patients with stroke concordance (paresis of pre-stroke dominant side) (N = 202, 49%) (Alberts 2004; Myint 2008; Ploughman 2004; Taub 1993; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c). The main inclusion criteria reported were as follows. 1. Movement capacity of the upper arm: i) ability to extend the metacarpophalangeal and interphalangeal joints at least 10, and the wrist 20 (Alberts 2004; Atteya 2004; Boake 2007; Dahl 2008; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Taub 1993; Wittenberg 2003; Wolf 2006; Wu 2007a); ii) ability to extend the metacarpophalangeal and interphalangeal joints of two digits and the wrist 10, plus 10 of thumb abduction/extension (Alberts 2004; Wolf 2006); iii) score 1 to 3 on the motor arm items of the National Institute of Health Stroke Scale (NIHSS) (Boake 2007; Dromerick 2000); iv) stage 3 or above in the reach Brunnstrom for the proximal part of the upper extremity (UE) (Lin 2007; Wu 2007b; Wu 2007c); v) Chedoke McMaster Impairment Inventory (CMII) (Ploughman 2004); vi) score 0 to 2 on Modified Rankin Scale before the stroke (Dahl 2008). 2. Absence of cognitive impairment: i) Mini Mental State Examination (MMSE) or modified MMSE more than 24 or 70 respectively (Alberts 2004; Atteya 2004; Dahl 2008; Lin 2007; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004; Taub 1993; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c); ii) no neglect or speech comprehension (Boake 2007; Dahl 2008; Taub 1993); iii) score 1 on the consciousness, communication and neglect item of the NIHSS (Dromerick 2000). 3. Non-use of the affected arm in the real world: score < 2.5 on the MAL (Alberts 2004; Lin 2007; Page 2005a; Page 2008; Wittenberg 2003; Wolf 2006; Wu 2007a; Wu 2007b). 4. No balance problems including walking (Alberts 2004; Lin 2007; Myint 2008; Taub 1993; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c). 5. No excessive pain in the affected arm: score < 4 on the visual analogue scale (VAS) (Alberts 2004; Atteya 2004; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Wolf 2006). 6. No excessive spasticity: score 2 (in any joint) respectively on the Ashworth Scale (AS) or on the modified AS (Atteya 2004; Lin 2007; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Taub 1993; Wu 2007a; Wu 2007c). 7. No joint limitation of the affected arm (Alberts 2004; Boake 2007; Wolf 2006). Intervention 6

9 Seven studies, with a total of 353 participants, focused on the efficacy of CIMT (Alberts 2004; Boake 2007; Dahl 2008, Myint 2008, Taub 1993, Wittenberg 2003; Wolf 2006), and 11 studies, with a total of 243 participants, focused on the efficacy of mcimt (Atteya 2004; Dromerick 2000; Lin 2007; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Wu 2007a; Wu 2007b; Wu 2007c). One study, with 23 patients, investigated the efficacy of FU (Ploughman 2004). Time of restraint: 90% of waking hours for six studies (Alberts 2004; Boake 2007; Dahl 2008; Myint 2008; Taub 1993; Wolf 2006); six hours per day for six studies (Dromerick 2000; Lin 2007; Wittenberg 2003; Wu 2007a; Wu 2007b; Wu 2007c); five hours per day for six studies (Atteya 2004; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008); a mean effective restraint time of 2.7 hours per was reported by one study (Ploughman 2004). Time of exercise with the affected arm: between 30 and 45 hours/week in five studies (Alberts 2004; Dahl 2008; Taub 1993; Wittenberg 2003; Wolf 2006); between 10 and 20 hours/week in seven studies (Boake 2007; Dromerick 2000; Lin 2007; Myint 2008; Wu 2007a; Wu 2007b; Wu 2007c); less than 5 hours/week in seven studies (Atteya 2004; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004). Treatment duration: two weeks in eight studies (Alberts 2004; Boake 2007; Dahl 2008; Dromerick 2000; Myint 2008; Taub 1993; Wittenberg 2003; Wolf 2006); three weeks for four studies (Lin 2007; Wu 2007a; Wu 2007b; Wu 2007c); 10 weeks for six studies (Atteya 2004; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008). Types of exercise: all studies used functional or ADL tasks: in nine studies this was done through shaping techniques (Alberts 2004; Boake 2007; Myint 2008; Page 2002b; Page 2004; Page 2005a; Page 2008; Wolf 2006; Wu 2007a); two studies included proprioceptive neuromuscular facilitation (PNF) (Atteya 2004; Page 2001); one study used conventional treatment for upper extremity, which involved the facilitation of proximal motor control progressing to skilled-task training, without shaping therapy (Ploughman 2004). Anatomical region restraint: both hand and arm in nine studies (Atteya 2004; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2008; Ploughman 2004; Taub 1993; Wittenberg 2003); only the hand in 10 studies (Alberts 2004; Boake 2007; Dahl 2008; Dromerick 2000; Lin 2007; Page 2005a; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c). Intervention delivery In all studies the interventions were delivered and supervised by trained physiotherapists or occupational therapists, and each patient assigned to an intervention group participated in individual therapy sessions, except in one study (Dahl 2008) where the participants exercised in groups of four. Constraint at home was checked by questioning the patient every two weeks about satisfaction with the protocol (Atteya 2004), keeping a log of the hours of restraint per day (Myint 2008; Page 2002a; Page 2004; Page 2005a; Ploughman 2004), and through a physical sensor and timer placed in the mitt and by a home diary (Wolf 2006). Supervision of the constraint was not described in other studies. Outcomes All studies considered pre-treatment and post-treatment outcome measures. Five studies had longer follow up: Boake 2007 and Myint 2008 at four months; Dahl 2008 at six months; Wolf 2006 at four, eight and 12 months; and Taub 1993 up to three years. The 19 included trials considered similar outcome categories. The measures selected by the review team for each outcome category are detailed below and in Table 2. Primary outcomes 1. Disability: i) FIM: five studies (Dahl 2008; Lin 2007; Ploughman 2004; Wu 2007a; Wu 2007c); ii) BI: one study (Myint 2008). Secondary outcomes 1. Arm motor function : i) ARAT: nine studies (Atteya 2004; Dromerick 2000; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004); ii) WMFT: six studies (Alberts 2004; Atteya 2004; Dahl 2008; Page 2001; Wittenberg 2003; Wolf 2006); iii) EMF: one study (Taub 1993). 2. Perceived motor function, amount of use and quality of use : i) MAL: 16 studies (Atteya 2004; Boake 2007; Dahl 2008; Lin 2007; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Taub 1993; Wittenberg 2003; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c). 3. Arm motor impairment: 7

10 i) FMA: 10 studies (Alberts 2004; Atteya 2004; Boake 2007; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Wu 2007b; Wu 2007c); ii) CMII: one study (Ploughman 2004); iii) Jamar hand dynamometer: one study (Ploughman 2004); iv) maximal grip strength with a force transducer: one study (Alberts 2004). 4. Dexterity: i) GPT: one study (Boake 2007); ii) NHPT: one study (Myint 2008). 5. Quality of life : 6. SIS: three studies (Dahl 2008; Wolf 2006; Wu 2007c). Three studies (Boake 2007; Myint 2008; Wolf 2006) reported data from a follow up of four months: Wolf 2006 also reported follow up at eight and 12 months. At long-term follow up, Myint 2008 lost less than 10% of patients, Boake 2007 and Wolf 2006 lost between 10% and 20% of patients. Two studies performed intention-to-treat analyses (Alberts 2004; Wolf 2006). Eleven studies that did not have apparent withdrawals performed de facto intention-to-treat analyses (Atteya 2004; Dahl 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Wittenberg 2003; Wu 2007a; Wu 2007b; Wu 2007c). One study mixed intention-to-treat and per-protocol analyses (Boake 2007). The others have performed per-protocol analyses (Dromerick 2000; Lin 2007; Myint 2008; Ploughman 2004; Taub 1993). Risk of bias in included studies Randomisation The sequence of randomisation was described and appropriate in 11 studies (Alberts 2004; Dromerick 2000; Lin 2007; Page 2004; Page 2005a; Page 2008; Ploughman 2004; Wittenberg 2003; Wolf 2006; Wu 2007b; Wu 2007c). Lin 2007 used a randomisation stratified by side of stroke; Alberts 2004 and Wolf 2006 balanced the randomisation with respect to gender, pre-morbid handedness, side of stroke and level of function, and Boake 2007 stratified by age and NIHSS score. Validity of scales All scales used in the studies for primary and secondary outcomes are supported by references to their psychometric properties, and are considered able to quantify performance in individuals after stroke with motor characteristics similar to the patients enrolled in the included studies. The study on clinimetric properties of the MAL scale reports relatively stable internal consistency in a population of chronic stroke patients, a correlation with ARA score at baseline (Spearman s rho was 0.63 for AoU and QoU), but considerable doubts remain about the longitudinal construct validity of the instrument, and the study does not recommend its use as a primary outcome measure in trials (Van der Lee 2004). Allocation concealment The allocation concealment was described and appropriate in three studies (Alberts 2004; Dahl 2008; Wolf 2006). Blinding Outcome assessors were blinded in all studies. Follow up and withdrawals Five studies provided complete information about patients who withdrew and the reasons (Boake 2007; Lin 2007; Myint 2008; Ploughman 2004; Wolf 2006); one study (Dromerick 2000) provided numbers of withdrawals but not reasons; eight studies presented unclear information about withdrawals: none of these eight clearly stated that there were no dropouts (Alberts 2004; Atteya 2004; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Taub 1993). At post-treatment follow up two studies lost less than 10% of patients (Lin 2007; Wolf 2006); three studies lost between 10% and 20% of patients (Dromerick 2000; Myint 2008; Ploughman 2004); and one study lost more than 20% of patients (Boake 2007). Sample size and power The sample size was based on prior statistical power calculations in only one trial (Wolf 2006). Most studies were very small. The median sample size was 15 randomised patients (interquartile range 10 to 23). Small sample sizes are related to type 2 errors (Altman 1990; Hotopf 1997; Hotopf 1999). If the median number of patients randomised is 15, then the complete analysis will only include seven to eight patients per group. These studies do not have the power to detect even quite important differences between the intervention and no intervention, or between two different treatments. Assuming, for instance, that CIMT increases the performance of the affected arm by 10% on a motor function scale - a difference of interest to a physiotherapist - 37 participants would be needed in each trial arm to detect such a difference at 80% power and 95% confidence in a two-tailed test. Only one trial (Wolf 2006) considered here would be able to detect such a difference. Even with a difference of 40% between the two treatments, five patients completing the trial would be needed for each group. Six of the included trials would have been able to detect such a difference. Studies awaiting assessment 8

11 There are three studies awaiting assessment because the information currently available is insufficient to determine whether they would be eligible for inclusion in this review. Two studies (Dromerick 2007; Edwards 2007) are published as abstracts in an international conference, and one (Jansa 2007) as a poster. Effects of interventions We could only conduct meta-analyses when at least two studies provided sufficient data. We included trials comparing the intervention against no treatment or no active treatment in a specific subgroup on account of the potential heterogeneity of the control group baselines (Alberts 2004; Taub 1993; Wittenberg 2003). In consideration of the clinical heterogeneity among studies, related to variability in the interventions included and in the patient case-mix, we considered it appropriate to perform random-effects meta-analyses to incorporate heterogeneity irrespective of I 2 values, except for subgroup analyses. Primary outcomes Comparison 1.1: Disability post-intervention Six studies recruiting 184 participants measured disability immediately after the experimental and control interventions (Dahl 2008; Lin 2007; Myint 2008; Ploughman 2004; Wu 2007a; Wu 2007c). The impact of CIMT on disability indicated a significant effect: SMD 0.36, 95% CI 0.06 to 0.65 (Analysis 1.1). Sixty-nine patients contributing to this meta-analysis were recruited from studies with more than 10% loss to follow up. Comparison 2.1: Disability - three- to six-month follow up Two studies recruiting 73 participants measured disability at three (Myint 2008) and at six months after treatment (Dahl 2008). The impact of CIMT on disability indicated a non-significant effect: SMD -0.07, 95% CI to 0.40 (Analysis 2.1). Subgroup analysis: Disability We carried out analyses for the following three subgroups considering data availability. 1. Dosage of task practice : we classified trials according to whether they provided more than 30 hours of exercise, or 30 hours or less. 2. Anatomical region restraint: we divided trials according to whether both arm and hand were restrained, or only a hand. 3. Time since stroke: we classified trials according to whether they recruited within three months, three to nine months, or more than nine months post stroke. Comparison 3.1: Amount of task practice Over 30 hours of exercise: two studies with 73 participants (Dahl 2008; Myint 2008) reported over 30 hours of exercise; four studies with 111 participants (Lin 2007; Ploughman 2004; Wu 2007a; Wu 2007c) reported 30 or less hours of exercise. Longer exercise for upper limb function showed no statistically significant effect size: SMD 0.02 (95% CI to 0.49); shorter exercise had significant effect size: SMD 0.58 (95% CI 0.20 to 0.97) (Analysis 3.1). The difference between the effect estimates for the two groups of trials is near to being significant (P = 0.07). This interaction term determines high heterogeneity among studies (I 2 = 70%). Comparison 3.2: Anatomical region restraint One study of 43 participants (Myint 2008) reported both arm and hand restriction; five studies including 141 participants (Dahl 2008; Lin 2007; Ploughman 2004; Wu 2007a; Wu 2007c) reported only hand restriction. The restriction of both arm and hand for upper limb function showed a non-statistically significant effect size: SMD 0.02, 95% CI to 0.51; restriction only of the hand was significant: SMD 0.47, 95% CI 0.13 to 0.81 (Analysis 3.2). The difference between the effect estimates for the two groups of trials is not significant (P = 0.20), although this result may reflect low power and unbalance in these subgroups. We found moderate heterogeneity among studies (I 2 = 39.1%). Comparison 3.3: Time since stroke Zero to three months: two studies involving 66 participants (Myint 2008; Ploughman 2004); more than nine months: two studies involving 62 participants (Lin 2007; Wu 2007a). No studies measured disability on subacute stroke patients at three to nine months. We did not include two studies (Dahl 2008; Wu 2007c) in this subgroup analysis because of the wide range of chronicity of included patients. Acute and chronic stroke patients showed no statistically significant effect size: zero to three months: SMD 0.18, 95% CI to 0.67; more than nine months: SMD 0.49, 95% CI to 1.00 (Analysis 3.3). The difference between the effect estimates for the two groups of trials was not significant (P = 0.39). We did not find heterogeneity among studies (I 2 = 0%). Secondary outcomes Comparison 4.1: Arm motor function Fourteen studies recruiting 436 participants measured arm motor function (Alberts 2004; Atteya 2004; Dahl 2008; Dromerick 2000; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004; Taub 1993; Wittenberg 2003; Wolf 2006). Data were available for 373 participants (86%). 9

12 The impact of CIMT on upper limb function indicated a significant effect size: SMD 0.72, 95% CI 0.32 to 1.12 (Analysis 4.1). We found moderate heterogeneity among studies (I 2 = 42%). Comparison 4.2: Perceived arm motor function (amount of use) Sixteen studies recruiting 541 participants measured perceived arm motor function (AoU) (Atteya 2004; Boake 2007; Dahl 2008; Lin 2007; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Taub 1993; Wittenberg 2003; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c); data were available for 457 participants (84%). The impact of CIMT on perceived upper limb function AoU indicated a large and significant effect: WMD 1.16, 95% CI 1.05 to 1.27 (Analysis 4.2). Comparison 4.3: Perceived arm motor function (quality of use) Sixteen studies recruiting 541 participants measured perceived arm motor function (QoU) (Atteya 2004; Boake 2007; Dahl 2008; Lin 2007; Myint 2008; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Taub 1993; Wittenberg 2003; Wolf 2006; Wu 2007a; Wu 2007b; Wu 2007c); data were available for 450 participants (83%). The impact of CIMT on perceived upper limb function QoU indicated a large and significant effect: WMD 0.87, 95% CI 0.75 to 0.98 (Analysis 4.3). Comparison 4.4: Arm motor impairment Eleven studies recruiting 192 participants measured arm motor impairment (Alberts 2004; Atteya 2004; Boake 2007; Page 2001; Page 2002b; Page 2004; Page 2005a; Page 2008; Ploughman 2004; Wu 2007b; Wu 2007c) data were available for 161 participants (84%). The impact of CIMT on upper limb impairment indicated a significant effect: SMD 0.59, 95% CI 0.14 to 1.05 (Analysis 4.4). Comparison 4.5: Quality of life Three studies recruiting 278 participants measured quality of life (Dahl 2008; Wolf 2006; Wu 2007c); data were available for 56 participants (20%). The impact of CIMT on quality of life indicated a non-significant effect: WMD 6.68, 95% CI to (Analysis 4.5). Comparison 4.6: Dexterity Two studies (Boake 2007; Myint 2008) included measures of dexterity. Unfortunately, data reported by Boake 2007 were poorly reported and uninterpretable. D I S C U S S I O N Summary of main results We included 19 RCTs with a total of 619 participants in the current review. The majority of identified studies were in people who had compromised but residual ability of the upper arm and the hand, with limited pain or spasticity and absence of cognitive impairment. The impact of CIMT on the disability (six studies recruiting 184 participants) indicated a modest significant benefit (SMD 0.36, 95% CI 0.06 to 0.65), although 69 of these 184 patients included in the analysis were recruited from studies with more than 10% loss to follow up. This result is consistent across secondary outcomes (arm motor function and impairment). Although this evidence is promising, the included studies had several methodological weaknesses, did not assess potential harms, had only short-term follow up, were possibly subject to conflicts of interest, and publication bias may have influenced the overall estimate (see below for further discussion of these aspects). Only two small studies assessed the duration of benefit by assessing disability a few months after treatment; there was no evidence that benefit was maintained. Overall completeness and applicability of evidence To quantify the change in the disability after CIMT, we classified the magnitude of effect sizes as proposed by Juni An effect size of 0.20 standard deviation units is considered small. It corresponds to an overlap in the distribution in patients allocated to the experimental or control intervention in about 85% of cases; an effect size of 0.50 is considered to be moderate, corresponding to an overlap in approximately 67% of cases; 0.36 is an intermediate value, with an overlap in about 76% of patients - 24 would indeed benefit. The difference in changes of disability scores (FIM scale) is 0 in the worst scenario (estimated by Dahl 2008) and 14 in the best scenario (estimated by Wu 2007c). We found that studies were heterogeneous in terms of patient case-mix and how the CIMT intervention has been implemented. We explored possible sources of heterogeneity in our subgroup analyses: dosage of task practice, time since stroke, and anatomical region constraint. These analyses are observational and should be interpreted with caution. Marginally significant effects were obtained in the equal or less than 30-hours-practice subgroup on the disability, but not when practice was extended over 30 hours. This means that the dosage of task practice may be an important factor associated with the success of the intervention. The other subgroup analyses did not reveal significant patterns. This may be due to the small size of the subgroups. It is not possible yet to define which subgroups of patients may mostly benefit from 10

13 CIMT interventions in relation to anatomical region and time since stroke. Potential biases in the review process We chose disability as the primary outcome, which was considered only by a minority of eligible studies (six studies recruiting 184 participants), whereas arm motor function was used as the primary outcome in the majority of included studies (14 studies recruiting 411 participants). Compared with disability, the average benefit measured on arm motor function scales indicated a larger effect size, SMD 0.72 (95% CI 0.32 to 1.12), with an overlap in about 53% of cases (Juni 2006). We argue that both disability and arm motor function are important outcomes. Differences in magnitude of effect size between these measures may be related to differences in responsiveness, consistency and precision between tests (Croarkin 2004). Although the analysis showed a larger effect of CIMT on perceived arm motor function (AoU and QoU), caution is needed in interpreting this result because of the lack of consistency in the MAL scale, as described in the Risk of bias in included studies section. Its clinimetric properties need further investigation so as to define its use better in longitudinal studies. Most trials were very small, with someone enrolling only six or 10 patients. This is unacceptable, given the high incidence of stroke and the opportunity to recruit a large sample easily. Our sample of trials may therefore have been influenced by publication bias, which tends to exaggerate the effect of treatment. The randomisation methods were described only in about half of the included trials. It is likely that some studies excluded patients after randomisation, or that the blinding was not adequately maintained. These are defects that could be expected to lead to bias in favour of a treatment effect. The reporting of the data was poor; for example, many trials only reported that there were no significant differences between the intervention and the control groups. This lack of proper reporting could also be expected to lead to bias in favour of a treatment effect. It should also be noted that many authors of trials have a cultural and professional interest in disseminating positive results about the rehabilitative techniques they propose. Although CIMT appeared to have moderately positive effects on disability assessed at the end of treatment, many studies were underpowered and at high risk of small-trial and publication bias. On the other hand, these apparent benefits were consistent across other outcomes (improving arm motor function and motor impairment). We found no evidence of benefit from CIMT on disability at six months. It is not clear whether CIMT maintains efficacy in the long run. CIMT can be considered a multifaceted intervention where the restriction is accompanied by a certain amount and quality of exercise. The advantage may be due not only to the constraint itself, but also to the type and amount of exercise. However, this review could not identify which of these factors contributes most. Patients who might benefit from this treatment are those with active wrist and finger extension, with limited pain or spasticity but who underuse or do not use this movement in daily living. Implications for research It is likely that additional trials investigating CIMT as a rehabilitation technique would be worthwhile if they: 1. consider disability or arm motor function as the primary outcomes; 2. include a validated quality of life measure as one of the outcomes; 3. recruit a larger sample (minimum of 74 participants for arm motor function measures); 4. involve a control group under active treatment, since CIMT involves a certain amount of exercise; 5. explore benefits for longer follow-up periods (one year); and 6. include all randomised patients in their analyses. As CIMT is a multifaceted intervention (restraint of the unaffected arm plus several hours of different exercises), it is not clear whether the effects are due to one or more characteristics of the technique, so further studies should focus on detecting the origin of these improvements. It will be useful to establish which subgroup of patients are most likely to respond, and which would be the best constraint technique in relation to anatomical region, exercise time, and time since stroke. A C K N O W L E D G E M E N T S A U T H O R S C O N C L U S I O N S Implications for practice We would like to thank Mrs Hazel Fraser for her help and assistance, Mrs Brenda Thomas for her help with developing the search strategy, Mrs Elena Telaro for the elaboration to the first protocol, Prof Piergiorgio Duca for his precious help with statistical analyses. Our thanks to the Italian Cochrane Centre for its methodological support. Thanks to Dr Judith Baggott for her precious help with the translation of the manuscript in English. Thanks to Profs Alberts, Boake, Dromerick, Lin, Myint, Ploughman, Wittenberg, Wolf and Wu for their availability and collaboration. Special thanks to Prof Edward Taub for his support. 11